iommu: Add gfp parameter to iommu_ops::map
[linux-2.6-block.git] / drivers / iommu / iommu.c
CommitLineData
45051539 1// SPDX-License-Identifier: GPL-2.0-only
fc2100eb
JR
2/*
3 * Copyright (C) 2007-2008 Advanced Micro Devices, Inc.
63ce3ae8 4 * Author: Joerg Roedel <jroedel@suse.de>
fc2100eb
JR
5 */
6
92e7066f 7#define pr_fmt(fmt) "iommu: " fmt
7d3002cc 8
905d66c1 9#include <linux/device.h>
40998188 10#include <linux/kernel.h>
fc2100eb
JR
11#include <linux/bug.h>
12#include <linux/types.h>
c1af7b40
PG
13#include <linux/init.h>
14#include <linux/export.h>
60db4027 15#include <linux/slab.h>
fc2100eb
JR
16#include <linux/errno.h>
17#include <linux/iommu.h>
d72e31c9
AW
18#include <linux/idr.h>
19#include <linux/notifier.h>
20#include <linux/err.h>
104a1c13 21#include <linux/pci.h>
f096c061 22#include <linux/bitops.h>
57f98d2f 23#include <linux/property.h>
eab03e2a 24#include <linux/fsl/mc.h>
7f6db171 25#include <trace/events/iommu.h>
d72e31c9
AW
26
27static struct kset *iommu_group_kset;
e38d1f13 28static DEFINE_IDA(iommu_group_ida);
22bb182c
JR
29
30static unsigned int iommu_def_domain_type __read_mostly;
68a6efe8 31static bool iommu_dma_strict __read_mostly = true;
faf14989 32static u32 iommu_cmd_line __read_mostly;
d72e31c9
AW
33
34struct iommu_group {
35 struct kobject kobj;
36 struct kobject *devices_kobj;
37 struct list_head devices;
38 struct mutex mutex;
39 struct blocking_notifier_head notifier;
40 void *iommu_data;
41 void (*iommu_data_release)(void *iommu_data);
42 char *name;
43 int id;
53723dc5 44 struct iommu_domain *default_domain;
e39cb8a3 45 struct iommu_domain *domain;
d72e31c9
AW
46};
47
c09e22d5 48struct group_device {
d72e31c9
AW
49 struct list_head list;
50 struct device *dev;
51 char *name;
52};
53
54struct iommu_group_attribute {
55 struct attribute attr;
56 ssize_t (*show)(struct iommu_group *group, char *buf);
57 ssize_t (*store)(struct iommu_group *group,
58 const char *buf, size_t count);
59};
60
bc7d12b9 61static const char * const iommu_group_resv_type_string[] = {
adfd3738
EA
62 [IOMMU_RESV_DIRECT] = "direct",
63 [IOMMU_RESV_DIRECT_RELAXABLE] = "direct-relaxable",
64 [IOMMU_RESV_RESERVED] = "reserved",
65 [IOMMU_RESV_MSI] = "msi",
66 [IOMMU_RESV_SW_MSI] = "msi",
bc7d12b9
EA
67};
68
faf14989
JR
69#define IOMMU_CMD_LINE_DMA_API BIT(0)
70
71static void iommu_set_cmd_line_dma_api(void)
72{
73 iommu_cmd_line |= IOMMU_CMD_LINE_DMA_API;
74}
75
22bb182c 76static bool iommu_cmd_line_dma_api(void)
faf14989
JR
77{
78 return !!(iommu_cmd_line & IOMMU_CMD_LINE_DMA_API);
79}
80
d72e31c9
AW
81#define IOMMU_GROUP_ATTR(_name, _mode, _show, _store) \
82struct iommu_group_attribute iommu_group_attr_##_name = \
83 __ATTR(_name, _mode, _show, _store)
fc2100eb 84
d72e31c9
AW
85#define to_iommu_group_attr(_attr) \
86 container_of(_attr, struct iommu_group_attribute, attr)
87#define to_iommu_group(_kobj) \
88 container_of(_kobj, struct iommu_group, kobj)
fc2100eb 89
b0119e87
JR
90static LIST_HEAD(iommu_device_list);
91static DEFINE_SPINLOCK(iommu_device_lock);
92
5fa9e7c5
JR
93/*
94 * Use a function instead of an array here because the domain-type is a
95 * bit-field, so an array would waste memory.
96 */
97static const char *iommu_domain_type_str(unsigned int t)
98{
99 switch (t) {
100 case IOMMU_DOMAIN_BLOCKED:
101 return "Blocked";
102 case IOMMU_DOMAIN_IDENTITY:
103 return "Passthrough";
104 case IOMMU_DOMAIN_UNMANAGED:
105 return "Unmanaged";
106 case IOMMU_DOMAIN_DMA:
107 return "Translated";
108 default:
109 return "Unknown";
110 }
111}
112
113static int __init iommu_subsys_init(void)
114{
22bb182c
JR
115 bool cmd_line = iommu_cmd_line_dma_api();
116
117 if (!cmd_line) {
118 if (IS_ENABLED(CONFIG_IOMMU_DEFAULT_PASSTHROUGH))
119 iommu_set_default_passthrough(false);
120 else
121 iommu_set_default_translated(false);
2cc13bb4 122
2896ba40
JR
123 if (iommu_default_passthrough() && mem_encrypt_active()) {
124 pr_info("Memory encryption detected - Disabling default IOMMU Passthrough\n");
2cc13bb4
JR
125 iommu_set_default_translated(false);
126 }
22bb182c
JR
127 }
128
129 pr_info("Default domain type: %s %s\n",
130 iommu_domain_type_str(iommu_def_domain_type),
131 cmd_line ? "(set via kernel command line)" : "");
5fa9e7c5
JR
132
133 return 0;
134}
135subsys_initcall(iommu_subsys_init);
136
b0119e87
JR
137int iommu_device_register(struct iommu_device *iommu)
138{
139 spin_lock(&iommu_device_lock);
140 list_add_tail(&iommu->list, &iommu_device_list);
141 spin_unlock(&iommu_device_lock);
b0119e87
JR
142 return 0;
143}
144
145void iommu_device_unregister(struct iommu_device *iommu)
146{
147 spin_lock(&iommu_device_lock);
148 list_del(&iommu->list);
149 spin_unlock(&iommu_device_lock);
150}
151
0c830e6b
JP
152static struct iommu_param *iommu_get_dev_param(struct device *dev)
153{
154 struct iommu_param *param = dev->iommu_param;
155
156 if (param)
157 return param;
158
159 param = kzalloc(sizeof(*param), GFP_KERNEL);
160 if (!param)
161 return NULL;
162
163 mutex_init(&param->lock);
164 dev->iommu_param = param;
165 return param;
166}
167
168static void iommu_free_dev_param(struct device *dev)
169{
170 kfree(dev->iommu_param);
171 dev->iommu_param = NULL;
172}
173
cc5aed44
JR
174int iommu_probe_device(struct device *dev)
175{
176 const struct iommu_ops *ops = dev->bus->iommu_ops;
0c830e6b 177 int ret;
cc5aed44
JR
178
179 WARN_ON(dev->iommu_group);
0c830e6b
JP
180 if (!ops)
181 return -EINVAL;
cc5aed44 182
0c830e6b
JP
183 if (!iommu_get_dev_param(dev))
184 return -ENOMEM;
cc5aed44 185
0c830e6b
JP
186 ret = ops->add_device(dev);
187 if (ret)
188 iommu_free_dev_param(dev);
dc9de8a2
JR
189
190 return ret;
cc5aed44
JR
191}
192
193void iommu_release_device(struct device *dev)
194{
195 const struct iommu_ops *ops = dev->bus->iommu_ops;
196
197 if (dev->iommu_group)
198 ops->remove_device(dev);
0c830e6b
JP
199
200 iommu_free_dev_param(dev);
cc5aed44
JR
201}
202
53723dc5
JR
203static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
204 unsigned type);
e39cb8a3
JR
205static int __iommu_attach_device(struct iommu_domain *domain,
206 struct device *dev);
207static int __iommu_attach_group(struct iommu_domain *domain,
208 struct iommu_group *group);
209static void __iommu_detach_group(struct iommu_domain *domain,
210 struct iommu_group *group);
53723dc5 211
fccb4e3b
WD
212static int __init iommu_set_def_domain_type(char *str)
213{
214 bool pt;
7f9584df 215 int ret;
fccb4e3b 216
7f9584df
AS
217 ret = kstrtobool(str, &pt);
218 if (ret)
219 return ret;
fccb4e3b 220
adab0b07
JR
221 if (pt)
222 iommu_set_default_passthrough(true);
223 else
224 iommu_set_default_translated(true);
faf14989 225
fccb4e3b
WD
226 return 0;
227}
228early_param("iommu.passthrough", iommu_set_def_domain_type);
229
68a6efe8
ZL
230static int __init iommu_dma_setup(char *str)
231{
232 return kstrtobool(str, &iommu_dma_strict);
233}
234early_param("iommu.strict", iommu_dma_setup);
235
d72e31c9
AW
236static ssize_t iommu_group_attr_show(struct kobject *kobj,
237 struct attribute *__attr, char *buf)
1460432c 238{
d72e31c9
AW
239 struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
240 struct iommu_group *group = to_iommu_group(kobj);
241 ssize_t ret = -EIO;
1460432c 242
d72e31c9
AW
243 if (attr->show)
244 ret = attr->show(group, buf);
245 return ret;
246}
247
248static ssize_t iommu_group_attr_store(struct kobject *kobj,
249 struct attribute *__attr,
250 const char *buf, size_t count)
251{
252 struct iommu_group_attribute *attr = to_iommu_group_attr(__attr);
253 struct iommu_group *group = to_iommu_group(kobj);
254 ssize_t ret = -EIO;
1460432c 255
d72e31c9
AW
256 if (attr->store)
257 ret = attr->store(group, buf, count);
258 return ret;
1460432c 259}
1460432c 260
d72e31c9
AW
261static const struct sysfs_ops iommu_group_sysfs_ops = {
262 .show = iommu_group_attr_show,
263 .store = iommu_group_attr_store,
264};
1460432c 265
d72e31c9
AW
266static int iommu_group_create_file(struct iommu_group *group,
267 struct iommu_group_attribute *attr)
268{
269 return sysfs_create_file(&group->kobj, &attr->attr);
1460432c 270}
1460432c 271
d72e31c9
AW
272static void iommu_group_remove_file(struct iommu_group *group,
273 struct iommu_group_attribute *attr)
274{
275 sysfs_remove_file(&group->kobj, &attr->attr);
276}
277
278static ssize_t iommu_group_show_name(struct iommu_group *group, char *buf)
279{
280 return sprintf(buf, "%s\n", group->name);
281}
282
6c65fb31
EA
283/**
284 * iommu_insert_resv_region - Insert a new region in the
285 * list of reserved regions.
286 * @new: new region to insert
287 * @regions: list of regions
288 *
4dbd258f
EA
289 * Elements are sorted by start address and overlapping segments
290 * of the same type are merged.
