xref: /openbmc/linux/drivers/dax/bus.c (revision bbaf1ff0)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright(c) 2017-2018 Intel Corporation. All rights reserved. */
3 #include <linux/memremap.h>
4 #include <linux/device.h>
5 #include <linux/mutex.h>
6 #include <linux/list.h>
7 #include <linux/slab.h>
8 #include <linux/dax.h>
9 #include <linux/io.h>
10 #include "dax-private.h"
11 #include "bus.h"
12 
13 static DEFINE_MUTEX(dax_bus_lock);
14 
15 #define DAX_NAME_LEN 30
16 struct dax_id {
17 	struct list_head list;
18 	char dev_name[DAX_NAME_LEN];
19 };
20 
21 static int dax_bus_uevent(const struct device *dev, struct kobj_uevent_env *env)
22 {
23 	/*
24 	 * We only ever expect to handle device-dax instances, i.e. the
25 	 * @type argument to MODULE_ALIAS_DAX_DEVICE() is always zero
26 	 */
27 	return add_uevent_var(env, "MODALIAS=" DAX_DEVICE_MODALIAS_FMT, 0);
28 }
29 
30 static struct dax_device_driver *to_dax_drv(struct device_driver *drv)
31 {
32 	return container_of(drv, struct dax_device_driver, drv);
33 }
34 
35 static struct dax_id *__dax_match_id(struct dax_device_driver *dax_drv,
36 		const char *dev_name)
37 {
38 	struct dax_id *dax_id;
39 
40 	lockdep_assert_held(&dax_bus_lock);
41 
42 	list_for_each_entry(dax_id, &dax_drv->ids, list)
43 		if (sysfs_streq(dax_id->dev_name, dev_name))
44 			return dax_id;
45 	return NULL;
46 }
47 
48 static int dax_match_id(struct dax_device_driver *dax_drv, struct device *dev)
49 {
50 	int match;
51 
52 	mutex_lock(&dax_bus_lock);
53 	match = !!__dax_match_id(dax_drv, dev_name(dev));
54 	mutex_unlock(&dax_bus_lock);
55 
56 	return match;
57 }
58 
59 static int dax_match_type(struct dax_device_driver *dax_drv, struct device *dev)
60 {
61 	enum dax_driver_type type = DAXDRV_DEVICE_TYPE;
62 	struct dev_dax *dev_dax = to_dev_dax(dev);
63 
64 	if (dev_dax->region->res.flags & IORESOURCE_DAX_KMEM)
65 		type = DAXDRV_KMEM_TYPE;
66 
67 	if (dax_drv->type == type)
68 		return 1;
69 
70 	/* default to device mode if dax_kmem is disabled */
71 	if (dax_drv->type == DAXDRV_DEVICE_TYPE &&
72 	    !IS_ENABLED(CONFIG_DEV_DAX_KMEM))
73 		return 1;
74 
75 	return 0;
76 }
77 
78 enum id_action {
79 	ID_REMOVE,
80 	ID_ADD,
81 };
82 
83 static ssize_t do_id_store(struct device_driver *drv, const char *buf,
84 		size_t count, enum id_action action)
85 {
86 	struct dax_device_driver *dax_drv = to_dax_drv(drv);
87 	unsigned int region_id, id;
88 	char devname[DAX_NAME_LEN];
89 	struct dax_id *dax_id;
90 	ssize_t rc = count;
91 	int fields;
92 
93 	fields = sscanf(buf, "dax%d.%d", &region_id, &id);
94 	if (fields != 2)
95 		return -EINVAL;
96 	sprintf(devname, "dax%d.%d", region_id, id);
97 	if (!sysfs_streq(buf, devname))
98 		return -EINVAL;
99 
100 	mutex_lock(&dax_bus_lock);
101 	dax_id = __dax_match_id(dax_drv, buf);
102 	if (!dax_id) {
103 		if (action == ID_ADD) {
104 			dax_id = kzalloc(sizeof(*dax_id), GFP_KERNEL);
105 			if (dax_id) {
106 				strncpy(dax_id->dev_name, buf, DAX_NAME_LEN);
107 				list_add(&dax_id->list, &dax_drv->ids);
108 			} else
109 				rc = -ENOMEM;
110 		}
111 	} else if (action == ID_REMOVE) {
112 		list_del(&dax_id->list);
113 		kfree(dax_id);
114 	}
115 	mutex_unlock(&dax_bus_lock);
116 
117 	if (rc < 0)
118 		return rc;
119 	if (action == ID_ADD)
120 		rc = driver_attach(drv);
121 	if (rc)
122 		return rc;
123 	return count;
124 }
125 
126 static ssize_t new_id_store(struct device_driver *drv, const char *buf,
127 		size_t count)
128 {
129 	return do_id_store(drv, buf, count, ID_ADD);
130 }
131 static DRIVER_ATTR_WO(new_id);
132 
133 static ssize_t remove_id_store(struct device_driver *drv, const char *buf,
134 		size_t count)
135 {
136 	return do_id_store(drv, buf, count, ID_REMOVE);
137 }
138 static DRIVER_ATTR_WO(remove_id);
139 
140 static struct attribute *dax_drv_attrs[] = {
141 	&driver_attr_new_id.attr,
142 	&driver_attr_remove_id.attr,
143 	NULL,
144 };
145 ATTRIBUTE_GROUPS(dax_drv);
146 
147 static int dax_bus_match(struct device *dev, struct device_driver *drv);
148 
149 /*
150  * Static dax regions are regions created by an external subsystem
151  * nvdimm where a single range is assigned. Its boundaries are by the external
152  * subsystem and are usually limited to one physical memory range. For example,
153  * for PMEM it is usually defined by NVDIMM Namespace boundaries (i.e. a
154  * single contiguous range)
155  *
156  * On dynamic dax regions, the assigned region can be partitioned by dax core
157  * into multiple subdivisions. A subdivision is represented into one
158  * /dev/daxN.M device composed by one or more potentially discontiguous ranges.
159  *
160  * When allocating a dax region, drivers must set whether it's static
161  * (IORESOURCE_DAX_STATIC).  On static dax devices, the @pgmap is pre-assigned
162  * to dax core when calling devm_create_dev_dax(), whereas in dynamic dax
163  * devices it is NULL but afterwards allocated by dax core on device ->probe().
164  * Care is needed to make sure that dynamic dax devices are torn down with a
165  * cleared @pgmap field (see kill_dev_dax()).
