projects/linux·blob master · refs

1625 lines · 38.4 KiB

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 struct resource dax_regions = DEFINE_RES_MEM_NAMED(0, -1, "DAX Regions");
14static DEFINE_MUTEX(dax_bus_lock);
15
16/*
17 * All changes to the dax region configuration occur with this lock held
18 * for write.
19 */
20DECLARE_RWSEM(dax_region_rwsem);
21
22/*
23 * All changes to the dax device configuration occur with this lock held
24 * for write.
25 */
26DECLARE_RWSEM(dax_dev_rwsem);
27
28#define DAX_NAME_LEN 30
29struct dax_id {
30 struct list_head list;
31 char dev_name[DAX_NAME_LEN];
32};
33
34static int dax_bus_uevent(const struct device *dev, struct kobj_uevent_env *env)
35{
36 /*
37 * We only ever expect to handle device-dax instances, i.e. the
38 * @type argument to MODULE_ALIAS_DAX_DEVICE() is always zero
39 */
40 return add_uevent_var(env, "MODALIAS=" DAX_DEVICE_MODALIAS_FMT, 0);
41}
42
43static struct dax_id *__dax_match_id(const struct dax_device_driver *dax_drv,
44 const char *dev_name)
45{
46 struct dax_id *dax_id;
47
48 lockdep_assert_held(&dax_bus_lock);
49
50 list_for_each_entry(dax_id, &dax_drv->ids, list)
51 if (sysfs_streq(dax_id->dev_name, dev_name))
52 return dax_id;
53 return NULL;
54}
55
56static int dax_match_id(const struct dax_device_driver *dax_drv, struct device *dev)
57{
58 int match;
59
60 mutex_lock(&dax_bus_lock);
61 match = !!__dax_match_id(dax_drv, dev_name(dev));
62 mutex_unlock(&dax_bus_lock);
63
64 return match;
65}
66
67static int dax_match_type(const struct dax_device_driver *dax_drv, struct device *dev)
68{
69 enum dax_driver_type type = DAXDRV_DEVICE_TYPE;
70 struct dev_dax *dev_dax = to_dev_dax(dev);
71
72 if (dev_dax->region->res.flags & IORESOURCE_DAX_KMEM)
73 type = DAXDRV_KMEM_TYPE;
74
75 if (dax_drv->type == type)
76 return 1;
77
78 /* default to device mode if dax_kmem is disabled */
79 if (dax_drv->type == DAXDRV_DEVICE_TYPE &&
80 !IS_ENABLED(CONFIG_DEV_DAX_KMEM))
81 return 1;
82
83 return 0;
84}
85
86enum id_action {
87 ID_REMOVE,
88 ID_ADD,
89};
90
91static ssize_t do_id_store(struct device_driver *drv, const char *buf,
92 size_t count, enum id_action action)
93{
94 struct dax_device_driver *dax_drv = to_dax_drv(drv);
95 unsigned int region_id, id;
96 char devname[DAX_NAME_LEN];
97 struct dax_id *dax_id;
98 ssize_t rc = count;
99 int fields;
100
101 fields = sscanf(buf, "dax%d.%d", &region_id, &id);
102 if (fields != 2)
103 return -EINVAL;
104 sprintf(devname, "dax%d.%d", region_id, id);
105 if (!sysfs_streq(buf, devname))
106 return -EINVAL;
107
108 mutex_lock(&dax_bus_lock);
109 dax_id = __dax_match_id(dax_drv, buf);
110 if (!dax_id) {
111 if (action == ID_ADD) {
112 dax_id = kzalloc_obj(*dax_id);
113 if (dax_id) {
114 strscpy(dax_id->dev_name, buf, DAX_NAME_LEN);
115 list_add(&dax_id->list, &dax_drv->ids);
116 } else
117 rc = -ENOMEM;
118 }
119 } else if (action == ID_REMOVE) {
120 list_del(&dax_id->list);
121 kfree(dax_id);
122 }
123 mutex_unlock(&dax_bus_lock);
124
125 if (rc < 0)
126 return rc;
127 if (action == ID_ADD)
128 rc = driver_attach(drv);
129 if (rc)
130 return rc;
131 return count;
132}
133
134static ssize_t new_id_store(struct device_driver *drv, const char *buf,
135 size_t count)
136{
137 return do_id_store(drv, buf, count, ID_ADD);
138}
139static DRIVER_ATTR_WO(new_id);
140
141static ssize_t remove_id_store(struct device_driver *drv, const char *buf,
142 size_t count)
143{
144 return do_id_store(drv, buf, count, ID_REMOVE);
145}
146static DRIVER_ATTR_WO(remove_id);
147
148static struct attribute *dax_drv_attrs[] = {
149 &driver_attr_new_id.attr,
150 &driver_attr_remove_id.attr,
151 NULL,
152};
153ATTRIBUTE_GROUPS(dax_drv);
154
155static int dax_bus_match(struct device *dev, const struct device_driver *drv);
156
157/*
158 * Static dax regions are regions created by an external subsystem
159 * nvdimm where a single range is assigned. Its boundaries are by the external
160 * subsystem and are usually limited to one physical memory range. For example,
161 * for PMEM it is usually defined by NVDIMM Namespace boundaries (i.e. a
162 * single contiguous range)
163 *
164 * On dynamic dax regions, the assigned region can be partitioned by dax core
165 * into multiple subdivisions. A subdivision is represented into one
166 * /dev/daxN.M device composed by one or more potentially discontiguous ranges.
167 *
168 * When allocating a dax region, drivers must set whether it's static
169 * (IORESOURCE_DAX_STATIC). On static dax devices, the @pgmap is pre-assigned
170 * to dax core when calling devm_create_dev_dax(), whereas in dynamic dax
171 * devices it is NULL but afterwards allocated by dax core on device ->probe().
172 * Care is needed to make sure that dynamic dax devices are torn down with a
173 * cleared @pgmap field (see kill_dev_dax()).
174 */
175static bool is_static(struct dax_region *dax_region)
176{
177 return (dax_region->res.flags & IORESOURCE_DAX_STATIC) != 0;
178}
179
180bool static_dev_dax(struct dev_dax *dev_dax)
181{
182 return is_static(dev_dax->region);
183}
184EXPORT_SYMBOL_GPL(static_dev_dax);
185
186static u64 dev_dax_size(struct dev_dax *dev_dax)
187{
188 u64 size = 0;
189 int i;
190
191 lockdep_assert_held(&dax_dev_rwsem);
192
193 for (i = 0; i < dev_dax->nr_range; i++)
194 size += range_len(&dev_dax->ranges[i].range);
195
196 return size;
197}
198
199static int dax_bus_probe(struct device *dev)
200{
201 struct dax_device_driver *dax_drv = to_dax_drv(dev->driver);
202 struct dev_dax *dev_dax = to_dev_dax(dev);
203 struct dax_region *dax_region = dev_dax->region;
204 int rc;
205 u64 size;
206
207 rc = down_read_interruptible(&dax_dev_rwsem);
208 if (rc)
209 return rc;
210 size = dev_dax_size(dev_dax);
211 up_read(&dax_dev_rwsem);
212
213 if (size == 0 || dev_dax->id < 0)
214 return -ENXIO;
215
216 rc = dax_drv->probe(dev_dax);
217
218 if (rc || is_static(dax_region))
219 return rc;
220
221 /*
222 * Track new seed creation only after successful probe of the
223 * previous seed.