6c65fb31 291 */
4dbd258f
EA
292int iommu_insert_resv_region(struct iommu_resv_region *new,
293 struct list_head *regions)
6c65fb31 294{
4dbd258f
EA
295 struct iommu_resv_region *iter, *tmp, *nr, *top;
296 LIST_HEAD(stack);
297
298 nr = iommu_alloc_resv_region(new->start, new->length,
299 new->prot, new->type);
300 if (!nr)
301 return -ENOMEM;
302
303 /* First add the new element based on start address sorting */
304 list_for_each_entry(iter, regions, list) {
305 if (nr->start < iter->start ||
306 (nr->start == iter->start && nr->type <= iter->type))
307 break;
308 }
309 list_add_tail(&nr->list, &iter->list);
310
311 /* Merge overlapping segments of type nr->type in @regions, if any */
312 list_for_each_entry_safe(iter, tmp, regions, list) {
313 phys_addr_t top_end, iter_end = iter->start + iter->length - 1;
314
315 /* no merge needed on elements of different types than @nr */
316 if (iter->type != nr->type) {
317 list_move_tail(&iter->list, &stack);
318 continue;
319 }
320
321 /* look for the last stack element of same type as @iter */
322 list_for_each_entry_reverse(top, &stack, list)
323 if (top->type == iter->type)
324 goto check_overlap;
325
326 list_move_tail(&iter->list, &stack);
327 continue;
328
329check_overlap:
330 top_end = top->start + top->length - 1;
331
332 if (iter->start > top_end + 1) {
333 list_move_tail(&iter->list, &stack);
6c65fb31 334 } else {
4dbd258f
EA
335 top->length = max(top_end, iter_end) - top->start + 1;
336 list_del(&iter->list);
337 kfree(iter);
6c65fb31
EA
338 }
339 }
4dbd258f 340 list_splice(&stack, regions);
6c65fb31
EA
341 return 0;
342}
343
344static int
345iommu_insert_device_resv_regions(struct list_head *dev_resv_regions,
346 struct list_head *group_resv_regions)
347{
348 struct iommu_resv_region *entry;
a514a6e2 349 int ret = 0;
6c65fb31
EA
350
351 list_for_each_entry(entry, dev_resv_regions, list) {
352 ret = iommu_insert_resv_region(entry, group_resv_regions);
353 if (ret)
354 break;
355 }
356 return ret;
357}
358
359int iommu_get_group_resv_regions(struct iommu_group *group,
360 struct list_head *head)
361{
8d2932dd 362 struct group_device *device;
6c65fb31
EA
363 int ret = 0;
364
365 mutex_lock(&group->mutex);
366 list_for_each_entry(device, &group->devices, list) {
367 struct list_head dev_resv_regions;
368
369 INIT_LIST_HEAD(&dev_resv_regions);
370 iommu_get_resv_regions(device->dev, &dev_resv_regions);
371 ret = iommu_insert_device_resv_regions(&dev_resv_regions, head);
372 iommu_put_resv_regions(device->dev, &dev_resv_regions);
373 if (ret)
374 break;
375 }
376 mutex_unlock(&group->mutex);
377 return ret;
378}
379EXPORT_SYMBOL_GPL(iommu_get_group_resv_regions);
380
bc7d12b9
EA
381static ssize_t iommu_group_show_resv_regions(struct iommu_group *group,
382 char *buf)
383{
384 struct iommu_resv_region *region, *next;
385 struct list_head group_resv_regions;
386 char *str = buf;
387
388 INIT_LIST_HEAD(&group_resv_regions);
389 iommu_get_group_resv_regions(group, &group_resv_regions);
390
391 list_for_each_entry_safe(region, next, &group_resv_regions, list) {
392 str += sprintf(str, "0x%016llx 0x%016llx %s\n",
393 (long long int)region->start,
394 (long long int)(region->start +
395 region->length - 1),
396 iommu_group_resv_type_string[region->type]);
397 kfree(region);
398 }
399
400 return (str - buf);
401}
402
c52c72d3
OJ
403static ssize_t iommu_group_show_type(struct iommu_group *group,
404 char *buf)
405{
406 char *type = "unknown\n";
407
408 if (group->default_domain) {
409 switch (group->default_domain->type) {
410 case IOMMU_DOMAIN_BLOCKED:
411 type = "blocked\n";
412 break;
413 case IOMMU_DOMAIN_IDENTITY:
414 type = "identity\n";
415 break;
416 case IOMMU_DOMAIN_UNMANAGED:
417 type = "unmanaged\n";
418 break;
419 case IOMMU_DOMAIN_DMA:
24f307d8 420 type = "DMA\n";
c52c72d3
OJ
421 break;
422 }
423 }
424 strcpy(buf, type);
425
426 return strlen(type);
427}
428
d72e31c9
AW
429static IOMMU_GROUP_ATTR(name, S_IRUGO, iommu_group_show_name, NULL);
430
bc7d12b9
EA
431static IOMMU_GROUP_ATTR(reserved_regions, 0444,
432 iommu_group_show_resv_regions, NULL);
433
c52c72d3
OJ
434static IOMMU_GROUP_ATTR(type, 0444, iommu_group_show_type, NULL);
435
d72e31c9
AW
436static void iommu_group_release(struct kobject *kobj)
437{
438 struct iommu_group *group = to_iommu_group(kobj);
439
269aa808
JR
440 pr_debug("Releasing group %d\n", group->id);
441
d72e31c9
AW
442 if (group->iommu_data_release)
443 group->iommu_data_release(group->iommu_data);
444
feccf398 445 ida_simple_remove(&iommu_group_ida, group->id);
d72e31c9 446
53723dc5
JR
447 if (group->default_domain)
448 iommu_domain_free(group->default_domain);
449
d72e31c9
AW
450 kfree(group->name);
451 kfree(group);
452}
453
454static struct kobj_type iommu_group_ktype = {
455 .sysfs_ops = &iommu_group_sysfs_ops,
456 .release = iommu_group_release,
457};
458
459/**
460 * iommu_group_alloc - Allocate a new group
d72e31c9
AW
461 *
462 * This function is called by an iommu driver to allocate a new iommu
463 * group. The iommu group represents the minimum granularity of the iommu.
464 * Upon successful return, the caller holds a reference to the supplied
465 * group in order to hold the group until devices are added. Use
466 * iommu_group_put() to release this extra reference count, allowing the
467 * group to be automatically reclaimed once it has no devices or external
468 * references.
469 */
470struct iommu_group *iommu_group_alloc(void)
1460432c 471{
d72e31c9
AW
472 struct iommu_group *group;
473 int ret;
474
475 group = kzalloc(sizeof(*group), GFP_KERNEL);
476 if (!group)
477 return ERR_PTR(-ENOMEM);
478
479 group->kobj.kset = iommu_group_kset;
480 mutex_init(&group->mutex);
481 INIT_LIST_HEAD(&group->devices);
482 BLOCKING_INIT_NOTIFIER_HEAD(&group->notifier);
483
feccf398
HK
484 ret = ida_simple_get(&iommu_group_ida, 0, 0, GFP_KERNEL);
485 if (ret < 0) {
d72e31c9 486 kfree(group);
feccf398 487 return ERR_PTR(ret);
d72e31c9 488 }
feccf398 489 group->id = ret;
1460432c 490
d72e31c9
AW
491 ret = kobject_init_and_add(&group->kobj, &iommu_group_ktype,
492 NULL, "%d", group->id);
493 if (ret) {
feccf398 494 ida_simple_remove(&iommu_group_ida, group->id);
d72e31c9
AW
495 kfree(group);
496 return ERR_PTR(ret);
497 }
498
499 group->devices_kobj = kobject_create_and_add("devices", &group->kobj);
500 if (!group->devices_kobj) {
501 kobject_put(&group->kobj); /* triggers .release & free */
502 return ERR_PTR(-ENOMEM);
503 }
504
505 /*
506 * The devices_kobj holds a reference on the group kobject, so
507 * as long as that exists so will the group. We can therefore
508 * use the devices_kobj for reference counting.
509 */
510 kobject_put(&group->kobj);
511
bc7d12b9
EA
512 ret = iommu_group_create_file(group,
513 &iommu_group_attr_reserved_regions);
514 if (ret)
515 return ERR_PTR(ret);
516
c52c72d3
OJ
517 ret = iommu_group_create_file(group, &iommu_group_attr_type);
518 if (ret)
519 return ERR_PTR(ret);
520
269aa808
JR
521 pr_debug("Allocated group %d\n", group->id);
522
d72e31c9
AW
523 return group;
524}
525EXPORT_SYMBOL_GPL(iommu_group_alloc);
526
aa16bea9
AK
527struct iommu_group *iommu_group_get_by_id(int id)
528{
529 struct kobject *group_kobj;
530 struct iommu_group *group;
531 const char *name;
532
533 if (!iommu_group_kset)
534 return NULL;
535
536 name = kasprintf(GFP_KERNEL, "%d", id);
537 if (!name)
538 return NULL;
539
540 group_kobj = kset_find_obj(iommu_group_kset, name);
541 kfree(name);
542
543 if (!group_kobj)
544 return NULL;
545
546 group = container_of(group_kobj, struct iommu_group, kobj);
547 BUG_ON(group->id != id);
548
549 kobject_get(group->devices_kobj);
550 kobject_put(&group->kobj);
551
552 return group;
553}
554EXPORT_SYMBOL_GPL(iommu_group_get_by_id);
555
d72e31c9
AW
556/**
557 * iommu_group_get_iommudata - retrieve iommu_data registered for a group
558 * @group: the group
559 *
560 * iommu drivers can store data in the group for use when doing iommu
561 * operations. This function provides a way to retrieve it. Caller
562 * should hold a group reference.
563 */
564void *iommu_group_get_iommudata(struct iommu_group *group)
565{
566 return group->iommu_data;
567}
568EXPORT_SYMBOL_GPL(iommu_group_get_iommudata);
569
570/**
571 * iommu_group_set_iommudata - set iommu_data for a group
572 * @group: the group
573 * @iommu_data: new data
574 * @release: release function for iommu_data
575 *
576 * iommu drivers can store data in the group for use when doing iommu
577 * operations. This function provides a way to set the data after
578 * the group has been allocated. Caller should hold a group reference.
579 */
580void iommu_group_set_iommudata(struct iommu_group *group, void *iommu_data,
581 void (*release)(void *iommu_data))
1460432c 582{
d72e31c9
AW
583 group->iommu_data = iommu_data;
584 group->iommu_data_release = release;
585}
586EXPORT_SYMBOL_GPL(iommu_group_set_iommudata);
1460432c 587
d72e31c9
AW
588/**
589 * iommu_group_set_name - set name for a group
590 * @group: the group
591 * @name: name
592 *
593 * Allow iommu driver to set a name for a group. When set it will
594 * appear in a name attribute file under the group in sysfs.
595 */
596int iommu_group_set_name(struct iommu_group *group, const char *name)
597{
598 int ret;
599
600 if (group->name) {
601 iommu_group_remove_file(group, &iommu_group_attr_name);
602 kfree(group->name);
603 group->name = NULL;
604 if (!name)
605 return 0;
606 }
607
608 group->name = kstrdup(name, GFP_KERNEL);
609 if (!group->name)
610 return -ENOMEM;
611
612 ret = iommu_group_create_file(group, &iommu_group_attr_name);
613 if (ret) {
614 kfree(group->name);
615 group->name = NULL;
616 return ret;
617 }
1460432c
AW
618
619 return 0;
620}
d72e31c9 621EXPORT_SYMBOL_GPL(iommu_group_set_name);
1460432c 622
beed2821
JR
623static int iommu_group_create_direct_mappings(struct iommu_group *group,
624 struct device *dev)
625{
626 struct iommu_domain *domain = group->default_domain;
e5b5234a 627 struct iommu_resv_region *entry;
beed2821
JR
628 struct list_head mappings;
629 unsigned long pg_size;
630 int ret = 0;
631
632 if (!domain || domain->type != IOMMU_DOMAIN_DMA)
633 return 0;
634
d16e0faa 635 BUG_ON(!domain->pgsize_bitmap);
beed2821 636
d16e0faa 637 pg_size = 1UL << __ffs(domain->pgsize_bitmap);
beed2821
JR
638 INIT_LIST_HEAD(&mappings);
639
e5b5234a 640 iommu_get_resv_regions(dev, &mappings);
beed2821
JR
641
642 /* We need to consider overlapping regions for different devices */
643 list_for_each_entry(entry, &mappings, list) {
644 dma_addr_t start, end, addr;
645
e5b5234a
EA
646 if (domain->ops->apply_resv_region)
647 domain->ops->apply_resv_region(dev, domain, entry);
33b21a6b 648
beed2821
JR
649 start = ALIGN(entry->start, pg_size);
650 end = ALIGN(entry->start + entry->length, pg_size);
651
adfd3738
EA
652 if (entry->type != IOMMU_RESV_DIRECT &&
653 entry->type != IOMMU_RESV_DIRECT_RELAXABLE)
544a25d9
EA
654 continue;
655
beed2821
JR
656 for (addr = start; addr < end; addr += pg_size) {
657 phys_addr_t phys_addr;
658
659 phys_addr = iommu_iova_to_phys(domain, addr);
660 if (phys_addr)
661 continue;
662
663 ret = iommu_map(domain, addr, addr, pg_size, entry->prot);
664 if (ret)
665 goto out;
666 }
667
668 }
669
add02cfd
JR
670 iommu_flush_tlb_all(domain);
671
beed2821 672out:
e5b5234a 673 iommu_put_resv_regions(dev, &mappings);
beed2821
JR
674
675 return ret;
676}
677
d72e31c9
AW
678/**
679 * iommu_group_add_device - add a device to an iommu group
680 * @group: the group into which to add the device (reference should be held)
681 * @dev: the device
682 *
683 * This function is called by an iommu driver to add a device into a
684 * group. Adding a device increments the group reference count.