166  */
167 static bool is_static(struct dax_region *dax_region)
168 {
169 	return (dax_region->res.flags & IORESOURCE_DAX_STATIC) != 0;
170 }
171 
172 bool static_dev_dax(struct dev_dax *dev_dax)
173 {
174 	return is_static(dev_dax->region);
175 }
176 EXPORT_SYMBOL_GPL(static_dev_dax);
177 
178 static u64 dev_dax_size(struct dev_dax *dev_dax)
179 {
180 	u64 size = 0;
181 	int i;
182 
183 	device_lock_assert(&dev_dax->dev);
184 
185 	for (i = 0; i < dev_dax->nr_range; i++)
186 		size += range_len(&dev_dax->ranges[i].range);
187 
188 	return size;
189 }
190 
191 static int dax_bus_probe(struct device *dev)
192 {
193 	struct dax_device_driver *dax_drv = to_dax_drv(dev->driver);
194 	struct dev_dax *dev_dax = to_dev_dax(dev);
195 	struct dax_region *dax_region = dev_dax->region;
196 	int rc;
197 
198 	if (dev_dax_size(dev_dax) == 0 || dev_dax->id < 0)
199 		return -ENXIO;
200 
201 	rc = dax_drv->probe(dev_dax);
202 
203 	if (rc || is_static(dax_region))
204 		return rc;
205 
206 	/*
207 	 * Track new seed creation only after successful probe of the
208 	 * previous seed.
209 	 */
210 	if (dax_region->seed == dev)
211 		dax_region->seed = NULL;
212 
213 	return 0;
214 }
215 
216 static void dax_bus_remove(struct device *dev)
217 {
218 	struct dax_device_driver *dax_drv = to_dax_drv(dev->driver);
219 	struct dev_dax *dev_dax = to_dev_dax(dev);
220 
221 	if (dax_drv->remove)
222 		dax_drv->remove(dev_dax);
223 }
224 
225 static struct bus_type dax_bus_type = {
226 	.name = "dax",
227 	.uevent = dax_bus_uevent,
228 	.match = dax_bus_match,
229 	.probe = dax_bus_probe,
230 	.remove = dax_bus_remove,
231 	.drv_groups = dax_drv_groups,
232 };
233 
234 static int dax_bus_match(struct device *dev, struct device_driver *drv)
235 {
236 	struct dax_device_driver *dax_drv = to_dax_drv(drv);
237 
238 	if (dax_match_id(dax_drv, dev))
239 		return 1;
240 	return dax_match_type(dax_drv, dev);
241 }
242 
243 /*
244  * Rely on the fact that drvdata is set before the attributes are
245  * registered, and that the attributes are unregistered before drvdata
246  * is cleared to assume that drvdata is always valid.
247  */
248 static ssize_t id_show(struct device *dev,
249 		struct device_attribute *attr, char *buf)
250 {
251 	struct dax_region *dax_region = dev_get_drvdata(dev);
252 
253 	return sprintf(buf, "%d\n", dax_region->id);
254 }
255 static DEVICE_ATTR_RO(id);
256 
257 static ssize_t region_size_show(struct device *dev,
258 		struct device_attribute *attr, char *buf)
259 {
260 	struct dax_region *dax_region = dev_get_drvdata(dev);
261 
262 	return sprintf(buf, "%llu\n", (unsigned long long)
263 			resource_size(&dax_region->res));
264 }
265 static struct device_attribute dev_attr_region_size = __ATTR(size, 0444,
266 		region_size_show, NULL);
267 
268 static ssize_t region_align_show(struct device *dev,
269 		struct device_attribute *attr, char *buf)
270 {
271 	struct dax_region *dax_region = dev_get_drvdata(dev);
272 
273 	return sprintf(buf, "%u\n", dax_region->align);
274 }
275 static struct device_attribute dev_attr_region_align =
276 		__ATTR(align, 0400, region_align_show, NULL);
277 
278 #define for_each_dax_region_resource(dax_region, res) \
279 	for (res = (dax_region)->res.child; res; res = res->sibling)
280 
281 static unsigned long long dax_region_avail_size(struct dax_region *dax_region)
282 {
283 	resource_size_t size = resource_size(&dax_region->res);
284 	struct resource *res;
285 
286 	device_lock_assert(dax_region->dev);
287 
288 	for_each_dax_region_resource(dax_region, res)
289 		size -= resource_size(res);
290 	return size;
291 }
292 
293 static ssize_t available_size_show(struct device *dev,
294 		struct device_attribute *attr, char *buf)
295 {
296 	struct dax_region *dax_region = dev_get_drvdata(dev);
297 	unsigned long long size;
298 
299 	device_lock(dev);
300 	size = dax_region_avail_size(dax_region);
301 	device_unlock(dev);
302 
303 	return sprintf(buf, "%llu\n", size);
304 }
305 static DEVICE_ATTR_RO(available_size);
306 
307 static ssize_t seed_show(struct device *dev,
308 		struct device_attribute *attr, char *buf)
309 {
310 	struct dax_region *dax_region = dev_get_drvdata(dev);
311 	struct device *seed;
312 	ssize_t rc;
313 
314 	if (is_static(dax_region))
315 		return -EINVAL;
316 
317 	device_lock(dev);
318 	seed = dax_region->seed;
319 	rc = sprintf(buf, "%s\n", seed ? dev_name(seed) : "");
320 	device_unlock(dev);
321 
322 	return rc;
323 }
324 static DEVICE_ATTR_RO(seed);
325 
326 static ssize_t create_show(struct device *dev,
327 		struct device_attribute *attr, char *buf)
328 {
329 	struct dax_region *dax_region = dev_get_drvdata(dev);
330 	struct device *youngest;
331 	ssize_t rc;
332 
333 	if (is_static(dax_region))
334 		return -EINVAL;
335 
336 	device_lock(dev);
337 	youngest = dax_region->youngest;
338 	rc = sprintf(buf, "%s\n", youngest ? dev_name(youngest) : "");
339 	device_unlock(dev);
340 
341 	return rc;
342 }
343 
344 static ssize_t create_store(struct device *dev, struct device_attribute *attr,
345 		const char *buf, size_t len)
346 {
347 	struct dax_region *dax_region = dev_get_drvdata(dev);
348 	unsigned long long avail;
349 	ssize_t rc;
350 	int val;
351 
352 	if (is_static(dax_region))
353 		return -EINVAL;
354 
355 	rc = kstrtoint(buf, 0, &val);
356 	if (rc)
357 		return rc;
358 	if (val != 1)
359 		return -EINVAL;
360 
361 	device_lock(dev);
362 	avail = dax_region_avail_size(dax_region);
363 	if (avail == 0)
364 		rc = -ENOSPC;
365 	else {
366 		struct dev_dax_data data = {
367 			.dax_region = dax_region,
368 			.size = 0,
369 			.id = -1,
370 		};
371 		struct dev_dax *dev_dax = devm_create_dev_dax(&data);
372 
373 		if (IS_ERR(dev_dax))
374 			rc = PTR_ERR(dev_dax);
375 		else {
376 			/*
377 			 * In support of crafting multiple new devices
378 			 * simultaneously multiple seeds can be created,
379 			 * but only the first one that has not been
380 			 * successfully bound is tracked as the region
381 			 * seed.