224 */
225 if (dax_region->seed == dev)
226 dax_region->seed = NULL;
227
228 return 0;
229}
230
231static void dax_bus_remove(struct device *dev)
232{
233 struct dax_device_driver *dax_drv = to_dax_drv(dev->driver);
234 struct dev_dax *dev_dax = to_dev_dax(dev);
235
236 if (dax_drv->remove)
237 dax_drv->remove(dev_dax);
238}
239
240static const struct bus_type dax_bus_type = {
241 .name = "dax",
242 .uevent = dax_bus_uevent,
243 .match = dax_bus_match,
244 .probe = dax_bus_probe,
245 .remove = dax_bus_remove,
246 .drv_groups = dax_drv_groups,
247};
248
249static int dax_bus_match(struct device *dev, const struct device_driver *drv)
250{
251 const struct dax_device_driver *dax_drv = to_dax_drv(drv);
252
253 if (dax_match_id(dax_drv, dev))
254 return 1;
255 return dax_match_type(dax_drv, dev);
256}
257
258/*
259 * Rely on the fact that drvdata is set before the attributes are
260 * registered, and that the attributes are unregistered before drvdata
261 * is cleared to assume that drvdata is always valid.
262 */
263static ssize_t id_show(struct device *dev,
264 struct device_attribute *attr, char *buf)
265{
266 struct dax_region *dax_region = dev_get_drvdata(dev);
267
268 return sysfs_emit(buf, "%d\n", dax_region->id);
269}
270static DEVICE_ATTR_RO(id);
271
272static ssize_t region_size_show(struct device *dev,
273 struct device_attribute *attr, char *buf)
274{
275 struct dax_region *dax_region = dev_get_drvdata(dev);
276
277 return sysfs_emit(buf, "%llu\n",
278 (unsigned long long)resource_size(&dax_region->res));
279}
280static struct device_attribute dev_attr_region_size = __ATTR(size, 0444,
281 region_size_show, NULL);
282
283static ssize_t region_align_show(struct device *dev,
284 struct device_attribute *attr, char *buf)
285{
286 struct dax_region *dax_region = dev_get_drvdata(dev);
287
288 return sysfs_emit(buf, "%u\n", dax_region->align);
289}
290static struct device_attribute dev_attr_region_align =
291 __ATTR(align, 0400, region_align_show, NULL);
292
293#define for_each_dax_region_resource(dax_region, res) \
294 for (res = (dax_region)->res.child; res; res = res->sibling)
295
296static unsigned long long dax_region_avail_size(struct dax_region *dax_region)
297{
298 resource_size_t size = resource_size(&dax_region->res);
299 struct resource *res;
300
301 lockdep_assert_held(&dax_region_rwsem);
302
303 for_each_dax_region_resource(dax_region, res)
304 size -= resource_size(res);
305 return size;
306}
307
308static ssize_t available_size_show(struct device *dev,
309 struct device_attribute *attr, char *buf)
310{
311 struct dax_region *dax_region = dev_get_drvdata(dev);
312 unsigned long long size;
313 int rc;
314
315 rc = down_read_interruptible(&dax_region_rwsem);
316 if (rc)
317 return rc;
318 size = dax_region_avail_size(dax_region);
319 up_read(&dax_region_rwsem);
320
321 return sysfs_emit(buf, "%llu\n", size);
322}
323static DEVICE_ATTR_RO(available_size);
324
325static ssize_t seed_show(struct device *dev,
326 struct device_attribute *attr, char *buf)
327{
328 struct dax_region *dax_region = dev_get_drvdata(dev);
329 struct device *seed;
330 ssize_t rc;
331
332 if (is_static(dax_region))
333 return -EINVAL;
334
335 rc = down_read_interruptible(&dax_region_rwsem);
336 if (rc)
337 return rc;
338 seed = dax_region->seed;
339 rc = sysfs_emit(buf, "%s\n", seed ? dev_name(seed) : "");
340 up_read(&dax_region_rwsem);
341
342 return rc;
343}
344static DEVICE_ATTR_RO(seed);
345
346static ssize_t create_show(struct device *dev,
347 struct device_attribute *attr, char *buf)
348{
349 struct dax_region *dax_region = dev_get_drvdata(dev);
350 struct device *youngest;
351 ssize_t rc;
352
353 if (is_static(dax_region))
354 return -EINVAL;
355
356 rc = down_read_interruptible(&dax_region_rwsem);
357 if (rc)
358 return rc;
359 youngest = dax_region->youngest;
360 rc = sysfs_emit(buf, "%s\n", youngest ? dev_name(youngest) : "");
361 up_read(&dax_region_rwsem);
362
363 return rc;
364}
365
366static struct dev_dax *__devm_create_dev_dax(struct dev_dax_data *data);
367
368static ssize_t create_store(struct device *dev, struct device_attribute *attr,
369 const char *buf, size_t len)
370{
371 struct dax_region *dax_region = dev_get_drvdata(dev);
372 unsigned long long avail;
373 ssize_t rc;
374 int val;
375
376 if (is_static(dax_region))
377 return -EINVAL;
378
379 rc = kstrtoint(buf, 0, &val);
380 if (rc)
381 return rc;
382 if (val != 1)
383 return -EINVAL;
384
385 rc = down_write_killable(&dax_region_rwsem);
386 if (rc)
387 return rc;
388 avail = dax_region_avail_size(dax_region);
389 if (avail == 0)
390 rc = -ENOSPC;
391 else {
392 struct dev_dax_data data = {
393 .dax_region = dax_region,
394 .size = 0,
395 .id = -1,
396 .memmap_on_memory = false,
397 };
398 struct dev_dax *dev_dax = __devm_create_dev_dax(&data);
399
400 if (IS_ERR(dev_dax))
401 rc = PTR_ERR(dev_dax);
402 else {
403 /*
404 * In support of crafting multiple new devices
405 * simultaneously multiple seeds can be created,
406 * but only the first one that has not been
407 * successfully bound is tracked as the region
408 * seed.