685 */
686int iommu_group_add_device(struct iommu_group *group, struct device *dev)
1460432c 687{
d72e31c9 688 int ret, i = 0;
c09e22d5 689 struct group_device *device;
d72e31c9
AW
690
691 device = kzalloc(sizeof(*device), GFP_KERNEL);
692 if (!device)
693 return -ENOMEM;
694
695 device->dev = dev;
1460432c 696
d72e31c9 697 ret = sysfs_create_link(&dev->kobj, &group->kobj, "iommu_group");
797a8b4d
RM
698 if (ret)
699 goto err_free_device;
d72e31c9
AW
700
701 device->name = kasprintf(GFP_KERNEL, "%s", kobject_name(&dev->kobj));
702rename:
703 if (!device->name) {
797a8b4d
RM
704 ret = -ENOMEM;
705 goto err_remove_link;
d72e31c9 706 }
1460432c 707
d72e31c9
AW
708 ret = sysfs_create_link_nowarn(group->devices_kobj,
709 &dev->kobj, device->name);
710 if (ret) {
d72e31c9
AW
711 if (ret == -EEXIST && i >= 0) {
712 /*
713 * Account for the slim chance of collision
714 * and append an instance to the name.
715 */
797a8b4d 716 kfree(device->name);
d72e31c9
AW
717 device->name = kasprintf(GFP_KERNEL, "%s.%d",
718 kobject_name(&dev->kobj), i++);
719 goto rename;
720 }
797a8b4d 721 goto err_free_name;
d72e31c9
AW
722 }
723
724 kobject_get(group->devices_kobj);
725
726 dev->iommu_group = group;
727
beed2821
JR
728 iommu_group_create_direct_mappings(group, dev);
729
d72e31c9
AW
730 mutex_lock(&group->mutex);
731 list_add_tail(&device->list, &group->devices);
e39cb8a3 732 if (group->domain)
797a8b4d 733 ret = __iommu_attach_device(group->domain, dev);
d72e31c9 734 mutex_unlock(&group->mutex);
797a8b4d
RM
735 if (ret)
736 goto err_put_group;
d72e31c9
AW
737
738 /* Notify any listeners about change to group. */
739 blocking_notifier_call_chain(&group->notifier,
740 IOMMU_GROUP_NOTIFY_ADD_DEVICE, dev);
d1cf7e82
SK
741
742 trace_add_device_to_group(group->id, dev);
269aa808 743
780da9e4 744 dev_info(dev, "Adding to iommu group %d\n", group->id);
269aa808 745
1460432c 746 return 0;
797a8b4d
RM
747
748err_put_group:
749 mutex_lock(&group->mutex);
750 list_del(&device->list);
751 mutex_unlock(&group->mutex);
752 dev->iommu_group = NULL;
753 kobject_put(group->devices_kobj);
754err_free_name:
755 kfree(device->name);
756err_remove_link:
757 sysfs_remove_link(&dev->kobj, "iommu_group");
758err_free_device:
759 kfree(device);
780da9e4 760 dev_err(dev, "Failed to add to iommu group %d: %d\n", group->id, ret);
797a8b4d 761 return ret;
1460432c 762}
d72e31c9 763EXPORT_SYMBOL_GPL(iommu_group_add_device);
1460432c 764
d72e31c9
AW
765/**
766 * iommu_group_remove_device - remove a device from it's current group
767 * @dev: device to be removed
768 *
769 * This function is called by an iommu driver to remove the device from
770 * it's current group. This decrements the iommu group reference count.
771 */
772void iommu_group_remove_device(struct device *dev)
773{
774 struct iommu_group *group = dev->iommu_group;
c09e22d5 775 struct group_device *tmp_device, *device = NULL;
d72e31c9 776
780da9e4 777 dev_info(dev, "Removing from iommu group %d\n", group->id);
269aa808 778
d72e31c9
AW
779 /* Pre-notify listeners that a device is being removed. */
780 blocking_notifier_call_chain(&group->notifier,
781 IOMMU_GROUP_NOTIFY_DEL_DEVICE, dev);
782
783 mutex_lock(&group->mutex);
784 list_for_each_entry(tmp_device, &group->devices, list) {
785 if (tmp_device->dev == dev) {
786 device = tmp_device;
787 list_del(&device->list);
788 break;
789 }
790 }
791 mutex_unlock(&group->mutex);
792
793 if (!device)
794 return;
795
796 sysfs_remove_link(group->devices_kobj, device->name);
797 sysfs_remove_link(&dev->kobj, "iommu_group");
798
2e757086
SK
799 trace_remove_device_from_group(group->id, dev);
800
d72e31c9
AW
801 kfree(device->name);
802 kfree(device);
803 dev->iommu_group = NULL;
804 kobject_put(group->devices_kobj);
805}
806EXPORT_SYMBOL_GPL(iommu_group_remove_device);
807
426a2738
JR
808static int iommu_group_device_count(struct iommu_group *group)
809{
c09e22d5 810 struct group_device *entry;
426a2738
JR
811 int ret = 0;
812
813 list_for_each_entry(entry, &group->devices, list)
814 ret++;
815
816 return ret;
817}
818
d72e31c9
AW
819/**
820 * iommu_group_for_each_dev - iterate over each device in the group
821 * @group: the group
822 * @data: caller opaque data to be passed to callback function
823 * @fn: caller supplied callback function
824 *
825 * This function is called by group users to iterate over group devices.
826 * Callers should hold a reference count to the group during callback.
827 * The group->mutex is held across callbacks, which will block calls to
828 * iommu_group_add/remove_device.
829 */
e39cb8a3
JR
830static int __iommu_group_for_each_dev(struct iommu_group *group, void *data,
831 int (*fn)(struct device *, void *))
d72e31c9 832{
c09e22d5 833 struct group_device *device;
d72e31c9
AW
834 int ret = 0;
835
d72e31c9
AW
836 list_for_each_entry(device, &group->devices, list) {
837 ret = fn(device->dev, data);
838 if (ret)
839 break;
840 }
e39cb8a3
JR
841 return ret;
842}
843
844
845int iommu_group_for_each_dev(struct iommu_group *group, void *data,
846 int (*fn)(struct device *, void *))
847{
848 int ret;
849
850 mutex_lock(&group->mutex);
851 ret = __iommu_group_for_each_dev(group, data, fn);
d72e31c9 852 mutex_unlock(&group->mutex);
e39cb8a3 853
d72e31c9
AW
854 return ret;
855}
856EXPORT_SYMBOL_GPL(iommu_group_for_each_dev);
857
858/**
859 * iommu_group_get - Return the group for a device and increment reference
860 * @dev: get the group that this device belongs to
861 *
862 * This function is called by iommu drivers and users to get the group
863 * for the specified device. If found, the group is returned and the group
864 * reference in incremented, else NULL.
865 */
866struct iommu_group *iommu_group_get(struct device *dev)
867{
868 struct iommu_group *group = dev->iommu_group;
869
870 if (group)
871 kobject_get(group->devices_kobj);
872
873 return group;
874}
875EXPORT_SYMBOL_GPL(iommu_group_get);
876
13f59a78
RM
877/**
878 * iommu_group_ref_get - Increment reference on a group
879 * @group: the group to use, must not be NULL
880 *
881 * This function is called by iommu drivers to take additional references on an
882 * existing group. Returns the given group for convenience.
883 */
884struct iommu_group *iommu_group_ref_get(struct iommu_group *group)
885{
886 kobject_get(group->devices_kobj);
887 return group;
888}
889
d72e31c9
AW
890/**
891 * iommu_group_put - Decrement group reference
892 * @group: the group to use
893 *
894 * This function is called by iommu drivers and users to release the
895 * iommu group. Once the reference count is zero, the group is released.
896 */
897void iommu_group_put(struct iommu_group *group)
898{
899 if (group)
900 kobject_put(group->devices_kobj);
901}
902EXPORT_SYMBOL_GPL(iommu_group_put);
903
904/**
905 * iommu_group_register_notifier - Register a notifier for group changes
906 * @group: the group to watch
907 * @nb: notifier block to signal
908 *
909 * This function allows iommu group users to track changes in a group.
910 * See include/linux/iommu.h for actions sent via this notifier. Caller
911 * should hold a reference to the group throughout notifier registration.
912 */
913int iommu_group_register_notifier(struct iommu_group *group,
914 struct notifier_block *nb)
915{
916 return blocking_notifier_chain_register(&group->notifier, nb);
917}
918EXPORT_SYMBOL_GPL(iommu_group_register_notifier);
919
920/**
921 * iommu_group_unregister_notifier - Unregister a notifier
922 * @group: the group to watch
923 * @nb: notifier block to signal
924 *
925 * Unregister a previously registered group notifier block.
926 */
927int iommu_group_unregister_notifier(struct iommu_group *group,
928 struct notifier_block *nb)
929{
930 return blocking_notifier_chain_unregister(&group->notifier, nb);
931}
932EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier);
933
0c830e6b
JP
934/**
935 * iommu_register_device_fault_handler() - Register a device fault handler
936 * @dev: the device
937 * @handler: the fault handler
938 * @data: private data passed as argument to the handler
939 *
940 * When an IOMMU fault event is received, this handler gets called with the
bf3255b3
JPB
941 * fault event and data as argument. The handler should return 0 on success. If
942 * the fault is recoverable (IOMMU_FAULT_PAGE_REQ), the consumer should also
943 * complete the fault by calling iommu_page_response() with one of the following
944 * response code:
945 * - IOMMU_PAGE_RESP_SUCCESS: retry the translation
946 * - IOMMU_PAGE_RESP_INVALID: terminate the fault
947 * - IOMMU_PAGE_RESP_FAILURE: terminate the fault and stop reporting
948 * page faults if possible.
0c830e6b
JP
949 *
950 * Return 0 if the fault handler was installed successfully, or an error.
951 */
952int iommu_register_device_fault_handler(struct device *dev,
953 iommu_dev_fault_handler_t handler,
954 void *data)
955{
956 struct iommu_param *param = dev->iommu_param;
957 int ret = 0;
958
959 if (!param)
960 return -EINVAL;
961
962 mutex_lock(&param->lock);
963 /* Only allow one fault handler registered for each device */
964 if (param->fault_param) {
965 ret = -EBUSY;
966 goto done_unlock;
967 }
968
969 get_device(dev);
970 param->fault_param = kzalloc(sizeof(*param->fault_param), GFP_KERNEL);
971 if (!param->fault_param) {
972 put_device(dev);
973 ret = -ENOMEM;
974 goto done_unlock;
975 }
976 param->fault_param->handler = handler;
977 param->fault_param->data = data;
bf3255b3
JPB
978 mutex_init(&param->fault_param->lock);
979 INIT_LIST_HEAD(&param->fault_param->faults);
0c830e6b
JP
980
981done_unlock:
982 mutex_unlock(&param->lock);
983
984 return ret;
985}
986EXPORT_SYMBOL_GPL(iommu_register_device_fault_handler);
987
988/**
989 * iommu_unregister_device_fault_handler() - Unregister the device fault handler
990 * @dev: the device
991 *
992 * Remove the device fault handler installed with
993 * iommu_register_device_fault_handler().
994 *
995 * Return 0 on success, or an error.
996 */
997int iommu_unregister_device_fault_handler(struct device *dev)
998{
999 struct iommu_param *param = dev->iommu_param;
1000 int ret = 0;
1001
1002 if (!param)
1003 return -EINVAL;
1004
1005 mutex_lock(&param->lock);
1006
1007 if (!param->fault_param)
1008 goto unlock;
1009
bf3255b3
JPB
1010 /* we cannot unregister handler if there are pending faults */
1011 if (!list_empty(&param->fault_param->faults)) {
1012 ret = -EBUSY;
1013 goto unlock;
1014 }
1015
0c830e6b
JP
1016 kfree(param->fault_param);
1017 param->fault_param = NULL;
1018 put_device(dev);
1019unlock:
1020 mutex_unlock(&param->lock);
1021
1022 return ret;
1023}
1024EXPORT_SYMBOL_GPL(iommu_unregister_device_fault_handler);
1025
1026/**
1027 * iommu_report_device_fault() - Report fault event to device driver
1028 * @dev: the device
1029 * @evt: fault event data
1030 *
1031 * Called by IOMMU drivers when a fault is detected, typically in a threaded IRQ
bf3255b3
JPB
1032 * handler. When this function fails and the fault is recoverable, it is the
1033 * caller's responsibility to complete the fault.
