xref: /openbmc/linux/drivers/iommu/iommu.c (revision 8730046c)
1 /*
2  * Copyright (C) 2007-2008 Advanced Micro Devices, Inc.
3  * Author: Joerg Roedel <jroedel@suse.de>
4  *
5  * This program is free software; you can redistribute it and/or modify it
6  * under the terms of the GNU General Public License version 2 as published
7  * by the Free Software Foundation.
8  *
9  * This program is distributed in the hope that it will be useful,
10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
12  * GNU General Public License for more details.
13  *
14  * You should have received a copy of the GNU General Public License
15  * along with this program; if not, write to the Free Software
16  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
17  */
18 
19 #define pr_fmt(fmt)    "iommu: " fmt
20 
21 #include <linux/device.h>
22 #include <linux/kernel.h>
23 #include <linux/bug.h>
24 #include <linux/types.h>
25 #include <linux/module.h>
26 #include <linux/slab.h>
27 #include <linux/errno.h>
28 #include <linux/iommu.h>
29 #include <linux/idr.h>
30 #include <linux/notifier.h>
31 #include <linux/err.h>
32 #include <linux/pci.h>
33 #include <linux/bitops.h>
34 #include <linux/property.h>
35 #include <trace/events/iommu.h>
36 
37 static struct kset *iommu_group_kset;
38 static DEFINE_IDA(iommu_group_ida);
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 	ida_simple_remove(&iommu_group_ida, group->id);
148 
149 	if (group->default_domain)
150 		iommu_domain_free(group->default_domain);
151 
152 	kfree(group->name);
153 	kfree(group);
154 }
155 
156 static struct kobj_type iommu_group_ktype = {
157 	.sysfs_ops = &iommu_group_sysfs_ops,
158 	.release = iommu_group_release,
159 };
160 
161 /**
162  * iommu_group_alloc - Allocate a new group
163  * @name: Optional name to associate with group, visible in sysfs
164  *
165  * This function is called by an iommu driver to allocate a new iommu
166  * group.  The iommu group represents the minimum granularity of the iommu.
167  * Upon successful return, the caller holds a reference to the supplied
168  * group in order to hold the group until devices are added.  Use
169  * iommu_group_put() to release this extra reference count, allowing the
170  * group to be automatically reclaimed once it has no devices or external
171  * references.
172  */
173 struct iommu_group *iommu_group_alloc(void)
174 {
175 	struct iommu_group *group;
176 	int ret;
177 
178 	group = kzalloc(sizeof(*group), GFP_KERNEL);
179 	if (!group)
180 		return ERR_PTR(-ENOMEM);
181 
182 	group->kobj.kset = iommu_group_kset;
183 	mutex_init(&group->mutex);
184 	INIT_LIST_HEAD(&group->devices);
185 	BLOCKING_INIT_NOTIFIER_HEAD(&group->notifier);
186 
187 	ret = ida_simple_get(&iommu_group_ida, 0, 0, GFP_KERNEL);
188 	if (ret < 0) {
189 		kfree(group);
190 		return ERR_PTR(ret);
191 	}
192 	group->id = ret;
193 
194 	ret = kobject_init_and_add(&group->kobj, &iommu_group_ktype,
195 				   NULL, "%d", group->id);
196 	if (ret) {
197 		ida_simple_remove(&iommu_group_ida, group->id);
198 		kfree(group);
199 		return ERR_PTR(ret);
200 	}
201 
202 	group->devices_kobj = kobject_create_and_add("devices", &group->kobj);
203 	if (!group->devices_kobj) {
204 		kobject_put(&group->kobj); /* triggers .release & free */
205 		return ERR_PTR(-ENOMEM);
206 	}
207 
208 	/*
209 	 * The devices_kobj holds a reference on the group kobject, so
210 	 * as long as that exists so will the group.  We can therefore
211 	 * use the devices_kobj for reference counting.
212 	 */
213 	kobject_put(&group->kobj);
214 
215 	pr_debug("Allocated group %d\n", group->id);
216 
217 	return group;
218 }
219 EXPORT_SYMBOL_GPL(iommu_group_alloc);
220 
221 struct iommu_group *iommu_group_get_by_id(int id)
222 {
223 	struct kobject *group_kobj;
224 	struct iommu_group *group;
225 	const char *name;
226 
227 	if (!iommu_group_kset)
228 		return NULL;
229 
230 	name = kasprintf(GFP_KERNEL, "%d", id);
231 	if (!name)
232 		return NULL;
233 
234 	group_kobj = kset_find_obj(iommu_group_kset, name);
235 	kfree(name);
236 
237 	if (!group_kobj)
238 		return NULL;
239 
240 	group = container_of(group_kobj, struct iommu_group, kobj);
241 	BUG_ON(group->id != id);
242 
243 	kobject_get(group->devices_kobj);
244 	kobject_put(&group->kobj);
245 
246 	return group;
247 }
248 EXPORT_SYMBOL_GPL(iommu_group_get_by_id);
249 
250 /**
251  * iommu_group_get_iommudata - retrieve iommu_data registered for a group
252  * @group: the group
253  *
254  * iommu drivers can store data in the group for use when doing iommu
255  * operations.  This function provides a way to retrieve it.  Caller
256  * should hold a group reference.
257  */
258 void *iommu_group_get_iommudata(struct iommu_group *group)
259 {
260 	return group->iommu_data;
261 }
262 EXPORT_SYMBOL_GPL(iommu_group_get_iommudata);
263 
264 /**
265  * iommu_group_set_iommudata - set iommu_data for a group
266  * @group: the group
267  * @iommu_data: new data
268  * @release: release function for iommu_data
269  *
270  * iommu drivers can store data in the group for use when doing iommu
271  * operations.  This function provides a way to set the data after
272  * the group has been allocated.  Caller should hold a group reference.
273  */
274 void iommu_group_set_iommudata(struct iommu_group *group, void *iommu_data,
275 			       void (*release)(void *iommu_data))
276 {
277 	group->iommu_data = iommu_data;
278 	group->iommu_data_release = release;
279 }
280 EXPORT_SYMBOL_GPL(iommu_group_set_iommudata);
281 
282 /**
283  * iommu_group_set_name - set name for a group
284  * @group: the group
285  * @name: name
286  *
287  * Allow iommu driver to set a name for a group.  When set it will
288  * appear in a name attribute file under the group in sysfs.