409 */
410 if (!dax_region->seed)
411 dax_region->seed = &dev_dax->dev;
412 dax_region->youngest = &dev_dax->dev;
413 rc = len;
414 }
415 }
416 up_write(&dax_region_rwsem);
417
418 return rc;
419}
420static DEVICE_ATTR_RW(create);
421
422void kill_dev_dax(struct dev_dax *dev_dax)
423{
424 struct dax_device *dax_dev = dev_dax->dax_dev;
425 struct inode *inode = dax_inode(dax_dev);
426
427 kill_dax(dax_dev);
428 unmap_mapping_range(inode->i_mapping, 0, 0, 1);
429
430 /*
431 * Dynamic dax region have the pgmap allocated via dev_kzalloc()
432 * and thus freed by devm. Clear the pgmap to not have stale pgmap
433 * ranges on probe() from previous reconfigurations of region devices.
434 */
435 if (!static_dev_dax(dev_dax))
436 dev_dax->pgmap = NULL;
437}
438EXPORT_SYMBOL_GPL(kill_dev_dax);
439
440static void trim_dev_dax_range(struct dev_dax *dev_dax)
441{
442 int i = dev_dax->nr_range - 1;
443 struct range *range = &dev_dax->ranges[i].range;
444 struct dax_region *dax_region = dev_dax->region;
445
446 lockdep_assert_held_write(&dax_region_rwsem);
447 dev_dbg(&dev_dax->dev, "delete range[%d]: %#llx:%#llx\n", i,
448 (unsigned long long)range->start,
449 (unsigned long long)range->end);
450
451 __release_region(&dax_region->res, range->start, range_len(range));
452 if (--dev_dax->nr_range == 0) {
453 kfree(dev_dax->ranges);
454 dev_dax->ranges = NULL;
455 }
456}
457
458static void free_dev_dax_ranges(struct dev_dax *dev_dax)
459{
460 while (dev_dax->nr_range)
461 trim_dev_dax_range(dev_dax);
462}
463
464static void unregister_dev_dax(void *dev)
465{
466 struct dev_dax *dev_dax = to_dev_dax(dev);
467
468 dev_dbg(dev, "%s\n", __func__);
469
470 down_write(&dax_region_rwsem);
471 kill_dev_dax(dev_dax);
472 device_del(dev);
473 free_dev_dax_ranges(dev_dax);
474 put_device(dev);
475 up_write(&dax_region_rwsem);
476}
477
478static void dax_region_free(struct kref *kref)
479{
480 struct dax_region *dax_region;
481
482 dax_region = container_of(kref, struct dax_region, kref);
483 kfree(dax_region);
484}
485
486static void dax_region_put(struct dax_region *dax_region)
487{
488 kref_put(&dax_region->kref, dax_region_free);
489}
490
491/* a return value >= 0 indicates this invocation invalidated the id */
492static int __free_dev_dax_id(struct dev_dax *dev_dax)
493{
494 struct dax_region *dax_region;
495 int rc = dev_dax->id;
496
497 lockdep_assert_held_write(&dax_dev_rwsem);
498
499 if (!dev_dax->dyn_id || dev_dax->id < 0)
500 return -1;
501 dax_region = dev_dax->region;
502 ida_free(&dax_region->ida, dev_dax->id);
503 dax_region_put(dax_region);
504 dev_dax->id = -1;
505 return rc;
506}
507
508static int free_dev_dax_id(struct dev_dax *dev_dax)
509{
510 int rc;
511
512 rc = down_write_killable(&dax_dev_rwsem);
513 if (rc)
514 return rc;
515 rc = __free_dev_dax_id(dev_dax);
516 up_write(&dax_dev_rwsem);
517 return rc;
518}
519
520static int alloc_dev_dax_id(struct dev_dax *dev_dax)
521{
522 struct dax_region *dax_region = dev_dax->region;
523 int id;
524
525 id = ida_alloc(&dax_region->ida, GFP_KERNEL);
526 if (id < 0)
527 return id;
528 kref_get(&dax_region->kref);
529 dev_dax->dyn_id = true;
530 dev_dax->id = id;
531 return id;
532}
533
534static ssize_t delete_store(struct device *dev, struct device_attribute *attr,
535 const char *buf, size_t len)
536{
537 struct dax_region *dax_region = dev_get_drvdata(dev);
538 struct dev_dax *dev_dax;
539 struct device *victim;
540 bool do_del = false;
541 int rc;
542
543 if (is_static(dax_region))
544 return -EINVAL;
545
546 victim = device_find_child_by_name(dax_region->dev, buf);
547 if (!victim)
548 return -ENXIO;
549
550 device_lock(dev);
551 device_lock(victim);
552 dev_dax = to_dev_dax(victim);
553 down_write(&dax_dev_rwsem);
554 if (victim->driver || dev_dax_size(dev_dax))
555 rc = -EBUSY;
556 else {
557 /*
558 * Invalidate the device so it does not become active
559 * again, but always preserve device-id-0 so that
560 * /sys/bus/dax/ is guaranteed to be populated while any
561 * dax_region is registered.
562 */
563 if (dev_dax->id > 0) {
564 do_del = __free_dev_dax_id(dev_dax) >= 0;
565 rc = len;
566 if (dax_region->seed == victim)
567 dax_region->seed = NULL;
568 if (dax_region->youngest == victim)
569 dax_region->youngest = NULL;
570 } else
571 rc = -EBUSY;
572 }
573 up_write(&dax_dev_rwsem);
574 device_unlock(victim);
575
576 /* won the race to invalidate the device, clean it up */
577 if (do_del)
578 devm_release_action(dev, unregister_dev_dax, victim);
579 device_unlock(dev);
580 put_device(victim);
581
582 return rc;
583}
584static DEVICE_ATTR_WO(delete);
585
586static umode_t dax_region_visible(struct kobject *kobj, struct attribute *a,
587 int n)
588{
589 struct device *dev = container_of(kobj, struct device, kobj);
590 struct dax_region *dax_region = dev_get_drvdata(dev);
591
592 if (is_static(dax_region))
593 if (a == &dev_attr_available_size.attr
594 || a == &dev_attr_create.attr
595 || a == &dev_attr_seed.attr
596 || a == &dev_attr_delete.attr)
597 return 0;
598 return a->mode;
599}
600
601static struct attribute *dax_region_attributes[] = {
602 &dev_attr_available_size.attr,
603 &dev_attr_region_size.attr,
604 &dev_attr_region_align.attr,
605 &dev_attr_create.attr,
606 &dev_attr_seed.attr,
607 &dev_attr_delete.attr,
608 &dev_attr_id.attr,
609 NULL,
610};
611
612static const struct attribute_group dax_region_attribute_group = {
613 .name = "dax_region",
614 .attrs = dax_region_attributes,
615 .is_visible = dax_region_visible,
616};
617
618static const struct attribute_group *dax_region_attribute_groups[] = {
619 &dax_region_attribute_group,
620 NULL,
621};
622
623static void dax_region_unregister(void *region)
624{
625 struct dax_region *dax_region = region;
626
627 sysfs_remove_groups(&dax_region->dev->kobj,
628 dax_region_attribute_groups);
629 release_resource(&dax_region->res);
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 int rc;
639
640 /*
641 * The DAX core assumes that it can store its private data in
642 * parent->driver_data. This WARN is a reminder / safeguard for
643 * developers of device-dax drivers.