0c830e6b
JP
1034 *
1035 * Return 0 on success, or an error.
1036 */
1037int iommu_report_device_fault(struct device *dev, struct iommu_fault_event *evt)
1038{
1039 struct iommu_param *param = dev->iommu_param;
bf3255b3 1040 struct iommu_fault_event *evt_pending = NULL;
0c830e6b
JP
1041 struct iommu_fault_param *fparam;
1042 int ret = 0;
1043
1044 if (!param || !evt)
1045 return -EINVAL;
1046
1047 /* we only report device fault if there is a handler registered */
1048 mutex_lock(&param->lock);
1049 fparam = param->fault_param;
1050 if (!fparam || !fparam->handler) {
1051 ret = -EINVAL;
1052 goto done_unlock;
1053 }
bf3255b3
JPB
1054
1055 if (evt->fault.type == IOMMU_FAULT_PAGE_REQ &&
1056 (evt->fault.prm.flags & IOMMU_FAULT_PAGE_REQUEST_LAST_PAGE)) {
1057 evt_pending = kmemdup(evt, sizeof(struct iommu_fault_event),
1058 GFP_KERNEL);
1059 if (!evt_pending) {
1060 ret = -ENOMEM;
1061 goto done_unlock;
1062 }
1063 mutex_lock(&fparam->lock);
1064 list_add_tail(&evt_pending->list, &fparam->faults);
1065 mutex_unlock(&fparam->lock);
1066 }
1067
0c830e6b 1068 ret = fparam->handler(&evt->fault, fparam->data);
bf3255b3
JPB
1069 if (ret && evt_pending) {
1070 mutex_lock(&fparam->lock);
1071 list_del(&evt_pending->list);
1072 mutex_unlock(&fparam->lock);
1073 kfree(evt_pending);
1074 }
0c830e6b
JP
1075done_unlock:
1076 mutex_unlock(&param->lock);
1077 return ret;
1078}
1079EXPORT_SYMBOL_GPL(iommu_report_device_fault);
1080
bf3255b3
JPB
1081int iommu_page_response(struct device *dev,
1082 struct iommu_page_response *msg)
1083{
1084 bool pasid_valid;
1085 int ret = -EINVAL;
1086 struct iommu_fault_event *evt;
1087 struct iommu_fault_page_request *prm;
1088 struct iommu_param *param = dev->iommu_param;
1089 struct iommu_domain *domain = iommu_get_domain_for_dev(dev);
1090
1091 if (!domain || !domain->ops->page_response)
1092 return -ENODEV;
1093
1094 if (!param || !param->fault_param)
1095 return -EINVAL;
1096
1097 if (msg->version != IOMMU_PAGE_RESP_VERSION_1 ||
1098 msg->flags & ~IOMMU_PAGE_RESP_PASID_VALID)
1099 return -EINVAL;
1100
1101 /* Only send response if there is a fault report pending */
1102 mutex_lock(&param->fault_param->lock);
1103 if (list_empty(&param->fault_param->faults)) {
1104 dev_warn_ratelimited(dev, "no pending PRQ, drop response\n");
1105 goto done_unlock;
1106 }
1107 /*
1108 * Check if we have a matching page request pending to respond,
1109 * otherwise return -EINVAL
1110 */
1111 list_for_each_entry(evt, &param->fault_param->faults, list) {
1112 prm = &evt->fault.prm;
1113 pasid_valid = prm->flags & IOMMU_FAULT_PAGE_REQUEST_PASID_VALID;
1114
1115 if ((pasid_valid && prm->pasid != msg->pasid) ||
1116 prm->grpid != msg->grpid)
1117 continue;
1118
1119 /* Sanitize the reply */
1120 msg->flags = pasid_valid ? IOMMU_PAGE_RESP_PASID_VALID : 0;
1121
1122 ret = domain->ops->page_response(dev, evt, msg);
1123 list_del(&evt->list);
1124 kfree(evt);
1125 break;
1126 }
1127
1128done_unlock:
1129 mutex_unlock(&param->fault_param->lock);
1130 return ret;
1131}
1132EXPORT_SYMBOL_GPL(iommu_page_response);
1133
d72e31c9
AW
1134/**
1135 * iommu_group_id - Return ID for a group
1136 * @group: the group to ID
1137 *
1138 * Return the unique ID for the group matching the sysfs group number.
1139 */
1140int iommu_group_id(struct iommu_group *group)
1141{
1142 return group->id;
1143}
1144EXPORT_SYMBOL_GPL(iommu_group_id);
1460432c 1145
f096c061
AW
1146static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
1147 unsigned long *devfns);
1148
104a1c13
AW
1149/*
1150 * To consider a PCI device isolated, we require ACS to support Source
1151 * Validation, Request Redirection, Completer Redirection, and Upstream
1152 * Forwarding. This effectively means that devices cannot spoof their
1153 * requester ID, requests and completions cannot be redirected, and all
1154 * transactions are forwarded upstream, even as it passes through a
1155 * bridge where the target device is downstream.
1156 */
1157#define REQ_ACS_FLAGS (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
1158
f096c061
AW
1159/*
1160 * For multifunction devices which are not isolated from each other, find
1161 * all the other non-isolated functions and look for existing groups. For
1162 * each function, we also need to look for aliases to or from other devices
1163 * that may already have a group.
1164 */
1165static struct iommu_group *get_pci_function_alias_group(struct pci_dev *pdev,
1166 unsigned long *devfns)
1167{
1168 struct pci_dev *tmp = NULL;
1169 struct iommu_group *group;
1170
1171 if (!pdev->multifunction || pci_acs_enabled(pdev, REQ_ACS_FLAGS))
1172 return NULL;
1173
1174 for_each_pci_dev(tmp) {
1175 if (tmp == pdev || tmp->bus != pdev->bus ||
1176 PCI_SLOT(tmp->devfn) != PCI_SLOT(pdev->devfn) ||
1177 pci_acs_enabled(tmp, REQ_ACS_FLAGS))
1178 continue;
1179
1180 group = get_pci_alias_group(tmp, devfns);
1181 if (group) {
1182 pci_dev_put(tmp);
1183 return group;
1184 }
1185 }
1186
1187 return NULL;
1188}
1189
1190/*
338c3149
JL
1191 * Look for aliases to or from the given device for existing groups. DMA
1192 * aliases are only supported on the same bus, therefore the search
f096c061
AW
1193 * space is quite small (especially since we're really only looking at pcie
1194 * device, and therefore only expect multiple slots on the root complex or
1195 * downstream switch ports). It's conceivable though that a pair of
1196 * multifunction devices could have aliases between them that would cause a
1197 * loop. To prevent this, we use a bitmap to track where we've been.
1198 */
1199static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
1200 unsigned long *devfns)
1201{
1202 struct pci_dev *tmp = NULL;
1203 struct iommu_group *group;
1204
1205 if (test_and_set_bit(pdev->devfn & 0xff, devfns))
1206 return NULL;
1207
1208 group = iommu_group_get(&pdev->dev);
1209 if (group)
1210 return group;
1211
1212 for_each_pci_dev(tmp) {
1213 if (tmp == pdev || tmp->bus != pdev->bus)
1214 continue;
1215
1216 /* We alias them or they alias us */
338c3149 1217 if (pci_devs_are_dma_aliases(pdev, tmp)) {
f096c061
AW
1218 group = get_pci_alias_group(tmp, devfns);
1219 if (group) {
1220 pci_dev_put(tmp);
1221 return group;
1222 }
1223
1224 group = get_pci_function_alias_group(tmp, devfns);
1225 if (group) {
1226 pci_dev_put(tmp);
1227 return group;
1228 }
1229 }
1230 }
1231
1232 return NULL;
1233}
1234
104a1c13
AW
1235struct group_for_pci_data {
1236 struct pci_dev *pdev;
1237 struct iommu_group *group;
1238};
1239
1240/*
1241 * DMA alias iterator callback, return the last seen device. Stop and return
1242 * the IOMMU group if we find one along the way.
1243 */
1244static int get_pci_alias_or_group(struct pci_dev *pdev, u16 alias, void *opaque)
1245{
1246 struct group_for_pci_data *data = opaque;
1247
1248 data->pdev = pdev;
1249 data->group = iommu_group_get(&pdev->dev);
1250
1251 return data->group != NULL;
1252}
1253
6eab556a
JR
1254/*
1255 * Generic device_group call-back function. It just allocates one
1256 * iommu-group per device.
1257 */
1258struct iommu_group *generic_device_group(struct device *dev)
1259{
7f7a2304 1260 return iommu_group_alloc();
6eab556a
JR
1261}
1262
104a1c13
AW
1263/*
1264 * Use standard PCI bus topology, isolation features, and DMA alias quirks
1265 * to find or create an IOMMU group for a device.
1266 */
5e62292b 1267struct iommu_group *pci_device_group(struct device *dev)
104a1c13 1268{
5e62292b 1269 struct pci_dev *pdev = to_pci_dev(dev);
104a1c13
AW
1270 struct group_for_pci_data data;
1271 struct pci_bus *bus;
1272 struct iommu_group *group = NULL;
f096c061 1273 u64 devfns[4] = { 0 };
104a1c13 1274
5e62292b
JR
1275 if (WARN_ON(!dev_is_pci(dev)))
1276 return ERR_PTR(-EINVAL);
1277
104a1c13
AW
1278 /*
1279 * Find the upstream DMA alias for the device. A device must not
1280 * be aliased due to topology in order to have its own IOMMU group.
1281 * If we find an alias along the way that already belongs to a
1282 * group, use it.
1283 */
1284 if (pci_for_each_dma_alias(pdev, get_pci_alias_or_group, &data))
1285 return data.group;
1286
1287 pdev = data.pdev;
1288
1289 /*
1290 * Continue upstream from the point of minimum IOMMU granularity
1291 * due to aliases to the point where devices are protected from
1292 * peer-to-peer DMA by PCI ACS. Again, if we find an existing
1293 * group, use it.
1294 */
1295 for (bus = pdev->bus; !pci_is_root_bus(bus); bus = bus->parent) {
1296 if (!bus->self)
1297 continue;
1298
1299 if (pci_acs_path_enabled(bus->self, NULL, REQ_ACS_FLAGS))
1300 break;
1301
1302 pdev = bus->self;
1303
1304 group = iommu_group_get(&pdev->dev);
1305 if (group)
1306 return group;
1307 }
1308
1309 /*
f096c061
AW
1310 * Look for existing groups on device aliases. If we alias another
1311 * device or another device aliases us, use the same group.
104a1c13 1312 */
f096c061
AW
1313 group = get_pci_alias_group(pdev, (unsigned long *)devfns);
1314 if (group)
1315 return group;
104a1c13
AW
1316
1317 /*
f096c061
AW
1318 * Look for existing groups on non-isolated functions on the same
1319 * slot and aliases of those funcions, if any. No need to clear
1320 * the search bitmap, the tested devfns are still valid.
104a1c13 1321 */
f096c061
AW
1322 group = get_pci_function_alias_group(pdev, (unsigned long *)devfns);
1323 if (group)
1324 return group;
104a1c13
AW
1325
1326 /* No shared group found, allocate new */
7f7a2304 1327 return iommu_group_alloc();
104a1c13
AW
1328}
1329
eab03e2a
NG
1330/* Get the IOMMU group for device on fsl-mc bus */
1331struct iommu_group *fsl_mc_device_group(struct device *dev)
1332{
1333 struct device *cont_dev = fsl_mc_cont_dev(dev);
1334 struct iommu_group *group;
1335
1336 group = iommu_group_get(cont_dev);
1337 if (!group)
1338 group = iommu_group_alloc();
1339 return group;
1340}
1341
104a1c13
AW
1342/**
1343 * iommu_group_get_for_dev - Find or create the IOMMU group for a device
1344 * @dev: target device
1345 *
1346 * This function is intended to be called by IOMMU drivers and extended to
1347 * support common, bus-defined algorithms when determining or creating the
1348 * IOMMU group for a device. On success, the caller will hold a reference
1349 * to the returned IOMMU group, which will already include the provided
1350 * device. The reference should be released with iommu_group_put().