382 			 */
383 			if (!dax_region->seed)
384 				dax_region->seed = &dev_dax->dev;
385 			dax_region->youngest = &dev_dax->dev;
386 			rc = len;
387 		}
388 	}
389 	device_unlock(dev);
390 
391 	return rc;
392 }
393 static DEVICE_ATTR_RW(create);
394 
395 void kill_dev_dax(struct dev_dax *dev_dax)
396 {
397 	struct dax_device *dax_dev = dev_dax->dax_dev;
398 	struct inode *inode = dax_inode(dax_dev);
399 
400 	kill_dax(dax_dev);
401 	unmap_mapping_range(inode->i_mapping, 0, 0, 1);
402 
403 	/*
404 	 * Dynamic dax region have the pgmap allocated via dev_kzalloc()
405 	 * and thus freed by devm. Clear the pgmap to not have stale pgmap
406 	 * ranges on probe() from previous reconfigurations of region devices.
407 	 */
408 	if (!static_dev_dax(dev_dax))
409 		dev_dax->pgmap = NULL;
410 }
411 EXPORT_SYMBOL_GPL(kill_dev_dax);
412 
413 static void trim_dev_dax_range(struct dev_dax *dev_dax)
414 {
415 	int i = dev_dax->nr_range - 1;
416 	struct range *range = &dev_dax->ranges[i].range;
417 	struct dax_region *dax_region = dev_dax->region;
418 
419 	device_lock_assert(dax_region->dev);
420 	dev_dbg(&dev_dax->dev, "delete range[%d]: %#llx:%#llx\n", i,
421 		(unsigned long long)range->start,
422 		(unsigned long long)range->end);
423 
424 	__release_region(&dax_region->res, range->start, range_len(range));
425 	if (--dev_dax->nr_range == 0) {
426 		kfree(dev_dax->ranges);
427 		dev_dax->ranges = NULL;
428 	}
429 }
430 
431 static void free_dev_dax_ranges(struct dev_dax *dev_dax)
432 {
433 	while (dev_dax->nr_range)
434 		trim_dev_dax_range(dev_dax);
435 }
436 
437 static void unregister_dev_dax(void *dev)
438 {
439 	struct dev_dax *dev_dax = to_dev_dax(dev);
440 
441 	dev_dbg(dev, "%s\n", __func__);
442 
443 	kill_dev_dax(dev_dax);
444 	device_del(dev);
445 	free_dev_dax_ranges(dev_dax);
446 	put_device(dev);
447 }
448 
449 /* a return value >= 0 indicates this invocation invalidated the id */
450 static int __free_dev_dax_id(struct dev_dax *dev_dax)
451 {
452 	struct dax_region *dax_region = dev_dax->region;
453 	struct device *dev = &dev_dax->dev;
454 	int rc = dev_dax->id;
455 
456 	device_lock_assert(dev);
457 
458 	if (is_static(dax_region) || dev_dax->id < 0)
459 		return -1;
460 	ida_free(&dax_region->ida, dev_dax->id);
461 	dev_dax->id = -1;
462 	return rc;
463 }
464 
465 static int free_dev_dax_id(struct dev_dax *dev_dax)
466 {
467 	struct device *dev = &dev_dax->dev;
468 	int rc;
469 
470 	device_lock(dev);
471 	rc = __free_dev_dax_id(dev_dax);
472 	device_unlock(dev);
473 	return rc;
474 }
475 
476 static ssize_t delete_store(struct device *dev, struct device_attribute *attr,
477 		const char *buf, size_t len)
478 {
479 	struct dax_region *dax_region = dev_get_drvdata(dev);
480 	struct dev_dax *dev_dax;
481 	struct device *victim;
482 	bool do_del = false;
483 	int rc;
484 
485 	if (is_static(dax_region))
486 		return -EINVAL;
487 
488 	victim = device_find_child_by_name(dax_region->dev, buf);
489 	if (!victim)
490 		return -ENXIO;
491 
492 	device_lock(dev);
493 	device_lock(victim);
494 	dev_dax = to_dev_dax(victim);
495 	if (victim->driver || dev_dax_size(dev_dax))
496 		rc = -EBUSY;
497 	else {
498 		/*
499 		 * Invalidate the device so it does not become active
500 		 * again, but always preserve device-id-0 so that
501 		 * /sys/bus/dax/ is guaranteed to be populated while any
502 		 * dax_region is registered.
503 		 */
504 		if (dev_dax->id > 0) {
505 			do_del = __free_dev_dax_id(dev_dax) >= 0;
506 			rc = len;
507 			if (dax_region->seed == victim)
508 				dax_region->seed = NULL;
509 			if (dax_region->youngest == victim)
510 				dax_region->youngest = NULL;
511 		} else
512 			rc = -EBUSY;
513 	}
514 	device_unlock(victim);
515 
516 	/* won the race to invalidate the device, clean it up */
517 	if (do_del)
518 		devm_release_action(dev, unregister_dev_dax, victim);
519 	device_unlock(dev);
520 	put_device(victim);
521 
522 	return rc;
523 }
524 static DEVICE_ATTR_WO(delete);
525 
526 static umode_t dax_region_visible(struct kobject *kobj, struct attribute *a,
527 		int n)
528 {
529 	struct device *dev = container_of(kobj, struct device, kobj);
530 	struct dax_region *dax_region = dev_get_drvdata(dev);
531 
532 	if (is_static(dax_region))
533 		if (a == &dev_attr_available_size.attr
534 				|| a == &dev_attr_create.attr
535 				|| a == &dev_attr_seed.attr
536 				|| a == &dev_attr_delete.attr)
537 			return 0;
538 	return a->mode;
539 }
540 
541 static struct attribute *dax_region_attributes[] = {
542 	&dev_attr_available_size.attr,
543 	&dev_attr_region_size.attr,
544 	&dev_attr_region_align.attr,
545 	&dev_attr_create.attr,
546 	&dev_attr_seed.attr,
547 	&dev_attr_delete.attr,
548 	&dev_attr_id.attr,
549 	NULL,
550 };
551 
552 static const struct attribute_group dax_region_attribute_group = {
553 	.name = "dax_region",
554 	.attrs = dax_region_attributes,
555 	.is_visible = dax_region_visible,
556 };
557 
558 static const struct attribute_group *dax_region_attribute_groups[] = {
559 	&dax_region_attribute_group,
560 	NULL,
561 };
562 
563 static void dax_region_free(struct kref *kref)
564 {
565 	struct dax_region *dax_region;
566 
567 	dax_region = container_of(kref, struct dax_region, kref);
568 	kfree(dax_region);
569 }
570 
571 void dax_region_put(struct dax_region *dax_region)
572 {
573 	kref_put(&dax_region->kref, dax_region_free);
574 }
575 EXPORT_SYMBOL_GPL(dax_region_put);
576 
577 static void dax_region_unregister(void *region)
578 {
579 	struct dax_region *dax_region = region;
580 
581 	sysfs_remove_groups(&dax_region->dev->kobj,
582 			dax_region_attribute_groups);
583 	dax_region_put(dax_region);
584 }
585 
586 struct dax_region *alloc_dax_region(struct device *parent, int region_id,
587 		struct range *range, int target_node, unsigned int align,
588 		unsigned long flags)
589 {
590 	struct dax_region *dax_region;
591 
592 	/*
593 	 * The DAX core assumes that it can store its private data in
594 	 * parent->driver_data. This WARN is a reminder / safeguard for
595 	 * developers of device-dax drivers.