644 */
645 if (dev_get_drvdata(parent)) {
646 dev_WARN(parent, "dax core failed to setup private data\n");
647 return NULL;
648 }
649
650 if (!IS_ALIGNED(range->start, align)
651 || !IS_ALIGNED(range_len(range), align))
652 return NULL;
653
654 dax_region = kzalloc_obj(*dax_region);
655 if (!dax_region)
656 return NULL;
657
658 dev_set_drvdata(parent, dax_region);
659 kref_init(&dax_region->kref);
660 dax_region->id = region_id;
661 dax_region->align = align;
662 dax_region->dev = parent;
663 dax_region->target_node = target_node;
664 ida_init(&dax_region->ida);
665 dax_region->res = (struct resource) {
666 .start = range->start,
667 .end = range->end,
668 .flags = IORESOURCE_MEM | flags,
669 };
670
671 rc = request_resource(&dax_regions, &dax_region->res);
672 if (rc) {
673 dev_err(parent, "dax_region resource conflict for %pR\n",
674 &dax_region->res);
675 goto err_res;
676 }
677
678 if (sysfs_create_groups(&parent->kobj, dax_region_attribute_groups))
679 goto err_sysfs;
680
681 if (devm_add_action_or_reset(parent, dax_region_unregister, dax_region))
682 return NULL;
683 return dax_region;
684
685err_sysfs:
686 release_resource(&dax_region->res);
687err_res:
688 dax_region_put(dax_region);
689 return NULL;
690}
691EXPORT_SYMBOL_GPL(alloc_dax_region);
692
693static void dax_mapping_release(struct device *dev)
694{
695 struct dax_mapping *mapping = to_dax_mapping(dev);
696 struct device *parent = dev->parent;
697 struct dev_dax *dev_dax = to_dev_dax(parent);
698
699 ida_free(&dev_dax->ida, mapping->id);
700 kfree(mapping);
701 put_device(parent);
702}
703
704static void unregister_dax_mapping(void *data)
705{
706 struct device *dev = data;
707 struct dax_mapping *mapping = to_dax_mapping(dev);
708 struct dev_dax *dev_dax = to_dev_dax(dev->parent);
709
710 dev_dbg(dev, "%s\n", __func__);
711
712 dev_dax->ranges[mapping->range_id].mapping = NULL;
713 mapping->range_id = -1;
714
715 device_unregister(dev);
716}
717
718static struct dev_dax_range *get_dax_range(struct device *dev)
719{
720 struct dax_mapping *mapping = to_dax_mapping(dev);
721 struct dev_dax *dev_dax = to_dev_dax(dev->parent);
722 int rc;
723
724 rc = down_write_killable(&dax_region_rwsem);
725 if (rc)
726 return NULL;
727 if (mapping->range_id < 0) {
728 up_write(&dax_region_rwsem);
729 return NULL;
730 }
731
732 return &dev_dax->ranges[mapping->range_id];
733}
734
735static void put_dax_range(void)
736{
737 up_write(&dax_region_rwsem);
738}
739
740static ssize_t start_show(struct device *dev,
741 struct device_attribute *attr, char *buf)
742{
743 struct dev_dax_range *dax_range;
744 ssize_t rc;
745
746 dax_range = get_dax_range(dev);
747 if (!dax_range)
748 return -ENXIO;
749 rc = sysfs_emit(buf, "%#llx\n", dax_range->range.start);
750 put_dax_range();
751
752 return rc;
753}
754static DEVICE_ATTR(start, 0400, start_show, NULL);
755
756static ssize_t end_show(struct device *dev,
757 struct device_attribute *attr, char *buf)
758{
759 struct dev_dax_range *dax_range;
760 ssize_t rc;
761
762 dax_range = get_dax_range(dev);
763 if (!dax_range)
764 return -ENXIO;
765 rc = sysfs_emit(buf, "%#llx\n", dax_range->range.end);
766 put_dax_range();
767
768 return rc;
769}
770static DEVICE_ATTR(end, 0400, end_show, NULL);
771
772static ssize_t pgoff_show(struct device *dev,
773 struct device_attribute *attr, char *buf)
774{
775 struct dev_dax_range *dax_range;
776 ssize_t rc;
777
778 dax_range = get_dax_range(dev);
779 if (!dax_range)
780 return -ENXIO;
781 rc = sysfs_emit(buf, "%#lx\n", dax_range->pgoff);
782 put_dax_range();
783
784 return rc;
785}
786static DEVICE_ATTR(page_offset, 0400, pgoff_show, NULL);
787
788static struct attribute *dax_mapping_attributes[] = {
789 &dev_attr_start.attr,
790 &dev_attr_end.attr,
791 &dev_attr_page_offset.attr,
792 NULL,
793};
794
795static const struct attribute_group dax_mapping_attribute_group = {
796 .attrs = dax_mapping_attributes,
797};
798
799static const struct attribute_group *dax_mapping_attribute_groups[] = {
800 &dax_mapping_attribute_group,
801 NULL,
802};
803
804static const struct device_type dax_mapping_type = {
805 .release = dax_mapping_release,
806 .groups = dax_mapping_attribute_groups,
807};
808
809static int devm_register_dax_mapping(struct dev_dax *dev_dax, int range_id)
810{
811 struct dax_region *dax_region = dev_dax->region;
812 struct dax_mapping *mapping;
813 struct device *dev;
814 int rc;
815
816 lockdep_assert_held_write(&dax_region_rwsem);
817
818 if (dev_WARN_ONCE(&dev_dax->dev, !dax_region->dev->driver,
819 "region disabled\n"))
820 return -ENXIO;
821
822 mapping = kzalloc_obj(*mapping);
823 if (!mapping)
824 return -ENOMEM;
825 mapping->range_id = range_id;
826 mapping->id = ida_alloc(&dev_dax->ida, GFP_KERNEL);
827 if (mapping->id < 0) {
828 kfree(mapping);
829 return -ENOMEM;
830 }
831 dev_dax->ranges[range_id].mapping = mapping;
832 dev = &mapping->dev;
833 device_initialize(dev);
834 dev->parent = &dev_dax->dev;
835 get_device(dev->parent);
836 dev->type = &dax_mapping_type;
837 dev_set_name(dev, "mapping%d", mapping->id);
838 rc = device_add(dev);
839 if (rc) {
840 put_device(dev);
841 return rc;
842 }
843
844 rc = devm_add_action_or_reset(dax_region->dev, unregister_dax_mapping,
845 dev);
846 if (rc)
847 return rc;
848 return 0;
849}
850
851static int alloc_dev_dax_range(struct dev_dax *dev_dax, u64 start,
852 resource_size_t size)
853{
854 struct dax_region *dax_region = dev_dax->region;
855 struct resource *res = &dax_region->res;
856 struct device *dev = &dev_dax->dev;
857 struct dev_dax_range *ranges;
858 unsigned long pgoff = 0;
859 struct resource *alloc;
860 int i, rc;
861
862 lockdep_assert_held_write(&dax_region_rwsem);
863
864 /* handle the seed alloc special case */
865 if (!size) {
866 if (dev_WARN_ONCE(dev, dev_dax->nr_range,
867 "0-size allocation must be first\n"))
868 return -EBUSY;
869 /* nr_range == 0 is elsewhere special cased as 0-size device */
870 return 0;
871 }
872
873 alloc = __request_region(res, start, size, dev_name(dev), 0);
874 if (!alloc)
875 return -ENOMEM;
876
877 ranges = krealloc(dev_dax->ranges, sizeof(*ranges)
878 * (dev_dax->nr_range + 1), GFP_KERNEL);
879 if (!ranges) {
880 __release_region(res, alloc->start, resource_size(alloc));
881 return -ENOMEM;
882 }
883
884 for (i = 0; i < dev_dax->nr_range; i++)
885 pgoff += PHYS_PFN(range_len(&ranges[i].range));
886 dev_dax->ranges = ranges;
887 ranges[dev_dax->nr_range++] = (struct dev_dax_range) {
888 .pgoff = pgoff,
889 .range = {
890 .start = alloc->start,
891 .end = alloc->end,
892 },
893 };
894
895 dev_dbg(dev, "alloc range[%d]: %pa:%pa\n", dev_dax->nr_range - 1,
896 &alloc->start, &alloc->end);
897 /*
898 * A dev_dax instance must be registered before mapping device
899 * children can be added. Defer to devm_create_dev_dax() to add
900 * the initial mapping device.