1351 */
1352struct iommu_group *iommu_group_get_for_dev(struct device *dev)
1353{
46c6b2bc 1354 const struct iommu_ops *ops = dev->bus->iommu_ops;
c4a783b8 1355 struct iommu_group *group;
104a1c13
AW
1356 int ret;
1357
1358 group = iommu_group_get(dev);
1359 if (group)
1360 return group;
1361
05f80300
RM
1362 if (!ops)
1363 return ERR_PTR(-EINVAL);
104a1c13 1364
05f80300 1365 group = ops->device_group(dev);
72dcac63
JR
1366 if (WARN_ON_ONCE(group == NULL))
1367 return ERR_PTR(-EINVAL);
1368
104a1c13
AW
1369 if (IS_ERR(group))
1370 return group;
1371
1228236d
JR
1372 /*
1373 * Try to allocate a default domain - needs support from the
1374 * IOMMU driver.
1375 */
1376 if (!group->default_domain) {
fccb4e3b
WD
1377 struct iommu_domain *dom;
1378
1379 dom = __iommu_domain_alloc(dev->bus, iommu_def_domain_type);
1380 if (!dom && iommu_def_domain_type != IOMMU_DOMAIN_DMA) {
fccb4e3b 1381 dom = __iommu_domain_alloc(dev->bus, IOMMU_DOMAIN_DMA);
8bc32a28
JR
1382 if (dom) {
1383 dev_warn(dev,
1384 "failed to allocate default IOMMU domain of type %u; falling back to IOMMU_DOMAIN_DMA",
1385 iommu_def_domain_type);
1386 }
fccb4e3b
WD
1387 }
1388
1389 group->default_domain = dom;
eebb8034 1390 if (!group->domain)
fccb4e3b 1391 group->domain = dom;
68a6efe8
ZL
1392
1393 if (dom && !iommu_dma_strict) {
1394 int attr = 1;
1395 iommu_domain_set_attr(dom,
1396 DOMAIN_ATTR_DMA_USE_FLUSH_QUEUE,
1397 &attr);
1398 }
1228236d
JR
1399 }
1400
104a1c13
AW
1401 ret = iommu_group_add_device(group, dev);
1402 if (ret) {
1403 iommu_group_put(group);
1404 return ERR_PTR(ret);
1405 }
1406
1407 return group;
1408}
1409
6827ca83
JR
1410struct iommu_domain *iommu_group_default_domain(struct iommu_group *group)
1411{
1412 return group->default_domain;
1413}
1414
d72e31c9 1415static int add_iommu_group(struct device *dev, void *data)
1460432c 1416{
cc5aed44 1417 int ret = iommu_probe_device(dev);
38667f18
JR
1418
1419 /*
1420 * We ignore -ENODEV errors for now, as they just mean that the
1421 * device is not translated by an IOMMU. We still care about
1422 * other errors and fail to initialize when they happen.
1423 */
1424 if (ret == -ENODEV)
1425 ret = 0;
1426
1427 return ret;
1460432c
AW
1428}
1429
8da30142
JR
1430static int remove_iommu_group(struct device *dev, void *data)
1431{
cc5aed44 1432 iommu_release_device(dev);
1460432c
AW
1433
1434 return 0;
1435}
1436
d72e31c9
AW
1437static int iommu_bus_notifier(struct notifier_block *nb,
1438 unsigned long action, void *data)
1460432c 1439{
cc5aed44 1440 unsigned long group_action = 0;
1460432c 1441 struct device *dev = data;
d72e31c9 1442 struct iommu_group *group;
d72e31c9
AW
1443
1444 /*
1445 * ADD/DEL call into iommu driver ops if provided, which may
1446 * result in ADD/DEL notifiers to group->notifier
1447 */
1448 if (action == BUS_NOTIFY_ADD_DEVICE) {
cc5aed44 1449 int ret;
3ba8775f 1450
cc5aed44
JR
1451 ret = iommu_probe_device(dev);
1452 return (ret) ? NOTIFY_DONE : NOTIFY_OK;
843cb6dc 1453 } else if (action == BUS_NOTIFY_REMOVED_DEVICE) {
cc5aed44
JR
1454 iommu_release_device(dev);
1455 return NOTIFY_OK;
d72e31c9 1456 }
1460432c 1457
d72e31c9
AW
1458 /*
1459 * Remaining BUS_NOTIFYs get filtered and republished to the
1460 * group, if anyone is listening
1461 */
1462 group = iommu_group_get(dev);
1463 if (!group)
1464 return 0;
1460432c 1465
d72e31c9
AW
1466 switch (action) {
1467 case BUS_NOTIFY_BIND_DRIVER:
1468 group_action = IOMMU_GROUP_NOTIFY_BIND_DRIVER;
1469 break;
1470 case BUS_NOTIFY_BOUND_DRIVER:
1471 group_action = IOMMU_GROUP_NOTIFY_BOUND_DRIVER;
1472 break;
1473 case BUS_NOTIFY_UNBIND_DRIVER:
1474 group_action = IOMMU_GROUP_NOTIFY_UNBIND_DRIVER;
1475 break;
1476 case BUS_NOTIFY_UNBOUND_DRIVER:
1477 group_action = IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER;
1478 break;
1479 }
1460432c 1480
d72e31c9
AW
1481 if (group_action)
1482 blocking_notifier_call_chain(&group->notifier,
1483 group_action, dev);
1460432c 1484
d72e31c9 1485 iommu_group_put(group);
1460432c
AW
1486 return 0;
1487}
1488
fb3e3065 1489static int iommu_bus_init(struct bus_type *bus, const struct iommu_ops *ops)
ff21776d 1490{
fb3e3065
MS
1491 int err;
1492 struct notifier_block *nb;
b22f6434 1493
fb3e3065
MS
1494 nb = kzalloc(sizeof(struct notifier_block), GFP_KERNEL);
1495 if (!nb)
1496 return -ENOMEM;
1497
1498 nb->notifier_call = iommu_bus_notifier;
1499
1500 err = bus_register_notifier(bus, nb);
8da30142
JR
1501 if (err)
1502 goto out_free;
d7da6bdc 1503
8cec63e5 1504 err = bus_for_each_dev(bus, NULL, NULL, add_iommu_group);
8da30142
JR
1505 if (err)
1506 goto out_err;
1507
d7da6bdc
HS
1508
1509 return 0;
8da30142
JR
1510
1511out_err:
1512 /* Clean up */
8cec63e5 1513 bus_for_each_dev(bus, NULL, NULL, remove_iommu_group);
8da30142
JR
1514 bus_unregister_notifier(bus, nb);
1515
1516out_free:
1517 kfree(nb);
1518
1519 return err;
ff21776d 1520}
fc2100eb 1521
ff21776d
JR
1522/**
1523 * bus_set_iommu - set iommu-callbacks for the bus
1524 * @bus: bus.
1525 * @ops: the callbacks provided by the iommu-driver
1526 *
1527 * This function is called by an iommu driver to set the iommu methods
1528 * used for a particular bus. Drivers for devices on that bus can use
1529 * the iommu-api after these ops are registered.
1530 * This special function is needed because IOMMUs are usually devices on
1531 * the bus itself, so the iommu drivers are not initialized when the bus
1532 * is set up. With this function the iommu-driver can set the iommu-ops
1533 * afterwards.
1534 */
b22f6434 1535int bus_set_iommu(struct bus_type *bus, const struct iommu_ops *ops)
fc2100eb 1536{
d7da6bdc
HS
1537 int err;
1538
ff21776d
JR
1539 if (bus->iommu_ops != NULL)
1540 return -EBUSY;
fc2100eb 1541
ff21776d
JR
1542 bus->iommu_ops = ops;
1543
1544 /* Do IOMMU specific setup for this bus-type */
d7da6bdc
HS
1545 err = iommu_bus_init(bus, ops);
1546 if (err)
1547 bus->iommu_ops = NULL;
1548
1549 return err;
fc2100eb 1550}
ff21776d 1551EXPORT_SYMBOL_GPL(bus_set_iommu);
fc2100eb 1552
a1b60c1c 1553bool iommu_present(struct bus_type *bus)
fc2100eb 1554{
94441c3b 1555 return bus->iommu_ops != NULL;
fc2100eb 1556}
a1b60c1c 1557EXPORT_SYMBOL_GPL(iommu_present);
fc2100eb 1558
3c0e0ca0
JR
1559bool iommu_capable(struct bus_type *bus, enum iommu_cap cap)
1560{
1561 if (!bus->iommu_ops || !bus->iommu_ops->capable)
1562 return false;
1563
1564 return bus->iommu_ops->capable(cap);
1565}
1566EXPORT_SYMBOL_GPL(iommu_capable);
1567
4f3f8d9d
OBC
1568/**
1569 * iommu_set_fault_handler() - set a fault handler for an iommu domain
1570 * @domain: iommu domain
1571 * @handler: fault handler
77ca2332 1572 * @token: user data, will be passed back to the fault handler
0ed6d2d2
OBC
1573 *
1574 * This function should be used by IOMMU users which want to be notified
1575 * whenever an IOMMU fault happens.
1576 *
1577 * The fault handler itself should return 0 on success, and an appropriate
1578 * error code otherwise.
4f3f8d9d
OBC
1579 */
1580void iommu_set_fault_handler(struct iommu_domain *domain,
77ca2332
OBC
1581 iommu_fault_handler_t handler,
1582 void *token)
4f3f8d9d
OBC
1583{
1584 BUG_ON(!domain);
1585
1586 domain->handler = handler;
77ca2332 1587 domain->handler_token = token;
4f3f8d9d 1588}
30bd918c 1589EXPORT_SYMBOL_GPL(iommu_set_fault_handler);
4f3f8d9d 1590
53723dc5
JR
1591static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
1592 unsigned type)
fc2100eb
JR
1593{
1594 struct iommu_domain *domain;
fc2100eb 1595
94441c3b 1596 if (bus == NULL || bus->iommu_ops == NULL)
905d66c1
JR
1597 return NULL;
1598
53723dc5 1599 domain = bus->iommu_ops->domain_alloc(type);
fc2100eb
JR
1600 if (!domain)
1601 return NULL;
1602
8539c7c1 1603 domain->ops = bus->iommu_ops;
53723dc5 1604 domain->type = type;
d16e0faa
RM
1605 /* Assume all sizes by default; the driver may override this later */
1606 domain->pgsize_bitmap = bus->iommu_ops->pgsize_bitmap;
905d66c1 1607
fc2100eb 1608 return domain;
fc2100eb 1609}
fc2100eb 1610
53723dc5
JR
1611struct iommu_domain *iommu_domain_alloc(struct bus_type *bus)
1612{
1613 return __iommu_domain_alloc(bus, IOMMU_DOMAIN_UNMANAGED);
fc2100eb
JR
1614}
1615EXPORT_SYMBOL_GPL(iommu_domain_alloc);
1616
1617void iommu_domain_free(struct iommu_domain *domain)
1618{
89be34a1 1619 domain->ops->domain_free(domain);
fc2100eb
JR
1620}
1621EXPORT_SYMBOL_GPL(iommu_domain_free);
1622
426a2738
JR
1623static int __iommu_attach_device(struct iommu_domain *domain,
1624 struct device *dev)
fc2100eb 1625{
b54db778 1626 int ret;
e01d1913
BH
1627 if ((domain->ops->is_attach_deferred != NULL) &&
1628 domain->ops->is_attach_deferred(domain, dev))
1629 return 0;
1630
e5aa7f00
JR
1631 if (unlikely(domain->ops->attach_dev == NULL))
1632 return -ENODEV;
1633
b54db778
SK
1634 ret = domain->ops->attach_dev(domain, dev);
1635 if (!ret)
1636 trace_attach_device_to_domain(dev);
1637 return ret;
fc2100eb 1638}
426a2738
JR
1639
1640int iommu_attach_device(struct iommu_domain *domain, struct device *dev)
1641{
1642 struct iommu_group *group;
1643 int ret;
1644
1645 group = iommu_group_get(dev);
9ae9df03
JC
1646 if (!group)
1647 return -ENODEV;
1648
426a2738 1649 /*
05f80300 1650 * Lock the group to make sure the device-count doesn't
426a2738
JR
1651 * change while we are attaching
1652 */
1653 mutex_lock(&group->mutex);
1654 ret = -EINVAL;
1655 if (iommu_group_device_count(group) != 1)
1656 goto out_unlock;
1657
e39cb8a3 1658 ret = __iommu_attach_group(domain, group);
426a2738
JR
1659
1660out_unlock:
1661 mutex_unlock(&group->mutex);
1662 iommu_group_put(group);
1663
1664 return ret;
1665}
fc2100eb
JR
1666EXPORT_SYMBOL_GPL(iommu_attach_device);
1667
426a2738
JR
1668static void __iommu_detach_device(struct iommu_domain *domain,
1669 struct device *dev)
fc2100eb 1670{
e01d1913
BH
1671 if ((domain->ops->is_attach_deferred != NULL) &&
1672 domain->ops->is_attach_deferred(domain, dev))
1673 return;
1674
e5aa7f00
JR
1675 if (unlikely(domain->ops->detach_dev == NULL))
1676 return;
1677
1678 domain->ops->detach_dev(domain, dev);
69980630 1679 trace_detach_device_from_domain(dev);
fc2100eb 1680}
426a2738
JR
1681
1682void iommu_detach_device(struct iommu_domain *domain, struct device *dev)
1683{
1684 struct iommu_group *group;
1685
1686 group = iommu_group_get(dev);
9ae9df03
JC
1687 if (!group)
1688 return;
426a2738
JR
1689
1690 mutex_lock(&group->mutex);
1691 if (iommu_group_device_count(group) != 1) {
1692 WARN_ON(1);
1693 goto out_unlock;
1694 }
1695
e39cb8a3 1696 __iommu_detach_group(domain, group);
426a2738
JR
1697
1698out_unlock:
1699 mutex_unlock(&group->mutex);
1700 iommu_group_put(group);
1701}
fc2100eb
JR
1702EXPORT_SYMBOL_GPL(iommu_detach_device);
1703
2c1296d9
JR
1704struct iommu_domain *iommu_get_domain_for_dev(struct device *dev)
1705{
1706 struct iommu_domain *domain;
1707 struct iommu_group *group;
1708
1709 group = iommu_group_get(dev);
1464d0b1 1710 if (!group)
2c1296d9
JR
1711 return NULL;
1712
1713 domain = group->domain;
1714
1715 iommu_group_put(group);
1716
1717 return domain;
1718}
1719EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev);
fc2100eb 1720
d72e31c9 1721/*
6af588fe
RM
1722 * For IOMMU_DOMAIN_DMA implementations which already provide their own
1723 * guarantees that the group and its default domain are valid and correct.
