Loading...
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
13static DEFINE_MUTEX(dax_bus_lock);
14
15#define DAX_NAME_LEN 30
16struct dax_id {
17 struct list_head list;
18 char dev_name[DAX_NAME_LEN];
19};
20
21static int dax_bus_uevent(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
30static struct dax_device_driver *to_dax_drv(struct device_driver *drv)
31{
32 return container_of(drv, struct dax_device_driver, drv);
33}
34
35static 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
48static 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
59enum id_action {
60 ID_REMOVE,
61 ID_ADD,
62};
63
64static ssize_t do_id_store(struct device_driver *drv, const char *buf,
65 size_t count, enum id_action action)
66{
67 struct dax_device_driver *dax_drv = to_dax_drv(drv);
68 unsigned int region_id, id;
69 char devname[DAX_NAME_LEN];
70 struct dax_id *dax_id;
71 ssize_t rc = count;
72 int fields;
73
74 fields = sscanf(buf, "dax%d.%d", ®ion_id, &id);
75 if (fields != 2)
76 return -EINVAL;
77 sprintf(devname, "dax%d.%d", region_id, id);
78 if (!sysfs_streq(buf, devname))
79 return -EINVAL;
80
81 mutex_lock(&dax_bus_lock);
82 dax_id = __dax_match_id(dax_drv, buf);
83 if (!dax_id) {
84 if (action == ID_ADD) {
85 dax_id = kzalloc(sizeof(*dax_id), GFP_KERNEL);
86 if (dax_id) {
87 strncpy(dax_id->dev_name, buf, DAX_NAME_LEN);
88 list_add(&dax_id->list, &dax_drv->ids);
89 } else
90 rc = -ENOMEM;
91 }
92 } else if (action == ID_REMOVE) {
93 list_del(&dax_id->list);
94 kfree(dax_id);
95 }
96 mutex_unlock(&dax_bus_lock);
97
98 if (rc < 0)
99 return rc;
100 if (action == ID_ADD)
101 rc = driver_attach(drv);
102 if (rc)
103 return rc;
104 return count;
105}
106
107static ssize_t new_id_store(struct device_driver *drv, const char *buf,
108 size_t count)
109{
110 return do_id_store(drv, buf, count, ID_ADD);
111}
112static DRIVER_ATTR_WO(new_id);
113
114static ssize_t remove_id_store(struct device_driver *drv, const char *buf,
115 size_t count)
116{
117 return do_id_store(drv, buf, count, ID_REMOVE);
118}
119static DRIVER_ATTR_WO(remove_id);
120
121static struct attribute *dax_drv_attrs[] = {
122 &driver_attr_new_id.attr,
123 &driver_attr_remove_id.attr,
124 NULL,
125};
126ATTRIBUTE_GROUPS(dax_drv);
127
128static int dax_bus_match(struct device *dev, struct device_driver *drv);
129
130/*
131 * Static dax regions are regions created by an external subsystem
132 * nvdimm where a single range is assigned. Its boundaries are by the external
133 * subsystem and are usually limited to one physical memory range. For example,
134 * for PMEM it is usually defined by NVDIMM Namespace boundaries (i.e. a
135 * single contiguous range)
136 *
137 * On dynamic dax regions, the assigned region can be partitioned by dax core
138 * into multiple subdivisions. A subdivision is represented into one
139 * /dev/daxN.M device composed by one or more potentially discontiguous ranges.
140 *
141 * When allocating a dax region, drivers must set whether it's static
142 * (IORESOURCE_DAX_STATIC). On static dax devices, the @pgmap is pre-assigned
143 * to dax core when calling devm_create_dev_dax(), whereas in dynamic dax
144 * devices it is NULL but afterwards allocated by dax core on device ->probe().
145 * Care is needed to make sure that dynamic dax devices are torn down with a
146 * cleared @pgmap field (see kill_dev_dax()).
147 */
148static bool is_static(struct dax_region *dax_region)
149{
150 return (dax_region->res.flags & IORESOURCE_DAX_STATIC) != 0;
151}
152
153bool static_dev_dax(struct dev_dax *dev_dax)
154{
155 return is_static(dev_dax->region);
156}
157EXPORT_SYMBOL_GPL(static_dev_dax);
158
159static u64 dev_dax_size(struct dev_dax *dev_dax)
160{
161 u64 size = 0;
162 int i;
163
164 device_lock_assert(&dev_dax->dev);
165
166 for (i = 0; i < dev_dax->nr_range; i++)
167 size += range_len(&dev_dax->ranges[i].range);
168
169 return size;
170}
171
172static int dax_bus_probe(struct device *dev)
173{
174 struct dax_device_driver *dax_drv = to_dax_drv(dev->driver);
175 struct dev_dax *dev_dax = to_dev_dax(dev);
176 struct dax_region *dax_region = dev_dax->region;
177 int rc;
178
179 if (dev_dax_size(dev_dax) == 0 || dev_dax->id < 0)
180 return -ENXIO;
181
182 rc = dax_drv->probe(dev_dax);
183
184 if (rc || is_static(dax_region))
185 return rc;
186
187 /*
188 * Track new seed creation only after successful probe of the
189 * previous seed.
190 */
191 if (dax_region->seed == dev)
192 dax_region->seed = NULL;
193
194 return 0;
195}
196
197static void dax_bus_remove(struct device *dev)
198{
199 struct dax_device_driver *dax_drv = to_dax_drv(dev->driver);
200 struct dev_dax *dev_dax = to_dev_dax(dev);
201
202 if (dax_drv->remove)
203 dax_drv->remove(dev_dax);
204}
205
206static struct bus_type dax_bus_type = {
207 .name = "dax",
208 .uevent = dax_bus_uevent,
209 .match = dax_bus_match,
210 .probe = dax_bus_probe,
211 .remove = dax_bus_remove,
212 .drv_groups = dax_drv_groups,
213};
214
215static int dax_bus_match(struct device *dev, struct device_driver *drv)
216{
217 struct dax_device_driver *dax_drv = to_dax_drv(drv);
218
219 /*
220 * All but the 'device-dax' driver, which has 'match_always'
221 * set, requires an exact id match.
222 */
223 if (dax_drv->match_always)
224 return 1;
225
226 return dax_match_id(dax_drv, dev);
227}
228
229/*
230 * Rely on the fact that drvdata is set before the attributes are
231 * registered, and that the attributes are unregistered before drvdata
232 * is cleared to assume that drvdata is always valid.
233 */
234static ssize_t id_show(struct device *dev,
235 struct device_attribute *attr, char *buf)
236{
237 struct dax_region *dax_region = dev_get_drvdata(dev);
238
239 return sprintf(buf, "%d\n", dax_region->id);
240}
241static DEVICE_ATTR_RO(id);
242
243static ssize_t region_size_show(struct device *dev,
244 struct device_attribute *attr, char *buf)
245{
246 struct dax_region *dax_region = dev_get_drvdata(dev);
247
248 return sprintf(buf, "%llu\n", (unsigned long long)
249 resource_size(&dax_region->res));
250}
251static struct device_attribute dev_attr_region_size = __ATTR(size, 0444,
252 region_size_show, NULL);
253
254static ssize_t region_align_show(struct device *dev,
255 struct device_attribute *attr, char *buf)
256{
257 struct dax_region *dax_region = dev_get_drvdata(dev);
258
259 return sprintf(buf, "%u\n", dax_region->align);
260}
261static struct device_attribute dev_attr_region_align =
262 __ATTR(align, 0400, region_align_show, NULL);
263
264#define for_each_dax_region_resource(dax_region, res) \
265 for (res = (dax_region)->res.child; res; res = res->sibling)
266
267static unsigned long long dax_region_avail_size(struct dax_region *dax_region)
268{
269 resource_size_t size = resource_size(&dax_region->res);
270 struct resource *res;
271
272 device_lock_assert(dax_region->dev);
273
274 for_each_dax_region_resource(dax_region, res)
275 size -= resource_size(res);
276 return size;
277}
278
279static ssize_t available_size_show(struct device *dev,
280 struct device_attribute *attr, char *buf)
281{
282 struct dax_region *dax_region = dev_get_drvdata(dev);
283 unsigned long long size;
284
285 device_lock(dev);
286 size = dax_region_avail_size(dax_region);
287 device_unlock(dev);
288
289 return sprintf(buf, "%llu\n", size);
290}
291static DEVICE_ATTR_RO(available_size);
292
293static ssize_t seed_show(struct device *dev,
294 struct device_attribute *attr, char *buf)
295{
296 struct dax_region *dax_region = dev_get_drvdata(dev);
297 struct device *seed;
298 ssize_t rc;
299
300 if (is_static(dax_region))
301 return -EINVAL;
302
303 device_lock(dev);
304 seed = dax_region->seed;
305 rc = sprintf(buf, "%s\n", seed ? dev_name(seed) : "");
306 device_unlock(dev);
307
308 return rc;
309}
310static DEVICE_ATTR_RO(seed);
311
312static ssize_t create_show(struct device *dev,
313 struct device_attribute *attr, char *buf)
314{
315 struct dax_region *dax_region = dev_get_drvdata(dev);
316 struct device *youngest;
317 ssize_t rc;
318
319 if (is_static(dax_region))
320 return -EINVAL;
321
322 device_lock(dev);
323 youngest = dax_region->youngest;
324 rc = sprintf(buf, "%s\n", youngest ? dev_name(youngest) : "");
325 device_unlock(dev);
326
327 return rc;
328}
329
330static ssize_t create_store(struct device *dev, struct device_attribute *attr,
331 const char *buf, size_t len)
332{
333 struct dax_region *dax_region = dev_get_drvdata(dev);
334 unsigned long long avail;
335 ssize_t rc;
336 int val;
337
338 if (is_static(dax_region))
339 return -EINVAL;
340
341 rc = kstrtoint(buf, 0, &val);
342 if (rc)
343 return rc;
344 if (val != 1)
345 return -EINVAL;
346
347 device_lock(dev);
348 avail = dax_region_avail_size(dax_region);
349 if (avail == 0)
350 rc = -ENOSPC;
351 else {
352 struct dev_dax_data data = {
353 .dax_region = dax_region,
354 .size = 0,
355 .id = -1,
356 };
357 struct dev_dax *dev_dax = devm_create_dev_dax(&data);
358
359 if (IS_ERR(dev_dax))
360 rc = PTR_ERR(dev_dax);
361 else {
362 /*
363 * In support of crafting multiple new devices
364 * simultaneously multiple seeds can be created,
365 * but only the first one that has not been
366 * successfully bound is tracked as the region
367 * seed.
368 */
369 if (!dax_region->seed)
370 dax_region->seed = &dev_dax->dev;
371 dax_region->youngest = &dev_dax->dev;
372 rc = len;
373 }
374 }
375 device_unlock(dev);
376
377 return rc;
378}
379static DEVICE_ATTR_RW(create);
380
381void kill_dev_dax(struct dev_dax *dev_dax)
382{
383 struct dax_device *dax_dev = dev_dax->dax_dev;
384 struct inode *inode = dax_inode(dax_dev);
385
386 kill_dax(dax_dev);
387 unmap_mapping_range(inode->i_mapping, 0, 0, 1);
388
389 /*
390 * Dynamic dax region have the pgmap allocated via dev_kzalloc()
391 * and thus freed by devm. Clear the pgmap to not have stale pgmap
392 * ranges on probe() from previous reconfigurations of region devices.