901 */
902 if (!device_is_registered(&dev_dax->dev))
903 return 0;
904
905 rc = devm_register_dax_mapping(dev_dax, dev_dax->nr_range - 1);
906 if (rc)
907 trim_dev_dax_range(dev_dax);
908
909 return rc;
910}
911
912static int adjust_dev_dax_range(struct dev_dax *dev_dax, struct resource *res, resource_size_t size)
913{
914 int last_range = dev_dax->nr_range - 1;
915 struct dev_dax_range *dax_range = &dev_dax->ranges[last_range];
916 bool is_shrink = resource_size(res) > size;
917 struct range *range = &dax_range->range;
918 struct device *dev = &dev_dax->dev;
919 int rc;
920
921 lockdep_assert_held_write(&dax_region_rwsem);
922
923 if (dev_WARN_ONCE(dev, !size, "deletion is handled by dev_dax_shrink\n"))
924 return -EINVAL;
925
926 rc = adjust_resource(res, range->start, size);
927 if (rc)
928 return rc;
929
930 *range = (struct range) {
931 .start = range->start,
932 .end = range->start + size - 1,
933 };
934
935 dev_dbg(dev, "%s range[%d]: %#llx:%#llx\n", is_shrink ? "shrink" : "extend",
936 last_range, (unsigned long long) range->start,
937 (unsigned long long) range->end);
938
939 return 0;
940}
941
942static ssize_t size_show(struct device *dev,
943 struct device_attribute *attr, char *buf)
944{
945 struct dev_dax *dev_dax = to_dev_dax(dev);
946 unsigned long long size;
947 int rc;
948
949 rc = down_read_interruptible(&dax_dev_rwsem);
950 if (rc)
951 return rc;
952 size = dev_dax_size(dev_dax);
953 up_read(&dax_dev_rwsem);
954
955 return sysfs_emit(buf, "%llu\n", size);
956}
957
958static bool alloc_is_aligned(struct dev_dax *dev_dax, resource_size_t size)
959{
960 /*
961 * The minimum mapping granularity for a device instance is a
962 * single subsection, unless the arch says otherwise.
963 */
964 return IS_ALIGNED(size, max_t(unsigned long, dev_dax->align, memremap_compat_align()));
965}
966
967static int dev_dax_shrink(struct dev_dax *dev_dax, resource_size_t size)
968{
969 resource_size_t to_shrink = dev_dax_size(dev_dax) - size;
970 struct dax_region *dax_region = dev_dax->region;
971 struct device *dev = &dev_dax->dev;
972 int i;
973
974 for (i = dev_dax->nr_range - 1; i >= 0; i--) {
975 struct range *range = &dev_dax->ranges[i].range;
976 struct dax_mapping *mapping = dev_dax->ranges[i].mapping;
977 struct resource *adjust = NULL, *res;
978 resource_size_t shrink;
979
980 shrink = min_t(u64, to_shrink, range_len(range));
981 if (shrink >= range_len(range)) {
982 devm_release_action(dax_region->dev,
983 unregister_dax_mapping, &mapping->dev);
984 trim_dev_dax_range(dev_dax);
985 to_shrink -= shrink;
986 if (!to_shrink)
987 break;
988 continue;
989 }
990
991 for_each_dax_region_resource(dax_region, res)
992 if (strcmp(res->name, dev_name(dev)) == 0
993 && res->start == range->start) {
994 adjust = res;
995 break;
996 }
997
998 if (dev_WARN_ONCE(dev, !adjust || i != dev_dax->nr_range - 1,
999 "failed to find matching resource\n"))
1000 return -ENXIO;
1001 return adjust_dev_dax_range(dev_dax, adjust, range_len(range)
1002 - shrink);
1003 }
1004 return 0;
1005}
1006
1007/*
1008 * Only allow adjustments that preserve the relative pgoff of existing
1009 * allocations. I.e. the dev_dax->ranges array is ordered by increasing pgoff.
1010 */
1011static bool adjust_ok(struct dev_dax *dev_dax, struct resource *res)
1012{
1013 struct dev_dax_range *last;
1014 int i;
1015
1016 if (dev_dax->nr_range == 0)
1017 return false;
1018 if (strcmp(res->name, dev_name(&dev_dax->dev)) != 0)
1019 return false;
1020 last = &dev_dax->ranges[dev_dax->nr_range - 1];
1021 if (last->range.start != res->start || last->range.end != res->end)
1022 return false;
1023 for (i = 0; i < dev_dax->nr_range - 1; i++) {
1024 struct dev_dax_range *dax_range = &dev_dax->ranges[i];
1025
1026 if (dax_range->pgoff > last->pgoff)
1027 return false;
1028 }
1029
1030 return true;
1031}
1032
1033static ssize_t dev_dax_resize(struct dax_region *dax_region,
1034 struct dev_dax *dev_dax, resource_size_t size)
1035{
1036 resource_size_t avail = dax_region_avail_size(dax_region), to_alloc;
1037 resource_size_t dev_size = dev_dax_size(dev_dax);
1038 struct resource *region_res = &dax_region->res;
1039 struct device *dev = &dev_dax->dev;
1040 struct resource *res, *first;
1041 resource_size_t alloc = 0;
1042 int rc;
1043
1044 if (dev->driver)
1045 return -EBUSY;
1046 if (size == dev_size)
1047 return 0;
1048 if (size > dev_size && size - dev_size > avail)
1049 return -ENOSPC;
1050 if (size < dev_size)
1051 return dev_dax_shrink(dev_dax, size);
1052
1053 to_alloc = size - dev_size;
1054 if (dev_WARN_ONCE(dev, !alloc_is_aligned(dev_dax, to_alloc),
1055 "resize of %pa misaligned\n", &to_alloc))
1056 return -ENXIO;
1057
1058 /*
1059 * Expand the device into the unused portion of the region. This
1060 * may involve adjusting the end of an existing resource, or
1061 * allocating a new resource.