289  */
290 int iommu_group_set_name(struct iommu_group *group, const char *name)
291 {
292 	int ret;
293 
294 	if (group->name) {
295 		iommu_group_remove_file(group, &iommu_group_attr_name);
296 		kfree(group->name);
297 		group->name = NULL;
298 		if (!name)
299 			return 0;
300 	}
301 
302 	group->name = kstrdup(name, GFP_KERNEL);
303 	if (!group->name)
304 		return -ENOMEM;
305 
306 	ret = iommu_group_create_file(group, &iommu_group_attr_name);
307 	if (ret) {
308 		kfree(group->name);
309 		group->name = NULL;
310 		return ret;
311 	}
312 
313 	return 0;
314 }
315 EXPORT_SYMBOL_GPL(iommu_group_set_name);
316 
317 static int iommu_group_create_direct_mappings(struct iommu_group *group,
318 					      struct device *dev)
319 {
320 	struct iommu_domain *domain = group->default_domain;
321 	struct iommu_dm_region *entry;
322 	struct list_head mappings;
323 	unsigned long pg_size;
324 	int ret = 0;
325 
326 	if (!domain || domain->type != IOMMU_DOMAIN_DMA)
327 		return 0;
328 
329 	BUG_ON(!domain->pgsize_bitmap);
330 
331 	pg_size = 1UL << __ffs(domain->pgsize_bitmap);
332 	INIT_LIST_HEAD(&mappings);
333 
334 	iommu_get_dm_regions(dev, &mappings);
335 
336 	/* We need to consider overlapping regions for different devices */
337 	list_for_each_entry(entry, &mappings, list) {
338 		dma_addr_t start, end, addr;
339 
340 		if (domain->ops->apply_dm_region)
341 			domain->ops->apply_dm_region(dev, domain, entry);
342 
343 		start = ALIGN(entry->start, pg_size);
344 		end   = ALIGN(entry->start + entry->length, pg_size);
345 
346 		for (addr = start; addr < end; addr += pg_size) {
347 			phys_addr_t phys_addr;
348 
349 			phys_addr = iommu_iova_to_phys(domain, addr);
350 			if (phys_addr)
351 				continue;
352 
353 			ret = iommu_map(domain, addr, addr, pg_size, entry->prot);
354 			if (ret)
355 				goto out;
356 		}
357 
358 	}
359 
360 out:
361 	iommu_put_dm_regions(dev, &mappings);
362 
363 	return ret;
364 }
365 
366 /**
367  * iommu_group_add_device - add a device to an iommu group
368  * @group: the group into which to add the device (reference should be held)
369  * @dev: the device
370  *
371  * This function is called by an iommu driver to add a device into a
372  * group.  Adding a device increments the group reference count.
373  */
374 int iommu_group_add_device(struct iommu_group *group, struct device *dev)
375 {
376 	int ret, i = 0;
377 	struct iommu_device *device;
378 
379 	device = kzalloc(sizeof(*device), GFP_KERNEL);
380 	if (!device)
381 		return -ENOMEM;
382 
383 	device->dev = dev;
384 
385 	ret = sysfs_create_link(&dev->kobj, &group->kobj, "iommu_group");
386 	if (ret) {
387 		kfree(device);
388 		return ret;
389 	}
390 
391 	device->name = kasprintf(GFP_KERNEL, "%s", kobject_name(&dev->kobj));
392 rename:
393 	if (!device->name) {
394 		sysfs_remove_link(&dev->kobj, "iommu_group");
395 		kfree(device);
396 		return -ENOMEM;
397 	}
398 
399 	ret = sysfs_create_link_nowarn(group->devices_kobj,
400 				       &dev->kobj, device->name);
401 	if (ret) {
402 		kfree(device->name);
403 		if (ret == -EEXIST && i >= 0) {
404 			/*
405 			 * Account for the slim chance of collision
406 			 * and append an instance to the name.
407 			 */
408 			device->name = kasprintf(GFP_KERNEL, "%s.%d",
409 						 kobject_name(&dev->kobj), i++);
410 			goto rename;
411 		}
412 
413 		sysfs_remove_link(&dev->kobj, "iommu_group");
414 		kfree(device);
415 		return ret;
416 	}
417 
418 	kobject_get(group->devices_kobj);
419 
420 	dev->iommu_group = group;
421 
422 	iommu_group_create_direct_mappings(group, dev);
423 
424 	mutex_lock(&group->mutex);
425 	list_add_tail(&device->list, &group->devices);
426 	if (group->domain)
427 		__iommu_attach_device(group->domain, dev);
428 	mutex_unlock(&group->mutex);
429 
430 	/* Notify any listeners about change to group. */
431 	blocking_notifier_call_chain(&group->notifier,
432 				     IOMMU_GROUP_NOTIFY_ADD_DEVICE, dev);
433 
434 	trace_add_device_to_group(group->id, dev);
435 
436 	pr_info("Adding device %s to group %d\n", dev_name(dev), group->id);
437 
438 	return 0;
439 }
440 EXPORT_SYMBOL_GPL(iommu_group_add_device);
441 
442 /**
443  * iommu_group_remove_device - remove a device from it's current group
444  * @dev: device to be removed
445  *
446  * This function is called by an iommu driver to remove the device from
447  * it's current group.  This decrements the iommu group reference count.
448  */
449 void iommu_group_remove_device(struct device *dev)
450 {
451 	struct iommu_group *group = dev->iommu_group;
452 	struct iommu_device *tmp_device, *device = NULL;
453 
454 	pr_info("Removing device %s from group %d\n", dev_name(dev), group->id);
455 
456 	/* Pre-notify listeners that a device is being removed. */
457 	blocking_notifier_call_chain(&group->notifier,
458 				     IOMMU_GROUP_NOTIFY_DEL_DEVICE, dev);
459 
460 	mutex_lock(&group->mutex);
461 	list_for_each_entry(tmp_device, &group->devices, list) {
462 		if (tmp_device->dev == dev) {
463 			device = tmp_device;
464 			list_del(&device->list);
465 			break;
466 		}
467 	}
468 	mutex_unlock(&group->mutex);
469 
470 	if (!device)
471 		return;
472 
473 	sysfs_remove_link(group->devices_kobj, device->name);
474 	sysfs_remove_link(&dev->kobj, "iommu_group");
475 
476 	trace_remove_device_from_group(group->id, dev);
477 
478 	kfree(device->name);
479 	kfree(device);
480 	dev->iommu_group = NULL;
481 	kobject_put(group->devices_kobj);
482 }
483 EXPORT_SYMBOL_GPL(iommu_group_remove_device);
484 
485 static int iommu_group_device_count(struct iommu_group *group)
486 {
487 	struct iommu_device *entry;
488 	int ret = 0;
489 
490 	list_for_each_entry(entry, &group->devices, list)
491 		ret++;
492 
493 	return ret;
494 }
495 
496 /**
497  * iommu_group_for_each_dev - iterate over each device in the group
498  * @group: the group
499  * @data: caller opaque data to be passed to callback function
500  * @fn: caller supplied callback function
501  *
502  * This function is called by group users to iterate over group devices.
503  * Callers should hold a reference count to the group during callback.
504  * The group->mutex is held across callbacks, which will block calls to
505  * iommu_group_add/remove_device.
506  */
507 static int __iommu_group_for_each_dev(struct iommu_group *group, void *data,
508 				      int (*fn)(struct device *, void *))
509 {
510 	struct iommu_device *device;
511 	int ret = 0;
512 
513 	list_for_each_entry(device, &group->devices, list) {
514 		ret = fn(device->dev, data);
515 		if (ret)
516 			break;
517 	}
518 	return ret;
519 }
520 
521 
522 int iommu_group_for_each_dev(struct iommu_group *group, void *data,
523 			     int (*fn)(struct device *, void *))
524 {
525 	int ret;
526 
527 	mutex_lock(&group->mutex);
528 	ret = __iommu_group_for_each_dev(group, data, fn);
529 	mutex_unlock(&group->mutex);
530 
531 	return ret;
532 }
533 EXPORT_SYMBOL_GPL(iommu_group_for_each_dev);
534 
535 /**
536  * iommu_group_get - Return the group for a device and increment reference
537  * @dev: get the group that this device belongs to
538  *
539  * This function is called by iommu drivers and users to get the group
540  * for the specified device.  If found, the group is returned and the group
541  * reference in incremented, else NULL.