1724 */
1725struct iommu_domain *iommu_get_dma_domain(struct device *dev)
1726{
1727 return dev->iommu_group->default_domain;
1728}
1729
d72e31c9 1730/*
35449adc 1731 * IOMMU groups are really the natural working unit of the IOMMU, but
d72e31c9
AW
1732 * the IOMMU API works on domains and devices. Bridge that gap by
1733 * iterating over the devices in a group. Ideally we'd have a single
1734 * device which represents the requestor ID of the group, but we also
1735 * allow IOMMU drivers to create policy defined minimum sets, where
1736 * the physical hardware may be able to distiguish members, but we
1737 * wish to group them at a higher level (ex. untrusted multi-function
1738 * PCI devices). Thus we attach each device.
1739 */
1740static int iommu_group_do_attach_device(struct device *dev, void *data)
1741{
1742 struct iommu_domain *domain = data;
1743
426a2738 1744 return __iommu_attach_device(domain, dev);
d72e31c9
AW
1745}
1746
e39cb8a3
JR
1747static int __iommu_attach_group(struct iommu_domain *domain,
1748 struct iommu_group *group)
1749{
1750 int ret;
1751
1752 if (group->default_domain && group->domain != group->default_domain)
1753 return -EBUSY;
1754
1755 ret = __iommu_group_for_each_dev(group, domain,
1756 iommu_group_do_attach_device);
1757 if (ret == 0)
1758 group->domain = domain;
1759
1760 return ret;
d72e31c9
AW
1761}
1762
1763int iommu_attach_group(struct iommu_domain *domain, struct iommu_group *group)
1764{
e39cb8a3
JR
1765 int ret;
1766
1767 mutex_lock(&group->mutex);
1768 ret = __iommu_attach_group(domain, group);
1769 mutex_unlock(&group->mutex);
1770
1771 return ret;
d72e31c9
AW
1772}
1773EXPORT_SYMBOL_GPL(iommu_attach_group);
1774
1775static int iommu_group_do_detach_device(struct device *dev, void *data)
1776{
1777 struct iommu_domain *domain = data;
1778
426a2738 1779 __iommu_detach_device(domain, dev);
d72e31c9
AW
1780
1781 return 0;
1782}
1783
e39cb8a3
JR
1784static void __iommu_detach_group(struct iommu_domain *domain,
1785 struct iommu_group *group)
1786{
1787 int ret;
1788
1789 if (!group->default_domain) {
1790 __iommu_group_for_each_dev(group, domain,
1791 iommu_group_do_detach_device);
1792 group->domain = NULL;
1793 return;
1794 }
1795
1796 if (group->domain == group->default_domain)
1797 return;
1798
1799 /* Detach by re-attaching to the default domain */
1800 ret = __iommu_group_for_each_dev(group, group->default_domain,
1801 iommu_group_do_attach_device);
1802 if (ret != 0)
1803 WARN_ON(1);
1804 else
1805 group->domain = group->default_domain;
1806}
1807
d72e31c9
AW
1808void iommu_detach_group(struct iommu_domain *domain, struct iommu_group *group)
1809{
e39cb8a3
JR
1810 mutex_lock(&group->mutex);
1811 __iommu_detach_group(domain, group);
1812 mutex_unlock(&group->mutex);
d72e31c9
AW
1813}
1814EXPORT_SYMBOL_GPL(iommu_detach_group);
1815
bb5547ac 1816phys_addr_t iommu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova)
fc2100eb 1817{
e5aa7f00
JR
1818 if (unlikely(domain->ops->iova_to_phys == NULL))
1819 return 0;
1820
1821 return domain->ops->iova_to_phys(domain, iova);
fc2100eb
JR
1822}
1823EXPORT_SYMBOL_GPL(iommu_iova_to_phys);
dbb9fd86 1824
bd13969b
AW
1825static size_t iommu_pgsize(struct iommu_domain *domain,
1826 unsigned long addr_merge, size_t size)
1827{
1828 unsigned int pgsize_idx;
1829 size_t pgsize;
1830
1831 /* Max page size that still fits into 'size' */
1832 pgsize_idx = __fls(size);
1833
1834 /* need to consider alignment requirements ? */
1835 if (likely(addr_merge)) {
1836 /* Max page size allowed by address */
1837 unsigned int align_pgsize_idx = __ffs(addr_merge);
1838 pgsize_idx = min(pgsize_idx, align_pgsize_idx);
1839 }
1840
1841 /* build a mask of acceptable page sizes */
1842 pgsize = (1UL << (pgsize_idx + 1)) - 1;
1843
1844 /* throw away page sizes not supported by the hardware */
d16e0faa 1845 pgsize &= domain->pgsize_bitmap;
bd13969b
AW
1846
1847 /* make sure we're still sane */
1848 BUG_ON(!pgsize);
1849
1850 /* pick the biggest page */
1851 pgsize_idx = __fls(pgsize);
1852 pgsize = 1UL << pgsize_idx;
1853
1854 return pgsize;
1855}
1856
781ca2de
TM
1857int __iommu_map(struct iommu_domain *domain, unsigned long iova,
1858 phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
cefc53c7 1859{
1d7ae53b 1860 const struct iommu_ops *ops = domain->ops;
7d3002cc
OBC
1861 unsigned long orig_iova = iova;
1862 unsigned int min_pagesz;
1863 size_t orig_size = size;
06bfcaa9 1864 phys_addr_t orig_paddr = paddr;
7d3002cc 1865 int ret = 0;
cefc53c7 1866
1d7ae53b 1867 if (unlikely(ops->map == NULL ||
d16e0faa 1868 domain->pgsize_bitmap == 0UL))
e5aa7f00 1869 return -ENODEV;
cefc53c7 1870
a10315e5
JR
1871 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1872 return -EINVAL;
1873
7d3002cc 1874 /* find out the minimum page size supported */
d16e0faa 1875 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
7d3002cc
OBC
1876
1877 /*
1878 * both the virtual address and the physical one, as well as
1879 * the size of the mapping, must be aligned (at least) to the
1880 * size of the smallest page supported by the hardware
1881 */
1882 if (!IS_ALIGNED(iova | paddr | size, min_pagesz)) {
abedb049 1883 pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
6197ca82 1884 iova, &paddr, size, min_pagesz);
7d3002cc
OBC
1885 return -EINVAL;
1886 }
1887
abedb049 1888 pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova, &paddr, size);
7d3002cc
OBC
1889
1890 while (size) {
bd13969b 1891 size_t pgsize = iommu_pgsize(domain, iova | paddr, size);
7d3002cc 1892
abedb049 1893 pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx\n",
6197ca82 1894 iova, &paddr, pgsize);
781ca2de 1895 ret = ops->map(domain, iova, paddr, pgsize, prot, gfp);
7d3002cc 1896
7d3002cc
OBC
1897 if (ret)
1898 break;
1899
1900 iova += pgsize;
1901 paddr += pgsize;
1902 size -= pgsize;
1903 }
1904
1d7ae53b
DO
1905 if (ops->iotlb_sync_map)
1906 ops->iotlb_sync_map(domain);
1907
7d3002cc
OBC
1908 /* unroll mapping in case something went wrong */
1909 if (ret)
1910 iommu_unmap(domain, orig_iova, orig_size - size);
e0be7c86 1911 else
06bfcaa9 1912 trace_map(orig_iova, orig_paddr, orig_size);
7d3002cc
OBC
1913
1914 return ret;
cefc53c7 1915}
781ca2de
TM
1916
1917int iommu_map(struct iommu_domain *domain, unsigned long iova,
1918 phys_addr_t paddr, size_t size, int prot)
1919{
1920 might_sleep();
1921 return __iommu_map(domain, iova, paddr, size, prot, GFP_KERNEL);
1922}
cefc53c7
JR
1923EXPORT_SYMBOL_GPL(iommu_map);
1924
781ca2de
TM
1925int iommu_map_atomic(struct iommu_domain *domain, unsigned long iova,
1926 phys_addr_t paddr, size_t size, int prot)
1927{
1928 return __iommu_map(domain, iova, paddr, size, prot, GFP_ATOMIC);
1929}
1930EXPORT_SYMBOL_GPL(iommu_map_atomic);
1931
add02cfd
JR
1932static size_t __iommu_unmap(struct iommu_domain *domain,
1933 unsigned long iova, size_t size,
a7d20dc1 1934 struct iommu_iotlb_gather *iotlb_gather)
cefc53c7 1935{
add02cfd 1936 const struct iommu_ops *ops = domain->ops;
7d3002cc 1937 size_t unmapped_page, unmapped = 0;
6fd492fd 1938 unsigned long orig_iova = iova;
add02cfd 1939 unsigned int min_pagesz;
cefc53c7 1940
add02cfd 1941 if (unlikely(ops->unmap == NULL ||
d16e0faa 1942 domain->pgsize_bitmap == 0UL))
c5611a87 1943 return 0;
e5aa7f00 1944
a10315e5 1945 if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
c5611a87 1946 return 0;
a10315e5 1947
7d3002cc 1948 /* find out the minimum page size supported */
d16e0faa 1949 min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
7d3002cc
OBC
1950
1951 /*
1952 * The virtual address, as well as the size of the mapping, must be
1953 * aligned (at least) to the size of the smallest page supported
1954 * by the hardware
1955 */
1956 if (!IS_ALIGNED(iova | size, min_pagesz)) {
6197ca82
JP
1957 pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
1958 iova, size, min_pagesz);
c5611a87 1959 return 0;
7d3002cc
OBC
1960 }
1961
6197ca82 1962 pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova, size);
7d3002cc
OBC
1963
1964 /*
1965 * Keep iterating until we either unmap 'size' bytes (or more)
1966 * or we hit an area that isn't mapped.