1062 */
1063retry:
1064 first = region_res->child;
1065 if (!first)
1066 return alloc_dev_dax_range(dev_dax, dax_region->res.start, to_alloc);
1067
1068 rc = -ENOSPC;
1069 for (res = first; res; res = res->sibling) {
1070 struct resource *next = res->sibling;
1071
1072 /* space at the beginning of the region */
1073 if (res == first && res->start > dax_region->res.start) {
1074 alloc = min(res->start - dax_region->res.start, to_alloc);
1075 rc = alloc_dev_dax_range(dev_dax, dax_region->res.start, alloc);
1076 break;
1077 }
1078
1079 alloc = 0;
1080 /* space between allocations */
1081 if (next && next->start > res->end + 1)
1082 alloc = min(next->start - (res->end + 1), to_alloc);
1083
1084 /* space at the end of the region */
1085 if (!alloc && !next && res->end < region_res->end)
1086 alloc = min(region_res->end - res->end, to_alloc);
1087
1088 if (!alloc)
1089 continue;
1090
1091 if (adjust_ok(dev_dax, res)) {
1092 rc = adjust_dev_dax_range(dev_dax, res, resource_size(res) + alloc);
1093 break;
1094 }
1095 rc = alloc_dev_dax_range(dev_dax, res->end + 1, alloc);
1096 break;
1097 }
1098 if (rc)
1099 return rc;
1100 to_alloc -= alloc;
1101 if (to_alloc)
1102 goto retry;
1103 return 0;
1104}
1105
1106static ssize_t size_store(struct device *dev, struct device_attribute *attr,
1107 const char *buf, size_t len)
1108{
1109 ssize_t rc;
1110 unsigned long long val;
1111 struct dev_dax *dev_dax = to_dev_dax(dev);
1112 struct dax_region *dax_region = dev_dax->region;
1113
1114 rc = kstrtoull(buf, 0, &val);
1115 if (rc)
1116 return rc;
1117
1118 if (!alloc_is_aligned(dev_dax, val)) {
1119 dev_dbg(dev, "%s: size: %lld misaligned\n", __func__, val);
1120 return -EINVAL;
1121 }
1122
1123 rc = down_write_killable(&dax_region_rwsem);
1124 if (rc)
1125 return rc;
1126 if (!dax_region->dev->driver) {
1127 rc = -ENXIO;
1128 goto err_region;
1129 }
1130 rc = down_write_killable(&dax_dev_rwsem);
1131 if (rc)
1132 goto err_dev;
1133
1134 rc = dev_dax_resize(dax_region, dev_dax, val);
1135
1136err_dev:
1137 up_write(&dax_dev_rwsem);
1138err_region:
1139 up_write(&dax_region_rwsem);
1140
1141 if (rc == 0)
1142 return len;
1143 return rc;
1144}
1145static DEVICE_ATTR_RW(size);
1146
1147static ssize_t range_parse(const char *opt, size_t len, struct range *range)
1148{
1149 unsigned long long addr = 0;
1150 char *start, *end, *str;
1151 ssize_t rc = -EINVAL;
1152
1153 str = kstrdup(opt, GFP_KERNEL);
1154 if (!str)
1155 return rc;
1156
1157 end = str;
1158 start = strsep(&end, "-");
1159 if (!start || !end)
1160 goto err;
1161
1162 rc = kstrtoull(start, 16, &addr);
1163 if (rc)
1164 goto err;
1165 range->start = addr;
1166
1167 rc = kstrtoull(end, 16, &addr);
1168 if (rc)
1169 goto err;
1170 range->end = addr;
1171
1172err:
1173 kfree(str);
1174 return rc;
1175}
1176
1177static ssize_t mapping_store(struct device *dev, struct device_attribute *attr,
1178 const char *buf, size_t len)
1179{
1180 struct dev_dax *dev_dax = to_dev_dax(dev);
1181 struct dax_region *dax_region = dev_dax->region;
1182 size_t to_alloc;
1183 struct range r;
1184 ssize_t rc;
1185
1186 rc = range_parse(buf, len, &r);
1187 if (rc)
1188 return rc;
1189
1190 rc = down_write_killable(&dax_region_rwsem);
1191 if (rc)
1192 return rc;
1193 if (!dax_region->dev->driver) {
1194 up_write(&dax_region_rwsem);
1195 return rc;
1196 }
1197 rc = down_write_killable(&dax_dev_rwsem);
1198 if (rc) {
1199 up_write(&dax_region_rwsem);
1200 return rc;
1201 }
1202
1203 to_alloc = range_len(&r);
1204 if (alloc_is_aligned(dev_dax, to_alloc))
1205 rc = alloc_dev_dax_range(dev_dax, r.start, to_alloc);
1206 up_write(&dax_dev_rwsem);
1207 up_write(&dax_region_rwsem);
1208
1209 return rc == 0 ? len : rc;
1210}
1211static DEVICE_ATTR_WO(mapping);
1212
1213static ssize_t align_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 sysfs_emit(buf, "%d\n", dev_dax->align);
1219}
1220
1221static ssize_t dev_dax_validate_align(struct dev_dax *dev_dax)
1222{
1223 struct device *dev = &dev_dax->dev;
1224 int i;
1225
1226 for (i = 0; i < dev_dax->nr_range; i++) {
1227 size_t len = range_len(&dev_dax->ranges[i].range);
1228
1229 if (!alloc_is_aligned(dev_dax, len)) {
1230 dev_dbg(dev, "%s: align %u invalid for range %d\n",
1231 __func__, dev_dax->align, i);
1232 return -EINVAL;
1233 }
1234 }
1235
1236 return 0;
1237}
1238
1239static ssize_t align_store(struct device *dev, struct device_attribute *attr,
1240 const char *buf, size_t len)
1241{
1242 struct dev_dax *dev_dax = to_dev_dax(dev);
1243 struct dax_region *dax_region = dev_dax->region;
1244 unsigned long val, align_save;
1245 ssize_t rc;
1246
1247 rc = kstrtoul(buf, 0, &val);
1248 if (rc)
1249 return -ENXIO;
1250
1251 if (!dax_align_valid(val))
1252 return -EINVAL;
1253
1254 rc = down_write_killable(&dax_region_rwsem);
1255 if (rc)
1256 return rc;
1257 if (!dax_region->dev->driver) {
1258 up_write(&dax_region_rwsem);
1259 return -ENXIO;
1260 }
1261
1262 rc = down_write_killable(&dax_dev_rwsem);
1263 if (rc) {