393 */
394 if (!static_dev_dax(dev_dax))
395 dev_dax->pgmap = NULL;
396}
397EXPORT_SYMBOL_GPL(kill_dev_dax);
398
399static void trim_dev_dax_range(struct dev_dax *dev_dax)
400{
401 int i = dev_dax->nr_range - 1;
402 struct range *range = &dev_dax->ranges[i].range;
403 struct dax_region *dax_region = dev_dax->region;
404
405 device_lock_assert(dax_region->dev);
406 dev_dbg(&dev_dax->dev, "delete range[%d]: %#llx:%#llx\n", i,
407 (unsigned long long)range->start,
408 (unsigned long long)range->end);
409
410 __release_region(&dax_region->res, range->start, range_len(range));
411 if (--dev_dax->nr_range == 0) {
412 kfree(dev_dax->ranges);
413 dev_dax->ranges = NULL;
414 }
415}
416
417static void free_dev_dax_ranges(struct dev_dax *dev_dax)
418{
419 while (dev_dax->nr_range)
420 trim_dev_dax_range(dev_dax);
421}
422
423static void unregister_dev_dax(void *dev)
424{
425 struct dev_dax *dev_dax = to_dev_dax(dev);
426
427 dev_dbg(dev, "%s\n", __func__);
428
429 kill_dev_dax(dev_dax);
430 free_dev_dax_ranges(dev_dax);
431 device_del(dev);
432 put_device(dev);
433}
434
435/* a return value >= 0 indicates this invocation invalidated the id */
436static int __free_dev_dax_id(struct dev_dax *dev_dax)
437{
438 struct dax_region *dax_region = dev_dax->region;
439 struct device *dev = &dev_dax->dev;
440 int rc = dev_dax->id;
441
442 device_lock_assert(dev);
443
444 if (is_static(dax_region) || dev_dax->id < 0)
445 return -1;
446 ida_free(&dax_region->ida, dev_dax->id);
447 dev_dax->id = -1;
448 return rc;
449}
450
451static int free_dev_dax_id(struct dev_dax *dev_dax)
452{
453 struct device *dev = &dev_dax->dev;
454 int rc;
455
456 device_lock(dev);
457 rc = __free_dev_dax_id(dev_dax);
458 device_unlock(dev);
459 return rc;
460}
461
462static ssize_t delete_store(struct device *dev, struct device_attribute *attr,
463 const char *buf, size_t len)
464{
465 struct dax_region *dax_region = dev_get_drvdata(dev);
466 struct dev_dax *dev_dax;
467 struct device *victim;
468 bool do_del = false;
469 int rc;
470
471 if (is_static(dax_region))
472 return -EINVAL;
473
474 victim = device_find_child_by_name(dax_region->dev, buf);
475 if (!victim)
476 return -ENXIO;
477
478 device_lock(dev);
479 device_lock(victim);
480 dev_dax = to_dev_dax(victim);
481 if (victim->driver || dev_dax_size(dev_dax))
482 rc = -EBUSY;
483 else {
484 /*
485 * Invalidate the device so it does not become active
486 * again, but always preserve device-id-0 so that
487 * /sys/bus/dax/ is guaranteed to be populated while any
488 * dax_region is registered.
489 */
490 if (dev_dax->id > 0) {
491 do_del = __free_dev_dax_id(dev_dax) >= 0;
492 rc = len;
493 if (dax_region->seed == victim)
494 dax_region->seed = NULL;
495 if (dax_region->youngest == victim)
496 dax_region->youngest = NULL;
497 } else
498 rc = -EBUSY;
499 }
500 device_unlock(victim);
501
502 /* won the race to invalidate the device, clean it up */
503 if (do_del)
504 devm_release_action(dev, unregister_dev_dax, victim);
505 device_unlock(dev);
506 put_device(victim);
507
508 return rc;
509}
510static DEVICE_ATTR_WO(delete);
511
512static umode_t dax_region_visible(struct kobject *kobj, struct attribute *a,
513 int n)
514{
515 struct device *dev = container_of(kobj, struct device, kobj);
516 struct dax_region *dax_region = dev_get_drvdata(dev);
517
518 if (is_static(dax_region))
519 if (a == &dev_attr_available_size.attr
520 || a == &dev_attr_create.attr
521 || a == &dev_attr_seed.attr
522 || a == &dev_attr_delete.attr)
523 return 0;
524 return a->mode;
525}
526
527static struct attribute *dax_region_attributes[] = {
528 &dev_attr_available_size.attr,
529 &dev_attr_region_size.attr,
530 &dev_attr_region_align.attr,
531 &dev_attr_create.attr,
532 &dev_attr_seed.attr,
533 &dev_attr_delete.attr,
534 &dev_attr_id.attr,
535 NULL,
536};
537
538static const struct attribute_group dax_region_attribute_group = {
539 .name = "dax_region",
540 .attrs = dax_region_attributes,
541 .is_visible = dax_region_visible,
542};
543
544static const struct attribute_group *dax_region_attribute_groups[] = {
545 &dax_region_attribute_group,
546 NULL,
547};
548
549static void dax_region_free(struct kref *kref)
550{
551 struct dax_region *dax_region;
552
553 dax_region = container_of(kref, struct dax_region, kref);
554 kfree(dax_region);
555}
556
557void dax_region_put(struct dax_region *dax_region)
558{
559 kref_put(&dax_region->kref, dax_region_free);
560}
561EXPORT_SYMBOL_GPL(dax_region_put);
562
563static void dax_region_unregister(void *region)
564{
565 struct dax_region *dax_region = region;
566
567 sysfs_remove_groups(&dax_region->dev->kobj,
568 dax_region_attribute_groups);
569 dax_region_put(dax_region);
570}
571
572struct dax_region *alloc_dax_region(struct device *parent, int region_id,
573 struct range *range, int target_node, unsigned int align,
574 unsigned long flags)
575{
576 struct dax_region *dax_region;
577
578 /*
579 * The DAX core assumes that it can store its private data in
580 * parent->driver_data. This WARN is a reminder / safeguard for
581 * developers of device-dax drivers.
582 */
583 if (dev_get_drvdata(parent)) {
584 dev_WARN(parent, "dax core failed to setup private data\n");
585 return NULL;
586 }
587
588 if (!IS_ALIGNED(range->start, align)
589 || !IS_ALIGNED(range_len(range), align))
590 return NULL;
591
592 dax_region = kzalloc(sizeof(*dax_region), GFP_KERNEL);
593 if (!dax_region)
594 return NULL;
595
596 dev_set_drvdata(parent, dax_region);
597 kref_init(&dax_region->kref);
598 dax_region->id = region_id;
599 dax_region->align = align;
600 dax_region->dev = parent;
601 dax_region->target_node = target_node;
602 ida_init(&dax_region->ida);
603 dax_region->res = (struct resource) {
604 .start = range->start,
605 .end = range->end,
606 .flags = IORESOURCE_MEM | flags,
607 };
608
609 if (sysfs_create_groups(&parent->kobj, dax_region_attribute_groups)) {
610 kfree(dax_region);
611 return NULL;
612 }
613
614 kref_get(&dax_region->kref);
615 if (devm_add_action_or_reset(parent, dax_region_unregister, dax_region))
616 return NULL;
617 return dax_region;
618}
619EXPORT_SYMBOL_GPL(alloc_dax_region);
620
621static void dax_mapping_release(struct device *dev)
622{
623 struct dax_mapping *mapping = to_dax_mapping(dev);
624 struct dev_dax *dev_dax = to_dev_dax(dev->parent);
625
626 ida_free(&dev_dax->ida, mapping->id);
627 kfree(mapping);
628}
629
630static void unregister_dax_mapping(void *data)
631{
632 struct device *dev = data;
633 struct dax_mapping *mapping = to_dax_mapping(dev);
634 struct dev_dax *dev_dax = to_dev_dax(dev->parent);
635 struct dax_region *dax_region = dev_dax->region;
636
637 dev_dbg(dev, "%s\n", __func__);
638
639 device_lock_assert(dax_region->dev);
640
641 dev_dax->ranges[mapping->range_id].mapping = NULL;
642 mapping->range_id = -1;
643
644 device_del(dev);
645 put_device(dev);
646}
647
648static struct dev_dax_range *get_dax_range(struct device *dev)
649{
650 struct dax_mapping *mapping = to_dax_mapping(dev);
651 struct dev_dax *dev_dax = to_dev_dax(dev->parent);
652 struct dax_region *dax_region = dev_dax->region;
653
654 device_lock(dax_region->dev);
655 if (mapping->range_id < 0) {
656 device_unlock(dax_region->dev);
657 return NULL;
658 }
659
660 return &dev_dax->ranges[mapping->range_id];
661}
662
663static void put_dax_range(struct dev_dax_range *dax_range)
664{
665 struct dax_mapping *mapping = dax_range->mapping;
666 struct dev_dax *dev_dax = to_dev_dax(mapping->dev.parent);
667 struct dax_region *dax_region = dev_dax->region;
668
669 device_unlock(dax_region->dev);
670}
671
672static ssize_t start_show(struct device *dev,
673 struct device_attribute *attr, char *buf)
674{
675 struct dev_dax_range *dax_range;
676 ssize_t rc;
677
678 dax_range = get_dax_range(dev);
679 if (!dax_range)
680 return -ENXIO;
681 rc = sprintf(buf, "%#llx\n", dax_range->range.start);
682 put_dax_range(dax_range);
683
684 return rc;
685}
686static DEVICE_ATTR(start, 0400, start_show, NULL);
687
688static ssize_t end_show(struct device *dev,
689 struct device_attribute *attr, char *buf)
690{
691 struct dev_dax_range *dax_range;
692 ssize_t rc;
693
694 dax_range = get_dax_range(dev);
695 if (!dax_range)
696 return -ENXIO;
697 rc = sprintf(buf, "%#llx\n", dax_range->range.end);
698 put_dax_range(dax_range);
699
700 return rc;
701}
702static DEVICE_ATTR(end, 0400, end_show, NULL);
703
704static ssize_t pgoff_show(struct device *dev,
705 struct device_attribute *attr, char *buf)
706{
707 struct dev_dax_range *dax_range;
708 ssize_t rc;
709
710 dax_range = get_dax_range(dev);
711 if (!dax_range)
712 return -ENXIO;
713 rc = sprintf(buf, "%#lx\n", dax_range->pgoff);
714 put_dax_range(dax_range);
715
716 return rc;
717}
718static DEVICE_ATTR(page_offset, 0400, pgoff_show, NULL);
719
720static struct attribute *dax_mapping_attributes[] = {
721 &dev_attr_start.attr,
722 &dev_attr_end.attr,
723 &dev_attr_page_offset.attr,
724 NULL,
725};
726
727static const struct attribute_group dax_mapping_attribute_group = {
728 .attrs = dax_mapping_attributes,
729};
730
731static const struct attribute_group *dax_mapping_attribute_groups[] = {
732 &dax_mapping_attribute_group,
733 NULL,
734};
735
736static struct device_type dax_mapping_type = {
737 .release = dax_mapping_release,
738 .groups = dax_mapping_attribute_groups,
739};
740
741static int devm_register_dax_mapping(struct dev_dax *dev_dax, int range_id)
742{
743 struct dax_region *dax_region = dev_dax->region;
744 struct dax_mapping *mapping;
745 struct device *dev;
746 int rc;
747
748 device_lock_assert(dax_region->dev);
749
750 if (dev_WARN_ONCE(&dev_dax->dev, !dax_region->dev->driver,
751 "region disabled\n"))
752 return -ENXIO;
753
754 mapping = kzalloc(sizeof(*mapping), GFP_KERNEL);
755 if (!mapping)
756 return -ENOMEM;
757 mapping->range_id = range_id;
758 mapping->id = ida_alloc(&dev_dax->ida, GFP_KERNEL);
759 if (mapping->id < 0) {
760 kfree(mapping);
761 return -ENOMEM;
762 }
763 dev_dax->ranges[range_id].mapping = mapping;
764 dev = &mapping->dev;
765 device_initialize(dev);
766 dev->parent = &dev_dax->dev;
767 dev->type = &dax_mapping_type;
768 dev_set_name(dev, "mapping%d", mapping->id);
769 rc = device_add(dev);
770 if (rc) {
771 put_device(dev);
772 return rc;
773 }
774
775 rc = devm_add_action_or_reset(dax_region->dev, unregister_dax_mapping,
776 dev);
777 if (rc)
778 return rc;
779 return 0;
780}
781
782static int alloc_dev_dax_range(struct dev_dax *dev_dax, u64 start,
783 resource_size_t size)
784{
785 struct dax_region *dax_region = dev_dax->region;
786 struct resource *res = &dax_region->res;
787 struct device *dev = &dev_dax->dev;
788 struct dev_dax_range *ranges;
789 unsigned long pgoff = 0;
790 struct resource *alloc;
791 int i, rc;
792
793 device_lock_assert(dax_region->dev);
794
795 /* handle the seed alloc special case */
796 if (!size) {
797 if (dev_WARN_ONCE(dev, dev_dax->nr_range,
798 "0-size allocation must be first\n"))
799 return -EBUSY;
800 /* nr_range == 0 is elsewhere special cased as 0-size device */
801 return 0;
802 }
803
804 alloc = __request_region(res, start, size, dev_name(dev), 0);
805 if (!alloc)
806 return -ENOMEM;
807
808 ranges = krealloc(dev_dax->ranges, sizeof(*ranges)
809 * (dev_dax->nr_range + 1), GFP_KERNEL);
810 if (!ranges) {
811 __release_region(res, alloc->start, resource_size(alloc));
812 return -ENOMEM;
813 }
814
815 for (i = 0; i < dev_dax->nr_range; i++)
816 pgoff += PHYS_PFN(range_len(&ranges[i].range));
817 dev_dax->ranges = ranges;
818 ranges[dev_dax->nr_range++] = (struct dev_dax_range) {
819 .pgoff = pgoff,
820 .range = {
821 .start = alloc->start,
822 .end = alloc->end,
823 },
824 };
825
826 dev_dbg(dev, "alloc range[%d]: %pa:%pa\n", dev_dax->nr_range - 1,
827 &alloc->start, &alloc->end);
828 /*
829 * A dev_dax instance must be registered before mapping device
830 * children can be added. Defer to devm_create_dev_dax() to add
831 * the initial mapping device.