1967 */
1968 while (unmapped < size) {
bd13969b 1969 size_t pgsize = iommu_pgsize(domain, iova, size - unmapped);
7d3002cc 1970
56f8af5e 1971 unmapped_page = ops->unmap(domain, iova, pgsize, iotlb_gather);
7d3002cc
OBC
1972 if (!unmapped_page)
1973 break;
1974
6197ca82
JP
1975 pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
1976 iova, unmapped_page);
7d3002cc
OBC
1977
1978 iova += unmapped_page;
1979 unmapped += unmapped_page;
1980 }
1981
db8614d3 1982 trace_unmap(orig_iova, size, unmapped);
7d3002cc 1983 return unmapped;
cefc53c7 1984}
add02cfd
JR
1985
1986size_t iommu_unmap(struct iommu_domain *domain,
1987 unsigned long iova, size_t size)
1988{
a7d20dc1
WD
1989 struct iommu_iotlb_gather iotlb_gather;
1990 size_t ret;
1991
1992 iommu_iotlb_gather_init(&iotlb_gather);
1993 ret = __iommu_unmap(domain, iova, size, &iotlb_gather);
1994 iommu_tlb_sync(domain, &iotlb_gather);
1995
1996 return ret;
add02cfd 1997}
cefc53c7 1998EXPORT_SYMBOL_GPL(iommu_unmap);
1460432c 1999
add02cfd 2000size_t iommu_unmap_fast(struct iommu_domain *domain,
a7d20dc1
WD
2001 unsigned long iova, size_t size,
2002 struct iommu_iotlb_gather *iotlb_gather)
add02cfd 2003{
a7d20dc1 2004 return __iommu_unmap(domain, iova, size, iotlb_gather);
add02cfd
JR
2005}
2006EXPORT_SYMBOL_GPL(iommu_unmap_fast);
2007
781ca2de
TM
2008size_t __iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
2009 struct scatterlist *sg, unsigned int nents, int prot,
2010 gfp_t gfp)
315786eb 2011{
5d95f40e
RM
2012 size_t len = 0, mapped = 0;
2013 phys_addr_t start;
2014 unsigned int i = 0;
38ec010d 2015 int ret;
315786eb 2016
5d95f40e
RM
2017 while (i <= nents) {
2018 phys_addr_t s_phys = sg_phys(sg);
18f23409 2019
5d95f40e 2020 if (len && s_phys != start + len) {
781ca2de
TM
2021 ret = __iommu_map(domain, iova + mapped, start,
2022 len, prot, gfp);
2023
5d95f40e
RM
2024 if (ret)
2025 goto out_err;
18f23409 2026
5d95f40e
RM
2027 mapped += len;
2028 len = 0;
2029 }
38ec010d 2030
5d95f40e
RM
2031 if (len) {
2032 len += sg->length;
2033 } else {
2034 len = sg->length;
2035 start = s_phys;
2036 }
38ec010d 2037
5d95f40e
RM
2038 if (++i < nents)
2039 sg = sg_next(sg);
315786eb
OH
2040 }
2041
2042 return mapped;
38ec010d
JR
2043
2044out_err:
2045 /* undo mappings already done */
2046 iommu_unmap(domain, iova, mapped);
2047
2048 return 0;
2049
315786eb 2050}
781ca2de
TM
2051
2052size_t iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
2053 struct scatterlist *sg, unsigned int nents, int prot)
2054{
2055 might_sleep();
2056 return __iommu_map_sg(domain, iova, sg, nents, prot, GFP_KERNEL);
2057}
d88e61fa 2058EXPORT_SYMBOL_GPL(iommu_map_sg);
d7787d57 2059
781ca2de
TM
2060size_t iommu_map_sg_atomic(struct iommu_domain *domain, unsigned long iova,
2061 struct scatterlist *sg, unsigned int nents, int prot)
2062{
2063 return __iommu_map_sg(domain, iova, sg, nents, prot, GFP_ATOMIC);
2064}
2065EXPORT_SYMBOL_GPL(iommu_map_sg_atomic);
2066
d7787d57 2067int iommu_domain_window_enable(struct iommu_domain *domain, u32 wnd_nr,
80f97f0f 2068 phys_addr_t paddr, u64 size, int prot)
d7787d57
JR
2069{
2070 if (unlikely(domain->ops->domain_window_enable == NULL))
2071 return -ENODEV;
2072
80f97f0f
VS
2073 return domain->ops->domain_window_enable(domain, wnd_nr, paddr, size,
2074 prot);
d7787d57
JR
2075}
2076EXPORT_SYMBOL_GPL(iommu_domain_window_enable);
2077
2078void iommu_domain_window_disable(struct iommu_domain *domain, u32 wnd_nr)
2079{
2080 if (unlikely(domain->ops->domain_window_disable == NULL))
2081 return;
2082
2083 return domain->ops->domain_window_disable(domain, wnd_nr);
2084}
2085EXPORT_SYMBOL_GPL(iommu_domain_window_disable);
2086
207c6e36
JR
2087/**
2088 * report_iommu_fault() - report about an IOMMU fault to the IOMMU framework
2089 * @domain: the iommu domain where the fault has happened
2090 * @dev: the device where the fault has happened
2091 * @iova: the faulting address
2092 * @flags: mmu fault flags (e.g. IOMMU_FAULT_READ/IOMMU_FAULT_WRITE/...)
2093 *
2094 * This function should be called by the low-level IOMMU implementations
2095 * whenever IOMMU faults happen, to allow high-level users, that are
2096 * interested in such events, to know about them.
2097 *
2098 * This event may be useful for several possible use cases:
2099 * - mere logging of the event
2100 * - dynamic TLB/PTE loading
2101 * - if restarting of the faulting device is required
2102 *
2103 * Returns 0 on success and an appropriate error code otherwise (if dynamic
2104 * PTE/TLB loading will one day be supported, implementations will be able
2105 * to tell whether it succeeded or not according to this return value).
2106 *
2107 * Specifically, -ENOSYS is returned if a fault handler isn't installed
2108 * (though fault handlers can also return -ENOSYS, in case they want to
2109 * elicit the default behavior of the IOMMU drivers).
2110 */
2111int report_iommu_fault(struct iommu_domain *domain, struct device *dev,
2112 unsigned long iova, int flags)
2113{
2114 int ret = -ENOSYS;
2115
2116 /*
2117 * if upper layers showed interest and installed a fault handler,
2118 * invoke it.
2119 */
2120 if (domain->handler)
2121 ret = domain->handler(domain, dev, iova, flags,
2122 domain->handler_token);
2123
2124 trace_io_page_fault(dev, iova, flags);
2125 return ret;
2126}
2127EXPORT_SYMBOL_GPL(report_iommu_fault);
2128
d72e31c9 2129static int __init iommu_init(void)
1460432c 2130{
d72e31c9
AW
2131 iommu_group_kset = kset_create_and_add("iommu_groups",
2132 NULL, kernel_kobj);
d72e31c9
AW
2133 BUG_ON(!iommu_group_kset);
2134
bad614b2
GH
2135 iommu_debugfs_setup();
2136
d72e31c9 2137 return 0;
1460432c 2138}
d7ef9995 2139core_initcall(iommu_init);
0cd76dd1
JR
2140
2141int iommu_domain_get_attr(struct iommu_domain *domain,
2142 enum iommu_attr attr, void *data)
2143{
0ff64f80 2144 struct iommu_domain_geometry *geometry;
d2e12160 2145 bool *paging;
0ff64f80
JR
2146 int ret = 0;
2147
2148 switch (attr) {
2149 case DOMAIN_ATTR_GEOMETRY:
2150 geometry = data;
2151 *geometry = domain->geometry;
2152
d2e12160
JR
2153 break;
2154 case DOMAIN_ATTR_PAGING:
2155 paging = data;
d16e0faa 2156 *paging = (domain->pgsize_bitmap != 0UL);
0ff64f80
JR
2157 break;
2158 default:
2159 if (!domain->ops->domain_get_attr)
2160 return -EINVAL;
0cd76dd1 2161
0ff64f80
JR
2162 ret = domain->ops->domain_get_attr(domain, attr, data);
2163 }
2164
2165 return ret;
0cd76dd1
JR
2166}
2167EXPORT_SYMBOL_GPL(iommu_domain_get_attr);
2168
2169int iommu_domain_set_attr(struct iommu_domain *domain,
2170 enum iommu_attr attr, void *data)
2171{
69356712 2172 int ret = 0;
69356712
JR
2173
2174 switch (attr) {
69356712
JR
2175 default:
2176 if (domain->ops->domain_set_attr == NULL)
2177 return -EINVAL;
2178
2179 ret = domain->ops->domain_set_attr(domain, attr, data);
2180 }
2181
2182 return ret;
1460432c 2183}
0cd76dd1 2184EXPORT_SYMBOL_GPL(iommu_domain_set_attr);
a1015c2b 2185
e5b5234a 2186void iommu_get_resv_regions(struct device *dev, struct list_head *list)
a1015c2b
JR
2187{
2188 const struct iommu_ops *ops = dev->bus->iommu_ops;
2189
e5b5234a
EA
2190 if (ops && ops->get_resv_regions)
2191 ops->get_resv_regions(dev, list);
a1015c2b
JR
2192}
2193
e5b5234a 2194void iommu_put_resv_regions(struct device *dev, struct list_head *list)
a1015c2b
JR
2195{
2196 const struct iommu_ops *ops = dev->bus->iommu_ops;
2197
e5b5234a
EA
2198 if (ops && ops->put_resv_regions)
2199 ops->put_resv_regions(dev, list);
a1015c2b 2200}
d290f1e7 2201
2b20cbba 2202struct iommu_resv_region *iommu_alloc_resv_region(phys_addr_t start,
9d3a4de4
RM
2203 size_t length, int prot,
2204 enum iommu_resv_type type)
2b20cbba
EA
2205{
2206 struct iommu_resv_region *region;
2207
2208 region = kzalloc(sizeof(*region), GFP_KERNEL);
2209 if (!region)
2210 return NULL;
2211
2212 INIT_LIST_HEAD(&region->list);
2213 region->start = start;
2214 region->length = length;
2215 region->prot = prot;
2216 region->type = type;
2217 return region;
a1015c2b 2218}
d290f1e7 2219
7423e017
LB
2220static int
2221request_default_domain_for_dev(struct device *dev, unsigned long type)
d290f1e7 2222{
7423e017 2223 struct iommu_domain *domain;
d290f1e7
JR
2224 struct iommu_group *group;
2225 int ret;
2226
2227 /* Device must already be in a group before calling this function */
57274ea2
LB
2228 group = iommu_group_get(dev);
2229 if (!group)
2230 return -EINVAL;
d290f1e7
JR
2231
2232 mutex_lock(&group->mutex);
2233
d290f1e7 2234 ret = 0;
7423e017 2235 if (group->default_domain && group->default_domain->type == type)
d290f1e7
JR
2236 goto out;
2237
2238 /* Don't change mappings of existing devices */
2239 ret = -EBUSY;
2240 if (iommu_group_device_count(group) != 1)
2241 goto out;
2242
d290f1e7 2243 ret = -ENOMEM;
7423e017
LB
2244 domain = __iommu_domain_alloc(dev->bus, type);
2245 if (!domain)
d290f1e7
JR
2246 goto out;
2247
2248 /* Attach the device to the domain */
7423e017 2249 ret = __iommu_attach_group(domain, group);
d290f1e7 2250 if (ret) {
7423e017 2251 iommu_domain_free(domain);
d290f1e7
JR
2252 goto out;
2253 }
2254
7423e017
LB
2255 iommu_group_create_direct_mappings(group, dev);
2256
d127bc9b 2257 /* Make the domain the default for this group */
d290f1e7
JR
2258 if (group->default_domain)
2259 iommu_domain_free(group->default_domain);
7423e017 2260 group->default_domain = domain;
d290f1e7 2261
7423e017
LB
2262 dev_info(dev, "Using iommu %s mapping\n",
2263 type == IOMMU_DOMAIN_DMA ? "dma" : "direct");
d290f1e7
JR
2264
2265 ret = 0;
2266out:
2267 mutex_unlock(&group->mutex);
2268 iommu_group_put(group);
2269
2270 return ret;
2271}
57f98d2f 2272