542  */
543 struct iommu_group *iommu_group_get(struct device *dev)
544 {
545 	struct iommu_group *group = dev->iommu_group;
546 
547 	if (group)
548 		kobject_get(group->devices_kobj);
549 
550 	return group;
551 }
552 EXPORT_SYMBOL_GPL(iommu_group_get);
553 
554 /**
555  * iommu_group_ref_get - Increment reference on a group
556  * @group: the group to use, must not be NULL
557  *
558  * This function is called by iommu drivers to take additional references on an
559  * existing group.  Returns the given group for convenience.
560  */
561 struct iommu_group *iommu_group_ref_get(struct iommu_group *group)
562 {
563 	kobject_get(group->devices_kobj);
564 	return group;
565 }
566 
567 /**
568  * iommu_group_put - Decrement group reference
569  * @group: the group to use
570  *
571  * This function is called by iommu drivers and users to release the
572  * iommu group.  Once the reference count is zero, the group is released.
573  */
574 void iommu_group_put(struct iommu_group *group)
575 {
576 	if (group)
577 		kobject_put(group->devices_kobj);
578 }
579 EXPORT_SYMBOL_GPL(iommu_group_put);
580 
581 /**
582  * iommu_group_register_notifier - Register a notifier for group changes
583  * @group: the group to watch
584  * @nb: notifier block to signal
585  *
586  * This function allows iommu group users to track changes in a group.
587  * See include/linux/iommu.h for actions sent via this notifier.  Caller
588  * should hold a reference to the group throughout notifier registration.
589  */
590 int iommu_group_register_notifier(struct iommu_group *group,
591 				  struct notifier_block *nb)
592 {
593 	return blocking_notifier_chain_register(&group->notifier, nb);
594 }
595 EXPORT_SYMBOL_GPL(iommu_group_register_notifier);
596 
597 /**
598  * iommu_group_unregister_notifier - Unregister a notifier
599  * @group: the group to watch
600  * @nb: notifier block to signal
601  *
602  * Unregister a previously registered group notifier block.
603  */
604 int iommu_group_unregister_notifier(struct iommu_group *group,
605 				    struct notifier_block *nb)
606 {
607 	return blocking_notifier_chain_unregister(&group->notifier, nb);
608 }
609 EXPORT_SYMBOL_GPL(iommu_group_unregister_notifier);
610 
611 /**
612  * iommu_group_id - Return ID for a group
613  * @group: the group to ID
614  *
615  * Return the unique ID for the group matching the sysfs group number.
616  */
617 int iommu_group_id(struct iommu_group *group)
618 {
619 	return group->id;
620 }
621 EXPORT_SYMBOL_GPL(iommu_group_id);
622 
623 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
624 					       unsigned long *devfns);
625 
626 /*
627  * To consider a PCI device isolated, we require ACS to support Source
628  * Validation, Request Redirection, Completer Redirection, and Upstream
629  * Forwarding.  This effectively means that devices cannot spoof their
630  * requester ID, requests and completions cannot be redirected, and all
631  * transactions are forwarded upstream, even as it passes through a
632  * bridge where the target device is downstream.
633  */
634 #define REQ_ACS_FLAGS   (PCI_ACS_SV | PCI_ACS_RR | PCI_ACS_CR | PCI_ACS_UF)
635 
636 /*
637  * For multifunction devices which are not isolated from each other, find
638  * all the other non-isolated functions and look for existing groups.  For
639  * each function, we also need to look for aliases to or from other devices
640  * that may already have a group.
641  */
642 static struct iommu_group *get_pci_function_alias_group(struct pci_dev *pdev,
643 							unsigned long *devfns)
644 {
645 	struct pci_dev *tmp = NULL;
646 	struct iommu_group *group;
647 
648 	if (!pdev->multifunction || pci_acs_enabled(pdev, REQ_ACS_FLAGS))
649 		return NULL;
650 
651 	for_each_pci_dev(tmp) {
652 		if (tmp == pdev || tmp->bus != pdev->bus ||
653 		    PCI_SLOT(tmp->devfn) != PCI_SLOT(pdev->devfn) ||
654 		    pci_acs_enabled(tmp, REQ_ACS_FLAGS))
655 			continue;
656 
657 		group = get_pci_alias_group(tmp, devfns);
658 		if (group) {
659 			pci_dev_put(tmp);
660 			return group;
661 		}
662 	}
663 
664 	return NULL;
665 }
666 
667 /*
668  * Look for aliases to or from the given device for existing groups. DMA
669  * aliases are only supported on the same bus, therefore the search
670  * space is quite small (especially since we're really only looking at pcie
671  * device, and therefore only expect multiple slots on the root complex or
672  * downstream switch ports).  It's conceivable though that a pair of
673  * multifunction devices could have aliases between them that would cause a
674  * loop.  To prevent this, we use a bitmap to track where we've been.
675  */
676 static struct iommu_group *get_pci_alias_group(struct pci_dev *pdev,
677 					       unsigned long *devfns)
678 {
679 	struct pci_dev *tmp = NULL;
680 	struct iommu_group *group;
681 
682 	if (test_and_set_bit(pdev->devfn & 0xff, devfns))
683 		return NULL;
684 
685 	group = iommu_group_get(&pdev->dev);
686 	if (group)
687 		return group;
688 
689 	for_each_pci_dev(tmp) {
690 		if (tmp == pdev || tmp->bus != pdev->bus)
691 			continue;
692 
693 		/* We alias them or they alias us */
694 		if (pci_devs_are_dma_aliases(pdev, tmp)) {
695 			group = get_pci_alias_group(tmp, devfns);
696 			if (group) {
697 				pci_dev_put(tmp);
698 				return group;
699 			}
700 
701 			group = get_pci_function_alias_group(tmp, devfns);
702 			if (group) {
703 				pci_dev_put(tmp);
704 				return group;
705 			}
706 		}
707 	}
708 
709 	return NULL;
710 }
711 
712 struct group_for_pci_data {
713 	struct pci_dev *pdev;
714 	struct iommu_group *group;
715 };
716 
717 /*
718  * DMA alias iterator callback, return the last seen device.  Stop and return
719  * the IOMMU group if we find one along the way.
720  */
721 static int get_pci_alias_or_group(struct pci_dev *pdev, u16 alias, void *opaque)
722 {
723 	struct group_for_pci_data *data = opaque;
724 
725 	data->pdev = pdev;
726 	data->group = iommu_group_get(&pdev->dev);
727 
728 	return data->group != NULL;
729 }
730 
731 /*
732  * Generic device_group call-back function. It just allocates one
733  * iommu-group per device.