832 */
833 if (!device_is_registered(&dev_dax->dev))
834 return 0;
835
836 rc = devm_register_dax_mapping(dev_dax, dev_dax->nr_range - 1);
837 if (rc)
838 trim_dev_dax_range(dev_dax);
839
840 return rc;
841}
842
843static int adjust_dev_dax_range(struct dev_dax *dev_dax, struct resource *res, resource_size_t size)
844{
845 int last_range = dev_dax->nr_range - 1;
846 struct dev_dax_range *dax_range = &dev_dax->ranges[last_range];
847 struct dax_region *dax_region = dev_dax->region;
848 bool is_shrink = resource_size(res) > size;
849 struct range *range = &dax_range->range;
850 struct device *dev = &dev_dax->dev;
851 int rc;
852
853 device_lock_assert(dax_region->dev);
854
855 if (dev_WARN_ONCE(dev, !size, "deletion is handled by dev_dax_shrink\n"))
856 return -EINVAL;
857
858 rc = adjust_resource(res, range->start, size);
859 if (rc)
860 return rc;
861
862 *range = (struct range) {
863 .start = range->start,
864 .end = range->start + size - 1,
865 };
866
867 dev_dbg(dev, "%s range[%d]: %#llx:%#llx\n", is_shrink ? "shrink" : "extend",
868 last_range, (unsigned long long) range->start,
869 (unsigned long long) range->end);
870
871 return 0;
872}
873
874static ssize_t size_show(struct device *dev,
875 struct device_attribute *attr, char *buf)
876{
877 struct dev_dax *dev_dax = to_dev_dax(dev);
878 unsigned long long size;
879
880 device_lock(dev);
881 size = dev_dax_size(dev_dax);
882 device_unlock(dev);
883
884 return sprintf(buf, "%llu\n", size);
885}
886
887static bool alloc_is_aligned(struct dev_dax *dev_dax, resource_size_t size)
888{
889 /*
890 * The minimum mapping granularity for a device instance is a
891 * single subsection, unless the arch says otherwise.
892 */
893 return IS_ALIGNED(size, max_t(unsigned long, dev_dax->align, memremap_compat_align()));
894}
895
896static int dev_dax_shrink(struct dev_dax *dev_dax, resource_size_t size)
897{
898 resource_size_t to_shrink = dev_dax_size(dev_dax) - size;
899 struct dax_region *dax_region = dev_dax->region;
900 struct device *dev = &dev_dax->dev;
901 int i;
902
903 for (i = dev_dax->nr_range - 1; i >= 0; i--) {
904 struct range *range = &dev_dax->ranges[i].range;
905 struct dax_mapping *mapping = dev_dax->ranges[i].mapping;
906 struct resource *adjust = NULL, *res;
907 resource_size_t shrink;
908
909 shrink = min_t(u64, to_shrink, range_len(range));
910 if (shrink >= range_len(range)) {
911 devm_release_action(dax_region->dev,
912 unregister_dax_mapping, &mapping->dev);
913 trim_dev_dax_range(dev_dax);
914 to_shrink -= shrink;
915 if (!to_shrink)
916 break;
917 continue;
918 }
919
920 for_each_dax_region_resource(dax_region, res)
921 if (strcmp(res->name, dev_name(dev)) == 0
922 && res->start == range->start) {
923 adjust = res;
924 break;
925 }
926
927 if (dev_WARN_ONCE(dev, !adjust || i != dev_dax->nr_range - 1,
928 "failed to find matching resource\n"))
929 return -ENXIO;
930 return adjust_dev_dax_range(dev_dax, adjust, range_len(range)
931 - shrink);
932 }
933 return 0;
934}
935
936/*
937 * Only allow adjustments that preserve the relative pgoff of existing
938 * allocations. I.e. the dev_dax->ranges array is ordered by increasing pgoff.
939 */
940static bool adjust_ok(struct dev_dax *dev_dax, struct resource *res)
941{
942 struct dev_dax_range *last;
943 int i;
944
945 if (dev_dax->nr_range == 0)
946 return false;
947 if (strcmp(res->name, dev_name(&dev_dax->dev)) != 0)
948 return false;
949 last = &dev_dax->ranges[dev_dax->nr_range - 1];
950 if (last->range.start != res->start || last->range.end != res->end)
951 return false;
952 for (i = 0; i < dev_dax->nr_range - 1; i++) {
953 struct dev_dax_range *dax_range = &dev_dax->ranges[i];
954
955 if (dax_range->pgoff > last->pgoff)
956 return false;
957 }
958
959 return true;
960}
961
962static ssize_t dev_dax_resize(struct dax_region *dax_region,
963 struct dev_dax *dev_dax, resource_size_t size)
964{
965 resource_size_t avail = dax_region_avail_size(dax_region), to_alloc;
966 resource_size_t dev_size = dev_dax_size(dev_dax);
967 struct resource *region_res = &dax_region->res;
968 struct device *dev = &dev_dax->dev;
969 struct resource *res, *first;
970 resource_size_t alloc = 0;
971 int rc;
972
973 if (dev->driver)
974 return -EBUSY;
975 if (size == dev_size)
976 return 0;
977 if (size > dev_size && size - dev_size > avail)
978 return -ENOSPC;
979 if (size < dev_size)
980 return dev_dax_shrink(dev_dax, size);
981
982 to_alloc = size - dev_size;
983 if (dev_WARN_ONCE(dev, !alloc_is_aligned(dev_dax, to_alloc),
984 "resize of %pa misaligned\n", &to_alloc))
985 return -ENXIO;
986
987 /*
988 * Expand the device into the unused portion of the region. This
989 * may involve adjusting the end of an existing resource, or
990 * allocating a new resource.
991 */
992retry:
993 first = region_res->child;
994 if (!first)
995 return alloc_dev_dax_range(dev_dax, dax_region->res.start, to_alloc);
996
997 rc = -ENOSPC;
998 for (res = first; res; res = res->sibling) {
999 struct resource *next = res->sibling;
1000
1001 /* space at the beginning of the region */
1002 if (res == first && res->start > dax_region->res.start) {
1003 alloc = min(res->start - dax_region->res.start, to_alloc);
1004 rc = alloc_dev_dax_range(dev_dax, dax_region->res.start, alloc);
1005 break;
1006 }
1007
1008 alloc = 0;
1009 /* space between allocations */
1010 if (next && next->start > res->end + 1)
1011 alloc = min(next->start - (res->end + 1), to_alloc);
1012
1013 /* space at the end of the region */
1014 if (!alloc && !next && res->end < region_res->end)
1015 alloc = min(region_res->end - res->end, to_alloc);
1016
1017 if (!alloc)
1018 continue;
1019
1020 if (adjust_ok(dev_dax, res)) {
1021 rc = adjust_dev_dax_range(dev_dax, res, resource_size(res) + alloc);
1022 break;
1023 }
1024 rc = alloc_dev_dax_range(dev_dax, res->end + 1, alloc);
1025 break;
1026 }
1027 if (rc)
1028 return rc;
1029 to_alloc -= alloc;
1030 if (to_alloc)
1031 goto retry;
1032 return 0;
1033}
1034
1035static ssize_t size_store(struct device *dev, struct device_attribute *attr,
1036 const char *buf, size_t len)
1037{
1038 ssize_t rc;
1039 unsigned long long val;
1040 struct dev_dax *dev_dax = to_dev_dax(dev);
1041 struct dax_region *dax_region = dev_dax->region;
1042
1043 rc = kstrtoull(buf, 0, &val);
1044 if (rc)
1045 return rc;
1046
1047 if (!alloc_is_aligned(dev_dax, val)) {
1048 dev_dbg(dev, "%s: size: %lld misaligned\n", __func__, val);
1049 return -EINVAL;
1050 }
1051
1052 device_lock(dax_region->dev);
1053 if (!dax_region->dev->driver) {
1054 device_unlock(dax_region->dev);
1055 return -ENXIO;
1056 }
1057 device_lock(dev);
1058 rc = dev_dax_resize(dax_region, dev_dax, val);
1059 device_unlock(dev);
1060 device_unlock(dax_region->dev);
1061
1062 return rc == 0 ? len : rc;
1063}
1064static DEVICE_ATTR_RW(size);
1065
1066static ssize_t range_parse(const char *opt, size_t len, struct range *range)
1067{
1068 unsigned long long addr = 0;
1069 char *start, *end, *str;
1070 ssize_t rc = -EINVAL;
1071
1072 str = kstrdup(opt, GFP_KERNEL);
1073 if (!str)
1074 return rc;
1075
1076 end = str;
1077 start = strsep(&end, "-");
1078 if (!start || !end)
1079 goto err;
1080
1081 rc = kstrtoull(start, 16, &addr);
1082 if (rc)
1083 goto err;
1084 range->start = addr;
1085
1086 rc = kstrtoull(end, 16, &addr);
1087 if (rc)
1088 goto err;
1089 range->end = addr;
1090
1091err:
1092 kfree(str);
1093 return rc;
1094}
1095
1096static ssize_t mapping_store(struct device *dev, struct device_attribute *attr,
1097 const char *buf, size_t len)
1098{
1099 struct dev_dax *dev_dax = to_dev_dax(dev);
1100 struct dax_region *dax_region = dev_dax->region;
1101 size_t to_alloc;
1102 struct range r;
1103 ssize_t rc;
1104
1105 rc = range_parse(buf, len, &r);
1106 if (rc)
1107 return rc;
1108
1109 rc = -ENXIO;
1110 device_lock(dax_region->dev);
1111 if (!dax_region->dev->driver) {
1112 device_unlock(dax_region->dev);
1113 return rc;
1114 }
1115 device_lock(dev);
1116
1117 to_alloc = range_len(&r);
1118 if (alloc_is_aligned(dev_dax, to_alloc))
1119 rc = alloc_dev_dax_range(dev_dax, r.start, to_alloc);
1120 device_unlock(dev);
1121 device_unlock(dax_region->dev);
1122
1123 return rc == 0 ? len : rc;
1124}
1125static DEVICE_ATTR_WO(mapping);
1126
1127static ssize_t align_show(struct device *dev,
1128 struct device_attribute *attr, char *buf)
1129{
1130 struct dev_dax *dev_dax = to_dev_dax(dev);
1131
1132 return sprintf(buf, "%d\n", dev_dax->align);
1133}
1134
1135static ssize_t dev_dax_validate_align(struct dev_dax *dev_dax)
1136{
1137 struct device *dev = &dev_dax->dev;
1138 int i;
1139
1140 for (i = 0; i < dev_dax->nr_range; i++) {
1141 size_t len = range_len(&dev_dax->ranges[i].range);
1142
1143 if (!alloc_is_aligned(dev_dax, len)) {
1144 dev_dbg(dev, "%s: align %u invalid for range %d\n",
1145 __func__, dev_dax->align, i);
1146 return -EINVAL;
1147 }
1148 }
1149
1150 return 0;
1151}
1152
1153static ssize_t align_store(struct device *dev, struct device_attribute *attr,
1154 const char *buf, size_t len)
1155{
1156 struct dev_dax *dev_dax = to_dev_dax(dev);
1157 struct dax_region *dax_region = dev_dax->region;
1158 unsigned long val, align_save;
1159 ssize_t rc;
