xref: /openbmc/linux/drivers/iommu/iommu.c (revision abfbd895)
1 /*
2  * Copyright (C) 2007-2008 Advanced Micro Devices, Inc.
3  * Author: Joerg Roedel <jroedel@suse.de>
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms of the GNU General Public License version 2 as published
7  * by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write to the Free Software
16  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
17  */
18 
19 #define pr_fmt(fmt)    "iommu: " fmt
20 
21 #include <linux/device.h>
22 #include <linux/kernel.h>
23 #include <linux/bug.h>
24 #include <linux/types.h>
25 #include <linux/module.h>
26 #include <linux/slab.h>
27 #include <linux/errno.h>
28 #include <linux/iommu.h>
29 #include <linux/idr.h>
30 #include <linux/notifier.h>
31 #include <linux/err.h>
32 #include <linux/pci.h>
33 #include <linux/bitops.h>
34 #include <trace/events/iommu.h>
35 
36 static struct kset *iommu_group_kset;
37 static struct ida iommu_group_ida;
38 static struct mutex iommu_group_mutex;
39 
40 struct iommu_callback_data {
41 	const struct iommu_ops *ops;
42 };
43 
44 struct 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;
54 	struct iommu_domain *default_domain;
55 	struct iommu_domain *domain;
56 };
57 
58 struct iommu_device {
59 	struct list_head list;
60 	struct device *dev;
61 	char *name;
62 };
63 
64 struct 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)		\
72 struct iommu_group_attribute iommu_group_attr_##_name =		\
73 	__ATTR(_name, _mode, _show, _store)
74 
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)
79 
80 static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
81 						 unsigned type);
82 static int __iommu_attach_device(struct iommu_domain *domain,
83 				 struct device *dev);
84 static int __iommu_attach_group(struct iommu_domain *domain,
85 				struct iommu_group *group);
86 static void __iommu_detach_group(struct iommu_domain *domain,
87 				 struct iommu_group *group);
88 
89 static ssize_t iommu_group_attr_show(struct kobject *kobj,
90 				     struct attribute *__attr, char *buf)
91 {
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;
95 
96 	if (attr->show)
97 		ret = attr->show(group, buf);
98 	return ret;
99 }
100 
101 static 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;
108 
109 	if (attr->store)
110 		ret = attr->store(group, buf, count);
111 	return ret;
112 }
113 
114 static const struct sysfs_ops iommu_group_sysfs_ops = {
115 	.show = iommu_group_attr_show,
116 	.store = iommu_group_attr_store,
117 };
118 
119 static 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);
123 }
124 
125 static 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 
131 static ssize_t iommu_group_show_name(struct iommu_group *group, char *buf)
132 {
133 	return sprintf(buf, "%s\n", group->name);
134 }
135 
136 static IOMMU_GROUP_ATTR(name, S_IRUGO, iommu_group_show_name, NULL);
137 
138 static void iommu_group_release(struct kobject *kobj)
139 {
140 	struct iommu_group *group = to_iommu_group(kobj);
141 
142 	pr_debug("Releasing group %d\n", group->id);
143 
144 	if (group->iommu_data_release)
145 		group->iommu_data_release(group->iommu_data);
146 
147 	mutex_lock(&iommu_group_mutex);
148 	ida_remove(&iommu_group_ida, group->id);
149 	mutex_unlock(&iommu_group_mutex);
150 
151 	if (group->default_domain)
152 		iommu_domain_free(group->default_domain);
153 
154 	kfree(group->name);
155 	kfree(group);
156 }
157 
158 static struct kobj_type iommu_group_ktype = {
159 	.sysfs_ops = &iommu_group_sysfs_ops,
160 	.release = iommu_group_release,
161 };
162 
163 /**
164  * iommu_group_alloc - Allocate a new group
165  * @name: Optional name to associate with group, visible in sysfs
166  *
167  * This function is called by an iommu driver to allocate a new iommu
168  * group.  The iommu group represents the minimum granularity of the iommu.
169  * Upon successful return, the caller holds a reference to the supplied
170  * group in order to hold the group until devices are added.  Use
171  * iommu_group_put() to release this extra reference count, allowing the
172  * group to be automatically reclaimed once it has no devices or external
173  * references.
174  */
175 struct iommu_group *iommu_group_alloc(void)
176 {
177 	struct iommu_group *group;
178 	int ret;
179 
180 	group = kzalloc(sizeof(*group), GFP_KERNEL);
181 	if (!group)
182 		return ERR_PTR(-ENOMEM);
183 
184 	group->kobj.kset = iommu_group_kset;
185 	mutex_init(&group->mutex);
186 	INIT_LIST_HEAD(&group->devices);
187 	BLOCKING_INIT_NOTIFIER_HEAD(&group->notifier);
188 
189 	mutex_lock(&iommu_group_mutex);
190 
191 again:
192 	if (unlikely(0 == ida_pre_get(&iommu_group_ida, GFP_KERNEL))) {
193 		kfree(group);
194 		mutex_unlock(&iommu_group_mutex);
195 		return ERR_PTR(-ENOMEM);
196 	}
197 
198 	if (-EAGAIN == ida_get_new(&iommu_group_ida, &group->id))
199 		goto again;
200 
201 	mutex_unlock(&iommu_group_mutex);
202 
203 	ret = kobject_init_and_add(&group->kobj, &iommu_group_ktype,
204 				   NULL, "%d", group->id);
205 	if (ret) {
206 		mutex_lock(&iommu_group_mutex);
207 		ida_remove(&iommu_group_ida, group->id);
208 		mutex_unlock(&iommu_group_mutex);
209 		kfree(group);
210 		return ERR_PTR(ret);
211 	}
212 
213 	group->devices_kobj = kobject_create_and_add("devices", &group->kobj);
214 	if (!group->devices_kobj) {
215 		kobject_put(&group->kobj); /* triggers .release & free */
216 		return ERR_PTR(-ENOMEM);
217 	}
218 
219 	/*
220 	 * The devices_kobj holds a reference on the group kobject, so
221 	 * as long as that exists so will the group.  We can therefore
222 	 * use the devices_kobj for reference counting.
223 	 */
224 	kobject_put(&group->kobj);
225 
226 	pr_debug("Allocated group %d\n", group->id);
227 
228 	return group;
229 }
230 EXPORT_SYMBOL_GPL(iommu_group_alloc);
231 
232 struct iommu_group *iommu_group_get_by_id(int id)
233 {
234 	struct kobject *group_kobj;
235 	struct iommu_group *group;
236 	const char *name;
237 
238 	if (!iommu_group_kset)
239 		return NULL;
240 
241 	name = kasprintf(GFP_KERNEL, "%d", id);
242 	if (!name)
243 		return NULL;
244 
245 	group_kobj = kset_find_obj(iommu_group_kset, name);
246 	kfree(name);
247 
248 	if (!group_kobj)
249 		return NULL;
250 
251 	group = container_of(group_kobj, struct iommu_group, kobj);
252 	BUG_ON(group->id != id);
253 
254 	kobject_get(group->devices_kobj);
255 	kobject_put(&group->kobj);
256 
257 	return group;
258 }
259 EXPORT_SYMBOL_GPL(iommu_group_get_by_id);
260 
261 /**
262  * iommu_group_get_iommudata - retrieve iommu_data registered for a group
263  * @group: the group
264  *
265  * iommu drivers can store data in the group for use when doing iommu
266  * operations.  This function provides a way to retrieve it.  Caller
267  * should hold a group reference.