1264 up_write(&dax_region_rwsem);
1265 return rc;
1266 }
1267 if (dev->driver) {
1268 rc = -EBUSY;
1269 goto out_unlock;
1270 }
1271
1272 align_save = dev_dax->align;
1273 dev_dax->align = val;
1274 rc = dev_dax_validate_align(dev_dax);
1275 if (rc)
1276 dev_dax->align = align_save;
1277out_unlock:
1278 up_write(&dax_dev_rwsem);
1279 up_write(&dax_region_rwsem);
1280 return rc == 0 ? len : rc;
1281}
1282static DEVICE_ATTR_RW(align);
1283
1284static int dev_dax_target_node(struct dev_dax *dev_dax)
1285{
1286 struct dax_region *dax_region = dev_dax->region;
1287
1288 return dax_region->target_node;
1289}
1290
1291static ssize_t target_node_show(struct device *dev,
1292 struct device_attribute *attr, char *buf)
1293{
1294 struct dev_dax *dev_dax = to_dev_dax(dev);
1295
1296 return sysfs_emit(buf, "%d\n", dev_dax_target_node(dev_dax));
1297}
1298static DEVICE_ATTR_RO(target_node);
1299
1300static ssize_t resource_show(struct device *dev,
1301 struct device_attribute *attr, char *buf)
1302{
1303 struct dev_dax *dev_dax = to_dev_dax(dev);
1304 struct dax_region *dax_region = dev_dax->region;
1305 unsigned long long start;
1306
1307 if (dev_dax->nr_range < 1)
1308 start = dax_region->res.start;
1309 else
1310 start = dev_dax->ranges[0].range.start;
1311
1312 return sysfs_emit(buf, "%#llx\n", start);
1313}
1314static DEVICE_ATTR(resource, 0400, resource_show, NULL);
1315
1316static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
1317 char *buf)
1318{
1319 /*
1320 * We only ever expect to handle device-dax instances, i.e. the
1321 * @type argument to MODULE_ALIAS_DAX_DEVICE() is always zero
1322 */
1323 return sysfs_emit(buf, DAX_DEVICE_MODALIAS_FMT "\n", 0);
1324}
1325static DEVICE_ATTR_RO(modalias);
1326
1327static ssize_t numa_node_show(struct device *dev,
1328 struct device_attribute *attr, char *buf)
1329{
1330 return sysfs_emit(buf, "%d\n", dev_to_node(dev));
1331}
1332static DEVICE_ATTR_RO(numa_node);
1333
1334static ssize_t memmap_on_memory_show(struct device *dev,
1335 struct device_attribute *attr, char *buf)
1336{
1337 struct dev_dax *dev_dax = to_dev_dax(dev);
1338
1339 return sysfs_emit(buf, "%d\n", dev_dax->memmap_on_memory);
1340}
1341
1342static ssize_t memmap_on_memory_store(struct device *dev,
1343 struct device_attribute *attr,
1344 const char *buf, size_t len)
1345{
1346 struct dev_dax *dev_dax = to_dev_dax(dev);
1347 bool val;
1348 int rc;
1349
1350 rc = kstrtobool(buf, &val);
1351 if (rc)
1352 return rc;
1353
1354 if (val == true && !mhp_supports_memmap_on_memory()) {
1355 dev_dbg(dev, "memmap_on_memory is not available\n");
1356 return -EOPNOTSUPP;
1357 }
1358
1359 rc = down_write_killable(&dax_dev_rwsem);
1360 if (rc)
1361 return rc;
1362
1363 if (dev_dax->memmap_on_memory != val && dev->driver &&
1364 to_dax_drv(dev->driver)->type == DAXDRV_KMEM_TYPE) {
1365 up_write(&dax_dev_rwsem);
1366 return -EBUSY;
1367 }
1368
1369 dev_dax->memmap_on_memory = val;
1370 up_write(&dax_dev_rwsem);
1371
1372 return len;
1373}
1374static DEVICE_ATTR_RW(memmap_on_memory);
1375
1376static umode_t dev_dax_visible(struct kobject *kobj, struct attribute *a, int n)
1377{
1378 struct device *dev = container_of(kobj, struct device, kobj);
1379 struct dev_dax *dev_dax = to_dev_dax(dev);
1380 struct dax_region *dax_region = dev_dax->region;
1381
1382 if (a == &dev_attr_target_node.attr && dev_dax_target_node(dev_dax) < 0)
1383 return 0;
1384 if (a == &dev_attr_numa_node.attr && !IS_ENABLED(CONFIG_NUMA))
1385 return 0;
1386 if (a == &dev_attr_mapping.attr && is_static(dax_region))
1387 return 0;
1388 if ((a == &dev_attr_align.attr ||
1389 a == &dev_attr_size.attr) && is_static(dax_region))
1390 return 0444;
1391 return a->mode;
1392}
1393
1394static struct attribute *dev_dax_attributes[] = {
1395 &dev_attr_modalias.attr,
1396 &dev_attr_size.attr,
1397 &dev_attr_mapping.attr,
1398 &dev_attr_target_node.attr,
1399 &dev_attr_align.attr,
1400 &dev_attr_resource.attr,
1401 &dev_attr_numa_node.attr,
1402 &dev_attr_memmap_on_memory.attr,
1403 NULL,
1404};
1405
1406static const struct attribute_group dev_dax_attribute_group = {
1407 .attrs = dev_dax_attributes,
1408 .is_visible = dev_dax_visible,
1409};
1410
1411static const struct attribute_group *dax_attribute_groups[] = {
1412 &dev_dax_attribute_group,
1413 NULL,
1414};
1415
1416static void dev_dax_release(struct device *dev)
1417{
1418 struct dev_dax *dev_dax = to_dev_dax(dev);
1419 struct dax_device *dax_dev = dev_dax->dax_dev;
1420
1421 put_dax(dax_dev);
1422 free_dev_dax_id(dev_dax);
1423 kfree(dev_dax->pgmap);
1424 kfree(dev_dax);
1425}
1426
1427static const struct device_type dev_dax_type = {
1428 .release = dev_dax_release,
1429 .groups = dax_attribute_groups,
1430};
1431
1432/* see "strong" declaration in tools/testing/nvdimm/dax-dev.c */
1433__weak phys_addr_t dax_pgoff_to_phys(struct dev_dax *dev_dax, pgoff_t pgoff,
1434 unsigned long size)
1435{
1436 for (int i = 0; i < dev_dax->nr_range; i++) {