1160
1161 rc = kstrtoul(buf, 0, &val);
1162 if (rc)
1163 return -ENXIO;
1164
1165 if (!dax_align_valid(val))
1166 return -EINVAL;
1167
1168 device_lock(dax_region->dev);
1169 if (!dax_region->dev->driver) {
1170 device_unlock(dax_region->dev);
1171 return -ENXIO;
1172 }
1173
1174 device_lock(dev);
1175 if (dev->driver) {
1176 rc = -EBUSY;
1177 goto out_unlock;
1178 }
1179
1180 align_save = dev_dax->align;
1181 dev_dax->align = val;
1182 rc = dev_dax_validate_align(dev_dax);
1183 if (rc)
1184 dev_dax->align = align_save;
1185out_unlock:
1186 device_unlock(dev);
1187 device_unlock(dax_region->dev);
1188 return rc == 0 ? len : rc;
1189}
1190static DEVICE_ATTR_RW(align);
1191
1192static int dev_dax_target_node(struct dev_dax *dev_dax)
1193{
1194 struct dax_region *dax_region = dev_dax->region;
1195
1196 return dax_region->target_node;
1197}
1198
1199static ssize_t target_node_show(struct device *dev,
1200 struct device_attribute *attr, char *buf)
1201{
1202 struct dev_dax *dev_dax = to_dev_dax(dev);
1203
1204 return sprintf(buf, "%d\n", dev_dax_target_node(dev_dax));
1205}
1206static DEVICE_ATTR_RO(target_node);
1207
1208static ssize_t resource_show(struct device *dev,
1209 struct device_attribute *attr, char *buf)
1210{
1211 struct dev_dax *dev_dax = to_dev_dax(dev);
1212 struct dax_region *dax_region = dev_dax->region;
1213 unsigned long long start;
1214
1215 if (dev_dax->nr_range < 1)
1216 start = dax_region->res.start;
1217 else
1218 start = dev_dax->ranges[0].range.start;
1219
1220 return sprintf(buf, "%#llx\n", start);
1221}
1222static DEVICE_ATTR(resource, 0400, resource_show, NULL);
1223
1224static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
1225 char *buf)
1226{
1227 /*
1228 * We only ever expect to handle device-dax instances, i.e. the
1229 * @type argument to MODULE_ALIAS_DAX_DEVICE() is always zero
1230 */
1231 return sprintf(buf, DAX_DEVICE_MODALIAS_FMT "\n", 0);
1232}
1233static DEVICE_ATTR_RO(modalias);
1234
1235static ssize_t numa_node_show(struct device *dev,
1236 struct device_attribute *attr, char *buf)
1237{
1238 return sprintf(buf, "%d\n", dev_to_node(dev));
1239}
1240static DEVICE_ATTR_RO(numa_node);
1241
1242static umode_t dev_dax_visible(struct kobject *kobj, struct attribute *a, int n)
1243{
1244 struct device *dev = container_of(kobj, struct device, kobj);
1245 struct dev_dax *dev_dax = to_dev_dax(dev);
1246 struct dax_region *dax_region = dev_dax->region;
1247
1248 if (a == &dev_attr_target_node.attr && dev_dax_target_node(dev_dax) < 0)
1249 return 0;
1250 if (a == &dev_attr_numa_node.attr && !IS_ENABLED(CONFIG_NUMA))
1251 return 0;
1252 if (a == &dev_attr_mapping.attr && is_static(dax_region))
1253 return 0;
1254 if ((a == &dev_attr_align.attr ||
1255 a == &dev_attr_size.attr) && is_static(dax_region))
1256 return 0444;
1257 return a->mode;
1258}
1259
1260static struct attribute *dev_dax_attributes[] = {
1261 &dev_attr_modalias.attr,
1262 &dev_attr_size.attr,
1263 &dev_attr_mapping.attr,
1264 &dev_attr_target_node.attr,
1265 &dev_attr_align.attr,
1266 &dev_attr_resource.attr,
1267 &dev_attr_numa_node.attr,
1268 NULL,
1269};
1270
1271static const struct attribute_group dev_dax_attribute_group = {
1272 .attrs = dev_dax_attributes,
1273 .is_visible = dev_dax_visible,
1274};
1275
1276static const struct attribute_group *dax_attribute_groups[] = {
1277 &dev_dax_attribute_group,
1278 NULL,
1279};
1280
1281static void dev_dax_release(struct device *dev)
1282{
1283 struct dev_dax *dev_dax = to_dev_dax(dev);
1284 struct dax_region *dax_region = dev_dax->region;
1285 struct dax_device *dax_dev = dev_dax->dax_dev;
1286
1287 put_dax(dax_dev);
1288 free_dev_dax_id(dev_dax);
1289 dax_region_put(dax_region);
1290 kfree(dev_dax->pgmap);
1291 kfree(dev_dax);
1292}
1293
1294static const struct device_type dev_dax_type = {
1295 .release = dev_dax_release,
1296 .groups = dax_attribute_groups,
1297};
1298
1299struct dev_dax *devm_create_dev_dax(struct dev_dax_data *data)
1300{
1301 struct dax_region *dax_region = data->dax_region;
1302 struct device *parent = dax_region->dev;
1303 struct dax_device *dax_dev;
1304 struct dev_dax *dev_dax;
1305 struct inode *inode;
1306 struct device *dev;
1307 int rc;
1308
1309 dev_dax = kzalloc(sizeof(*dev_dax), GFP_KERNEL);
1310 if (!dev_dax)
1311 return ERR_PTR(-ENOMEM);
1312
1313 if (is_static(dax_region)) {
1314 if (dev_WARN_ONCE(parent, data->id < 0,
1315 "dynamic id specified to static region\n")) {
1316 rc = -EINVAL;
1317 goto err_id;
1318 }
1319
1320 dev_dax->id = data->id;
1321 } else {
1322 if (dev_WARN_ONCE(parent, data->id >= 0,
1323 "static id specified to dynamic region\n")) {
1324 rc = -EINVAL;
1325 goto err_id;
1326 }
1327
1328 rc = ida_alloc(&dax_region->ida, GFP_KERNEL);
1329 if (rc < 0)
1330 goto err_id;
1331 dev_dax->id = rc;
1332 }
1333
1334 dev_dax->region = dax_region;
1335 dev = &dev_dax->dev;
1336 device_initialize(dev);
1337 dev_set_name(dev, "dax%d.%d", dax_region->id, dev_dax->id);
1338
1339 rc = alloc_dev_dax_range(dev_dax, dax_region->res.start, data->size);
1340 if (rc)
1341 goto err_range;
1342
1343 if (data->pgmap) {
1344 dev_WARN_ONCE(parent, !is_static(dax_region),
1345 "custom dev_pagemap requires a static dax_region\n");
1346
1347 dev_dax->pgmap = kmemdup(data->pgmap,
1348 sizeof(struct dev_pagemap), GFP_KERNEL);
1349 if (!dev_dax->pgmap) {
1350 rc = -ENOMEM;
1351 goto err_pgmap;
1352 }
1353 }
1354
1355 /*
1356 * No dax_operations since there is no access to this device outside of
1357 * mmap of the resulting character device.
1358 */
1359 dax_dev = alloc_dax(dev_dax, NULL);
1360 if (IS_ERR(dax_dev)) {
1361 rc = PTR_ERR(dax_dev);
1362 goto err_alloc_dax;
1363 }
1364 set_dax_synchronous(dax_dev);
1365 set_dax_nocache(dax_dev);
1366 set_dax_nomc(dax_dev);
1367
1368 /* a device_dax instance is dead while the driver is not attached */
1369 kill_dax(dax_dev);
1370
1371 dev_dax->dax_dev = dax_dev;
1372 dev_dax->target_node = dax_region->target_node;
1373 dev_dax->align = dax_region->align;
1374 ida_init(&dev_dax->ida);
1375 kref_get(&dax_region->kref);
1376
1377 inode = dax_inode(dax_dev);
1378 dev->devt = inode->i_rdev;
1379 dev->bus = &dax_bus_type;
1380 dev->parent = parent;
1381 dev->type = &dev_dax_type;
1382
1383 rc = device_add(dev);
1384 if (rc) {
1385 kill_dev_dax(dev_dax);
1386 put_device(dev);
1387 return ERR_PTR(rc);
1388 }
1389
1390 rc = devm_add_action_or_reset(dax_region->dev, unregister_dev_dax, dev);
1391 if (rc)
1392 return ERR_PTR(rc);
1393
1394 /* register mapping device for the initial allocation range */
1395 if (dev_dax->nr_range && range_len(&dev_dax->ranges[0].range)) {
1396 rc = devm_register_dax_mapping(dev_dax, 0);
1397 if (rc)
1398 return ERR_PTR(rc);
1399 }
1400
1401 return dev_dax;
1402
1403err_alloc_dax:
1404 kfree(dev_dax->pgmap);
1405err_pgmap:
1406 free_dev_dax_ranges(dev_dax);
1407err_range:
1408 free_dev_dax_id(dev_dax);
1409err_id:
1410 kfree(dev_dax);
1411
1412 return ERR_PTR(rc);
1413}
1414EXPORT_SYMBOL_GPL(devm_create_dev_dax);
1415
1416static int match_always_count;
1417
1418int __dax_driver_register(struct dax_device_driver *dax_drv,
1419 struct module *module, const char *mod_name)
1420{
1421 struct device_driver *drv = &dax_drv->drv;
1422 int rc = 0;
1423
1424 /*
1425 * dax_bus_probe() calls dax_drv->probe() unconditionally.
1426 * So better be safe than sorry and ensure it is provided.
1427 */
1428 if (!dax_drv->probe)
1429 return -EINVAL;
1430
1431 INIT_LIST_HEAD(&dax_drv->ids);
1432 drv->owner = module;
1433 drv->name = mod_name;
1434 drv->mod_name = mod_name;
1435 drv->bus = &dax_bus_type;
1436
1437 /* there can only be one default driver */
1438 mutex_lock(&dax_bus_lock);
1439 match_always_count += dax_drv->match_always;
1440 if (match_always_count > 1) {
1441 match_always_count--;
1442 WARN_ON(1);
1443 rc = -EINVAL;
1444 }
1445 mutex_unlock(&dax_bus_lock);
1446 if (rc)
1447 return rc;
1448
1449 rc = driver_register(drv);
1450 if (rc && dax_drv->match_always) {
1451 mutex_lock(&dax_bus_lock);
1452 match_always_count -= dax_drv->match_always;
1453 mutex_unlock(&dax_bus_lock);
1454 }
1455
1456 return rc;
1457}
1458EXPORT_SYMBOL_GPL(__dax_driver_register);
1459
1460void dax_driver_unregister(struct dax_device_driver *dax_drv)
1461{
1462 struct device_driver *drv = &dax_drv->drv;
1463 struct dax_id *dax_id, *_id;
1464
1465 mutex_lock(&dax_bus_lock);
1466 match_always_count -= dax_drv->match_always;
1467 list_for_each_entry_safe(dax_id, _id, &dax_drv->ids, list) {
1468 list_del(&dax_id->list);
1469 kfree(dax_id);
1470 }
1471 mutex_unlock(&dax_bus_lock);
1472 driver_unregister(drv);
1473}
1474EXPORT_SYMBOL_GPL(dax_driver_unregister);
1475
1476int __init dax_bus_init(void)
1477{
1478 return bus_register(&dax_bus_type);
1479}
1480
1481void __exit dax_bus_exit(void)
1482{
1483 bus_unregister(&dax_bus_type);
1484}
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
13static DEFINE_MUTEX(dax_bus_lock);
14
15/*
16 * All changes to the dax region configuration occur with this lock held
17 * for write.
18 */
19DECLARE_RWSEM(dax_region_rwsem);
20
21/*
22 * All changes to the dax device configuration occur with this lock held
23 * for write.