268  */
269 void *iommu_group_get_iommudata(struct iommu_group *group)
270 {
271 	return group->iommu_data;
272 }
273 EXPORT_SYMBOL_GPL(iommu_group_get_iommudata);
274 
275 /**
276  * iommu_group_set_iommudata - set iommu_data for a group
277  * @group: the group
278  * @iommu_data: new data
279  * @release: release function for iommu_data
280  *
281  * iommu drivers can store data in the group for use when doing iommu
282  * operations.  This function provides a way to set the data after
283  * the group has been allocated.  Caller should hold a group reference.
284  */
285 void iommu_group_set_iommudata(struct iommu_group *group, void *iommu_data,
286 			       void (*release)(void *iommu_data))
287 {
288 	group->iommu_data = iommu_data;
289 	group->iommu_data_release = release;
290 }
291 EXPORT_SYMBOL_GPL(iommu_group_set_iommudata);
292 
293 /**
294  * iommu_group_set_name - set name for a group
295  * @group: the group
296  * @name: name
297  *
298  * Allow iommu driver to set a name for a group.  When set it will
299  * appear in a name attribute file under the group in sysfs.
300  */
301 int iommu_group_set_name(struct iommu_group *group, const char *name)
302 {
303 	int ret;
304 
305 	if (group->name) {
306 		iommu_group_remove_file(group, &iommu_group_attr_name);
307 		kfree(group->name);
308 		group->name = NULL;
309 		if (!name)
310 			return 0;
311 	}
312 
313 	group->name = kstrdup(name, GFP_KERNEL);
314 	if (!group->name)
315 		return -ENOMEM;
316 
317 	ret = iommu_group_create_file(group, &iommu_group_attr_name);
318 	if (ret) {
319 		kfree(group->name);
320 		group->name = NULL;
321 		return ret;
322 	}
323 
324 	return 0;
325 }
326 EXPORT_SYMBOL_GPL(iommu_group_set_name);
327 
328 static int iommu_group_create_direct_mappings(struct iommu_group *group,
329 					      struct device *dev)
330 {
331 	struct iommu_domain *domain = group->default_domain;
332 	struct iommu_dm_region *entry;
333 	struct list_head mappings;
334 	unsigned long pg_size;
335 	int ret = 0;
336 
337 	if (!domain || domain->type != IOMMU_DOMAIN_DMA)
338 		return 0;
339 
340 	BUG_ON(!domain->ops->pgsize_bitmap);
341 
342 	pg_size = 1UL << __ffs(domain->ops->pgsize_bitmap);
343 	INIT_LIST_HEAD(&mappings);
344 
345 	iommu_get_dm_regions(dev, &mappings);
346 
347 	/* We need to consider overlapping regions for different devices */
348 	list_for_each_entry(entry, &mappings, list) {
349 		dma_addr_t start, end, addr;
350 
351 		start = ALIGN(entry->start, pg_size);
352 		end   = ALIGN(entry->start + entry->length, pg_size);
353 
354 		for (addr = start; addr < end; addr += pg_size) {
355 			phys_addr_t phys_addr;
356 
357 			phys_addr = iommu_iova_to_phys(domain, addr);
358 			if (phys_addr)
359 				continue;
360 
361 			ret = iommu_map(domain, addr, addr, pg_size, entry->prot);
362 			if (ret)
363 				goto out;
364 		}
365 
366 	}
367 
368 out:
369 	iommu_put_dm_regions(dev, &mappings);
370 
371 	return ret;
372 }
373 
374 /**
375  * iommu_group_add_device - add a device to an iommu group
376  * @group: the group into which to add the device (reference should be held)
377  * @dev: the device
378  *
379  * This function is called by an iommu driver to add a device into a
380  * group.  Adding a device increments the group reference count.
381  */
382 int iommu_group_add_device(struct iommu_group *group, struct device *dev)
383 {
384 	int ret, i = 0;
385 	struct iommu_device *device;
386 
387 	device = kzalloc(sizeof(*device), GFP_KERNEL);
388 	if (!device)
389 		return -ENOMEM;
390 
391 	device->dev = dev;
392 
393 	ret = sysfs_create_link(&dev->kobj, &group->kobj, "iommu_group");
394 	if (ret) {
395 		kfree(device);
396 		return ret;
397 	}
398 
399 	device->name = kasprintf(GFP_KERNEL, "%s", kobject_name(&dev->kobj));
400 rename:
401 	if (!device->name) {
402 		sysfs_remove_link(&dev->kobj, "iommu_group");
403 		kfree(device);
404 		return -ENOMEM;
405 	}
406 
407 	ret = sysfs_create_link_nowarn(group->devices_kobj,
408 				       &dev->kobj, device->name);
409 	if (ret) {
410 		kfree(device->name);
411 		if (ret == -EEXIST && i >= 0) {
412 			/*
413 			 * Account for the slim chance of collision
414 			 * and append an instance to the name.
415 			 */
416 			device->name = kasprintf(GFP_KERNEL, "%s.%d",
417 						 kobject_name(&dev->kobj), i++);
418 			goto rename;
419 		}
420 
421 		sysfs_remove_link(&dev->kobj, "iommu_group");
422 		kfree(device);
423 		return ret;
424 	}
425 
426 	kobject_get(group->devices_kobj);
427 
428 	dev->iommu_group = group;
429 
430 	iommu_group_create_direct_mappings(group, dev);
431 
432 	mutex_lock(&group->mutex);
433 	list_add_tail(&device->list, &group->devices);
434 	if (group->domain)
435 		__iommu_attach_device(group->domain, dev);
436 	mutex_unlock(&group->mutex);
437 
438 	/* Notify any listeners about change to group. */
439 	blocking_notifier_call_chain(&group->notifier,
440 				     IOMMU_GROUP_NOTIFY_ADD_DEVICE, dev);
441 
442 	trace_add_device_to_group(group->id, dev);
443 
444 	pr_info("Adding device %s to group %d\n", dev_name(dev), group->id);
445 
446 	return 0;
447 }
448 EXPORT_SYMBOL_GPL(iommu_group_add_device);
449 
450 /**
451  * iommu_group_remove_device - remove a device from it's current group
452  * @dev: device to be removed
453  *
454  * This function is called by an iommu driver to remove the device from
455  * it's current group.  This decrements the iommu group reference count.