1437 struct dev_dax_range *dax_range = &dev_dax->ranges[i];
1438 struct range *range = &dax_range->range;
1439 phys_addr_t phys;
1440
1441 if (!in_range(pgoff, dax_range->pgoff, PHYS_PFN(range_len(range))))
1442 continue;
1443 phys = PFN_PHYS(pgoff - dax_range->pgoff) + range->start;
1444 if (phys + size - 1 <= range->end)
1445 return phys;
1446 break;
1447 }
1448 return -1;
1449}
1450EXPORT_SYMBOL_GPL(dax_pgoff_to_phys);
1451
1452static struct dev_dax *__devm_create_dev_dax(struct dev_dax_data *data)
1453{
1454 struct dax_region *dax_region = data->dax_region;
1455 struct device *parent = dax_region->dev;
1456 struct dax_device *dax_dev;
1457 struct dev_dax *dev_dax;
1458 struct inode *inode;
1459 struct device *dev;
1460 int rc;
1461
1462 dev_dax = kzalloc_obj(*dev_dax);
1463 if (!dev_dax)
1464 return ERR_PTR(-ENOMEM);
1465
1466 dev_dax->region = dax_region;
1467 if (is_static(dax_region)) {
1468 if (dev_WARN_ONCE(parent, data->id < 0,
1469 "dynamic id specified to static region\n")) {
1470 rc = -EINVAL;
1471 goto err_id;
1472 }
1473
1474 dev_dax->id = data->id;
1475 } else {
1476 if (dev_WARN_ONCE(parent, data->id >= 0,
1477 "static id specified to dynamic region\n")) {
1478 rc = -EINVAL;
1479 goto err_id;
1480 }
1481
1482 rc = alloc_dev_dax_id(dev_dax);
1483 if (rc < 0)
1484 goto err_id;
1485 }
1486
1487 dev = &dev_dax->dev;
1488 device_initialize(dev);
1489 dev_set_name(dev, "dax%d.%d", dax_region->id, dev_dax->id);
1490
1491 rc = alloc_dev_dax_range(dev_dax, dax_region->res.start, data->size);
1492 if (rc)
1493 goto err_range;
1494
1495 if (data->pgmap) {
1496 dev_WARN_ONCE(parent, !is_static(dax_region),
1497 "custom dev_pagemap requires a static dax_region\n");
1498
1499 dev_dax->pgmap = kmemdup(data->pgmap,
1500 sizeof(struct dev_pagemap), GFP_KERNEL);
1501 if (!dev_dax->pgmap) {
1502 rc = -ENOMEM;
1503 goto err_pgmap;
1504 }
1505 }
1506
1507 /*
1508 * No dax_operations since there is no access to this device outside of
1509 * mmap of the resulting character device.
1510 */
1511 dax_dev = alloc_dax(dev_dax, NULL);
1512 if (IS_ERR(dax_dev)) {
1513 rc = PTR_ERR(dax_dev);
1514 goto err_alloc_dax;
1515 }
1516 set_dax_synchronous(dax_dev);
1517 set_dax_nocache(dax_dev);
1518 set_dax_nomc(dax_dev);
1519
1520 /* a device_dax instance is dead while the driver is not attached */
1521 kill_dax(dax_dev);
1522
1523 dev_dax->dax_dev = dax_dev;
1524 dev_dax->target_node = dax_region->target_node;
1525 dev_dax->align = dax_region->align;
1526 ida_init(&dev_dax->ida);
1527
1528 dev_dax->memmap_on_memory = data->memmap_on_memory;
1529
1530 inode = dax_inode(dax_dev);
1531 dev->devt = inode->i_rdev;
1532 dev->bus = &dax_bus_type;
1533 dev->parent = parent;
1534 dev->type = &dev_dax_type;
1535
1536 rc = device_add(dev);
1537 if (rc) {
1538 kill_dev_dax(dev_dax);
1539 put_device(dev);
1540 return ERR_PTR(rc);
1541 }
1542
1543 rc = devm_add_action_or_reset(dax_region->dev, unregister_dev_dax, dev);
1544 if (rc)
1545 return ERR_PTR(rc);
1546
1547 /* register mapping device for the initial allocation range */
1548 if (dev_dax->nr_range && range_len(&dev_dax->ranges[0].range)) {
1549 rc = devm_register_dax_mapping(dev_dax, 0);
1550 if (rc)
1551 return ERR_PTR(rc);
1552 }
1553
1554 return dev_dax;
1555
1556err_alloc_dax:
1557 kfree(dev_dax->pgmap);
1558err_pgmap:
1559 free_dev_dax_ranges(dev_dax);
1560err_range:
1561 free_dev_dax_id(dev_dax);
1562err_id:
1563 kfree(dev_dax);
1564
1565 return ERR_PTR(rc);
1566}
1567
1568struct dev_dax *devm_create_dev_dax(struct dev_dax_data *data)
1569{
1570 struct dev_dax *dev_dax;
1571
1572 down_write(&dax_region_rwsem);
1573 dev_dax = __devm_create_dev_dax(data);
1574 up_write(&dax_region_rwsem);
1575
1576 return dev_dax;
1577}
1578EXPORT_SYMBOL_GPL(devm_create_dev_dax);
1579
1580int __dax_driver_register(struct dax_device_driver *dax_drv,
1581 struct module *module, const char *mod_name)
1582{
1583 struct device_driver *drv = &dax_drv->drv;
1584
1585 /*
1586 * dax_bus_probe() calls dax_drv->probe() unconditionally.
1587 * So better be safe than sorry and ensure it is provided.
1588 */
1589 if (!dax_drv->probe)
1590 return -EINVAL;
1591
1592 INIT_LIST_HEAD(&dax_drv->ids);
1593 drv->owner = module;
1594 drv->name = mod_name;
1595 drv->mod_name = mod_name;
1596 drv->bus = &dax_bus_type;
1597
1598 return driver_register(drv);
1599}
1600EXPORT_SYMBOL_GPL(__dax_driver_register);
1601
1602void dax_driver_unregister(struct dax_device_driver *dax_drv)
1603{
1604 struct device_driver *drv = &dax_drv->drv;
1605 struct dax_id *dax_id, *_id;
1606
1607 mutex_lock(&dax_bus_lock);
1608 list_for_each_entry_safe(dax_id, _id, &dax_drv->ids, list) {
1609 list_del(&dax_id->list);
1610 kfree(dax_id);
1611 }
1612 mutex_unlock(&dax_bus_lock);
1613 driver_unregister(drv);
1614}
1615EXPORT_SYMBOL_GPL(dax_driver_unregister);
1616
1617int __init dax_bus_init(void)
1618{
1619 return bus_register(&dax_bus_type);
1620}
1621
1622void __exit dax_bus_exit(void)
1623{
1624 bus_unregister(&dax_bus_type);
1625}