734  */
735 struct iommu_group *generic_device_group(struct device *dev)
736 {
737 	struct iommu_group *group;
738 
739 	group = iommu_group_alloc();
740 	if (IS_ERR(group))
741 		return NULL;
742 
743 	return group;
744 }
745 
746 /*
747  * Use standard PCI bus topology, isolation features, and DMA alias quirks
748  * to find or create an IOMMU group for a device.
749  */
750 struct iommu_group *pci_device_group(struct device *dev)
751 {
752 	struct pci_dev *pdev = to_pci_dev(dev);
753 	struct group_for_pci_data data;
754 	struct pci_bus *bus;
755 	struct iommu_group *group = NULL;
756 	u64 devfns[4] = { 0 };
757 
758 	if (WARN_ON(!dev_is_pci(dev)))
759 		return ERR_PTR(-EINVAL);
760 
761 	/*
762 	 * Find the upstream DMA alias for the device.  A device must not
763 	 * be aliased due to topology in order to have its own IOMMU group.
764 	 * If we find an alias along the way that already belongs to a
765 	 * group, use it.
766 	 */
767 	if (pci_for_each_dma_alias(pdev, get_pci_alias_or_group, &data))
768 		return data.group;
769 
770 	pdev = data.pdev;
771 
772 	/*
773 	 * Continue upstream from the point of minimum IOMMU granularity
774 	 * due to aliases to the point where devices are protected from
775 	 * peer-to-peer DMA by PCI ACS.  Again, if we find an existing
776 	 * group, use it.
777 	 */
778 	for (bus = pdev->bus; !pci_is_root_bus(bus); bus = bus->parent) {
779 		if (!bus->self)
780 			continue;
781 
782 		if (pci_acs_path_enabled(bus->self, NULL, REQ_ACS_FLAGS))
783 			break;
784 
785 		pdev = bus->self;
786 
787 		group = iommu_group_get(&pdev->dev);
788 		if (group)
789 			return group;
790 	}
791 
792 	/*
793 	 * Look for existing groups on device aliases.  If we alias another
794 	 * device or another device aliases us, use the same group.
795 	 */
796 	group = get_pci_alias_group(pdev, (unsigned long *)devfns);
797 	if (group)
798 		return group;
799 
800 	/*
801 	 * Look for existing groups on non-isolated functions on the same
802 	 * slot and aliases of those funcions, if any.  No need to clear
803 	 * the search bitmap, the tested devfns are still valid.
804 	 */
805 	group = get_pci_function_alias_group(pdev, (unsigned long *)devfns);
806 	if (group)
807 		return group;
808 
809 	/* No shared group found, allocate new */
810 	group = iommu_group_alloc();
811 	if (IS_ERR(group))
812 		return NULL;
813 
814 	return group;
815 }
816 
817 /**
818  * iommu_group_get_for_dev - Find or create the IOMMU group for a device
819  * @dev: target device
820  *
821  * This function is intended to be called by IOMMU drivers and extended to
822  * support common, bus-defined algorithms when determining or creating the
823  * IOMMU group for a device.  On success, the caller will hold a reference
824  * to the returned IOMMU group, which will already include the provided
825  * device.  The reference should be released with iommu_group_put().
826  */
827 struct iommu_group *iommu_group_get_for_dev(struct device *dev)
828 {
829 	const struct iommu_ops *ops = dev->bus->iommu_ops;
830 	struct iommu_group *group;
831 	int ret;
832 
833 	group = iommu_group_get(dev);
834 	if (group)
835 		return group;
836 
837 	group = ERR_PTR(-EINVAL);
838 
839 	if (ops && ops->device_group)
840 		group = ops->device_group(dev);
841 
842 	if (IS_ERR(group))
843 		return group;
844 
845 	/*
846 	 * Try to allocate a default domain - needs support from the
847 	 * IOMMU driver.
848 	 */
849 	if (!group->default_domain) {
850 		group->default_domain = __iommu_domain_alloc(dev->bus,
851 							     IOMMU_DOMAIN_DMA);
852 		if (!group->domain)
853 			group->domain = group->default_domain;
854 	}
855 
856 	ret = iommu_group_add_device(group, dev);
857 	if (ret) {
858 		iommu_group_put(group);
859 		return ERR_PTR(ret);
860 	}
861 
862 	return group;
863 }
864 
865 struct iommu_domain *iommu_group_default_domain(struct iommu_group *group)
866 {
867 	return group->default_domain;
868 }
869 
870 static int add_iommu_group(struct device *dev, void *data)
871 {
872 	struct iommu_callback_data *cb = data;
873 	const struct iommu_ops *ops = cb->ops;
874 	int ret;
875 
876 	if (!ops->add_device)
877 		return 0;
878 
879 	WARN_ON(dev->iommu_group);
880 
881 	ret = ops->add_device(dev);
882 
883 	/*
884 	 * We ignore -ENODEV errors for now, as they just mean that the
885 	 * device is not translated by an IOMMU. We still care about
886 	 * other errors and fail to initialize when they happen.
887 	 */
888 	if (ret == -ENODEV)
889 		ret = 0;
890 
891 	return ret;
892 }
893 
894 static int remove_iommu_group(struct device *dev, void *data)
895 {
896 	struct iommu_callback_data *cb = data;
897 	const struct iommu_ops *ops = cb->ops;
898 
899 	if (ops->remove_device && dev->iommu_group)
900 		ops->remove_device(dev);
901 
902 	return 0;
903 }
904 
905 static int iommu_bus_notifier(struct notifier_block *nb,
906 			      unsigned long action, void *data)
907 {
908 	struct device *dev = data;
909 	const struct iommu_ops *ops = dev->bus->iommu_ops;
910 	struct iommu_group *group;
911 	unsigned long group_action = 0;
912 
913 	/*
914 	 * ADD/DEL call into iommu driver ops if provided, which may
915 	 * result in ADD/DEL notifiers to group->notifier
916 	 */
917 	if (action == BUS_NOTIFY_ADD_DEVICE) {
918 		if (ops->add_device)
919 			return ops->add_device(dev);
920 	} else if (action == BUS_NOTIFY_REMOVED_DEVICE) {
921 		if (ops->remove_device && dev->iommu_group) {
922 			ops->remove_device(dev);
923 			return 0;
924 		}
925 	}
926 
927 	/*
928 	 * Remaining BUS_NOTIFYs get filtered and republished to the
929 	 * group, if anyone is listening
930 	 */
931 	group = iommu_group_get(dev);
932 	if (!group)
933 		return 0;
934 
935 	switch (action) {
936 	case BUS_NOTIFY_BIND_DRIVER:
937 		group_action = IOMMU_GROUP_NOTIFY_BIND_DRIVER;
938 		break;
939 	case BUS_NOTIFY_BOUND_DRIVER:
940 		group_action = IOMMU_GROUP_NOTIFY_BOUND_DRIVER;
941 		break;
942 	case BUS_NOTIFY_UNBIND_DRIVER:
943 		group_action = IOMMU_GROUP_NOTIFY_UNBIND_DRIVER;
944 		break;
945 	case BUS_NOTIFY_UNBOUND_DRIVER:
946 		group_action = IOMMU_GROUP_NOTIFY_UNBOUND_DRIVER;
947 		break;
948 	}
949 
950 	if (group_action)
951 		blocking_notifier_call_chain(&group->notifier,
952 					     group_action, dev);
953 
954 	iommu_group_put(group);
955 	return 0;
956 }
957 
958 static int iommu_bus_init(struct bus_type *bus, const struct iommu_ops *ops)
959 {
960 	int err;
961 	struct notifier_block *nb;
962 	struct iommu_callback_data cb = {
963 		.ops = ops,
964 	};
965 
966 	nb = kzalloc(sizeof(struct notifier_block), GFP_KERNEL);
967 	if (!nb)
968 		return -ENOMEM;
969 
970 	nb->notifier_call = iommu_bus_notifier;
971 
972 	err = bus_register_notifier(bus, nb);
973 	if (err)
974 		goto out_free;
975 
976 	err = bus_for_each_dev(bus, NULL, &cb, add_iommu_group);
977 	if (err)
978 		goto out_err;
979 
980 
981 	return 0;
982 
983 out_err:
984 	/* Clean up */
985 	bus_for_each_dev(bus, NULL, &cb, remove_iommu_group);
986 	bus_unregister_notifier(bus, nb);
987 
988 out_free:
989 	kfree(nb);
990 
991 	return err;
992 }
993 
994 /**
995  * bus_set_iommu - set iommu-callbacks for the bus
996  * @bus: bus.