596 	 */
597 	if (dev_get_drvdata(parent)) {
598 		dev_WARN(parent, "dax core failed to setup private data\n");
599 		return NULL;
600 	}
601 
602 	if (!IS_ALIGNED(range->start, align)
603 			|| !IS_ALIGNED(range_len(range), align))
604 		return NULL;
605 
606 	dax_region = kzalloc(sizeof(*dax_region), GFP_KERNEL);
607 	if (!dax_region)
608 		return NULL;
609 
610 	dev_set_drvdata(parent, dax_region);
611 	kref_init(&dax_region->kref);
612 	dax_region->id = region_id;
613 	dax_region->align = align;
614 	dax_region->dev = parent;
615 	dax_region->target_node = target_node;
616 	ida_init(&dax_region->ida);
617 	dax_region->res = (struct resource) {
618 		.start = range->start,
619 		.end = range->end,
620 		.flags = IORESOURCE_MEM | flags,
621 	};
622 
623 	if (sysfs_create_groups(&parent->kobj, dax_region_attribute_groups)) {
624 		kfree(dax_region);
625 		return NULL;
626 	}
627 
628 	kref_get(&dax_region->kref);
629 	if (devm_add_action_or_reset(parent, dax_region_unregister, dax_region))
630 		return NULL;
631 	return dax_region;
632 }
633 EXPORT_SYMBOL_GPL(alloc_dax_region);
634 
635 static void dax_mapping_release(struct device *dev)
636 {
637 	struct dax_mapping *mapping = to_dax_mapping(dev);
638 	struct dev_dax *dev_dax = to_dev_dax(dev->parent);
639 
640 	ida_free(&dev_dax->ida, mapping->id);
641 	kfree(mapping);
642 }
643 
644 static void unregister_dax_mapping(void *data)
645 {
646 	struct device *dev = data;
647 	struct dax_mapping *mapping = to_dax_mapping(dev);
648 	struct dev_dax *dev_dax = to_dev_dax(dev->parent);
649 	struct dax_region *dax_region = dev_dax->region;
650 
651 	dev_dbg(dev, "%s\n", __func__);
652 
653 	device_lock_assert(dax_region->dev);
654 
655 	dev_dax->ranges[mapping->range_id].mapping = NULL;
656 	mapping->range_id = -1;
657 
658 	device_del(dev);
659 	put_device(dev);
660 }
661 
662 static struct dev_dax_range *get_dax_range(struct device *dev)
663 {
664 	struct dax_mapping *mapping = to_dax_mapping(dev);
665 	struct dev_dax *dev_dax = to_dev_dax(dev->parent);
666 	struct dax_region *dax_region = dev_dax->region;
667 
668 	device_lock(dax_region->dev);
669 	if (mapping->range_id < 0) {
670 		device_unlock(dax_region->dev);
671 		return NULL;
672 	}
673 
674 	return &dev_dax->ranges[mapping->range_id];
675 }
676 
677 static void put_dax_range(struct dev_dax_range *dax_range)
678 {
679 	struct dax_mapping *mapping = dax_range->mapping;
680 	struct dev_dax *dev_dax = to_dev_dax(mapping->dev.parent);
681 	struct dax_region *dax_region = dev_dax->region;
682 
683 	device_unlock(dax_region->dev);
684 }
685 
686 static ssize_t start_show(struct device *dev,
687 		struct device_attribute *attr, char *buf)
688 {
689 	struct dev_dax_range *dax_range;
690 	ssize_t rc;
691 
692 	dax_range = get_dax_range(dev);
693 	if (!dax_range)
694 		return -ENXIO;
695 	rc = sprintf(buf, "%#llx\n", dax_range->range.start);
696 	put_dax_range(dax_range);
697 
698 	return rc;
699 }
700 static DEVICE_ATTR(start, 0400, start_show, NULL);
701 
702 static ssize_t end_show(struct device *dev,
703 		struct device_attribute *attr, char *buf)
704 {
705 	struct dev_dax_range *dax_range;
706 	ssize_t rc;
707 
708 	dax_range = get_dax_range(dev);
709 	if (!dax_range)
710 		return -ENXIO;
711 	rc = sprintf(buf, "%#llx\n", dax_range->range.end);
712 	put_dax_range(dax_range);
713 
714 	return rc;
715 }
716 static DEVICE_ATTR(end, 0400, end_show, NULL);
717 
718 static ssize_t pgoff_show(struct device *dev,
719 		struct device_attribute *attr, char *buf)
720 {
721 	struct dev_dax_range *dax_range;
722 	ssize_t rc;
723 
724 	dax_range = get_dax_range(dev);
725 	if (!dax_range)
726 		return -ENXIO;
727 	rc = sprintf(buf, "%#lx\n", dax_range->pgoff);
728 	put_dax_range(dax_range);
729 
730 	return rc;
731 }
732 static DEVICE_ATTR(page_offset, 0400, pgoff_show, NULL);
733 
734 static struct attribute *dax_mapping_attributes[] = {
735 	&dev_attr_start.attr,
736 	&dev_attr_end.attr,
737 	&dev_attr_page_offset.attr,
738 	NULL,
739 };
740 
741 static const struct attribute_group dax_mapping_attribute_group = {
742 	.attrs = dax_mapping_attributes,
743 };
744 
745 static const struct attribute_group *dax_mapping_attribute_groups[] = {
746 	&dax_mapping_attribute_group,
747 	NULL,
748 };
749 
750 static struct device_type dax_mapping_type = {
751 	.release = dax_mapping_release,
752 	.groups = dax_mapping_attribute_groups,
753 };
754 
755 static int devm_register_dax_mapping(struct dev_dax *dev_dax, int range_id)
756 {
757 	struct dax_region *dax_region = dev_dax->region;
758 	struct dax_mapping *mapping;
759 	struct device *dev;
760 	int rc;
761 
762 	device_lock_assert(dax_region->dev);
763 
764 	if (dev_WARN_ONCE(&dev_dax->dev, !dax_region->dev->driver,
765 				"region disabled\n"))
766 		return -ENXIO;
767 
768 	mapping = kzalloc(sizeof(*mapping), GFP_KERNEL);
769 	if (!mapping)
770 		return -ENOMEM;
771 	mapping->range_id = range_id;
772 	mapping->id = ida_alloc(&dev_dax->ida, GFP_KERNEL);
773 	if (mapping->id < 0) {
774 		kfree(mapping);
775 		return -ENOMEM;
776 	}
777 	dev_dax->ranges[range_id].mapping = mapping;
778 	dev = &mapping->dev;
779 	device_initialize(dev);
780 	dev->parent = &dev_dax->dev;
781 	dev->type = &dax_mapping_type;