456  */
457 void iommu_group_remove_device(struct device *dev)
458 {
459 	struct iommu_group *group = dev->iommu_group;
460 	struct iommu_device *tmp_device, *device = NULL;
461 
462 	pr_info("Removing device %s from group %d\n", dev_name(dev), group->id);
463 
464 	/* Pre-notify listeners that a device is being removed. */
465 	blocking_notifier_call_chain(&group->notifier,
466 				     IOMMU_GROUP_NOTIFY_DEL_DEVICE, dev);
467 
468 	mutex_lock(&group->mutex);
469 	list_for_each_entry(tmp_device, &group->devices, list) {
470 		if (tmp_device->dev == dev) {
471 			device = tmp_device;
472 			list_del(&device->list);
473 			break;
474 		}
475 	}
476 	mutex_unlock(&group->mutex);
477 
478 	if (!device)
479 		return;
480 
481 	sysfs_remove_link(group->devices_kobj, device->name);
482 	sysfs_remove_link(&dev->kobj, "iommu_group");
483 
484 	trace_remove_device_from_group(group->id, dev);
485 
486 	kfree(device->name);
487 	kfree(device);
488 	dev->iommu_group = NULL;
489 	kobject_put(group->devices_kobj);
490 }
491 EXPORT_SYMBOL_GPL(iommu_group_remove_device);
492 
493 static int iommu_group_device_count(struct iommu_group *group)
494 {
495 	struct iommu_device *entry;
496 	int ret = 0;
497 
498 	list_for_each_entry(entry, &group->devices, list)
499 		ret++;
500 
501 	return ret;
502 }
503 
504 /**
505  * iommu_group_for_each_dev - iterate over each device in the group
506  * @group: the group
507  * @data: caller opaque data to be passed to callback function
508  * @fn: caller supplied callback function
509  *
510  * This function is called by group users to iterate over group devices.
511  * Callers should hold a reference count to the group during callback.
512  * The group->mutex is held across callbacks, which will block calls to
513  * iommu_group_add/remove_device.
514  */
515 static int __iommu_group_for_each_dev(struct iommu_group *group, void *data,
516 				      int (*fn)(struct device *, void *))
517 {
518 	struct iommu_device *device;
519 	int ret = 0;
520 
521 	list_for_each_entry(device, &group->devices, list) {
522 		ret = fn(device->dev, data);
523 		if (ret)
524 			break;
525 	}
526 	return ret;
527 }
528 
529 
530 int iommu_group_for_each_dev(struct iommu_group *group, void *data,
531 			     int (*fn)(struct device *, void *))
532 {
533 	int ret;
534 
535 	mutex_lock(&group->mutex);
536 	ret = __iommu_group_for_each_dev(group, data, fn);
537 	mutex_unlock(&group->mutex);
538 
539 	return ret;
540 }
541 EXPORT_SYMBOL_GPL(iommu_group_for_each_dev);
542 
543 /**
544  * iommu_group_get - Return the group for a device and increment reference
545  * @dev: get the group that this device belongs to
546  *
547  * This function is called by iommu drivers and users to get the group
548  * for the specified device.  If found, the group is returned and the group
549  * reference in incremented, else NULL.
550  */
551 struct iommu_group *iommu_group_get(struct device *dev)
552 {
553 	struct iommu_group *group = dev->iommu_group;
554 
555 	if (group)
556 		kobject_get(group->devices_kobj);
557 
558 	return group;
559 }
560 EXPORT_SYMBOL_GPL(iommu_group_get);
561 
562 /**
563  * iommu_group_put - Decrement group reference
564  * @group: the group to use
565  *
566  * This function is called by iommu drivers and users to release the
567  * iommu group.  Once the reference count is zero, the group is released.
568  */
569 void iommu_group_put(struct iommu_group *group)
570 {
571 	if (group)
572 		kobject_put(group->devices_kobj);
573 }
574 EXPORT_SYMBOL_GPL(iommu_group_put);
575 
576 /**
577  * iommu_group_register_notifier - Register a notifier for group changes
578  * @group: the group to watch
579  * @nb: notifier block to signal
580  *
581  * This function allows iommu group users to track changes in a group.
582  * See include/linux/iommu.h for actions sent via this notifier.  Caller
583  * should hold a reference to the group throughout notifier registration.
584  */
585 int iommu_group_register_notifier(struct iommu_group *group,
586 				  struct notifier_block *nb)
587 {
588 	return blocking_notifier_chain_register(&group->notifier, nb);
589 }
590 EXPORT_SYMBOL_GPL(iommu_group_register_notifier);
591 
592 /**
593  * iommu_group_unregister_notifier - Unregister a notifier
594  * @group: the group to watch
595  * @nb: notifier block to signal
596  *
597  * Unregister a previously registered group notifier block.
598  */
599 int iommu_group_unregister_notifier(struct iommu_group *group,
600 				    struct notifier_block *nb)
601 {
602 	return blocking_notifier_chain_unregister(&group->notifier, nb);
603 }
604 EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier);
605 
606 /**
607  * iommu_group_id - Return ID for a group
608  * @group: the group to ID
609  *
610  * Return the unique ID for the group matching the sysfs group number.
611  */
612 int iommu_group_id(struct iommu_group *group)
613 {
614 	return group->id;
615 }
616 EXPORT_SYMBOL_GPL(iommu_group_id);
617 
618 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
619 					       unsigned long *devfns);
620 
621 /*
622  * To consider a PCI device isolated, we require ACS to support Source
623  * Validation, Request Redirection, Completer Redirection, and Upstream
624  * Forwarding.  This effectively means that devices cannot spoof their
625  * requester ID, requests and completions cannot be redirected, and all
626  * transactions are forwarded upstream, even as it passes through a
627  * bridge where the target device is downstream.
628  */
629 #define REQ_ACS_FLAGS   (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
630 
631 /*
632  * For multifunction devices which are not isolated from each other, find
633  * all the other non-isolated functions and look for existing groups.  For
634  * each function, we also need to look for aliases to or from other devices
635  * that may already have a group.