997  * @ops: the callbacks provided by the iommu-driver
998  *
999  * This function is called by an iommu driver to set the iommu methods
1000  * used for a particular bus. Drivers for devices on that bus can use
1001  * the iommu-api after these ops are registered.
1002  * This special function is needed because IOMMUs are usually devices on
1003  * the bus itself, so the iommu drivers are not initialized when the bus
1004  * is set up. With this function the iommu-driver can set the iommu-ops
1005  * afterwards.
1006  */
1007 int bus_set_iommu(struct bus_type *bus, const struct iommu_ops *ops)
1008 {
1009 	int err;
1010 
1011 	if (bus->iommu_ops != NULL)
1012 		return -EBUSY;
1013 
1014 	bus->iommu_ops = ops;
1015 
1016 	/* Do IOMMU specific setup for this bus-type */
1017 	err = iommu_bus_init(bus, ops);
1018 	if (err)
1019 		bus->iommu_ops = NULL;
1020 
1021 	return err;
1022 }
1023 EXPORT_SYMBOL_GPL(bus_set_iommu);
1024 
1025 bool iommu_present(struct bus_type *bus)
1026 {
1027 	return bus->iommu_ops != NULL;
1028 }
1029 EXPORT_SYMBOL_GPL(iommu_present);
1030 
1031 bool iommu_capable(struct bus_type *bus, enum iommu_cap cap)
1032 {
1033 	if (!bus->iommu_ops || !bus->iommu_ops->capable)
1034 		return false;
1035 
1036 	return bus->iommu_ops->capable(cap);
1037 }
1038 EXPORT_SYMBOL_GPL(iommu_capable);
1039 
1040 /**
1041  * iommu_set_fault_handler() - set a fault handler for an iommu domain
1042  * @domain: iommu domain
1043  * @handler: fault handler
1044  * @token: user data, will be passed back to the fault handler
1045  *
1046  * This function should be used by IOMMU users which want to be notified
1047  * whenever an IOMMU fault happens.
1048  *
1049  * The fault handler itself should return 0 on success, and an appropriate
1050  * error code otherwise.
1051  */
1052 void iommu_set_fault_handler(struct iommu_domain *domain,
1053 					iommu_fault_handler_t handler,
1054 					void *token)
1055 {
1056 	BUG_ON(!domain);
1057 
1058 	domain->handler = handler;
1059 	domain->handler_token = token;
1060 }
1061 EXPORT_SYMBOL_GPL(iommu_set_fault_handler);
1062 
1063 static struct iommu_domain *__iommu_domain_alloc(struct bus_type *bus,
1064 						 unsigned type)
1065 {
1066 	struct iommu_domain *domain;
1067 
1068 	if (bus == NULL || bus->iommu_ops == NULL)
1069 		return NULL;
1070 
1071 	domain = bus->iommu_ops->domain_alloc(type);
1072 	if (!domain)
1073 		return NULL;
1074 
1075 	domain->ops  = bus->iommu_ops;
1076 	domain->type = type;
1077 	/* Assume all sizes by default; the driver may override this later */
1078 	domain->pgsize_bitmap  = bus->iommu_ops->pgsize_bitmap;
1079 
1080 	return domain;
1081 }
1082 
1083 struct iommu_domain *iommu_domain_alloc(struct bus_type *bus)
1084 {
1085 	return __iommu_domain_alloc(bus, IOMMU_DOMAIN_UNMANAGED);
1086 }
1087 EXPORT_SYMBOL_GPL(iommu_domain_alloc);
1088 
1089 void iommu_domain_free(struct iommu_domain *domain)
1090 {
1091 	domain->ops->domain_free(domain);
1092 }
1093 EXPORT_SYMBOL_GPL(iommu_domain_free);
1094 
1095 static int __iommu_attach_device(struct iommu_domain *domain,
1096 				 struct device *dev)
1097 {
1098 	int ret;
1099 	if (unlikely(domain->ops->attach_dev == NULL))
1100 		return -ENODEV;
1101 
1102 	ret = domain->ops->attach_dev(domain, dev);
1103 	if (!ret)
1104 		trace_attach_device_to_domain(dev);
1105 	return ret;
1106 }
1107 
1108 int iommu_attach_device(struct iommu_domain *domain, struct device *dev)
1109 {
1110 	struct iommu_group *group;
1111 	int ret;
1112 
1113 	group = iommu_group_get(dev);
1114 	/* FIXME: Remove this when groups a mandatory for iommu drivers */
1115 	if (group == NULL)
1116 		return __iommu_attach_device(domain, dev);
1117 
1118 	/*
1119 	 * We have a group - lock it to make sure the device-count doesn't
1120 	 * change while we are attaching
1121 	 */
1122 	mutex_lock(&group->mutex);
1123 	ret = -EINVAL;
1124 	if (iommu_group_device_count(group) != 1)
1125 		goto out_unlock;
1126 
1127 	ret = __iommu_attach_group(domain, group);
1128 
1129 out_unlock:
1130 	mutex_unlock(&group->mutex);
1131 	iommu_group_put(group);
1132 
1133 	return ret;
1134 }
1135 EXPORT_SYMBOL_GPL(iommu_attach_device);
1136 
1137 static void __iommu_detach_device(struct iommu_domain *domain,
1138 				  struct device *dev)
1139 {
1140 	if (unlikely(domain->ops->detach_dev == NULL))
1141 		return;
1142 
1143 	domain->ops->detach_dev(domain, dev);
1144 	trace_detach_device_from_domain(dev);
1145 }
1146 
1147 void iommu_detach_device(struct iommu_domain *domain, struct device *dev)
1148 {
1149 	struct iommu_group *group;
1150 
1151 	group = iommu_group_get(dev);
1152 	/* FIXME: Remove this when groups a mandatory for iommu drivers */
1153 	if (group == NULL)
1154 		return __iommu_detach_device(domain, dev);
1155 
1156 	mutex_lock(&group->mutex);
1157 	if (iommu_group_device_count(group) != 1) {
1158 		WARN_ON(1);
1159 		goto out_unlock;
1160 	}
1161 
1162 	__iommu_detach_group(domain, group);
1163 
1164 out_unlock:
1165 	mutex_unlock(&group->mutex);
1166 	iommu_group_put(group);
1167 }
1168 EXPORT_SYMBOL_GPL(iommu_detach_device);
1169 
1170 struct iommu_domain *iommu_get_domain_for_dev(struct device *dev)
1171 {
1172 	struct iommu_domain *domain;
1173 	struct iommu_group *group;
1174 
1175 	group = iommu_group_get(dev);
1176 	/* FIXME: Remove this when groups a mandatory for iommu drivers */
1177 	if (group == NULL)
1178 		return NULL;
1179 
1180 	domain = group->domain;
1181 
1182 	iommu_group_put(group);
1183 
1184 	return domain;
1185 }
1186 EXPORT_SYMBOL_GPL(iommu_get_domain_for_dev);
1187 
1188 /*
1189  * IOMMU groups are really the natrual working unit of the IOMMU, but
1190  * the IOMMU API works on domains and devices.  Bridge that gap by
1191  * iterating over the devices in a group.  Ideally we'd have a single
1192  * device which represents the requestor ID of the group, but we also
1193  * allow IOMMU drivers to create policy defined minimum sets, where
1194  * the physical hardware may be able to distiguish members, but we
1195  * wish to group them at a higher level (ex. untrusted multi-function
1196  * PCI devices).  Thus we attach each device.