24 */
25DECLARE_RWSEM(dax_dev_rwsem);
26
27#define DAX_NAME_LEN 30
28struct dax_id {
29 struct list_head list;
30 char dev_name[DAX_NAME_LEN];
31};
32
33static int dax_bus_uevent(const struct device *dev, struct kobj_uevent_env *env)
34{
35 /*
36 * We only ever expect to handle device-dax instances, i.e. the
37 * @type argument to MODULE_ALIAS_DAX_DEVICE() is always zero
38 */
39 return add_uevent_var(env, "MODALIAS=" DAX_DEVICE_MODALIAS_FMT, 0);
40}
41
42#define to_dax_drv(__drv) container_of_const(__drv, struct dax_device_driver, drv)
43
44static struct dax_id *__dax_match_id(const struct dax_device_driver *dax_drv,
45 const char *dev_name)
46{
47 struct dax_id *dax_id;
48
49 lockdep_assert_held(&dax_bus_lock);
50
51 list_for_each_entry(dax_id, &dax_drv->ids, list)
52 if (sysfs_streq(dax_id->dev_name, dev_name))
53 return dax_id;
54 return NULL;
55}
56
57static int dax_match_id(const struct dax_device_driver *dax_drv, struct device *dev)
58{
59 int match;
60
61 mutex_lock(&dax_bus_lock);
62 match = !!__dax_match_id(dax_drv, dev_name(dev));
63 mutex_unlock(&dax_bus_lock);
64
65 return match;
66}
67
68static int dax_match_type(const struct dax_device_driver *dax_drv, struct device *dev)
69{
70 enum dax_driver_type type = DAXDRV_DEVICE_TYPE;
71 struct dev_dax *dev_dax = to_dev_dax(dev);
72
73 if (dev_dax->region->res.flags & IORESOURCE_DAX_KMEM)
74 type = DAXDRV_KMEM_TYPE;
75
76 if (dax_drv->type == type)
77 return 1;
78
79 /* default to device mode if dax_kmem is disabled */
80 if (dax_drv->type == DAXDRV_DEVICE_TYPE &&
81 !IS_ENABLED(CONFIG_DEV_DAX_KMEM))
82 return 1;
83
84 return 0;
85}
86
87enum id_action {
88 ID_REMOVE,
89 ID_ADD,
90};
91
92static ssize_t do_id_store(struct device_driver *drv, const char *buf,
93 size_t count, enum id_action action)
94{
95 struct dax_device_driver *dax_drv = to_dax_drv(drv);
96 unsigned int region_id, id;
97 char devname[DAX_NAME_LEN];
98 struct dax_id *dax_id;
99 ssize_t rc = count;
100 int fields;
101
102 fields = sscanf(buf, "dax%d.%d", ®ion_id, &id);
103 if (fields != 2)
104 return -EINVAL;
105 sprintf(devname, "dax%d.%d", region_id, id);
106 if (!sysfs_streq(buf, devname))
107 return -EINVAL;
108
109 mutex_lock(&dax_bus_lock);
110 dax_id = __dax_match_id(dax_drv, buf);
111 if (!dax_id) {
112 if (action == ID_ADD) {
113 dax_id = kzalloc(sizeof(*dax_id), GFP_KERNEL);
114 if (dax_id) {
115 strscpy(dax_id->dev_name, buf, DAX_NAME_LEN);
116 list_add(&dax_id->list, &dax_drv->ids);
117 } else
118 rc = -ENOMEM;
119 }
120 } else if (action == ID_REMOVE) {
121 list_del(&dax_id->list);
122 kfree(dax_id);
123 }
124 mutex_unlock(&dax_bus_lock);
125
126 if (rc < 0)
127 return rc;
128 if (action == ID_ADD)
129 rc = driver_attach(drv);
130 if (rc)
131 return rc;
132 return count;
133}
134
135static ssize_t new_id_store(struct device_driver *drv, const char *buf,
136 size_t count)
137{
138 return do_id_store(drv, buf, count, ID_ADD);
139}
140static DRIVER_ATTR_WO(new_id);
141
142static ssize_t remove_id_store(struct device_driver *drv, const char *buf,
143 size_t count)
144{
145 return do_id_store(drv, buf, count, ID_REMOVE);
146}
147static DRIVER_ATTR_WO(remove_id);
148
149static struct attribute *dax_drv_attrs[] = {
150 &driver_attr_new_id.attr,
151 &driver_attr_remove_id.attr,
152 NULL,
153};
154ATTRIBUTE_GROUPS(dax_drv);
155
156static int dax_bus_match(struct device *dev, const struct device_driver *drv);
157
158/*
159 * Static dax regions are regions created by an external subsystem
160 * nvdimm where a single range is assigned. Its boundaries are by the external
161 * subsystem and are usually limited to one physical memory range. For example,
162 * for PMEM it is usually defined by NVDIMM Namespace boundaries (i.e. a
163 * single contiguous range)
164 *
165 * On dynamic dax regions, the assigned region can be partitioned by dax core
166 * into multiple subdivisions. A subdivision is represented into one
167 * /dev/daxN.M device composed by one or more potentially discontiguous ranges.
168 *
169 * When allocating a dax region, drivers must set whether it's static
170 * (IORESOURCE_DAX_STATIC). On static dax devices, the @pgmap is pre-assigned
171 * to dax core when calling devm_create_dev_dax(), whereas in dynamic dax
172 * devices it is NULL but afterwards allocated by dax core on device ->probe().
173 * Care is needed to make sure that dynamic dax devices are torn down with a
174 * cleared @pgmap field (see kill_dev_dax()).
175 */
176static bool is_static(struct dax_region *dax_region)
177{
178 return (dax_region->res.flags & IORESOURCE_DAX_STATIC) != 0;
179}
180
181bool static_dev_dax(struct dev_dax *dev_dax)
182{
183 return is_static(dev_dax->region);
184}
185EXPORT_SYMBOL_GPL(static_dev_dax);
186
187static u64 dev_dax_size(struct dev_dax *dev_dax)
188{
189 u64 size = 0;
190 int i;
191
192 lockdep_assert_held(&dax_dev_rwsem);
193
194 for (i = 0; i < dev_dax->nr_range; i++)
195 size += range_len(&dev_dax->ranges[i].range);
196
197 return size;
198}
199
200static int dax_bus_probe(struct device *dev)
201{
202 struct dax_device_driver *dax_drv = to_dax_drv(dev->driver);
203 struct dev_dax *dev_dax = to_dev_dax(dev);
204 struct dax_region *dax_region = dev_dax->region;
205 int rc;
206 u64 size;
207
208 rc = down_read_interruptible(&dax_dev_rwsem);
209 if (rc)
210 return rc;
211 size = dev_dax_size(dev_dax);
212 up_read(&dax_dev_rwsem);
213
214 if (size == 0 || dev_dax->id < 0)
215 return -ENXIO;
216
217 rc = dax_drv->probe(dev_dax);
218
219 if (rc || is_static(dax_region))
220 return rc;
221
222 /*
223 * Track new seed creation only after successful probe of the
224 * previous seed.
225 */
226 if (dax_region->seed == dev)
227 dax_region->seed = NULL;
228
229 return 0;
230}
231
232static void dax_bus_remove(struct device *dev)
233{
234 struct dax_device_driver *dax_drv = to_dax_drv(dev->driver);
235 struct dev_dax *dev_dax = to_dev_dax(dev);
236
237 if (dax_drv->remove)
238 dax_drv->remove(dev_dax);
239}
240
241static const struct bus_type dax_bus_type = {
242 .name = "dax",
243 .uevent = dax_bus_uevent,
244 .match = dax_bus_match,
245 .probe = dax_bus_probe,
246 .remove = dax_bus_remove,
247 .drv_groups = dax_drv_groups,
248};
249
250static int dax_bus_match(struct device *dev, const struct device_driver *drv)
251{
252 const struct dax_device_driver *dax_drv = to_dax_drv(drv);
253
254 if (dax_match_id(dax_drv, dev))
255 return 1;
256 return dax_match_type(dax_drv, dev);
257}
258
259/*
260 * Rely on the fact that drvdata is set before the attributes are
261 * registered, and that the attributes are unregistered before drvdata
262 * is cleared to assume that drvdata is always valid.
263 */
264static ssize_t id_show(struct device *dev,
265 struct device_attribute *attr, char *buf)
266{
267 struct dax_region *dax_region = dev_get_drvdata(dev);
268
269 return sysfs_emit(buf, "%d\n", dax_region->id);
270}
271static DEVICE_ATTR_RO(id);
272
273static ssize_t region_size_show(struct device *dev,
274 struct device_attribute *attr, char *buf)
275{
276 struct dax_region *dax_region = dev_get_drvdata(dev);
277
278 return sysfs_emit(buf, "%llu\n",
279 (unsigned long long)resource_size(&dax_region->res));
280}
281static struct device_attribute dev_attr_region_size = __ATTR(size, 0444,
282 region_size_show, NULL);
283
284static ssize_t region_align_show(struct device *dev,
285 struct device_attribute *attr, char *buf)
286{
287 struct dax_region *dax_region = dev_get_drvdata(dev);
288
289 return sysfs_emit(buf, "%u\n", dax_region->align);
290}
291static struct device_attribute dev_attr_region_align =
292 __ATTR(align, 0400, region_align_show, NULL);
293
294#define for_each_dax_region_resource(dax_region, res) \
295 for (res = (dax_region)->res.child; res; res = res->sibling)
296
297static unsigned long long dax_region_avail_size(struct dax_region *dax_region)
298{
299 resource_size_t size = resource_size(&dax_region->res);
300 struct resource *res;
301
302 lockdep_assert_held(&dax_region_rwsem);
303
304 for_each_dax_region_resource(dax_region, res)
305 size -= resource_size(res);
306 return size;
307}
308
309static ssize_t available_size_show(struct device *dev,
310 struct device_attribute *attr, char *buf)
311{
312 struct dax_region *dax_region = dev_get_drvdata(dev);
313 unsigned long long size;
314 int rc;
315
316 rc = down_read_interruptible(&dax_region_rwsem);
317 if (rc)
318 return rc;
319 size = dax_region_avail_size(dax_region);
320 up_read(&dax_region_rwsem);
321
322 return sysfs_emit(buf, "%llu\n", size);
323}
324static DEVICE_ATTR_RO(available_size);
325
326static ssize_t seed_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 *seed;
331 ssize_t rc;
332
333 if (is_static(dax_region))
334 return -EINVAL;
335
336 rc = down_read_interruptible(&dax_region_rwsem);
337 if (rc)
338 return rc;
339 seed = dax_region->seed;
340 rc = sysfs_emit(buf, "%s\n", seed ? dev_name(seed) : "");
341 up_read(&dax_region_rwsem);
342
343 return rc;
344}
345static DEVICE_ATTR_RO(seed);
346
347static ssize_t create_show(struct device *dev,
348 struct device_attribute *attr, char *buf)
349{
350 struct dax_region *dax_region = dev_get_drvdata(dev);
351 struct device *youngest;
352 ssize_t rc;
353
354 if (is_static(dax_region))
355 return -EINVAL;
356
357 rc = down_read_interruptible(&dax_region_rwsem);
358 if (rc)
359 return rc;
360 youngest = dax_region->youngest;
361 rc = sysfs_emit(buf, "%s\n", youngest ? dev_name(youngest) : "");
362 up_read(&dax_region_rwsem);
363
364 return rc;
365}
366
367static struct dev_dax *__devm_create_dev_dax(struct dev_dax_data *data);
368
369static ssize_t create_store(struct device *dev, struct device_attribute *attr,
370 const char *buf, size_t len)
371{
372 struct dax_region *dax_region = dev_get_drvdata(dev);
373 unsigned long long avail;
374 ssize_t rc;
375 int val;
376
377 if (is_static(dax_region))
378 return -EINVAL;
379
380 rc = kstrtoint(buf, 0, &val);
381 if (rc)
382 return rc;
383 if (val != 1)
384 return -EINVAL;
385
386 rc = down_write_killable(&dax_region_rwsem);
387 if (rc)
388 return rc;
389 avail = dax_region_avail_size(dax_region);
390 if (avail == 0)
391 rc = -ENOSPC;
392 else {
393 struct dev_dax_data data = {
394 .dax_region = dax_region,
395 .size = 0,
396 .id = -1,
397 .memmap_on_memory = false,
398 };
399 struct dev_dax *dev_dax = __devm_create_dev_dax(&data);
400
401 if (IS_ERR(dev_dax))
402 rc = PTR_ERR(dev_dax);
403 else {
404 /*
405 * In support of crafting multiple new devices
406 * simultaneously multiple seeds can be created,
407 * but only the first one that has not been
408 * successfully bound is tracked as the region
409 * seed.
410 */
411 if (!dax_region->seed)
412 dax_region->seed = &dev_dax->dev;
413 dax_region->youngest = &dev_dax->dev;
414 rc = len;
415 }
416 }
417 up_write(&dax_region_rwsem);
418
419 return rc;
420}
421static DEVICE_ATTR_RW(create);
422
423void kill_dev_dax(struct dev_dax *dev_dax)
424{
425 struct dax_device *dax_dev = dev_dax->dax_dev;
426 struct inode *inode = dax_inode(dax_dev);
427
428 kill_dax(dax_dev);
429 unmap_mapping_range(inode->i_mapping, 0, 0, 1);
430
431 /*
432 * Dynamic dax region have the pgmap allocated via dev_kzalloc()
433 * and thus freed by devm. Clear the pgmap to not have stale pgmap
434 * ranges on probe() from previous reconfigurations of region devices.