636  */
637 static struct iommu_group *get_pci_function_alias_group(struct pci_dev *pdev,
638 							unsigned long *devfns)
639 {
640 	struct pci_dev *tmp = NULL;
641 	struct iommu_group *group;
642 
643 	if (!pdev->multifunction || pci_acs_enabled(pdev, REQ_ACS_FLAGS))
644 		return NULL;
645 
646 	for_each_pci_dev(tmp) {
647 		if (tmp == pdev || tmp->bus != pdev->bus ||
648 		    PCI_SLOT(tmp->devfn) != PCI_SLOT(pdev->devfn) ||
649 		    pci_acs_enabled(tmp, REQ_ACS_FLAGS))
650 			continue;
651 
652 		group = get_pci_alias_group(tmp, devfns);
653 		if (group) {
654 			pci_dev_put(tmp);
655 			return group;
656 		}
657 	}
658 
659 	return NULL;
660 }
661 
662 /*
663  * Look for aliases to or from the given device for exisiting groups.  The
664  * dma_alias_devfn only supports aliases on the same bus, therefore the search
665  * space is quite small (especially since we're really only looking at pcie
666  * device, and therefore only expect multiple slots on the root complex or
667  * downstream switch ports).  It's conceivable though that a pair of
668  * multifunction devices could have aliases between them that would cause a
669  * loop.  To prevent this, we use a bitmap to track where we've been.
670  */
671 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
672 					       unsigned long *devfns)
673 {
674 	struct pci_dev *tmp = NULL;
675 	struct iommu_group *group;
676 
677 	if (test_and_set_bit(pdev->devfn & 0xff, devfns))
678 		return NULL;
679 
680 	group = iommu_group_get(&pdev->dev);
681 	if (group)
682 		return group;
683 
684 	for_each_pci_dev(tmp) {
685 		if (tmp == pdev || tmp->bus != pdev->bus)
686 			continue;
687 
688 		/* We alias them or they alias us */
689 		if (((pdev->dev_flags & PCI_DEV_FLAGS_DMA_ALIAS_DEVFN) &&
690 		     pdev->dma_alias_devfn == tmp->devfn) ||
691 		    ((tmp->dev_flags & PCI_DEV_FLAGS_DMA_ALIAS_DEVFN) &&
692 		     tmp->dma_alias_devfn == pdev->devfn)) {
693 
694 			group = get_pci_alias_group(tmp, devfns);
695 			if (group) {
696 				pci_dev_put(tmp);
697 				return group;
698 			}
699 
700 			group = get_pci_function_alias_group(tmp, devfns);
701 			if (group) {
702 				pci_dev_put(tmp);
703 				return group;
704 			}
705 		}
706 	}
707 
708 	return NULL;
709 }
710 
711 struct group_for_pci_data {
712 	struct pci_dev *pdev;
713 	struct iommu_group *group;
714 };
715 
716 /*
717  * DMA alias iterator callback, return the last seen device.  Stop and return
718  * the IOMMU group if we find one along the way.
719  */
720 static int get_pci_alias_or_group(struct pci_dev *pdev, u16 alias, void *opaque)
721 {
722 	struct group_for_pci_data *data = opaque;
723 
724 	data->pdev = pdev;
725 	data->group = iommu_group_get(&pdev->dev);
726 
727 	return data->group != NULL;
728 }
729 
730 /*
731  * Generic device_group call-back function. It just allocates one
732  * iommu-group per device.
733  */
734 struct iommu_group *generic_device_group(struct device *dev)
735 {
736 	struct iommu_group *group;
737 
738 	group = iommu_group_alloc();
739 	if (IS_ERR(group))
740 		return NULL;
741 
742 	return group;
743 }
744 
745 /*
746  * Use standard PCI bus topology, isolation features, and DMA alias quirks
747  * to find or create an IOMMU group for a device.
748  */
749 struct iommu_group *pci_device_group(struct device *dev)
750 {
751 	struct pci_dev *pdev = to_pci_dev(dev);
752 	struct group_for_pci_data data;
753 	struct pci_bus *bus;
754 	struct iommu_group *group = NULL;
755 	u64 devfns[4] = { 0 };
756 
757 	if (WARN_ON(!dev_is_pci(dev)))
758 		return ERR_PTR(-EINVAL);
759 
760 	/*
761 	 * Find the upstream DMA alias for the device.  A device must not
762 	 * be aliased due to topology in order to have its own IOMMU group.
763 	 * If we find an alias along the way that already belongs to a
764 	 * group, use it.
765 	 */
766 	if (pci_for_each_dma_alias(pdev, get_pci_alias_or_group, &data))
767 		return data.group;
768 
769 	pdev = data.pdev;
770 
771 	/*
772 	 * Continue upstream from the point of minimum IOMMU granularity
773 	 * due to aliases to the point where devices are protected from
774 	 * peer-to-peer DMA by PCI ACS.  Again, if we find an existing
775 	 * group, use it.
776 	 */
777 	for (bus = pdev->bus; !pci_is_root_bus(bus); bus = bus->parent) {
778 		if (!bus->self)
779 			continue;
780 
781 		if (pci_acs_path_enabled(bus->self, NULL, REQ_ACS_FLAGS))
782 			break;
783 
784 		pdev = bus->self;
785 
786 		group = iommu_group_get(&pdev->dev);
787 		if (group)
788 			return group;
789 	}
790 
791 	/*
792 	 * Look for existing groups on device aliases.  If we alias another
793 	 * device or another device aliases us, use the same group.
794 	 */
795 	group = get_pci_alias_group(pdev, (unsigned long *)devfns);
796 	if (group)
797 		return group;
798 
799 	/*
800 	 * Look for existing groups on non-isolated functions on the same
801 	 * slot and aliases of those funcions, if any.  No need to clear
802 	 * the search bitmap, the tested devfns are still valid.
803 	 */
804 	group = get_pci_function_alias_group(pdev, (unsigned long *)devfns);
805 	if (group)
806 		return group;
807 
808 	/* No shared group found, allocate new */
809 	group = iommu_group_alloc();
810 	if (IS_ERR(group))
811 		return NULL;
812 
813 	return group;
814 }
815 
816 /**
817  * iommu_group_get_for_dev - Find or create the IOMMU group for a device
818  * @dev: target device
819  *
820  * This function is intended to be called by IOMMU drivers and extended to
821  * support common, bus-defined algorithms when determining or creating the
822  * IOMMU group for a device.  On success, the caller will hold a reference
823  * to the returned IOMMU group, which will already include the provided
824  * device.  The reference should be released with iommu_group_put().
825  */
826 struct iommu_group *iommu_group_get_for_dev(struct device *dev)
827 {
828 	const struct iommu_ops *ops = dev->bus->iommu_ops;
829 	struct iommu_group *group;
830 	int ret;
831 
832 	group = iommu_group_get(dev);
833 	if (group)
834 		return group;
835 
836 	group = ERR_PTR(-EINVAL);
837 
838 	if (ops && ops->device_group)
839 		group = ops->device_group(dev);
840 
841 	if (IS_ERR(group))
842 		return group;
843 
844 	/*
845 	 * Try to allocate a default domain - needs support from the
846 	 * IOMMU driver.