782 	dev_set_name(dev, "mapping%d", mapping->id);
783 	rc = device_add(dev);
784 	if (rc) {
785 		put_device(dev);
786 		return rc;
787 	}
788 
789 	rc = devm_add_action_or_reset(dax_region->dev, unregister_dax_mapping,
790 			dev);
791 	if (rc)
792 		return rc;
793 	return 0;
794 }
795 
796 static int alloc_dev_dax_range(struct dev_dax *dev_dax, u64 start,
797 		resource_size_t size)
798 {
799 	struct dax_region *dax_region = dev_dax->region;
800 	struct resource *res = &dax_region->res;
801 	struct device *dev = &dev_dax->dev;
802 	struct dev_dax_range *ranges;
803 	unsigned long pgoff = 0;
804 	struct resource *alloc;
805 	int i, rc;
806 
807 	device_lock_assert(dax_region->dev);
808 
809 	/* handle the seed alloc special case */
810 	if (!size) {
811 		if (dev_WARN_ONCE(dev, dev_dax->nr_range,
812 					"0-size allocation must be first\n"))
813 			return -EBUSY;
814 		/* nr_range == 0 is elsewhere special cased as 0-size device */
815 		return 0;
816 	}
817 
818 	alloc = __request_region(res, start, size, dev_name(dev), 0);
819 	if (!alloc)
820 		return -ENOMEM;
821 
822 	ranges = krealloc(dev_dax->ranges, sizeof(*ranges)
823 			* (dev_dax->nr_range + 1), GFP_KERNEL);
824 	if (!ranges) {
825 		__release_region(res, alloc->start, resource_size(alloc));
826 		return -ENOMEM;
827 	}
828 
829 	for (i = 0; i < dev_dax->nr_range; i++)
830 		pgoff += PHYS_PFN(range_len(&ranges[i].range));
831 	dev_dax->ranges = ranges;
832 	ranges[dev_dax->nr_range++] = (struct dev_dax_range) {
833 		.pgoff = pgoff,
834 		.range = {
835 			.start = alloc->start,
836 			.end = alloc->end,
837 		},
838 	};
839 
840 	dev_dbg(dev, "alloc range[%d]: %pa:%pa\n", dev_dax->nr_range - 1,
841 			&alloc->start, &alloc->end);
842 	/*
843 	 * A dev_dax instance must be registered before mapping device
844 	 * children can be added. Defer to devm_create_dev_dax() to add
845 	 * the initial mapping device.
846 	 */
847 	if (!device_is_registered(&dev_dax->dev))
848 		return 0;
849 
850 	rc = devm_register_dax_mapping(dev_dax, dev_dax->nr_range - 1);
851 	if (rc)
852 		trim_dev_dax_range(dev_dax);
853 
854 	return rc;
855 }
856 
857 static int adjust_dev_dax_range(struct dev_dax *dev_dax, struct resource *res, resource_size_t size)
858 {
859 	int last_range = dev_dax->nr_range - 1;
860 	struct dev_dax_range *dax_range = &dev_dax->ranges[last_range];
861 	struct dax_region *dax_region = dev_dax->region;
862 	bool is_shrink = resource_size(res) > size;
863 	struct range *range = &dax_range->range;
864 	struct device *dev = &dev_dax->dev;
865 	int rc;
866 
867 	device_lock_assert(dax_region->dev);
868 
869 	if (dev_WARN_ONCE(dev, !size, "deletion is handled by dev_dax_shrink\n"))
870 		return -EINVAL;
871 
872 	rc = adjust_resource(res, range->start, size);
873 	if (rc)
874 		return rc;
875 
876 	*range = (struct range) {
877 		.start = range->start,
878 		.end = range->start + size - 1,
879 	};
880 
881 	dev_dbg(dev, "%s range[%d]: %#llx:%#llx\n", is_shrink ? "shrink" : "extend",
882 			last_range, (unsigned long long) range->start,
883 			(unsigned long long) range->end);
884 
885 	return 0;
886 }
887 
888 static ssize_t size_show(struct device *dev,
889 		struct device_attribute *attr, char *buf)
890 {
891 	struct dev_dax *dev_dax = to_dev_dax(dev);
892 	unsigned long long size;
893 
894 	device_lock(dev);
895 	size = dev_dax_size(dev_dax);
896 	device_unlock(dev);
897 
898 	return sprintf(buf, "%llu\n", size);
899 }
900 
901 static bool alloc_is_aligned(struct dev_dax *dev_dax, resource_size_t size)
902 {
903 	/*
904 	 * The minimum mapping granularity for a device instance is a
905 	 * single subsection, unless the arch says otherwise.
906 	 */
907 	return IS_ALIGNED(size, max_t(unsigned long, dev_dax->align, memremap_compat_align()));
908 }
909 
910 static int dev_dax_shrink(struct dev_dax *dev_dax, resource_size_t size)
911 {
912 	resource_size_t to_shrink = dev_dax_size(dev_dax) - size;
913 	struct dax_region *dax_region = dev_dax->region;
914 	struct device *dev = &dev_dax->dev;
915 	int i;
916 
917 	for (i = dev_dax->nr_range - 1; i >= 0; i--) {
918 		struct range *range = &dev_dax->ranges[i].range;
919 		struct dax_mapping *mapping = dev_dax->ranges[i].mapping;
920 		struct resource *adjust = NULL, *res;
921 		resource_size_t shrink;
922 
923 		shrink = min_t(u64, to_shrink, range_len(range));
924 		if (shrink >= range_len(range)) {
925 			devm_release_action(dax_region->dev,
926 					unregister_dax_mapping, &mapping->dev);
927 			trim_dev_dax_range(dev_dax);
928 			to_shrink -= shrink;
929 			if (!to_shrink)
930 				break;
931 			continue;
932 		}
933 
934 		for_each_dax_region_resource(dax_region, res)
935 			if (strcmp(res->name, dev_name(dev)) == 0
936 					&& res->start == range->start) {
937 				adjust = res;
938 				break;
939 			}
940 
941 		if (dev_WARN_ONCE(dev, !adjust || i != dev_dax->nr_range - 1,
942 					"failed to find matching resource\n"))
943 			return -ENXIO;
944 		return adjust_dev_dax_range(dev_dax, adjust, range_len(range)
945 				- shrink);
946 	}
947 	return 0;
948 }
949 
950 /*
951  * Only allow adjustments that preserve the relative pgoff of existing
952  * allocations. I.e. the dev_dax->ranges array is ordered by increasing pgoff.