435 */
436 if (!static_dev_dax(dev_dax))
437 dev_dax->pgmap = NULL;
438}
439EXPORT_SYMBOL_GPL(kill_dev_dax);
440
441static void trim_dev_dax_range(struct dev_dax *dev_dax)
442{
443 int i = dev_dax->nr_range - 1;
444 struct range *range = &dev_dax->ranges[i].range;
445 struct dax_region *dax_region = dev_dax->region;
446
447 lockdep_assert_held_write(&dax_region_rwsem);
448 dev_dbg(&dev_dax->dev, "delete range[%d]: %#llx:%#llx\n", i,
449 (unsigned long long)range->start,
450 (unsigned long long)range->end);
451
452 __release_region(&dax_region->res, range->start, range_len(range));
453 if (--dev_dax->nr_range == 0) {
454 kfree(dev_dax->ranges);
455 dev_dax->ranges = NULL;
456 }
457}
458
459static void free_dev_dax_ranges(struct dev_dax *dev_dax)
460{
461 while (dev_dax->nr_range)
462 trim_dev_dax_range(dev_dax);
463}
464
465static void unregister_dev_dax(void *dev)
466{
467 struct dev_dax *dev_dax = to_dev_dax(dev);
468
469 dev_dbg(dev, "%s\n", __func__);
470
471 down_write(&dax_region_rwsem);
472 kill_dev_dax(dev_dax);
473 device_del(dev);
474 free_dev_dax_ranges(dev_dax);
475 put_device(dev);
476 up_write(&dax_region_rwsem);
477}
478
479static void dax_region_free(struct kref *kref)
480{
481 struct dax_region *dax_region;
482
483 dax_region = container_of(kref, struct dax_region, kref);
484 kfree(dax_region);
485}
486
487static void dax_region_put(struct dax_region *dax_region)
488{
489 kref_put(&dax_region->kref, dax_region_free);
490}
491
492/* a return value >= 0 indicates this invocation invalidated the id */
493static int __free_dev_dax_id(struct dev_dax *dev_dax)
494{
495 struct dax_region *dax_region;
496 int rc = dev_dax->id;
497
498 lockdep_assert_held_write(&dax_dev_rwsem);
499
500 if (!dev_dax->dyn_id || dev_dax->id < 0)
501 return -1;
502 dax_region = dev_dax->region;
503 ida_free(&dax_region->ida, dev_dax->id);
504 dax_region_put(dax_region);
505 dev_dax->id = -1;
506 return rc;
507}
508
509static int free_dev_dax_id(struct dev_dax *dev_dax)
510{
511 int rc;
512
513 rc = down_write_killable(&dax_dev_rwsem);
514 if (rc)
515 return rc;
516 rc = __free_dev_dax_id(dev_dax);
517 up_write(&dax_dev_rwsem);
518 return rc;
519}
520
521static int alloc_dev_dax_id(struct dev_dax *dev_dax)
522{
523 struct dax_region *dax_region = dev_dax->region;
524 int id;
525
526 id = ida_alloc(&dax_region->ida, GFP_KERNEL);
527 if (id < 0)
528 return id;
529 kref_get(&dax_region->kref);
530 dev_dax->dyn_id = true;
531 dev_dax->id = id;
532 return id;
533}
534
535static ssize_t delete_store(struct device *dev, struct device_attribute *attr,
536 const char *buf, size_t len)
537{
538 struct dax_region *dax_region = dev_get_drvdata(dev);
539 struct dev_dax *dev_dax;
540 struct device *victim;
541 bool do_del = false;
542 int rc;
543
544 if (is_static(dax_region))
545 return -EINVAL;
546
547 victim = device_find_child_by_name(dax_region->dev, buf);
548 if (!victim)
549 return -ENXIO;
550
551 device_lock(dev);
552 device_lock(victim);
553 dev_dax = to_dev_dax(victim);
554 down_write(&dax_dev_rwsem);
555 if (victim->driver || dev_dax_size(dev_dax))
556 rc = -EBUSY;
557 else {
558 /*
559 * Invalidate the device so it does not become active
560 * again, but always preserve device-id-0 so that
561 * /sys/bus/dax/ is guaranteed to be populated while any
562 * dax_region is registered.
563 */
564 if (dev_dax->id > 0) {
565 do_del = __free_dev_dax_id(dev_dax) >= 0;
566 rc = len;
567 if (dax_region->seed == victim)
568 dax_region->seed = NULL;
569 if (dax_region->youngest == victim)
570 dax_region->youngest = NULL;
571 } else
572 rc = -EBUSY;
573 }
574 up_write(&dax_dev_rwsem);
575 device_unlock(victim);
576
577 /* won the race to invalidate the device, clean it up */
578 if (do_del)
579 devm_release_action(dev, unregister_dev_dax, victim);
580 device_unlock(dev);
581 put_device(victim);
582
583 return rc;
584}
585static DEVICE_ATTR_WO(delete);
586
587static umode_t dax_region_visible(struct kobject *kobj, struct attribute *a,
588 int n)
589{
590 struct device *dev = container_of(kobj, struct device, kobj);
591 struct dax_region *dax_region = dev_get_drvdata(dev);
592
593 if (is_static(dax_region))
594 if (a == &dev_attr_available_size.attr
595 || a == &dev_attr_create.attr
596 || a == &dev_attr_seed.attr
597 || a == &dev_attr_delete.attr)
598 return 0;
599 return a->mode;
600}
601
602static struct attribute *dax_region_attributes[] = {
603 &dev_attr_available_size.attr,
604 &dev_attr_region_size.attr,
605 &dev_attr_region_align.attr,
606 &dev_attr_create.attr,
607 &dev_attr_seed.attr,
608 &dev_attr_delete.attr,
609 &dev_attr_id.attr,
610 NULL,
611};
612
613static const struct attribute_group dax_region_attribute_group = {
614 .name = "dax_region",
615 .attrs = dax_region_attributes,
616 .is_visible = dax_region_visible,
617};
618
619static const struct attribute_group *dax_region_attribute_groups[] = {
620 &dax_region_attribute_group,
621 NULL,
622};
623
624static void dax_region_unregister(void *region)
625{
626 struct dax_region *dax_region = region;
627
628 sysfs_remove_groups(&dax_region->dev->kobj,
629 dax_region_attribute_groups);
630 dax_region_put(dax_region);
631}
632
633struct dax_region *alloc_dax_region(struct device *parent, int region_id,
634 struct range *range, int target_node, unsigned int align,
635 unsigned long flags)
636{
637 struct dax_region *dax_region;
638
639 /*
640 * The DAX core assumes that it can store its private data in
641 * parent->driver_data. This WARN is a reminder / safeguard for
642 * developers of device-dax drivers.
643 */
644 if (dev_get_drvdata(parent)) {
645 dev_WARN(parent, "dax core failed to setup private data\n");
646 return NULL;
647 }
648
649 if (!IS_ALIGNED(range->start, align)
650 || !IS_ALIGNED(range_len(range), align))
651 return NULL;
652
653 dax_region = kzalloc(sizeof(*dax_region), GFP_KERNEL);
654 if (!dax_region)
655 return NULL;
656
657 dev_set_drvdata(parent, dax_region);
658 kref_init(&dax_region->kref);
659 dax_region->id = region_id;
660 dax_region->align = align;
661 dax_region->dev = parent;
662 dax_region->target_node = target_node;
663 ida_init(&dax_region->ida);
664 dax_region->res = (struct resource) {
665 .start = range->start,
666 .end = range->end,
667 .flags = IORESOURCE_MEM | flags,
668 };
669
670 if (sysfs_create_groups(&parent->kobj, dax_region_attribute_groups)) {
671 kfree(dax_region);
672 return NULL;
673 }
674
675 if (devm_add_action_or_reset(parent, dax_region_unregister, dax_region))
676 return NULL;
677 return dax_region;
678}
679EXPORT_SYMBOL_GPL(alloc_dax_region);
680
681static void dax_mapping_release(struct device *dev)
682{
683 struct dax_mapping *mapping = to_dax_mapping(dev);
684 struct device *parent = dev->parent;
685 struct dev_dax *dev_dax = to_dev_dax(parent);
686
687 ida_free(&dev_dax->ida, mapping->id);
688 kfree(mapping);
689 put_device(parent);
690}
691
692static void unregister_dax_mapping(void *data)
693{
694 struct device *dev = data;
695 struct dax_mapping *mapping = to_dax_mapping(dev);
696 struct dev_dax *dev_dax = to_dev_dax(dev->parent);
697
698 dev_dbg(dev, "%s\n", __func__);
699
700 dev_dax->ranges[mapping->range_id].mapping = NULL;
701 mapping->range_id = -1;
702
703 device_unregister(dev);
704}
705
706static struct dev_dax_range *get_dax_range(struct device *dev)
707{
708 struct dax_mapping *mapping = to_dax_mapping(dev);
709 struct dev_dax *dev_dax = to_dev_dax(dev->parent);
710 int rc;
711
712 rc = down_write_killable(&dax_region_rwsem);
713 if (rc)
714 return NULL;
715 if (mapping->range_id < 0) {
716 up_write(&dax_region_rwsem);
717 return NULL;
718 }
719
720 return &dev_dax->ranges[mapping->range_id];
721}
722
723static void put_dax_range(void)
724{
725 up_write(&dax_region_rwsem);
726}
727
728static ssize_t start_show(struct device *dev,
729 struct device_attribute *attr, char *buf)
730{
731 struct dev_dax_range *dax_range;
732 ssize_t rc;
733
734 dax_range = get_dax_range(dev);
735 if (!dax_range)
736 return -ENXIO;
737 rc = sysfs_emit(buf, "%#llx\n", dax_range->range.start);
738 put_dax_range();
739
740 return rc;
741}
742static DEVICE_ATTR(start, 0400, start_show, NULL);
743
744static ssize_t end_show(struct device *dev,
745 struct device_attribute *attr, char *buf)
746{
747 struct dev_dax_range *dax_range;
748 ssize_t rc;
749
750 dax_range = get_dax_range(dev);
751 if (!dax_range)
752 return -ENXIO;
753 rc = sysfs_emit(buf, "%#llx\n", dax_range->range.end);
754 put_dax_range();
755
756 return rc;
757}
758static DEVICE_ATTR(end, 0400, end_show, NULL);
759
760static ssize_t pgoff_show(struct device *dev,
761 struct device_attribute *attr, char *buf)
762{
763 struct dev_dax_range *dax_range;
764 ssize_t rc;
765
766 dax_range = get_dax_range(dev);
767 if (!dax_range)
768 return -ENXIO;
769 rc = sysfs_emit(buf, "%#lx\n", dax_range->pgoff);
770 put_dax_range();
771
772 return rc;
773}
774static DEVICE_ATTR(page_offset, 0400, pgoff_show, NULL);
775
776static struct attribute *dax_mapping_attributes[] = {
777 &dev_attr_start.attr,
778 &dev_attr_end.attr,
779 &dev_attr_page_offset.attr,
780 NULL,
781};
782
783static const struct attribute_group dax_mapping_attribute_group = {
784 .attrs = dax_mapping_attributes,
785};
786
787static const struct attribute_group *dax_mapping_attribute_groups[] = {
788 &dax_mapping_attribute_group,
789 NULL,
790};
791
792static const struct device_type dax_mapping_type = {
793 .release = dax_mapping_release,
794 .groups = dax_mapping_attribute_groups,
795};
796
797static int devm_register_dax_mapping(struct dev_dax *dev_dax, int range_id)
798{
799 struct dax_region *dax_region = dev_dax->region;
800 struct dax_mapping *mapping;
801 struct device *dev;
802 int rc;
803
804 lockdep_assert_held_write(&dax_region_rwsem);
805
806 if (dev_WARN_ONCE(&dev_dax->dev, !dax_region->dev->driver,
807 "region disabled\n"))
808 return -ENXIO;
809
810 mapping = kzalloc(sizeof(*mapping), GFP_KERNEL);
811 if (!mapping)
812 return -ENOMEM;
813 mapping->range_id = range_id;
814 mapping->id = ida_alloc(&dev_dax->ida, GFP_KERNEL);
815 if (mapping->id < 0) {
816 kfree(mapping);
817 return -ENOMEM;
818 }
819 dev_dax->ranges[range_id].mapping = mapping;
820 dev = &mapping->dev;
821 device_initialize(dev);
822 dev->parent = &dev_dax->dev;
823 get_device(dev->parent);
824 dev->type = &dax_mapping_type;
825 dev_set_name(dev, "mapping%d", mapping->id);
826 rc = device_add(dev);
827 if (rc) {
828 put_device(dev);
829 return rc;
830 }
831
832 rc = devm_add_action_or_reset(dax_region->dev, unregister_dax_mapping,
833 dev);
834 if (rc)
835 return rc;
836 return 0;
837}
838
839static int alloc_dev_dax_range(struct dev_dax *dev_dax, u64 start,
840 resource_size_t size)
841{
842 struct dax_region *dax_region = dev_dax->region;
843 struct resource *res = &dax_region->res;
844 struct device *dev = &dev_dax->dev;
845 struct dev_dax_range *ranges;
846 unsigned long pgoff = 0;
847 struct resource *alloc;
848 int i, rc;
849
850 lockdep_assert_held_write(&dax_region_rwsem);
851
852 /* handle the seed alloc special case */
853 if (!size) {
854 if (dev_WARN_ONCE(dev, dev_dax->nr_range,
855 "0-size allocation must be first\n"))
856 return -EBUSY;
857 /* nr_range == 0 is elsewhere special cased as 0-size device */
858 return 0;
859 }
860
861 alloc = __request_region(res, start, size, dev_name(dev), 0);
862 if (!alloc)
863 return -ENOMEM;
864
865 ranges = krealloc(dev_dax->ranges, sizeof(*ranges)
866 * (dev_dax->nr_range + 1), GFP_KERNEL);
867 if (!ranges) {
868 __release_region(res, alloc->start, resource_size(alloc));
869 return -ENOMEM;
870 }
871
872 for (i = 0; i < dev_dax->nr_range; i++)
873 pgoff += PHYS_PFN(range_len(&ranges[i].range));
874 dev_dax->ranges = ranges;
875 ranges[dev_dax->nr_range++] = (struct dev_dax_range) {
876 .pgoff = pgoff,
877 .range = {
878 .start = alloc->start,
879 .end = alloc->end,
880 },
881 };
882
883 dev_dbg(dev, "alloc range[%d]: %pa:%pa\n", dev_dax->nr_range - 1,
884 &alloc->start, &alloc->end);
885 /*
886 * A dev_dax instance must be registered before mapping device
887 * children can be added. Defer to devm_create_dev_dax() to add
888 * the initial mapping device.