847 	 */
848 	if (!group->default_domain) {
849 		group->default_domain = __iommu_domain_alloc(dev->bus,
850 							     IOMMU_DOMAIN_DMA);
851 		group->domain = group->default_domain;
852 	}
853 
854 	ret = iommu_group_add_device(group, dev);
855 	if (ret) {
856 		iommu_group_put(group);
857 		return ERR_PTR(ret);
858 	}
859 
860 	return group;
861 }
862 
863 struct iommu_domain *iommu_group_default_domain(struct iommu_group *group)
864 {
865 	return group->default_domain;
866 }
867 
868 static int add_iommu_group(struct device *dev, void *data)
869 {
870 	struct iommu_callback_data *cb = data;
871 	const struct iommu_ops *ops = cb->ops;
872 	int ret;
873 
874 	if (!ops->add_device)
875 		return 0;
876 
877 	WARN_ON(dev->iommu_group);
878 
879 	ret = ops->add_device(dev);
880 
881 	/*
882 	 * We ignore -ENODEV errors for now, as they just mean that the
883 	 * device is not translated by an IOMMU. We still care about
884 	 * other errors and fail to initialize when they happen.
885 	 */
886 	if (ret == -ENODEV)
887 		ret = 0;
888 
889 	return ret;
890 }
891 
892 static int remove_iommu_group(struct device *dev, void *data)
893 {
894 	struct iommu_callback_data *cb = data;
895 	const struct iommu_ops *ops = cb->ops;
896 
897 	if (ops->remove_device && dev->iommu_group)
898 		ops->remove_device(dev);
899 
900 	return 0;
901 }
902 
903 static int iommu_bus_notifier(struct notifier_block *nb,
904 			      unsigned long action, void *data)
905 {
906 	struct device *dev = data;
907 	const struct iommu_ops *ops = dev->bus->iommu_ops;
908 	struct iommu_group *group;
909 	unsigned long group_action = 0;
910 
911 	/*
912 	 * ADD/DEL call into iommu driver ops if provided, which may
913 	 * result in ADD/DEL notifiers to group->notifier
914 	 */
915 	if (action == BUS_NOTIFY_ADD_DEVICE) {
916 		if (ops->add_device)
917 			return ops->add_device(dev);
918 	} else if (action == BUS_NOTIFY_REMOVED_DEVICE) {
919 		if (ops->remove_device && dev->iommu_group) {
920 			ops->remove_device(dev);
921 			return 0;
922 		}
923 	}
924 
925 	/*
926 	 * Remaining BUS_NOTIFYs get filtered and republished to the
927 	 * group, if anyone is listening
928 	 */
929 	group = iommu_group_get(dev);
930 	if (!group)
931 		return 0;
932 
933 	switch (action) {
934 	case BUS_NOTIFY_BIND_DRIVER:
935 		group_action = IOMMU_GROUP_NOTIFY_BIND_DRIVER;
936 		break;
937 	case BUS_NOTIFY_BOUND_DRIVER:
938 		group_action = IOMMU_GROUP_NOTIFY_BOUND_DRIVER;
939 		break;
940 	case BUS_NOTIFY_UNBIND_DRIVER:
941 		group_action = IOMMU_GROUP_NOTIFY_UNBIND_DRIVER;
942 		break;
943 	case BUS_NOTIFY_UNBOUND_DRIVER:
944 		group_action = IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER;
945 		break;
946 	}
947 
948 	if (group_action)
949 		blocking_notifier_call_chain(&group->notifier,
950 					     group_action, dev);
951 
952 	iommu_group_put(group);
953 	return 0;
954 }
955 
956 static int iommu_bus_init(struct bus_type *bus, const struct iommu_ops *ops)
957 {
958 	int err;
959 	struct notifier_block *nb;
960 	struct iommu_callback_data cb = {
961 		.ops = ops,
962 	};
963 
964 	nb = kzalloc(sizeof(struct notifier_block), GFP_KERNEL);
965 	if (!nb)
966 		return -ENOMEM;
967 
968 	nb->notifier_call = iommu_bus_notifier;
969 
970 	err = bus_register_notifier(bus, nb);
971 	if (err)
972 		goto out_free;
973 
974 	err = bus_for_each_dev(bus, NULL, &cb, add_iommu_group);
975 	if (err)
976 		goto out_err;
977 
978 
979 	return 0;
980 
981 out_err:
982 	/* Clean up */
983 	bus_for_each_dev(bus, NULL, &cb, remove_iommu_group);
984 	bus_unregister_notifier(bus, nb);
985 
986 out_free:
987 	kfree(nb);
988 
989 	return err;
990 }
991 
992 /**
993  * bus_set_iommu - set iommu-callbacks for the bus
994  * @bus: bus.
995  * @ops: the callbacks provided by the iommu-driver
996  *
997  * This function is called by an iommu driver to set the iommu methods
998  * used for a particular bus. Drivers for devices on that bus can use
999  * the iommu-api after these ops are registered.
1000  * This special function is needed because IOMMUs are usually devices on
1001  * the bus itself, so the iommu drivers are not initialized when the bus
1002  * is set up. With this function the iommu-driver can set the iommu-ops
1003  * afterwards.
1004  */
1005 int bus_set_iommu(struct bus_type *bus, const struct iommu_ops *ops)
1006 {
1007 	int err;
1008 
1009 	if (bus->iommu_ops != NULL)
1010 		return -EBUSY;
1011 
1012 	bus->iommu_ops = ops;
1013 
1014 	/* Do IOMMU specific setup for this bus-type */
1015 	err = iommu_bus_init(bus, ops);
1016 	if (err)
1017 		bus->iommu_ops = NULL;
1018 
1019 	return err;
1020 }
1021 EXPORT_SYMBOL_GPL(bus_set_iommu);
1022 
1023 bool iommu_present(struct bus_type *bus)
1024 {
1025 	return bus->iommu_ops != NULL;
1026 }
1027 EXPORT_SYMBOL_GPL(iommu_present);
1028 
1029 bool iommu_capable(struct bus_type *bus, enum iommu_cap cap)
1030 {
1031 	if (!bus->iommu_ops || !bus->iommu_ops->capable)
1032 		return false;
1033 
1034 	return bus->iommu_ops->capable(cap);
1035 }
1036 EXPORT_SYMBOL_GPL(iommu_capable);
1037 
1038 /**
1039  * iommu_set_fault_handler() - set a fault handler for an iommu domain
1040  * @domain: iommu domain
1041  * @handler: fault handler
1042  * @token: user data, will be passed back to the fault handler
1043  *
1044  * This function should be used by IOMMU users which want to be notified
1045  * whenever an IOMMU fault happens.
1046  *
1047  * The fault handler itself should return 0 on success, and an appropriate
1048  * error code otherwise.