1197  */
1198 static int iommu_group_do_attach_device(struct device *dev, void *data)
1199 {
1200 	struct iommu_domain *domain = data;
1201 
1202 	return __iommu_attach_device(domain, dev);
1203 }
1204 
1205 static int __iommu_attach_group(struct iommu_domain *domain,
1206 				struct iommu_group *group)
1207 {
1208 	int ret;
1209 
1210 	if (group->default_domain && group->domain != group->default_domain)
1211 		return -EBUSY;
1212 
1213 	ret = __iommu_group_for_each_dev(group, domain,
1214 					 iommu_group_do_attach_device);
1215 	if (ret == 0)
1216 		group->domain = domain;
1217 
1218 	return ret;
1219 }
1220 
1221 int iommu_attach_group(struct iommu_domain *domain, struct iommu_group *group)
1222 {
1223 	int ret;
1224 
1225 	mutex_lock(&group->mutex);
1226 	ret = __iommu_attach_group(domain, group);
1227 	mutex_unlock(&group->mutex);
1228 
1229 	return ret;
1230 }
1231 EXPORT_SYMBOL_GPL(iommu_attach_group);
1232 
1233 static int iommu_group_do_detach_device(struct device *dev, void *data)
1234 {
1235 	struct iommu_domain *domain = data;
1236 
1237 	__iommu_detach_device(domain, dev);
1238 
1239 	return 0;
1240 }
1241 
1242 static void __iommu_detach_group(struct iommu_domain *domain,
1243 				 struct iommu_group *group)
1244 {
1245 	int ret;
1246 
1247 	if (!group->default_domain) {
1248 		__iommu_group_for_each_dev(group, domain,
1249 					   iommu_group_do_detach_device);
1250 		group->domain = NULL;
1251 		return;
1252 	}
1253 
1254 	if (group->domain == group->default_domain)
1255 		return;
1256 
1257 	/* Detach by re-attaching to the default domain */
1258 	ret = __iommu_group_for_each_dev(group, group->default_domain,
1259 					 iommu_group_do_attach_device);
1260 	if (ret != 0)
1261 		WARN_ON(1);
1262 	else
1263 		group->domain = group->default_domain;
1264 }
1265 
1266 void iommu_detach_group(struct iommu_domain *domain, struct iommu_group *group)
1267 {
1268 	mutex_lock(&group->mutex);
1269 	__iommu_detach_group(domain, group);
1270 	mutex_unlock(&group->mutex);
1271 }
1272 EXPORT_SYMBOL_GPL(iommu_detach_group);
1273 
1274 phys_addr_t iommu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova)
1275 {
1276 	if (unlikely(domain->ops->iova_to_phys == NULL))
1277 		return 0;
1278 
1279 	return domain->ops->iova_to_phys(domain, iova);
1280 }
1281 EXPORT_SYMBOL_GPL(iommu_iova_to_phys);
1282 
1283 static size_t iommu_pgsize(struct iommu_domain *domain,
1284 			   unsigned long addr_merge, size_t size)
1285 {
1286 	unsigned int pgsize_idx;
1287 	size_t pgsize;
1288 
1289 	/* Max page size that still fits into 'size' */
1290 	pgsize_idx = __fls(size);
1291 
1292 	/* need to consider alignment requirements ? */
1293 	if (likely(addr_merge)) {
1294 		/* Max page size allowed by address */
1295 		unsigned int align_pgsize_idx = __ffs(addr_merge);
1296 		pgsize_idx = min(pgsize_idx, align_pgsize_idx);
1297 	}
1298 
1299 	/* build a mask of acceptable page sizes */
1300 	pgsize = (1UL << (pgsize_idx + 1)) - 1;
1301 
1302 	/* throw away page sizes not supported by the hardware */
1303 	pgsize &= domain->pgsize_bitmap;
1304 
1305 	/* make sure we're still sane */
1306 	BUG_ON(!pgsize);
1307 
1308 	/* pick the biggest page */
1309 	pgsize_idx = __fls(pgsize);
1310 	pgsize = 1UL << pgsize_idx;
1311 
1312 	return pgsize;
1313 }
1314 
1315 int iommu_map(struct iommu_domain *domain, unsigned long iova,
1316 	      phys_addr_t paddr, size_t size, int prot)
1317 {
1318 	unsigned long orig_iova = iova;
1319 	unsigned int min_pagesz;
1320 	size_t orig_size = size;
1321 	phys_addr_t orig_paddr = paddr;
1322 	int ret = 0;
1323 
1324 	if (unlikely(domain->ops->map == NULL ||
1325 		     domain->pgsize_bitmap == 0UL))
1326 		return -ENODEV;
1327 
1328 	if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1329 		return -EINVAL;
1330 
1331 	/* find out the minimum page size supported */
1332 	min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
1333 
1334 	/*
1335 	 * both the virtual address and the physical one, as well as
1336 	 * the size of the mapping, must be aligned (at least) to the
1337 	 * size of the smallest page supported by the hardware
1338 	 */
1339 	if (!IS_ALIGNED(iova | paddr | size, min_pagesz)) {
1340 		pr_err("unaligned: iova 0x%lx pa %pa size 0x%zx min_pagesz 0x%x\n",
1341 		       iova, &paddr, size, min_pagesz);
1342 		return -EINVAL;
1343 	}
1344 
1345 	pr_debug("map: iova 0x%lx pa %pa size 0x%zx\n", iova, &paddr, size);
1346 
1347 	while (size) {
1348 		size_t pgsize = iommu_pgsize(domain, iova | paddr, size);
1349 
1350 		pr_debug("mapping: iova 0x%lx pa %pa pgsize 0x%zx\n",
1351 			 iova, &paddr, pgsize);
1352 
1353 		ret = domain->ops->map(domain, iova, paddr, pgsize, prot);
1354 		if (ret)
1355 			break;
1356 
1357 		iova += pgsize;
1358 		paddr += pgsize;
1359 		size -= pgsize;
1360 	}
1361 
1362 	/* unroll mapping in case something went wrong */
1363 	if (ret)
1364 		iommu_unmap(domain, orig_iova, orig_size - size);
1365 	else
1366 		trace_map(orig_iova, orig_paddr, orig_size);
1367 
1368 	return ret;
1369 }
1370 EXPORT_SYMBOL_GPL(iommu_map);
1371 
1372 size_t iommu_unmap(struct iommu_domain *domain, unsigned long iova, size_t size)
1373 {
1374 	size_t unmapped_page, unmapped = 0;
1375 	unsigned int min_pagesz;
1376 	unsigned long orig_iova = iova;
1377 
1378 	if (unlikely(domain->ops->unmap == NULL ||
1379 		     domain->pgsize_bitmap == 0UL))
1380 		return -ENODEV;
1381 
1382 	if (unlikely(!(domain->type & __IOMMU_DOMAIN_PAGING)))
1383 		return -EINVAL;
1384 
1385 	/* find out the minimum page size supported */
1386 	min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
1387 
1388 	/*
1389 	 * The virtual address, as well as the size of the mapping, must be
1390 	 * aligned (at least) to the size of the smallest page supported
1391 	 * by the hardware
1392 	 */
1393 	if (!IS_ALIGNED(iova | size, min_pagesz)) {
1394 		pr_err("unaligned: iova 0x%lx size 0x%zx min_pagesz 0x%x\n",
1395 		       iova, size, min_pagesz);
1396 		return -EINVAL;
1397 	}
1398 
1399 	pr_debug("unmap this: iova 0x%lx size 0x%zx\n", iova, size);
1400 
1401 	/*
1402 	 * Keep iterating until we either unmap 'size' bytes (or more)
1403 	 * or we hit an area that isn't mapped.