889 */
890 if (!device_is_registered(&dev_dax->dev))
891 return 0;
892
893 rc = devm_register_dax_mapping(dev_dax, dev_dax->nr_range - 1);
894 if (rc)
895 trim_dev_dax_range(dev_dax);
896
897 return rc;
898}
899
900static int adjust_dev_dax_range(struct dev_dax *dev_dax, struct resource *res, resource_size_t size)
901{
902 int last_range = dev_dax->nr_range - 1;
903 struct dev_dax_range *dax_range = &dev_dax->ranges[last_range];
904 bool is_shrink = resource_size(res) > size;
905 struct range *range = &dax_range->range;
906 struct device *dev = &dev_dax->dev;
907 int rc;
908
909 lockdep_assert_held_write(&dax_region_rwsem);
910
911 if (dev_WARN_ONCE(dev, !size, "deletion is handled by dev_dax_shrink\n"))
912 return -EINVAL;
913
914 rc = adjust_resource(res, range->start, size);
915 if (rc)
916 return rc;
917
918 *range = (struct range) {
919 .start = range->start,
920 .end = range->start + size - 1,
921 };
922
923 dev_dbg(dev, "%s range[%d]: %#llx:%#llx\n", is_shrink ? "shrink" : "extend",
924 last_range, (unsigned long long) range->start,
925 (unsigned long long) range->end);
926
927 return 0;
928}
929
930static ssize_t size_show(struct device *dev,
931 struct device_attribute *attr, char *buf)
932{
933 struct dev_dax *dev_dax = to_dev_dax(dev);
934 unsigned long long size;
935 int rc;
936
937 rc = down_read_interruptible(&dax_dev_rwsem);
938 if (rc)
939 return rc;
940 size = dev_dax_size(dev_dax);
941 up_read(&dax_dev_rwsem);
942
943 return sysfs_emit(buf, "%llu\n", size);
944}
945
946static bool alloc_is_aligned(struct dev_dax *dev_dax, resource_size_t size)
947{
948 /*
949 * The minimum mapping granularity for a device instance is a
950 * single subsection, unless the arch says otherwise.
951 */
952 return IS_ALIGNED(size, max_t(unsigned long, dev_dax->align, memremap_compat_align()));
953}
954
955static int dev_dax_shrink(struct dev_dax *dev_dax, resource_size_t size)
956{
957 resource_size_t to_shrink = dev_dax_size(dev_dax) - size;
958 struct dax_region *dax_region = dev_dax->region;
959 struct device *dev = &dev_dax->dev;
960 int i;
961
962 for (i = dev_dax->nr_range - 1; i >= 0; i--) {
963 struct range *range = &dev_dax->ranges[i].range;
964 struct dax_mapping *mapping = dev_dax->ranges[i].mapping;
965 struct resource *adjust = NULL, *res;
966 resource_size_t shrink;
967
968 shrink = min_t(u64, to_shrink, range_len(range));
969 if (shrink >= range_len(range)) {
970 devm_release_action(dax_region->dev,
971 unregister_dax_mapping, &mapping->dev);
972 trim_dev_dax_range(dev_dax);
973 to_shrink -= shrink;
974 if (!to_shrink)
975 break;
976 continue;
977 }
978
979 for_each_dax_region_resource(dax_region, res)
980 if (strcmp(res->name, dev_name(dev)) == 0
981 && res->start == range->start) {
982 adjust = res;
983 break;
984 }
985
986 if (dev_WARN_ONCE(dev, !adjust || i != dev_dax->nr_range - 1,
987 "failed to find matching resource\n"))
988 return -ENXIO;
989 return adjust_dev_dax_range(dev_dax, adjust, range_len(range)
990 - shrink);
991 }
992 return 0;
993}
994
995/*
996 * Only allow adjustments that preserve the relative pgoff of existing
997 * allocations. I.e. the dev_dax->ranges array is ordered by increasing pgoff.
998 */
999static bool adjust_ok(struct dev_dax *dev_dax, struct resource *res)
1000{
1001 struct dev_dax_range *last;
1002 int i;
1003
1004 if (dev_dax->nr_range == 0)
1005 return false;
1006 if (strcmp(res->name, dev_name(&dev_dax->dev)) != 0)
1007 return false;
1008 last = &dev_dax->ranges[dev_dax->nr_range - 1];
1009 if (last->range.start != res->start || last->range.end != res->end)
1010 return false;
1011 for (i = 0; i < dev_dax->nr_range - 1; i++) {
1012 struct dev_dax_range *dax_range = &dev_dax->ranges[i];
1013
1014 if (dax_range->pgoff > last->pgoff)
1015 return false;
1016 }
1017
1018 return true;
1019}
1020
1021static ssize_t dev_dax_resize(struct dax_region *dax_region,
1022 struct dev_dax *dev_dax, resource_size_t size)
1023{
1024 resource_size_t avail = dax_region_avail_size(dax_region), to_alloc;
1025 resource_size_t dev_size = dev_dax_size(dev_dax);
1026 struct resource *region_res = &dax_region->res;
1027 struct device *dev = &dev_dax->dev;
1028 struct resource *res, *first;
1029 resource_size_t alloc = 0;
1030 int rc;
1031
1032 if (dev->driver)
1033 return -EBUSY;
1034 if (size == dev_size)
1035 return 0;
1036 if (size > dev_size && size - dev_size > avail)
1037 return -ENOSPC;
1038 if (size < dev_size)
1039 return dev_dax_shrink(dev_dax, size);
1040
1041 to_alloc = size - dev_size;
1042 if (dev_WARN_ONCE(dev, !alloc_is_aligned(dev_dax, to_alloc),
1043 "resize of %pa misaligned\n", &to_alloc))
1044 return -ENXIO;
1045
1046 /*
1047 * Expand the device into the unused portion of the region. This
1048 * may involve adjusting the end of an existing resource, or
1049 * allocating a new resource.
1050 */
1051retry:
1052 first = region_res->child;
1053 if (!first)
1054 return alloc_dev_dax_range(dev_dax, dax_region->res.start, to_alloc);
1055
1056 rc = -ENOSPC;
1057 for (res = first; res; res = res->sibling) {
1058 struct resource *next = res->sibling;
1059
1060 /* space at the beginning of the region */
1061 if (res == first && res->start > dax_region->res.start) {
1062 alloc = min(res->start - dax_region->res.start, to_alloc);
1063 rc = alloc_dev_dax_range(dev_dax, dax_region->res.start, alloc);
1064 break;
1065 }
1066
1067 alloc = 0;
1068 /* space between allocations */
1069 if (next && next->start > res->end + 1)
1070 alloc = min(next->start - (res->end + 1), to_alloc);
1071
1072 /* space at the end of the region */
1073 if (!alloc && !next && res->end < region_res->end)
1074 alloc = min(region_res->end - res->end, to_alloc);
1075
1076 if (!alloc)
1077 continue;
1078
1079 if (adjust_ok(dev_dax, res)) {
1080 rc = adjust_dev_dax_range(dev_dax, res, resource_size(res) + alloc);
1081 break;
1082 }
1083 rc = alloc_dev_dax_range(dev_dax, res->end + 1, alloc);
1084 break;
1085 }
1086 if (rc)
1087 return rc;
1088 to_alloc -= alloc;
1089 if (to_alloc)
1090 goto retry;
1091 return 0;
1092}
1093
1094static ssize_t size_store(struct device *dev, struct device_attribute *attr,
1095 const char *buf, size_t len)
1096{
1097 ssize_t rc;
1098 unsigned long long val;
1099 struct dev_dax *dev_dax = to_dev_dax(dev);
1100 struct dax_region *dax_region = dev_dax->region;
1101
1102 rc = kstrtoull(buf, 0, &val);
1103 if (rc)
1104 return rc;
1105
1106 if (!alloc_is_aligned(dev_dax, val)) {
1107 dev_dbg(dev, "%s: size: %lld misaligned\n", __func__, val);
1108 return -EINVAL;
1109 }
1110
1111 rc = down_write_killable(&dax_region_rwsem);
1112 if (rc)
1113 return rc;
1114 if (!dax_region->dev->driver) {
1115 rc = -ENXIO;
1116 goto err_region;
1117 }
1118 rc = down_write_killable(&dax_dev_rwsem);
1119 if (rc)
1120 goto err_dev;
1121
1122 rc = dev_dax_resize(dax_region, dev_dax, val);
1123
1124err_dev:
1125 up_write(&dax_dev_rwsem);
1126err_region:
1127 up_write(&dax_region_rwsem);
1128
1129 if (rc == 0)
1130 return len;
1131 return rc;
1132}
1133static DEVICE_ATTR_RW(size);
1134
1135static ssize_t range_parse(const char *opt, size_t len, struct range *range)
1136{
1137 unsigned long long addr = 0;
1138 char *start, *end, *str;
1139 ssize_t rc = -EINVAL;
1140
1141 str = kstrdup(opt, GFP_KERNEL);
1142 if (!str)
1143 return rc;
1144
1145 end = str;
1146 start = strsep(&end, "-");
1147 if (!start || !end)
1148 goto err;
1149
1150 rc = kstrtoull(start, 16, &addr);
1151 if (rc)
1152 goto err;
1153 range->start = addr;
1154
1155 rc = kstrtoull(end, 16, &addr);
1156 if (rc)
1157 goto err;
1158 range->end = addr;
1159
1160err:
1161 kfree(str);
1162 return rc;
1163}
1164
1165static ssize_t mapping_store(struct device *dev, struct device_attribute *attr,
1166 const char *buf, size_t len)
1167{
1168 struct dev_dax *dev_dax = to_dev_dax(dev);
1169 struct dax_region *dax_region = dev_dax->region;
1170 size_t to_alloc;
1171 struct range r;
1172 ssize_t rc;
1173
1174 rc = range_parse(buf, len, &r);
1175 if (rc)
1176 return rc;
1177
1178 rc = down_write_killable(&dax_region_rwsem);
1179 if (rc)
1180 return rc;
1181 if (!dax_region->dev->driver) {
1182 up_write(&dax_region_rwsem);
1183 return rc;
1184 }
1185 rc = down_write_killable(&dax_dev_rwsem);
1186 if (rc) {
1187 up_write(&dax_region_rwsem);
1188 return rc;
1189 }
1190
1191 to_alloc = range_len(&r);
1192 if (alloc_is_aligned(dev_dax, to_alloc))
1193 rc = alloc_dev_dax_range(dev_dax, r.start, to_alloc);
1194 up_write(&dax_dev_rwsem);
1195 up_write(&dax_region_rwsem);
1196
1197 return rc == 0 ? len : rc;
1198}
1199static DEVICE_ATTR_WO(mapping);
1200
1201static ssize_t align_show(struct device *dev,
1202 struct device_attribute *attr, char *buf)
1203{
1204 struct dev_dax *dev_dax = to_dev_dax(dev);
1205
1206 return sysfs_emit(buf, "%d\n", dev_dax->align);
1207}
1208
1209static ssize_t dev_dax_validate_align(struct dev_dax *dev_dax)
1210{
1211 struct device *dev = &dev_dax->dev;
1212 int i;
1213
1214 for (i = 0; i < dev_dax->nr_range; i++) {
1215 size_t len = range_len(&dev_dax->ranges[i].range);
1216
1217 if (!alloc_is_aligned(dev_dax, len)) {
1218 dev_dbg(dev, "%s: align %u invalid for range %d\n",
1219 __func__, dev_dax->align, i);
1220 return -EINVAL;
1221 }
1222 }
1223
1224 return 0;
1225}
1226
1227static ssize_t align_store(struct device *dev, struct device_attribute *attr,
1228 const char *buf, size_t len)
1229{
1230 struct dev_dax *dev_dax = to_dev_dax(dev);
1231 struct dax_region *dax_region = dev_dax->region;
1232 unsigned long val, align_save;
1233 ssize_t rc;
1234
1235 rc = kstrtoul(buf, 0, &val);
1236 if (rc)
1237 return -ENXIO;
1238