1049  */
1050 void iommu_set_fault_handler(struct iommu_domain *domain,
1051 					iommu_fault_handler_t handler,
1052 					void *token)
1053 {
1054 	BUG_ON(!domain);
1055 
1056 	domain->handler = handler;
1057 	domain->handler_token = token;
1058 }
1059 EXPORT_SYMBOL_GPL(iommu_set_fault_handler);
1060 
1061 static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
1062 						 unsigned type)
1063 {
1064 	struct iommu_domain *domain;
1065 
1066 	if (bus == NULL || bus->iommu_ops == NULL)
1067 		return NULL;
1068 
1069 	domain = bus->iommu_ops->domain_alloc(type);
1070 	if (!domain)
1071 		return NULL;
1072 
1073 	domain->ops  = bus->iommu_ops;
1074 	domain->type = type;
1075 
1076 	return domain;
1077 }
1078 
1079 struct iommu_domain *iommu_domain_alloc(struct bus_type *bus)
1080 {
1081 	return __iommu_domain_alloc(bus, IOMMU_DOMAIN_UNMANAGED);
1082 }
1083 EXPORT_SYMBOL_GPL(iommu_domain_alloc);
1084 
1085 void iommu_domain_free(struct iommu_domain *domain)
1086 {
1087 	domain->ops->domain_free(domain);
1088 }
1089 EXPORT_SYMBOL_GPL(iommu_domain_free);
1090 
1091 static int __iommu_attach_device(struct iommu_domain *domain,
1092 				 struct device *dev)
1093 {
1094 	int ret;
1095 	if (unlikely(domain->ops->attach_dev == NULL))
1096 		return -ENODEV;
1097 
1098 	ret = domain->ops->attach_dev(domain, dev);
1099 	if (!ret)
1100 		trace_attach_device_to_domain(dev);
1101 	return ret;
1102 }
1103 
1104 int iommu_attach_device(struct iommu_domain *domain, struct device *dev)
1105 {
1106 	struct iommu_group *group;
1107 	int ret;
1108 
1109 	group = iommu_group_get(dev);
1110 	/* FIXME: Remove this when groups a mandatory for iommu drivers */
1111 	if (group == NULL)
1112 		return __iommu_attach_device(domain, dev);
1113 
1114 	/*
1115 	 * We have a group - lock it to make sure the device-count doesn't
1116 	 * change while we are attaching
1117 	 */
1118 	mutex_lock(&group->mutex);
1119 	ret = -EINVAL;
1120 	if (iommu_group_device_count(group) != 1)
1121 		goto out_unlock;
1122 
1123 	ret = __iommu_attach_group(domain, group);
1124 
1125 out_unlock:
1126 	mutex_unlock(&group->mutex);
1127 	iommu_group_put(group);
1128 
1129 	return ret;
1130 }
1131 EXPORT_SYMBOL_GPL(iommu_attach_device);
1132 
1133 static void __iommu_detach_device(struct iommu_domain *domain,
1134 				  struct device *dev)
1135 {
1136 	if (unlikely(domain->ops->detach_dev == NULL))
1137 		return;
1138 
1139 	domain->ops->detach_dev(domain, dev);
1140 	trace_detach_device_from_domain(dev);
1141 }
1142 
1143 void iommu_detach_device(struct iommu_domain *domain, struct device *dev)
1144 {
1145 	struct iommu_group *group;
1146 
1147 	group = iommu_group_get(dev);
1148 	/* FIXME: Remove this when groups a mandatory for iommu drivers */
1149 	if (group == NULL)
1150 		return __iommu_detach_device(domain, dev);
1151 
1152 	mutex_lock(&group->mutex);
1153 	if (iommu_group_device_count(group) != 1) {
1154 		WARN_ON(1);
1155 		goto out_unlock;
1156 	}
1157 
1158 	__iommu_detach_group(domain, group);
1159 
1160 out_unlock:
1161 	mutex_unlock(&group->mutex);
1162 	iommu_group_put(group);
1163 }
1164 EXPORT_SYMBOL_GPL(iommu_detach_device);
1165 
1166 struct iommu_domain *iommu_get_domain_for_dev(struct device *dev)
1167 {
1168 	struct iommu_domain *domain;
1169 	struct iommu_group *group;
1170 
1171 	group = iommu_group_get(dev);
1172 	/* FIXME: Remove this when groups a mandatory for iommu drivers */
1173 	if (group == NULL)
1174 		return NULL;
1175 
1176 	domain = group->domain;
1177 
1178 	iommu_group_put(group);
1179 
1180 	return domain;
1181 }
1182 EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev);
1183 
1184 /*
1185  * IOMMU groups are really the natrual working unit of the IOMMU, but
1186  * the IOMMU API works on domains and devices.  Bridge that gap by
1187  * iterating over the devices in a group.  Ideally we'd have a single
1188  * device which represents the requestor ID of the group, but we also
1189  * allow IOMMU drivers to create policy defined minimum sets, where
1190  * the physical hardware may be able to distiguish members, but we
1191  * wish to group them at a higher level (ex. untrusted multi-function
1192  * PCI devices).  Thus we attach each device.