1404 	 */
1405 	while (unmapped < size) {
1406 		size_t pgsize = iommu_pgsize(domain, iova, size - unmapped);
1407 
1408 		unmapped_page = domain->ops->unmap(domain, iova, pgsize);
1409 		if (!unmapped_page)
1410 			break;
1411 
1412 		pr_debug("unmapped: iova 0x%lx size 0x%zx\n",
1413 			 iova, unmapped_page);
1414 
1415 		iova += unmapped_page;
1416 		unmapped += unmapped_page;
1417 	}
1418 
1419 	trace_unmap(orig_iova, size, unmapped);
1420 	return unmapped;
1421 }
1422 EXPORT_SYMBOL_GPL(iommu_unmap);
1423 
1424 size_t default_iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
1425 			 struct scatterlist *sg, unsigned int nents, int prot)
1426 {
1427 	struct scatterlist *s;
1428 	size_t mapped = 0;
1429 	unsigned int i, min_pagesz;
1430 	int ret;
1431 
1432 	if (unlikely(domain->pgsize_bitmap == 0UL))
1433 		return 0;
1434 
1435 	min_pagesz = 1 << __ffs(domain->pgsize_bitmap);
1436 
1437 	for_each_sg(sg, s, nents, i) {
1438 		phys_addr_t phys = page_to_phys(sg_page(s)) + s->offset;
1439 
1440 		/*
1441 		 * We are mapping on IOMMU page boundaries, so offset within
1442 		 * the page must be 0. However, the IOMMU may support pages
1443 		 * smaller than PAGE_SIZE, so s->offset may still represent
1444 		 * an offset of that boundary within the CPU page.
1445 		 */
1446 		if (!IS_ALIGNED(s->offset, min_pagesz))
1447 			goto out_err;
1448 
1449 		ret = iommu_map(domain, iova + mapped, phys, s->length, prot);
1450 		if (ret)
1451 			goto out_err;
1452 
1453 		mapped += s->length;
1454 	}
1455 
1456 	return mapped;
1457 
1458 out_err:
1459 	/* undo mappings already done */
1460 	iommu_unmap(domain, iova, mapped);
1461 
1462 	return 0;
1463 
1464 }
1465 EXPORT_SYMBOL_GPL(default_iommu_map_sg);
1466 
1467 int iommu_domain_window_enable(struct iommu_domain *domain, u32 wnd_nr,
1468 			       phys_addr_t paddr, u64 size, int prot)
1469 {
1470 	if (unlikely(domain->ops->domain_window_enable == NULL))
1471 		return -ENODEV;
1472 
1473 	return domain->ops->domain_window_enable(domain, wnd_nr, paddr, size,
1474 						 prot);
1475 }
1476 EXPORT_SYMBOL_GPL(iommu_domain_window_enable);
1477 
1478 void iommu_domain_window_disable(struct iommu_domain *domain, u32 wnd_nr)
1479 {
1480 	if (unlikely(domain->ops->domain_window_disable == NULL))
1481 		return;
1482 
1483 	return domain->ops->domain_window_disable(domain, wnd_nr);
1484 }
1485 EXPORT_SYMBOL_GPL(iommu_domain_window_disable);
1486 
1487 static int __init iommu_init(void)
1488 {
1489 	iommu_group_kset = kset_create_and_add("iommu_groups",
1490 					       NULL, kernel_kobj);
1491 	BUG_ON(!iommu_group_kset);
1492 
1493 	return 0;
1494 }
1495 core_initcall(iommu_init);
1496 
1497 int iommu_domain_get_attr(struct iommu_domain *domain,
1498 			  enum iommu_attr attr, void *data)
1499 {
1500 	struct iommu_domain_geometry *geometry;
1501 	bool *paging;
1502 	int ret = 0;
1503 	u32 *count;
1504 
1505 	switch (attr) {
1506 	case DOMAIN_ATTR_GEOMETRY:
1507 		geometry  = data;
1508 		*geometry = domain->geometry;
1509 
1510 		break;
1511 	case DOMAIN_ATTR_PAGING:
1512 		paging  = data;
1513 		*paging = (domain->pgsize_bitmap != 0UL);
1514 		break;
1515 	case DOMAIN_ATTR_WINDOWS:
1516 		count = data;
1517 
1518 		if (domain->ops->domain_get_windows != NULL)
1519 			*count = domain->ops->domain_get_windows(domain);
1520 		else
1521 			ret = -ENODEV;
1522 
1523 		break;
1524 	default:
1525 		if (!domain->ops->domain_get_attr)
1526 			return -EINVAL;
1527 
1528 		ret = domain->ops->domain_get_attr(domain, attr, data);
1529 	}
1530 
1531 	return ret;
1532 }
1533 EXPORT_SYMBOL_GPL(iommu_domain_get_attr);
1534 
1535 int iommu_domain_set_attr(struct iommu_domain *domain,
1536 			  enum iommu_attr attr, void *data)
1537 {
1538 	int ret = 0;
1539 	u32 *count;
1540 
1541 	switch (attr) {
1542 	case DOMAIN_ATTR_WINDOWS:
1543 		count = data;
1544 
1545 		if (domain->ops->domain_set_windows != NULL)
1546 			ret = domain->ops->domain_set_windows(domain, *count);
1547 		else
1548 			ret = -ENODEV;
1549 
1550 		break;
1551 	default:
1552 		if (domain->ops->domain_set_attr == NULL)
1553 			return -EINVAL;
1554 
1555 		ret = domain->ops->domain_set_attr(domain, attr, data);
1556 	}
1557 
1558 	return ret;
1559 }
1560 EXPORT_SYMBOL_GPL(iommu_domain_set_attr);
1561 
1562 void iommu_get_dm_regions(struct device *dev, struct list_head *list)
1563 {
1564 	const struct iommu_ops *ops = dev->bus->iommu_ops;
1565 
1566 	if (ops && ops->get_dm_regions)
1567 		ops->get_dm_regions(dev, list);
1568 }
1569 