953  */
954 static bool adjust_ok(struct dev_dax *dev_dax, struct resource *res)
955 {
956 	struct dev_dax_range *last;
957 	int i;
958 
959 	if (dev_dax->nr_range == 0)
960 		return false;
961 	if (strcmp(res->name, dev_name(&dev_dax->dev)) != 0)
962 		return false;
963 	last = &dev_dax->ranges[dev_dax->nr_range - 1];
964 	if (last->range.start != res->start || last->range.end != res->end)
965 		return false;
966 	for (i = 0; i < dev_dax->nr_range - 1; i++) {
967 		struct dev_dax_range *dax_range = &dev_dax->ranges[i];
968 
969 		if (dax_range->pgoff > last->pgoff)
970 			return false;
971 	}
972 
973 	return true;
974 }
975 
976 static ssize_t dev_dax_resize(struct dax_region *dax_region,
977 		struct dev_dax *dev_dax, resource_size_t size)
978 {
979 	resource_size_t avail = dax_region_avail_size(dax_region), to_alloc;
980 	resource_size_t dev_size = dev_dax_size(dev_dax);
981 	struct resource *region_res = &dax_region->res;
982 	struct device *dev = &dev_dax->dev;
983 	struct resource *res, *first;
984 	resource_size_t alloc = 0;
985 	int rc;
986 
987 	if (dev->driver)
988 		return -EBUSY;
989 	if (size == dev_size)
990 		return 0;
991 	if (size > dev_size && size - dev_size > avail)
992 		return -ENOSPC;
993 	if (size < dev_size)
994 		return dev_dax_shrink(dev_dax, size);
995 
996 	to_alloc = size - dev_size;
997 	if (dev_WARN_ONCE(dev, !alloc_is_aligned(dev_dax, to_alloc),
998 			"resize of %pa misaligned\n", &to_alloc))
999 		return -ENXIO;
1000 
1001 	/*
1002 	 * Expand the device into the unused portion of the region. This
1003 	 * may involve adjusting the end of an existing resource, or
1004 	 * allocating a new resource.
1005 	 */
1006 retry:
1007 	first = region_res->child;
1008 	if (!first)
1009 		return alloc_dev_dax_range(dev_dax, dax_region->res.start, to_alloc);
1010 
1011 	rc = -ENOSPC;
1012 	for (res = first; res; res = res->sibling) {
1013 		struct resource *next = res->sibling;
1014 
1015 		/* space at the beginning of the region */
1016 		if (res == first && res->start > dax_region->res.start) {
1017 			alloc = min(res->start - dax_region->res.start, to_alloc);
1018 			rc = alloc_dev_dax_range(dev_dax, dax_region->res.start, alloc);
1019 			break;
1020 		}
1021 
1022 		alloc = 0;
1023 		/* space between allocations */
1024 		if (next && next->start > res->end + 1)
1025 			alloc = min(next->start - (res->end + 1), to_alloc);
1026 
1027 		/* space at the end of the region */
1028 		if (!alloc && !next && res->end < region_res->end)
1029 			alloc = min(region_res->end - res->end, to_alloc);
1030 
1031 		if (!alloc)
1032 			continue;
1033 
1034 		if (adjust_ok(dev_dax, res)) {
1035 			rc = adjust_dev_dax_range(dev_dax, res, resource_size(res) + alloc);
1036 			break;
1037 		}
1038 		rc = alloc_dev_dax_range(dev_dax, res->end + 1, alloc);
1039 		break;
1040 	}
1041 	if (rc)
1042 		return rc;
1043 	to_alloc -= alloc;
1044 	if (to_alloc)
1045 		goto retry;
1046 	return 0;
1047 }
1048 
1049 static ssize_t size_store(struct device *dev, struct device_attribute *attr,
1050 		const char *buf, size_t len)
1051 {
1052 	ssize_t rc;
1053 	unsigned long long val;
1054 	struct dev_dax *dev_dax = to_dev_dax(dev);
1055 	struct dax_region *dax_region = dev_dax->region;
1056 
1057 	rc = kstrtoull(buf, 0, &val);
1058 	if (rc)
1059 		return rc;
1060 
1061 	if (!alloc_is_aligned(dev_dax, val)) {
1062 		dev_dbg(dev, "%s: size: %lld misaligned\n", __func__, val);
1063 		return -EINVAL;
1064 	}
1065 
1066 	device_lock(dax_region->dev);
1067 	if (!dax_region->dev->driver) {
1068 		device_unlock(dax_region->dev);
1069 		return -ENXIO;
1070 	}
1071 	device_lock(dev);
1072 	rc = dev_dax_resize(dax_region, dev_dax, val);
1073 	device_unlock(dev);
1074 	device_unlock(dax_region->dev);
1075 
1076 	return rc == 0 ? len : rc;
1077 }
1078 static DEVICE_ATTR_RW(size);
1079 
1080 static ssize_t range_parse(const char *opt, size_t len, struct range *range)
1081 {
1082 	unsigned long long addr = 0;
1083 	char *start, *end, *str;
1084 	ssize_t rc = -EINVAL;
1085 
1086 	str = kstrdup(opt, GFP_KERNEL);
1087 	if (!str)
1088 		return rc;
1089 
1090 	end = str;
1091 	start = strsep(&end, "-");
1092 	if (!start || !end)
1093 		goto err;
1094 
1095 	rc = kstrtoull(start, 16, &addr);
1096 	if (rc)
1097 		goto err;
1098 	range->start = addr;
1099 
1100 	rc = kstrtoull(end, 16, &addr);
1101 	if (rc)
1102 		goto err;
1103 	range->end = addr;
1104 
1105 err:
1106 	kfree(str);
1107 	return rc;
1108 }
1109 
1110 static ssize_t mapping_store(struct device *dev, struct device_attribute *attr,
1111 		const char *buf, size_t len)
1112 {
1113 	struct dev_dax *dev_dax = to_dev_dax(dev);
1114 	struct dax_region *dax_region = dev_dax->region;
1115 	size_t to_alloc;
1116 	struct range r;
1117 	ssize_t rc;
1118 
1119 	rc = range_parse(buf, len, &r);
1120 	if (rc)
1121 		return rc;
1122 
1123 	rc = -ENXIO;
1124 	device_lock(dax_region->dev);
1125 	if (!dax_region->dev->driver) {
1126 		device_unlock(dax_region->dev);