1193  */
1194 static int iommu_group_do_attach_device(struct device *dev, void *data)
1195 {
1196 	struct iommu_domain *domain = data;
1197 
1198 	return __iommu_attach_device(domain, dev);
1199 }
1200 
1201 static int __iommu_attach_group(struct iommu_domain *domain,
1202 				struct iommu_group *group)
1203 {
1204 	int ret;
1205 
1206 	if (group->default_domain && group->domain != group->default_domain)
1207 		return -EBUSY;
1208 
1209 	ret = __iommu_group_for_each_dev(group, domain,
1210 					 iommu_group_do_attach_device);
1211 	if (ret == 0)
1212 		group->domain = domain;
1213 
1214 	return ret;
1215 }
1216 
1217 int iommu_attach_group(struct iommu_domain *domain, struct iommu_group *group)
1218 {
1219 	int ret;
1220 
1221 	mutex_lock(&group->mutex);
1222 	ret = __iommu_attach_group(domain, group);
1223 	mutex_unlock(&group->mutex);
1224 
1225 	return ret;
1226 }
1227 EXPORT_SYMBOL_GPL(iommu_attach_group);
1228 
1229 static int iommu_group_do_detach_device(struct device *dev, void *data)
1230 {
1231 	struct iommu_domain *domain = data;
1232 
1233 	__iommu_detach_device(domain, dev);
1234 
1235 	return 0;
1236 }
1237 
1238 static void __iommu_detach_group(struct iommu_domain *domain,
1239 				 struct iommu_group *group)
1240 {
1241 	int ret;
1242 
1243 	if (!group->default_domain) {
1244 		__iommu_group_for_each_dev(group, domain,
1245 					   iommu_group_do_detach_device);
1246 		group->domain = NULL;
1247 		return;
1248 	}
1249 
1250 	if (group->domain == group->default_domain)
1251 		return;
1252 
1253 	/* Detach by re-attaching to the default domain */
1254 	ret = __iommu_group_for_each_dev(group, group->default_domain,
1255 					 iommu_group_do_attach_device);
1256 	if (ret != 0)
1257 		WARN_ON(1);
1258 	else
1259 		group->domain = group->default_domain;
1260 }
1261 
1262 void iommu_detach_group(struct iommu_domain *domain, struct iommu_group *group)
1263 {
1264 	mutex_lock(&group->mutex);
1265 	__iommu_detach_group(domain, group);
1266 	mutex_unlock(&group->mutex);
1267 }
1268 EXPORT_SYMBOL_GPL(iommu_detach_group);
1269 
1270 phys_addr_t iommu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova)
1271 {
1272 	if (unlikely(domain->ops->iova_to_phys == NULL))
1273 		return 0;
1274 
1275 	return domain->ops->iova_to_phys(domain, iova);
1276 }
1277 EXPORT_SYMBOL_GPL(iommu_iova_to_phys);
1278 
1279 static size_t iommu_pgsize(struct iommu_domain *domain,
1280 			   unsigned long addr_merge, size_t size)
1281 {
1282 	unsigned int pgsize_idx;
1283 	size_t pgsize;
1284 
1285 	/* Max page size that still fits into 'size' */
1286 	pgsize_idx = __fls(size);
1287 
1288 	/* need to consider alignment requirements ? */
1289 	if (likely(addr_merge)) {
1290 		/* Max page size allowed by address */
1291 		unsigned int align_pgsize_idx = __ffs(addr_merge);
1292 		pgsize_idx = min(pgsize_idx, align_pgsize_idx);
1293 	}
1294 
1295 	/* build a mask of acceptable page sizes */
1296 	pgsize = (1UL << (pgsize_idx + 1)) - 1;
1297 
1298 	/* throw away page sizes not supported by the hardware */
1299 	pgsize &= domain->ops->pgsize_bitmap;
1300 
1301 	/* make sure we're still sane */
1302 	BUG_ON(!pgsize);
1303 
1304 	/* pick the biggest page */
1305 	pgsize_idx = __fls(pgsize);
1306 	pgsize = 1UL << pgsize_idx;
1307 
1308 	return pgsize;
1309 }
1310 
1311 int iommu_map(struct iommu_domain *domain, unsigned long iova,
1312 	      phys_addr_t paddr, size_t size, int prot)
1313 {
1314 	unsigned long orig_iova = iova;
1315 	unsigned int min_pagesz;
1316 	size_t orig_size = size;
1317 	int ret = 0;
1318 
1319 	if (unlikely(domain->ops->map == NULL ||
1320 		     domain->ops->pgsize_bitmap == 0UL))
1321 		return -ENODEV;
1322 
1323 	if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1324 		return -EINVAL;
1325 
1326 	/* find out the minimum page size supported */
1327 	min_pagesz = 1 << __ffs(domain->ops->pgsize_bitmap);
1328 
1329 	/*
1330 	 * both the virtual address and the physical one, as well as
1331 	 * the size of the mapping, must be aligned (at least) to the
1332 	 * size of the smallest page supported by the hardware
1333 	 */
1334 	if (!IS_ALIGNED(iova | paddr | size, min_pagesz)) {
1335 		pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
1336 		       iova, &paddr, size, min_pagesz);
1337 		return -EINVAL;
1338 	}
1339 
1340 	pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova, &paddr, size);
1341 
1342 	while (size) {
1343 		size_t pgsize = iommu_pgsize(domain, iova | paddr, size);
1344 
1345 		pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx\n",
1346 			 iova, &paddr, pgsize);
1347 
1348 		ret = domain->ops->map(domain, iova, paddr, pgsize, prot);
1349 		if (ret)
1350 			break;
1351 
1352 		iova += pgsize;
1353 		paddr += pgsize;
1354 		size -= pgsize;
1355 	}
1356 
1357 	/* unroll mapping in case something went wrong */
1358 	if (ret)
1359 		iommu_unmap(domain, orig_iova, orig_size - size);
1360 	else
1361 		trace_map(orig_iova, paddr, orig_size);
1362 
1363 	return ret;
1364 }
1365 EXPORT_SYMBOL_GPL(iommu_map);
1366 
1367 size_t iommu_unmap(struct iommu_domain *domain, unsigned long iova, size_t size)
1368 {
1369 	size_t unmapped_page, unmapped = 0;
1370 	unsigned int min_pagesz;
1371 	unsigned long orig_iova = iova;
1372 
1373 	if (unlikely(domain->ops->unmap == NULL ||
1374 		     domain->ops->pgsize_bitmap == 0UL))
1375 		return -ENODEV;
1376 
1377 	if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1378 		return -EINVAL;
1379 
1380 	/* find out the minimum page size supported */
1381 	min_pagesz = 1 << __ffs(domain->ops->pgsize_bitmap);
1382 
1383 	/*
1384 	 * The virtual address, as well as the size of the mapping, must be
1385 	 * aligned (at least) to the size of the smallest page supported
1386 	 * by the hardware
1387 	 */
1388 	if (!IS_ALIGNED(iova | size, min_pagesz)) {
1389 		pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
1390 		       iova, size, min_pagesz);
1391 		return -EINVAL;
1392 	}
1393 
1394 	pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova, size);
1395 
1396 	/*
1397 	 * Keep iterating until we either unmap 'size' bytes (or more)
1398 	 * or we hit an area that isn't mapped.
1399 	 */
1400 	while (unmapped < size) {
1401 		size_t pgsize = iommu_pgsize(domain, iova, size - unmapped);
1402 
1403 		unmapped_page = domain->ops->unmap(domain, iova, pgsize);
1404 		if (!unmapped_page)
1405 			break;
1406 
1407 		pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
1408 			 iova, unmapped_page);
1409 
1410 		iova += unmapped_page;
1411 		unmapped += unmapped_page;
1412 	}
1413 
1414 	trace_unmap(orig_iova, size, unmapped);
1415 	return unmapped;
1416 }
1417 EXPORT_SYMBOL_GPL(iommu_unmap);
1418 
1419 size_t default_iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
1420 			 struct scatterlist *sg, unsigned int nents, int prot)
1421 {
1422 	struct scatterlist *s;
1423 	size_t mapped = 0;
1424 	unsigned int i, min_pagesz;
1425 	int ret;
1426 
1427 	if (unlikely(domain->ops->pgsize_bitmap == 0UL))
1428 		return 0;
1429 
1430 	min_pagesz = 1 << __ffs(domain->ops->pgsize_bitmap);
1431 
1432 	for_each_sg(sg, s, nents, i) {
1433 		phys_addr_t phys = page_to_phys(sg_page(s)) + s->offset;
1434 
1435 		/*
1436 		 * We are mapping on IOMMU page boundaries, so offset within
1437 		 * the page must be 0. However, the IOMMU may support pages
1438 		 * smaller than PAGE_SIZE, so s->offset may still represent
1439 		 * an offset of that boundary within the CPU page.