7423e017
LB
2273/* Request that a device is direct mapped by the IOMMU */
2274int iommu_request_dm_for_dev(struct device *dev)
2275{
2276 return request_default_domain_for_dev(dev, IOMMU_DOMAIN_IDENTITY);
2277}
2278
2279/* Request that a device can't be direct mapped by the IOMMU */
2280int iommu_request_dma_domain_for_dev(struct device *dev)
2281{
2282 return request_default_domain_for_dev(dev, IOMMU_DOMAIN_DMA);
2283}
2284
8a69961c
JR
2285void iommu_set_default_passthrough(bool cmd_line)
2286{
2287 if (cmd_line)
2288 iommu_set_cmd_line_dma_api();
2289
2290 iommu_def_domain_type = IOMMU_DOMAIN_IDENTITY;
2291}
2292
2293void iommu_set_default_translated(bool cmd_line)
2294{
2295 if (cmd_line)
2296 iommu_set_cmd_line_dma_api();
2297
2298 iommu_def_domain_type = IOMMU_DOMAIN_DMA;
2299}
2300
2301bool iommu_default_passthrough(void)
2302{
2303 return iommu_def_domain_type == IOMMU_DOMAIN_IDENTITY;
2304}
2305EXPORT_SYMBOL_GPL(iommu_default_passthrough);
2306
534766df 2307const struct iommu_ops *iommu_ops_from_fwnode(struct fwnode_handle *fwnode)
e4f10ffe 2308{
e4f10ffe 2309 const struct iommu_ops *ops = NULL;
d0f6f583 2310 struct iommu_device *iommu;
e4f10ffe 2311
d0f6f583
JR
2312 spin_lock(&iommu_device_lock);
2313 list_for_each_entry(iommu, &iommu_device_list, list)
2314 if (iommu->fwnode == fwnode) {
2315 ops = iommu->ops;
e4f10ffe
LP
2316 break;
2317 }
d0f6f583 2318 spin_unlock(&iommu_device_lock);
e4f10ffe
LP
2319 return ops;
2320}
2321
57f98d2f
RM
2322int iommu_fwspec_init(struct device *dev, struct fwnode_handle *iommu_fwnode,
2323 const struct iommu_ops *ops)
2324{
b4ef725e 2325 struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
57f98d2f
RM
2326
2327 if (fwspec)
2328 return ops == fwspec->ops ? 0 : -EINVAL;
2329
2330 fwspec = kzalloc(sizeof(*fwspec), GFP_KERNEL);
2331 if (!fwspec)
2332 return -ENOMEM;
2333
2334 of_node_get(to_of_node(iommu_fwnode));
2335 fwspec->iommu_fwnode = iommu_fwnode;
2336 fwspec->ops = ops;
b4ef725e 2337 dev_iommu_fwspec_set(dev, fwspec);
57f98d2f
RM
2338 return 0;
2339}
2340EXPORT_SYMBOL_GPL(iommu_fwspec_init);
2341
2342void iommu_fwspec_free(struct device *dev)
2343{
b4ef725e 2344 struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
57f98d2f
RM
2345
2346 if (fwspec) {
2347 fwnode_handle_put(fwspec->iommu_fwnode);
2348 kfree(fwspec);
b4ef725e 2349 dev_iommu_fwspec_set(dev, NULL);
57f98d2f
RM
2350 }
2351}
2352EXPORT_SYMBOL_GPL(iommu_fwspec_free);
2353
2354int iommu_fwspec_add_ids(struct device *dev, u32 *ids, int num_ids)
2355{
b4ef725e 2356 struct iommu_fwspec *fwspec = dev_iommu_fwspec_get(dev);
57f98d2f
RM
2357 size_t size;
2358 int i;
2359
2360 if (!fwspec)
2361 return -EINVAL;
2362
2363 size = offsetof(struct iommu_fwspec, ids[fwspec->num_ids + num_ids]);
2364 if (size > sizeof(*fwspec)) {
b4ef725e 2365 fwspec = krealloc(fwspec, size, GFP_KERNEL);
57f98d2f
RM
2366 if (!fwspec)
2367 return -ENOMEM;
909111ba 2368
b4ef725e 2369 dev_iommu_fwspec_set(dev, fwspec);
57f98d2f
RM
2370 }
2371
2372 for (i = 0; i < num_ids; i++)
2373 fwspec->ids[fwspec->num_ids + i] = ids[i];
2374
2375 fwspec->num_ids += num_ids;
57f98d2f
RM
2376 return 0;
2377}
2378EXPORT_SYMBOL_GPL(iommu_fwspec_add_ids);
a3a19592
LB
2379
2380/*
2381 * Per device IOMMU features.
2382 */
2383bool iommu_dev_has_feature(struct device *dev, enum iommu_dev_features feat)
2384{
2385 const struct iommu_ops *ops = dev->bus->iommu_ops;
2386
2387 if (ops && ops->dev_has_feat)
2388 return ops->dev_has_feat(dev, feat);
2389
2390 return false;
2391}
2392EXPORT_SYMBOL_GPL(iommu_dev_has_feature);
2393
2394int iommu_dev_enable_feature(struct device *dev, enum iommu_dev_features feat)
2395{
2396 const struct iommu_ops *ops = dev->bus->iommu_ops;
2397
2398 if (ops && ops->dev_enable_feat)
2399 return ops->dev_enable_feat(dev, feat);
2400
2401 return -ENODEV;
2402}
2403EXPORT_SYMBOL_GPL(iommu_dev_enable_feature);
2404
2405/*
2406 * The device drivers should do the necessary cleanups before calling this.
2407 * For example, before disabling the aux-domain feature, the device driver
2408 * should detach all aux-domains. Otherwise, this will return -EBUSY.
2409 */
2410int iommu_dev_disable_feature(struct device *dev, enum iommu_dev_features feat)
2411{
2412 const struct iommu_ops *ops = dev->bus->iommu_ops;
2413
2414 if (ops && ops->dev_disable_feat)
2415 return ops->dev_disable_feat(dev, feat);
2416
2417 return -EBUSY;
2418}
2419EXPORT_SYMBOL_GPL(iommu_dev_disable_feature);
2420
2421bool iommu_dev_feature_enabled(struct device *dev, enum iommu_dev_features feat)
2422{
2423 const struct iommu_ops *ops = dev->bus->iommu_ops;
2424
2425 if (ops && ops->dev_feat_enabled)
2426 return ops->dev_feat_enabled(dev, feat);
2427
2428 return false;
2429}
2430EXPORT_SYMBOL_GPL(iommu_dev_feature_enabled);
2431
2432/*
2433 * Aux-domain specific attach/detach.
2434 *
2435 * Only works if iommu_dev_feature_enabled(dev, IOMMU_DEV_FEAT_AUX) returns
2436 * true. Also, as long as domains are attached to a device through this
2437 * interface, any tries to call iommu_attach_device() should fail
2438 * (iommu_detach_device() can't fail, so we fail when trying to re-attach).
2439 * This should make us safe against a device being attached to a guest as a
2440 * whole while there are still pasid users on it (aux and sva).
2441 */
2442int iommu_aux_attach_device(struct iommu_domain *domain, struct device *dev)
2443{
2444 int ret = -ENODEV;
2445
2446 if (domain->ops->aux_attach_dev)
2447 ret = domain->ops->aux_attach_dev(domain, dev);
2448
2449 if (!ret)
2450 trace_attach_device_to_domain(dev);
2451
2452 return ret;
2453}
2454EXPORT_SYMBOL_GPL(iommu_aux_attach_device);
2455
2456void iommu_aux_detach_device(struct iommu_domain *domain, struct device *dev)
2457{
2458 if (domain->ops->aux_detach_dev) {
2459 domain->ops->aux_detach_dev(domain, dev);
2460 trace_detach_device_from_domain(dev);
2461 }
2462}
2463EXPORT_SYMBOL_GPL(iommu_aux_detach_device);
2464
2465int iommu_aux_get_pasid(struct iommu_domain *domain, struct device *dev)
2466{
2467 int ret = -ENODEV;
2468
2469 if (domain->ops->aux_get_pasid)
2470 ret = domain->ops->aux_get_pasid(domain, dev);
2471
2472 return ret;
2473}
2474EXPORT_SYMBOL_GPL(iommu_aux_get_pasid);
26b25a2b
JPB
2475
2476/**
2477 * iommu_sva_bind_device() - Bind a process address space to a device
2478 * @dev: the device
2479 * @mm: the mm to bind, caller must hold a reference to it
2480 *
2481 * Create a bond between device and address space, allowing the device to access
2482 * the mm using the returned PASID. If a bond already exists between @device and
2483 * @mm, it is returned and an additional reference is taken. Caller must call
2484 * iommu_sva_unbind_device() to release each reference.
2485 *
2486 * iommu_dev_enable_feature(dev, IOMMU_DEV_FEAT_SVA) must be called first, to
2487 * initialize the required SVA features.
2488 *
2489 * On error, returns an ERR_PTR value.
2490 */
2491struct iommu_sva *
2492iommu_sva_bind_device(struct device *dev, struct mm_struct *mm, void *drvdata)
2493{
2494 struct iommu_group *group;
2495 struct iommu_sva *handle = ERR_PTR(-EINVAL);
2496 const struct iommu_ops *ops = dev->bus->iommu_ops;
2497
2498 if (!ops || !ops->sva_bind)
2499 return ERR_PTR(-ENODEV);
2500
2501 group = iommu_group_get(dev);
2502 if (!group)
2503 return ERR_PTR(-ENODEV);
2504
2505 /* Ensure device count and domain don't change while we're binding */
2506 mutex_lock(&group->mutex);
2507
2508 /*
2509 * To keep things simple, SVA currently doesn't support IOMMU groups
2510 * with more than one device. Existing SVA-capable systems are not
2511 * affected by the problems that required IOMMU groups (lack of ACS
2512 * isolation, device ID aliasing and other hardware issues).
2513 */
2514 if (iommu_group_device_count(group) != 1)
2515 goto out_unlock;
2516
2517 handle = ops->sva_bind(dev, mm, drvdata);
2518
2519out_unlock:
2520 mutex_unlock(&group->mutex);
2521 iommu_group_put(group);
2522
2523 return handle;
2524}
2525EXPORT_SYMBOL_GPL(iommu_sva_bind_device);
2526
2527/**
2528 * iommu_sva_unbind_device() - Remove a bond created with iommu_sva_bind_device
2529 * @handle: the handle returned by iommu_sva_bind_device()
2530 *
2531 * Put reference to a bond between device and address space. The device should
2532 * not be issuing any more transaction for this PASID. All outstanding page
2533 * requests for this PASID must have been flushed to the IOMMU.
2534 *
2535 * Returns 0 on success, or an error value
2536 */
2537void iommu_sva_unbind_device(struct iommu_sva *handle)
2538{
2539 struct iommu_group *group;
2540 struct device *dev = handle->dev;
2541 const struct iommu_ops *ops = dev->bus->iommu_ops;
2542
2543 if (!ops || !ops->sva_unbind)
2544 return;
2545
2546 group = iommu_group_get(dev);
2547 if (!group)
2548 return;
2549
2550 mutex_lock(&group->mutex);
2551 ops->sva_unbind(handle);
2552 mutex_unlock(&group->mutex);
2553
2554 iommu_group_put(group);
2555}
2556EXPORT_SYMBOL_GPL(iommu_sva_unbind_device);
2557
2558int iommu_sva_set_ops(struct iommu_sva *handle,
2559 const struct iommu_sva_ops *sva_ops)
2560{
2561 if (handle->ops && handle->ops != sva_ops)
2562 return -EEXIST;
2563
2564 handle->ops = sva_ops;
2565 return 0;
2566}
2567EXPORT_SYMBOL_GPL(iommu_sva_set_ops);
2568
2569int iommu_sva_get_pasid(struct iommu_sva *handle)
2570{
2571 const struct iommu_ops *ops = handle->dev->bus->iommu_ops;
2572
2573 if (!ops || !ops->sva_get_pasid)
2574 return IOMMU_PASID_INVALID;
2575
2576 return ops->sva_get_pasid(handle);
2577}
2578EXPORT_SYMBOL_GPL(iommu_sva_get_pasid);