1127 		return rc;
1128 	}
1129 	device_lock(dev);
1130 
1131 	to_alloc = range_len(&r);
1132 	if (alloc_is_aligned(dev_dax, to_alloc))
1133 		rc = alloc_dev_dax_range(dev_dax, r.start, to_alloc);
1134 	device_unlock(dev);
1135 	device_unlock(dax_region->dev);
1136 
1137 	return rc == 0 ? len : rc;
1138 }
1139 static DEVICE_ATTR_WO(mapping);
1140 
1141 static ssize_t align_show(struct device *dev,
1142 		struct device_attribute *attr, char *buf)
1143 {
1144 	struct dev_dax *dev_dax = to_dev_dax(dev);
1145 
1146 	return sprintf(buf, "%d\n", dev_dax->align);
1147 }
1148 
1149 static ssize_t dev_dax_validate_align(struct dev_dax *dev_dax)
1150 {
1151 	struct device *dev = &dev_dax->dev;
1152 	int i;
1153 
1154 	for (i = 0; i < dev_dax->nr_range; i++) {
1155 		size_t len = range_len(&dev_dax->ranges[i].range);
1156 
1157 		if (!alloc_is_aligned(dev_dax, len)) {
1158 			dev_dbg(dev, "%s: align %u invalid for range %d\n",
1159 				__func__, dev_dax->align, i);
1160 			return -EINVAL;
1161 		}
1162 	}
1163 
1164 	return 0;
1165 }
1166 
1167 static ssize_t align_store(struct device *dev, struct device_attribute *attr,
1168 		const char *buf, size_t len)
1169 {
1170 	struct dev_dax *dev_dax = to_dev_dax(dev);
1171 	struct dax_region *dax_region = dev_dax->region;
1172 	unsigned long val, align_save;
1173 	ssize_t rc;
1174 
1175 	rc = kstrtoul(buf, 0, &val);
1176 	if (rc)
1177 		return -ENXIO;
1178 
1179 	if (!dax_align_valid(val))
1180 		return -EINVAL;
1181 
1182 	device_lock(dax_region->dev);
1183 	if (!dax_region->dev->driver) {
1184 		device_unlock(dax_region->dev);
1185 		return -ENXIO;
1186 	}
1187 
1188 	device_lock(dev);
1189 	if (dev->driver) {
1190 		rc = -EBUSY;
1191 		goto out_unlock;
1192 	}
1193 
1194 	align_save = dev_dax->align;
1195 	dev_dax->align = val;
1196 	rc = dev_dax_validate_align(dev_dax);
1197 	if (rc)
1198 		dev_dax->align = align_save;
1199 out_unlock:
1200 	device_unlock(dev);
1201 	device_unlock(dax_region->dev);
1202 	return rc == 0 ? len : rc;
1203 }
1204 static DEVICE_ATTR_RW(align);
1205 
1206 static int dev_dax_target_node(struct dev_dax *dev_dax)
1207 {
1208 	struct dax_region *dax_region = dev_dax->region;
1209 
1210 	return dax_region->target_node;
1211 }
1212 
1213 static ssize_t target_node_show(struct device *dev,
1214 		struct device_attribute *attr, char *buf)
1215 {
1216 	struct dev_dax *dev_dax = to_dev_dax(dev);
1217 
1218 	return sprintf(buf, "%d\n", dev_dax_target_node(dev_dax));
1219 }
1220 static DEVICE_ATTR_RO(target_node);
1221 
1222 static ssize_t resource_show(struct device *dev,
1223 		struct device_attribute *attr, char *buf)
1224 {
1225 	struct dev_dax *dev_dax = to_dev_dax(dev);
1226 	struct dax_region *dax_region = dev_dax->region;
1227 	unsigned long long start;
1228 
1229 	if (dev_dax->nr_range < 1)
1230 		start = dax_region->res.start;
1231 	else
1232 		start = dev_dax->ranges[0].range.start;
1233 
1234 	return sprintf(buf, "%#llx\n", start);
1235 }
1236 static DEVICE_ATTR(resource, 0400, resource_show, NULL);
1237 
1238 static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
1239 		char *buf)
1240 {
1241 	/*
1242 	 * We only ever expect to handle device-dax instances, i.e. the
1243 	 * @type argument to MODULE_ALIAS_DAX_DEVICE() is always zero
1244 	 */
1245 	return sprintf(buf, DAX_DEVICE_MODALIAS_FMT "\n", 0);
1246 }
1247 static DEVICE_ATTR_RO(modalias);
1248 
1249 static ssize_t numa_node_show(struct device *dev,
1250 		struct device_attribute *attr, char *buf)
1251 {
1252 	return sprintf(buf, "%d\n", dev_to_node(dev));
1253 }
1254 static DEVICE_ATTR_RO(numa_node);
1255 
1256 static umode_t dev_dax_visible(struct kobject *kobj, struct attribute *a, int n)
1257 {
1258 	struct device *dev = container_of(kobj, struct device, kobj);
1259 	struct dev_dax *dev_dax = to_dev_dax(dev);
1260 	struct dax_region *dax_region = dev_dax->region;
1261 
1262 	if (a == &dev_attr_target_node.attr && dev_dax_target_node(dev_dax) < 0)
1263 		return 0;
1264 	if (a == &dev_attr_numa_node.attr && !IS_ENABLED(CONFIG_NUMA))
1265 		return 0;
1266 	if (a == &dev_attr_mapping.attr && is_static(dax_region))
1267 		return 0;
1268 	if ((a == &dev_attr_align.attr ||
1269 	     a == &dev_attr_size.attr) && is_static(dax_region))
1270 		return 0444;
1271 	return a->mode;
1272 }
1273 
1274 static struct attribute *dev_dax_attributes[] = {
1275 	&dev_attr_modalias.attr,
1276 	&dev_attr_size.attr,
1277 	&dev_attr_mapping.attr,
1278 	&dev_attr_target_node.attr,
1279 	&dev_attr_align.attr,
1280 	&dev_attr_resource.attr,
1281 	&dev_attr_numa_node.attr,
1282 	NULL,
1283 };
1284 
1285 static const struct attribute_group dev_dax_attribute_group = {
1286 	.attrs = dev_dax_attributes,
1287 	.is_visible = dev_dax_visible,
1288 };
1289 
1290 static const struct attribute_group *dax_attribute_groups[] = {
1291 	&dev_dax_attribute_group,
1292 	NULL,
1293 };
1294 
1295 static void dev_dax_release(struct device *dev)
1296 {
1297 	struct dev_dax *dev_dax = to_dev_dax(dev);