1440 		 */
1441 		if (!IS_ALIGNED(s->offset, min_pagesz))
1442 			goto out_err;
1443 
1444 		ret = iommu_map(domain, iova + mapped, phys, s->length, prot);
1445 		if (ret)
1446 			goto out_err;
1447 
1448 		mapped += s->length;
1449 	}
1450 
1451 	return mapped;
1452 
1453 out_err:
1454 	/* undo mappings already done */
1455 	iommu_unmap(domain, iova, mapped);
1456 
1457 	return 0;
1458 
1459 }
1460 EXPORT_SYMBOL_GPL(default_iommu_map_sg);
1461 
1462 int iommu_domain_window_enable(struct iommu_domain *domain, u32 wnd_nr,
1463 			       phys_addr_t paddr, u64 size, int prot)
1464 {
1465 	if (unlikely(domain->ops->domain_window_enable == NULL))
1466 		return -ENODEV;
1467 
1468 	return domain->ops->domain_window_enable(domain, wnd_nr, paddr, size,
1469 						 prot);
1470 }
1471 EXPORT_SYMBOL_GPL(iommu_domain_window_enable);
1472 
1473 void iommu_domain_window_disable(struct iommu_domain *domain, u32 wnd_nr)
1474 {
1475 	if (unlikely(domain->ops->domain_window_disable == NULL))
1476 		return;
1477 
1478 	return domain->ops->domain_window_disable(domain, wnd_nr);
1479 }
1480 EXPORT_SYMBOL_GPL(iommu_domain_window_disable);
1481 
1482 static int __init iommu_init(void)
1483 {
1484 	iommu_group_kset = kset_create_and_add("iommu_groups",
1485 					       NULL, kernel_kobj);
1486 	ida_init(&iommu_group_ida);
1487 	mutex_init(&iommu_group_mutex);
1488 
1489 	BUG_ON(!iommu_group_kset);
1490 
1491 	return 0;
1492 }
1493 core_initcall(iommu_init);
1494 
1495 int iommu_domain_get_attr(struct iommu_domain *domain,
1496 			  enum iommu_attr attr, void *data)
1497 {
1498 	struct iommu_domain_geometry *geometry;
1499 	bool *paging;
1500 	int ret = 0;
1501 	u32 *count;
1502 
1503 	switch (attr) {
1504 	case DOMAIN_ATTR_GEOMETRY:
1505 		geometry  = data;
1506 		*geometry = domain->geometry;
1507 
1508 		break;
1509 	case DOMAIN_ATTR_PAGING:
1510 		paging  = data;
1511 		*paging = (domain->ops->pgsize_bitmap != 0UL);
1512 		break;
1513 	case DOMAIN_ATTR_WINDOWS:
1514 		count = data;
1515 
1516 		if (domain->ops->domain_get_windows != NULL)
1517 			*count = domain->ops->domain_get_windows(domain);
1518 		else
1519 			ret = -ENODEV;
1520 
1521 		break;
1522 	default:
1523 		if (!domain->ops->domain_get_attr)
1524 			return -EINVAL;
1525 
1526 		ret = domain->ops->domain_get_attr(domain, attr, data);
1527 	}
1528 
1529 	return ret;
1530 }
1531 EXPORT_SYMBOL_GPL(iommu_domain_get_attr);
1532 
1533 int iommu_domain_set_attr(struct iommu_domain *domain,
1534 			  enum iommu_attr attr, void *data)
1535 {
1536 	int ret = 0;
1537 	u32 *count;
1538 
1539 	switch (attr) {
1540 	case DOMAIN_ATTR_WINDOWS:
1541 		count = data;
1542 
1543 		if (domain->ops->domain_set_windows != NULL)
1544 			ret = domain->ops->domain_set_windows(domain, *count);
1545 		else
1546 			ret = -ENODEV;
1547 
1548 		break;
1549 	default:
1550 		if (domain->ops->domain_set_attr == NULL)
1551 			return -EINVAL;
1552 
1553 		ret = domain->ops->domain_set_attr(domain, attr, data);
1554 	}
1555 
1556 	return ret;
1557 }
1558 EXPORT_SYMBOL_GPL(iommu_domain_set_attr);
1559 
1560 void iommu_get_dm_regions(struct device *dev, struct list_head *list)
1561 {
1562 	const struct iommu_ops *ops = dev->bus->iommu_ops;
1563 
1564 	if (ops && ops->get_dm_regions)
1565 		ops->get_dm_regions(dev, list);
1566 }
1567 
1568 void iommu_put_dm_regions(struct device *dev, struct list_head *list)
1569 {
1570 	const struct iommu_ops *ops = dev->bus->iommu_ops;
1571 
1572 	if (ops && ops->put_dm_regions)
1573 		ops->put_dm_regions(dev, list);
1574 }
1575 
1576 /* Request that a device is direct mapped by the IOMMU */
1577 int iommu_request_dm_for_dev(struct device *dev)
1578 {
1579 	struct iommu_domain *dm_domain;
1580 	struct iommu_group *group;
1581 	int ret;
1582 
1583 	/* Device must already be in a group before calling this function */
1584 	group = iommu_group_get_for_dev(dev);
1585 	if (IS_ERR(group))
1586 		return PTR_ERR(group);
1587 
1588 	mutex_lock(&group->mutex);
1589 
1590 	/* Check if the default domain is already direct mapped */
1591 	ret = 0;
1592 	if (group->default_domain &&
1593 	    group->default_domain->type == IOMMU_DOMAIN_IDENTITY)
1594 		goto out;
1595 
1596 	/* Don't change mappings of existing devices */
1597 	ret = -EBUSY;
1598 	if (iommu_group_device_count(group) != 1)
1599 		goto out;
1600 
1601 	/* Allocate a direct mapped domain */
1602 	ret = -ENOMEM;
1603 	dm_domain = __iommu_domain_alloc(dev->bus, IOMMU_DOMAIN_IDENTITY);
1604 	if (!dm_domain)
1605 		goto out;
1606 
1607 	/* Attach the device to the domain */
1608 	ret = __iommu_attach_group(dm_domain, group);
1609 	if (ret) {
1610 		iommu_domain_free(dm_domain);
1611 		goto out;
1612 	}
1613 
1614 	/* Make the direct mapped domain the default for this group */
1615 	if (group->default_domain)
1616 		iommu_domain_free(group->default_domain);
1617 	group->default_domain = dm_domain;
1618 
1619 	pr_info("Using direct mapping for device %s\n", dev_name(dev));
1620 
1621 	ret = 0;
1622 out:
1623 	mutex_unlock(&group->mutex);
1624 	iommu_group_put(group);
1625 
1626 	return ret;
1627 }
1628