1570 void iommu_put_dm_regions(struct device *dev, struct list_head *list)
1571 {
1572 	const struct iommu_ops *ops = dev->bus->iommu_ops;
1573 
1574 	if (ops && ops->put_dm_regions)
1575 		ops->put_dm_regions(dev, list);
1576 }
1577 
1578 /* Request that a device is direct mapped by the IOMMU */
1579 int iommu_request_dm_for_dev(struct device *dev)
1580 {
1581 	struct iommu_domain *dm_domain;
1582 	struct iommu_group *group;
1583 	int ret;
1584 
1585 	/* Device must already be in a group before calling this function */
1586 	group = iommu_group_get_for_dev(dev);
1587 	if (IS_ERR(group))
1588 		return PTR_ERR(group);
1589 
1590 	mutex_lock(&group->mutex);
1591 
1592 	/* Check if the default domain is already direct mapped */
1593 	ret = 0;
1594 	if (group->default_domain &&
1595 	    group->default_domain->type == IOMMU_DOMAIN_IDENTITY)
1596 		goto out;
1597 
1598 	/* Don't change mappings of existing devices */
1599 	ret = -EBUSY;
1600 	if (iommu_group_device_count(group) != 1)
1601 		goto out;
1602 
1603 	/* Allocate a direct mapped domain */
1604 	ret = -ENOMEM;
1605 	dm_domain = __iommu_domain_alloc(dev->bus, IOMMU_DOMAIN_IDENTITY);
1606 	if (!dm_domain)
1607 		goto out;
1608 
1609 	/* Attach the device to the domain */
1610 	ret = __iommu_attach_group(dm_domain, group);
1611 	if (ret) {
1612 		iommu_domain_free(dm_domain);
1613 		goto out;
1614 	}
1615 
1616 	/* Make the direct mapped domain the default for this group */
1617 	if (group->default_domain)
1618 		iommu_domain_free(group->default_domain);
1619 	group->default_domain = dm_domain;
1620 
1621 	pr_info("Using direct mapping for device %s\n", dev_name(dev));
1622 
1623 	ret = 0;
1624 out:
1625 	mutex_unlock(&group->mutex);
1626 	iommu_group_put(group);
1627 
1628 	return ret;
1629 }
1630 
1631 struct iommu_instance {
1632 	struct list_head list;
1633 	struct fwnode_handle *fwnode;
1634 	const struct iommu_ops *ops;
1635 };
1636 static LIST_HEAD(iommu_instance_list);
1637 static DEFINE_SPINLOCK(iommu_instance_lock);
1638 
1639 void iommu_register_instance(struct fwnode_handle *fwnode,
1640 			     const struct iommu_ops *ops)
1641 {
1642 	struct iommu_instance *iommu = kzalloc(sizeof(*iommu), GFP_KERNEL);
1643 
1644 	if (WARN_ON(!iommu))
1645 		return;
1646 
1647 	of_node_get(to_of_node(fwnode));
1648 	INIT_LIST_HEAD(&iommu->list);
1649 	iommu->fwnode = fwnode;
1650 	iommu->ops = ops;
1651 	spin_lock(&iommu_instance_lock);
1652 	list_add_tail(&iommu->list, &iommu_instance_list);
1653 	spin_unlock(&iommu_instance_lock);
1654 }
1655 
1656 const struct iommu_ops *iommu_get_instance(struct fwnode_handle *fwnode)
1657 {
1658 	struct iommu_instance *instance;
1659 	const struct iommu_ops *ops = NULL;
1660 
1661 	spin_lock(&iommu_instance_lock);
1662 	list_for_each_entry(instance, &iommu_instance_list, list)
1663 		if (instance->fwnode == fwnode) {
1664 			ops = instance->ops;
1665 			break;
1666 		}
1667 	spin_unlock(&iommu_instance_lock);
1668 	return ops;
1669 }
1670 
1671 int iommu_fwspec_init(struct device *dev, struct fwnode_handle *iommu_fwnode,
1672 		      const struct iommu_ops *ops)
1673 {
1674 	struct iommu_fwspec *fwspec = dev->iommu_fwspec;
1675 
1676 	if (fwspec)
1677 		return ops == fwspec->ops ? 0 : -EINVAL;
1678 
1679 	fwspec = kzalloc(sizeof(*fwspec), GFP_KERNEL);
1680 	if (!fwspec)
1681 		return -ENOMEM;
1682 
1683 	of_node_get(to_of_node(iommu_fwnode));
1684 	fwspec->iommu_fwnode = iommu_fwnode;
1685 	fwspec->ops = ops;
1686 	dev->iommu_fwspec = fwspec;
1687 	return 0;
1688 }
1689 EXPORT_SYMBOL_GPL(iommu_fwspec_init);
1690 
1691 void iommu_fwspec_free(struct device *dev)
1692 {
1693 	struct iommu_fwspec *fwspec = dev->iommu_fwspec;
1694 
1695 	if (fwspec) {
1696 		fwnode_handle_put(fwspec->iommu_fwnode);
1697 		kfree(fwspec);
1698 		dev->iommu_fwspec = NULL;
1699 	}
1700 }
1701 EXPORT_SYMBOL_GPL(iommu_fwspec_free);
1702 
1703 int iommu_fwspec_add_ids(struct device *dev, u32 *ids, int num_ids)
1704 {
1705 	struct iommu_fwspec *fwspec = dev->iommu_fwspec;
1706 	size_t size;
1707 	int i;
1708 
1709 	if (!fwspec)
1710 		return -EINVAL;
1711 
1712 	size = offsetof(struct iommu_fwspec, ids[fwspec->num_ids + num_ids]);
1713 	if (size > sizeof(*fwspec)) {
1714 		fwspec = krealloc(dev->iommu_fwspec, size, GFP_KERNEL);
1715 		if (!fwspec)
1716 			return -ENOMEM;
1717 	}
1718 
1719 	for (i = 0; i < num_ids; i++)
1720 		fwspec->ids[fwspec->num_ids + i] = ids[i];
1721 
1722 	fwspec->num_ids += num_ids;
1723 	dev->iommu_fwspec = fwspec;
1724 	return 0;
1725 }
1726 EXPORT_SYMBOL_GPL(iommu_fwspec_add_ids);
1727