1298 	struct dax_region *dax_region = dev_dax->region;
1299 	struct dax_device *dax_dev = dev_dax->dax_dev;
1300 
1301 	put_dax(dax_dev);
1302 	free_dev_dax_id(dev_dax);
1303 	dax_region_put(dax_region);
1304 	kfree(dev_dax->pgmap);
1305 	kfree(dev_dax);
1306 }
1307 
1308 static const struct device_type dev_dax_type = {
1309 	.release = dev_dax_release,
1310 	.groups = dax_attribute_groups,
1311 };
1312 
1313 struct dev_dax *devm_create_dev_dax(struct dev_dax_data *data)
1314 {
1315 	struct dax_region *dax_region = data->dax_region;
1316 	struct device *parent = dax_region->dev;
1317 	struct dax_device *dax_dev;
1318 	struct dev_dax *dev_dax;
1319 	struct inode *inode;
1320 	struct device *dev;
1321 	int rc;
1322 
1323 	dev_dax = kzalloc(sizeof(*dev_dax), GFP_KERNEL);
1324 	if (!dev_dax)
1325 		return ERR_PTR(-ENOMEM);
1326 
1327 	if (is_static(dax_region)) {
1328 		if (dev_WARN_ONCE(parent, data->id < 0,
1329 				"dynamic id specified to static region\n")) {
1330 			rc = -EINVAL;
1331 			goto err_id;
1332 		}
1333 
1334 		dev_dax->id = data->id;
1335 	} else {
1336 		if (dev_WARN_ONCE(parent, data->id >= 0,
1337 				"static id specified to dynamic region\n")) {
1338 			rc = -EINVAL;
1339 			goto err_id;
1340 		}
1341 
1342 		rc = ida_alloc(&dax_region->ida, GFP_KERNEL);
1343 		if (rc < 0)
1344 			goto err_id;
1345 		dev_dax->id = rc;
1346 	}
1347 
1348 	dev_dax->region = dax_region;
1349 	dev = &dev_dax->dev;
1350 	device_initialize(dev);
1351 	dev_set_name(dev, "dax%d.%d", dax_region->id, dev_dax->id);
1352 
1353 	rc = alloc_dev_dax_range(dev_dax, dax_region->res.start, data->size);
1354 	if (rc)
1355 		goto err_range;
1356 
1357 	if (data->pgmap) {
1358 		dev_WARN_ONCE(parent, !is_static(dax_region),
1359 			"custom dev_pagemap requires a static dax_region\n");
1360 
1361 		dev_dax->pgmap = kmemdup(data->pgmap,
1362 				sizeof(struct dev_pagemap), GFP_KERNEL);
1363 		if (!dev_dax->pgmap) {
1364 			rc = -ENOMEM;
1365 			goto err_pgmap;
1366 		}
1367 	}
1368 
1369 	/*
1370 	 * No dax_operations since there is no access to this device outside of
1371 	 * mmap of the resulting character device.
1372 	 */
1373 	dax_dev = alloc_dax(dev_dax, NULL);
1374 	if (IS_ERR(dax_dev)) {
1375 		rc = PTR_ERR(dax_dev);
1376 		goto err_alloc_dax;
1377 	}
1378 	set_dax_synchronous(dax_dev);
1379 	set_dax_nocache(dax_dev);
1380 	set_dax_nomc(dax_dev);
1381 
1382 	/* a device_dax instance is dead while the driver is not attached */
1383 	kill_dax(dax_dev);
1384 
1385 	dev_dax->dax_dev = dax_dev;
1386 	dev_dax->target_node = dax_region->target_node;
1387 	dev_dax->align = dax_region->align;
1388 	ida_init(&dev_dax->ida);
1389 	kref_get(&dax_region->kref);
1390 
1391 	inode = dax_inode(dax_dev);
1392 	dev->devt = inode->i_rdev;
1393 	dev->bus = &dax_bus_type;
1394 	dev->parent = parent;
1395 	dev->type = &dev_dax_type;
1396 
1397 	rc = device_add(dev);
1398 	if (rc) {
1399 		kill_dev_dax(dev_dax);
1400 		put_device(dev);
1401 		return ERR_PTR(rc);
1402 	}
1403 
1404 	rc = devm_add_action_or_reset(dax_region->dev, unregister_dev_dax, dev);
1405 	if (rc)
1406 		return ERR_PTR(rc);
1407 
1408 	/* register mapping device for the initial allocation range */
1409 	if (dev_dax->nr_range && range_len(&dev_dax->ranges[0].range)) {
1410 		rc = devm_register_dax_mapping(dev_dax, 0);
1411 		if (rc)
1412 			return ERR_PTR(rc);
1413 	}
1414 
1415 	return dev_dax;
1416 
1417 err_alloc_dax:
1418 	kfree(dev_dax->pgmap);
1419 err_pgmap:
1420 	free_dev_dax_ranges(dev_dax);
1421 err_range:
1422 	free_dev_dax_id(dev_dax);
1423 err_id:
1424 	kfree(dev_dax);
1425 
1426 	return ERR_PTR(rc);
1427 }
1428 EXPORT_SYMBOL_GPL(devm_create_dev_dax);
1429 
1430 int __dax_driver_register(struct dax_device_driver *dax_drv,
1431 		struct module *module, const char *mod_name)
1432 {
1433 	struct device_driver *drv = &dax_drv->drv;
1434 
1435 	/*
1436 	 * dax_bus_probe() calls dax_drv->probe() unconditionally.
1437 	 * So better be safe than sorry and ensure it is provided.
1438 	 */
1439 	if (!dax_drv->probe)
1440 		return -EINVAL;
1441 
1442 	INIT_LIST_HEAD(&dax_drv->ids);
1443 	drv->owner = module;
1444 	drv->name = mod_name;
1445 	drv->mod_name = mod_name;
1446 	drv->bus = &dax_bus_type;
1447 
1448 	return driver_register(drv);
1449 }
1450 EXPORT_SYMBOL_GPL(__dax_driver_register);
1451 
1452 void dax_driver_unregister(struct dax_device_driver *dax_drv)
1453 {
1454 	struct device_driver *drv = &dax_drv->drv;
1455 	struct dax_id *dax_id, *_id;
1456 
1457 	mutex_lock(&dax_bus_lock);
1458 	list_for_each_entry_safe(dax_id, _id, &dax_drv->ids, list) {
1459 		list_del(&dax_id->list);
1460 		kfree(dax_id);
1461 	}
1462 	mutex_unlock(&dax_bus_lock);
1463 	driver_unregister(drv);
1464 }
1465 EXPORT_SYMBOL_GPL(dax_driver_unregister);
1466 
1467 int __init dax_bus_init(void)
1468 {
1469 	return bus_register(&dax_bus_type);
1470 }
1471 
1472 void __exit dax_bus_exit(void)
1473 {
1474 	bus_unregister(&dax_bus_type);
1475 }
1476