1239 if (!dax_align_valid(val))
1240 return -EINVAL;
1241
1242 rc = down_write_killable(&dax_region_rwsem);
1243 if (rc)
1244 return rc;
1245 if (!dax_region->dev->driver) {
1246 up_write(&dax_region_rwsem);
1247 return -ENXIO;
1248 }
1249
1250 rc = down_write_killable(&dax_dev_rwsem);
1251 if (rc) {
1252 up_write(&dax_region_rwsem);
1253 return rc;
1254 }
1255 if (dev->driver) {
1256 rc = -EBUSY;
1257 goto out_unlock;
1258 }
1259
1260 align_save = dev_dax->align;
1261 dev_dax->align = val;
1262 rc = dev_dax_validate_align(dev_dax);
1263 if (rc)
1264 dev_dax->align = align_save;
1265out_unlock:
1266 up_write(&dax_dev_rwsem);
1267 up_write(&dax_region_rwsem);
1268 return rc == 0 ? len : rc;
1269}
1270static DEVICE_ATTR_RW(align);
1271
1272static int dev_dax_target_node(struct dev_dax *dev_dax)
1273{
1274 struct dax_region *dax_region = dev_dax->region;
1275
1276 return dax_region->target_node;
1277}
1278
1279static ssize_t target_node_show(struct device *dev,
1280 struct device_attribute *attr, char *buf)
1281{
1282 struct dev_dax *dev_dax = to_dev_dax(dev);
1283
1284 return sysfs_emit(buf, "%d\n", dev_dax_target_node(dev_dax));
1285}
1286static DEVICE_ATTR_RO(target_node);
1287
1288static ssize_t resource_show(struct device *dev,
1289 struct device_attribute *attr, char *buf)
1290{
1291 struct dev_dax *dev_dax = to_dev_dax(dev);
1292 struct dax_region *dax_region = dev_dax->region;
1293 unsigned long long start;
1294
1295 if (dev_dax->nr_range < 1)
1296 start = dax_region->res.start;
1297 else
1298 start = dev_dax->ranges[0].range.start;
1299
1300 return sysfs_emit(buf, "%#llx\n", start);
1301}
1302static DEVICE_ATTR(resource, 0400, resource_show, NULL);
1303
1304static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
1305 char *buf)
1306{
1307 /*
1308 * We only ever expect to handle device-dax instances, i.e. the
1309 * @type argument to MODULE_ALIAS_DAX_DEVICE() is always zero
1310 */
1311 return sysfs_emit(buf, DAX_DEVICE_MODALIAS_FMT "\n", 0);
1312}
1313static DEVICE_ATTR_RO(modalias);
1314
1315static ssize_t numa_node_show(struct device *dev,
1316 struct device_attribute *attr, char *buf)
1317{
1318 return sysfs_emit(buf, "%d\n", dev_to_node(dev));
1319}
1320static DEVICE_ATTR_RO(numa_node);
1321
1322static ssize_t memmap_on_memory_show(struct device *dev,
1323 struct device_attribute *attr, char *buf)
1324{
1325 struct dev_dax *dev_dax = to_dev_dax(dev);
1326
1327 return sysfs_emit(buf, "%d\n", dev_dax->memmap_on_memory);
1328}
1329
1330static ssize_t memmap_on_memory_store(struct device *dev,
1331 struct device_attribute *attr,
1332 const char *buf, size_t len)
1333{
1334 struct dev_dax *dev_dax = to_dev_dax(dev);
1335 bool val;
1336 int rc;
1337
1338 rc = kstrtobool(buf, &val);
1339 if (rc)
1340 return rc;
1341
1342 if (val == true && !mhp_supports_memmap_on_memory()) {
1343 dev_dbg(dev, "memmap_on_memory is not available\n");
1344 return -EOPNOTSUPP;
1345 }
1346
1347 rc = down_write_killable(&dax_dev_rwsem);
1348 if (rc)
1349 return rc;
1350
1351 if (dev_dax->memmap_on_memory != val && dev->driver &&
1352 to_dax_drv(dev->driver)->type == DAXDRV_KMEM_TYPE) {
1353 up_write(&dax_dev_rwsem);
1354 return -EBUSY;
1355 }
1356
1357 dev_dax->memmap_on_memory = val;
1358 up_write(&dax_dev_rwsem);
1359
1360 return len;
1361}
1362static DEVICE_ATTR_RW(memmap_on_memory);
1363
1364static umode_t dev_dax_visible(struct kobject *kobj, struct attribute *a, int n)
1365{
1366 struct device *dev = container_of(kobj, struct device, kobj);
1367 struct dev_dax *dev_dax = to_dev_dax(dev);
1368 struct dax_region *dax_region = dev_dax->region;
1369
1370 if (a == &dev_attr_target_node.attr && dev_dax_target_node(dev_dax) < 0)
1371 return 0;
1372 if (a == &dev_attr_numa_node.attr && !IS_ENABLED(CONFIG_NUMA))
1373 return 0;
1374 if (a == &dev_attr_mapping.attr && is_static(dax_region))
1375 return 0;
1376 if ((a == &dev_attr_align.attr ||
1377 a == &dev_attr_size.attr) && is_static(dax_region))
1378 return 0444;
1379 return a->mode;
1380}
1381
1382static struct attribute *dev_dax_attributes[] = {
1383 &dev_attr_modalias.attr,
1384 &dev_attr_size.attr,
1385 &dev_attr_mapping.attr,
1386 &dev_attr_target_node.attr,
1387 &dev_attr_align.attr,
1388 &dev_attr_resource.attr,
1389 &dev_attr_numa_node.attr,
1390 &dev_attr_memmap_on_memory.attr,
1391 NULL,
1392};
1393
1394static const struct attribute_group dev_dax_attribute_group = {
1395 .attrs = dev_dax_attributes,
1396 .is_visible = dev_dax_visible,
1397};
1398
1399static const struct attribute_group *dax_attribute_groups[] = {
1400 &dev_dax_attribute_group,
1401 NULL,
1402};
1403
1404static void dev_dax_release(struct device *dev)
1405{
1406 struct dev_dax *dev_dax = to_dev_dax(dev);
1407 struct dax_device *dax_dev = dev_dax->dax_dev;
1408
1409 put_dax(dax_dev);
1410 free_dev_dax_id(dev_dax);
1411 kfree(dev_dax->pgmap);
1412 kfree(dev_dax);
1413}
1414
1415static const struct device_type dev_dax_type = {
1416 .release = dev_dax_release,
1417 .groups = dax_attribute_groups,
1418};
1419
1420static struct dev_dax *__devm_create_dev_dax(struct dev_dax_data *data)
1421{
1422 struct dax_region *dax_region = data->dax_region;
1423 struct device *parent = dax_region->dev;
1424 struct dax_device *dax_dev;
1425 struct dev_dax *dev_dax;
1426 struct inode *inode;
1427 struct device *dev;
1428 int rc;
1429
1430 dev_dax = kzalloc(sizeof(*dev_dax), GFP_KERNEL);
1431 if (!dev_dax)
1432 return ERR_PTR(-ENOMEM);
1433
1434 dev_dax->region = dax_region;
1435 if (is_static(dax_region)) {
1436 if (dev_WARN_ONCE(parent, data->id < 0,
1437 "dynamic id specified to static region\n")) {
1438 rc = -EINVAL;
1439 goto err_id;
1440 }
1441
1442 dev_dax->id = data->id;
1443 } else {
1444 if (dev_WARN_ONCE(parent, data->id >= 0,
1445 "static id specified to dynamic region\n")) {
1446 rc = -EINVAL;
1447 goto err_id;
1448 }
1449
1450 rc = alloc_dev_dax_id(dev_dax);
1451 if (rc < 0)
1452 goto err_id;
1453 }
1454
1455 dev = &dev_dax->dev;
1456 device_initialize(dev);
1457 dev_set_name(dev, "dax%d.%d", dax_region->id, dev_dax->id);
1458
1459 rc = alloc_dev_dax_range(dev_dax, dax_region->res.start, data->size);
1460 if (rc)
1461 goto err_range;
1462
1463 if (data->pgmap) {
1464 dev_WARN_ONCE(parent, !is_static(dax_region),
1465 "custom dev_pagemap requires a static dax_region\n");
1466
1467 dev_dax->pgmap = kmemdup(data->pgmap,
1468 sizeof(struct dev_pagemap), GFP_KERNEL);
1469 if (!dev_dax->pgmap) {
1470 rc = -ENOMEM;
1471 goto err_pgmap;
1472 }
1473 }
1474
1475 /*
1476 * No dax_operations since there is no access to this device outside of
1477 * mmap of the resulting character device.
1478 */
1479 dax_dev = alloc_dax(dev_dax, NULL);
1480 if (IS_ERR(dax_dev)) {
1481 rc = PTR_ERR(dax_dev);
1482 goto err_alloc_dax;
1483 }
1484 set_dax_synchronous(dax_dev);
1485 set_dax_nocache(dax_dev);
1486 set_dax_nomc(dax_dev);
1487
1488 /* a device_dax instance is dead while the driver is not attached */
1489 kill_dax(dax_dev);
1490
1491 dev_dax->dax_dev = dax_dev;
1492 dev_dax->target_node = dax_region->target_node;
1493 dev_dax->align = dax_region->align;
1494 ida_init(&dev_dax->ida);
1495
1496 dev_dax->memmap_on_memory = data->memmap_on_memory;
1497
1498 inode = dax_inode(dax_dev);
1499 dev->devt = inode->i_rdev;
1500 dev->bus = &dax_bus_type;
1501 dev->parent = parent;
1502 dev->type = &dev_dax_type;
1503
1504 rc = device_add(dev);
1505 if (rc) {
1506 kill_dev_dax(dev_dax);
1507 put_device(dev);
1508 return ERR_PTR(rc);
1509 }
1510
1511 rc = devm_add_action_or_reset(dax_region->dev, unregister_dev_dax, dev);
1512 if (rc)
1513 return ERR_PTR(rc);
1514
1515 /* register mapping device for the initial allocation range */
1516 if (dev_dax->nr_range && range_len(&dev_dax->ranges[0].range)) {
1517 rc = devm_register_dax_mapping(dev_dax, 0);
1518 if (rc)
1519 return ERR_PTR(rc);
1520 }
1521
1522 return dev_dax;
1523
1524err_alloc_dax:
1525 kfree(dev_dax->pgmap);
1526err_pgmap:
1527 free_dev_dax_ranges(dev_dax);
1528err_range:
1529 free_dev_dax_id(dev_dax);
1530err_id:
1531 kfree(dev_dax);
1532
1533 return ERR_PTR(rc);
1534}
1535
1536struct dev_dax *devm_create_dev_dax(struct dev_dax_data *data)
1537{
1538 struct dev_dax *dev_dax;
1539
1540 down_write(&dax_region_rwsem);
1541 dev_dax = __devm_create_dev_dax(data);
1542 up_write(&dax_region_rwsem);
1543
1544 return dev_dax;
1545}
1546EXPORT_SYMBOL_GPL(devm_create_dev_dax);
1547
1548int __dax_driver_register(struct dax_device_driver *dax_drv,
1549 struct module *module, const char *mod_name)
1550{
1551 struct device_driver *drv = &dax_drv->drv;
1552
1553 /*
1554 * dax_bus_probe() calls dax_drv->probe() unconditionally.
1555 * So better be safe than sorry and ensure it is provided.
1556 */
1557 if (!dax_drv->probe)
1558 return -EINVAL;
1559
1560 INIT_LIST_HEAD(&dax_drv->ids);
1561 drv->owner = module;
1562 drv->name = mod_name;
1563 drv->mod_name = mod_name;
1564 drv->bus = &dax_bus_type;
1565
1566 return driver_register(drv);
1567}
1568EXPORT_SYMBOL_GPL(__dax_driver_register);
1569
1570void dax_driver_unregister(struct dax_device_driver *dax_drv)
1571{
1572 struct device_driver *drv = &dax_drv->drv;
1573 struct dax_id *dax_id, *_id;
1574
1575 mutex_lock(&dax_bus_lock);
1576 list_for_each_entry_safe(dax_id, _id, &dax_drv->ids, list) {
1577 list_del(&dax_id->list);
1578 kfree(dax_id);
1579 }
1580 mutex_unlock(&dax_bus_lock);
1581 driver_unregister(drv);
1582}
1583EXPORT_SYMBOL_GPL(dax_driver_unregister);
1584
1585int __init dax_bus_init(void)
1586{
1587 return bus_register(&dax_bus_type);
1588}
1589
1590void __exit dax_bus_exit(void)
1591{
1592 bus_unregister(&dax_bus_type);
1593}