Merge tag 'trace-v5.11-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/rostedt...
[linux/fpc-iii.git] / drivers / dax / bus.c
blob737b207c9e30d73f1378830462104f682d624da4
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"
13 static struct class *dax_class;
15 static DEFINE_MUTEX(dax_bus_lock);
17 #define DAX_NAME_LEN 30
18 struct dax_id {
19 struct list_head list;
20 char dev_name[DAX_NAME_LEN];
23 static int dax_bus_uevent(struct device *dev, struct kobj_uevent_env *env)
26 * We only ever expect to handle device-dax instances, i.e. the
27 * @type argument to MODULE_ALIAS_DAX_DEVICE() is always zero
29 return add_uevent_var(env, "MODALIAS=" DAX_DEVICE_MODALIAS_FMT, 0);
32 static struct dax_device_driver *to_dax_drv(struct device_driver *drv)
34 return container_of(drv, struct dax_device_driver, drv);
37 static struct dax_id *__dax_match_id(struct dax_device_driver *dax_drv,
38 const char *dev_name)
40 struct dax_id *dax_id;
42 lockdep_assert_held(&dax_bus_lock);
44 list_for_each_entry(dax_id, &dax_drv->ids, list)
45 if (sysfs_streq(dax_id->dev_name, dev_name))
46 return dax_id;
47 return NULL;
50 static int dax_match_id(struct dax_device_driver *dax_drv, struct device *dev)
52 int match;
54 mutex_lock(&dax_bus_lock);
55 match = !!__dax_match_id(dax_drv, dev_name(dev));
56 mutex_unlock(&dax_bus_lock);
58 return match;
61 enum id_action {
62 ID_REMOVE,
63 ID_ADD,
66 static ssize_t do_id_store(struct device_driver *drv, const char *buf,
67 size_t count, enum id_action action)
69 struct dax_device_driver *dax_drv = to_dax_drv(drv);
70 unsigned int region_id, id;
71 char devname[DAX_NAME_LEN];
72 struct dax_id *dax_id;
73 ssize_t rc = count;
74 int fields;
76 fields = sscanf(buf, "dax%d.%d", &region_id, &id);
77 if (fields != 2)
78 return -EINVAL;
79 sprintf(devname, "dax%d.%d", region_id, id);
80 if (!sysfs_streq(buf, devname))
81 return -EINVAL;
83 mutex_lock(&dax_bus_lock);
84 dax_id = __dax_match_id(dax_drv, buf);
85 if (!dax_id) {
86 if (action == ID_ADD) {
87 dax_id = kzalloc(sizeof(*dax_id), GFP_KERNEL);
88 if (dax_id) {
89 strncpy(dax_id->dev_name, buf, DAX_NAME_LEN);
90 list_add(&dax_id->list, &dax_drv->ids);
91 } else
92 rc = -ENOMEM;
93 } else
94 /* nothing to remove */;
95 } else if (action == ID_REMOVE) {
96 list_del(&dax_id->list);
97 kfree(dax_id);
98 } else
99 /* dax_id already added */;
100 mutex_unlock(&dax_bus_lock);
102 if (rc < 0)
103 return rc;
104 if (action == ID_ADD)
105 rc = driver_attach(drv);
106 if (rc)
107 return rc;
108 return count;
111 static ssize_t new_id_store(struct device_driver *drv, const char *buf,
112 size_t count)
114 return do_id_store(drv, buf, count, ID_ADD);
116 static DRIVER_ATTR_WO(new_id);
118 static ssize_t remove_id_store(struct device_driver *drv, const char *buf,
119 size_t count)
121 return do_id_store(drv, buf, count, ID_REMOVE);
123 static DRIVER_ATTR_WO(remove_id);
125 static struct attribute *dax_drv_attrs[] = {
126 &driver_attr_new_id.attr,
127 &driver_attr_remove_id.attr,
128 NULL,
130 ATTRIBUTE_GROUPS(dax_drv);
132 static int dax_bus_match(struct device *dev, struct device_driver *drv);
134 static bool is_static(struct dax_region *dax_region)
136 return (dax_region->res.flags & IORESOURCE_DAX_STATIC) != 0;
139 static u64 dev_dax_size(struct dev_dax *dev_dax)
141 u64 size = 0;
142 int i;
144 device_lock_assert(&dev_dax->dev);
146 for (i = 0; i < dev_dax->nr_range; i++)
147 size += range_len(&dev_dax->ranges[i].range);
149 return size;
152 static int dax_bus_probe(struct device *dev)
154 struct dax_device_driver *dax_drv = to_dax_drv(dev->driver);
155 struct dev_dax *dev_dax = to_dev_dax(dev);
156 struct dax_region *dax_region = dev_dax->region;
157 int rc;
159 if (dev_dax_size(dev_dax) == 0 || dev_dax->id < 0)
160 return -ENXIO;
162 rc = dax_drv->probe(dev_dax);
164 if (rc || is_static(dax_region))
165 return rc;
168 * Track new seed creation only after successful probe of the
169 * previous seed.
171 if (dax_region->seed == dev)
172 dax_region->seed = NULL;
174 return 0;
177 static int dax_bus_remove(struct device *dev)
179 struct dax_device_driver *dax_drv = to_dax_drv(dev->driver);
180 struct dev_dax *dev_dax = to_dev_dax(dev);
182 return dax_drv->remove(dev_dax);
185 static struct bus_type dax_bus_type = {
186 .name = "dax",
187 .uevent = dax_bus_uevent,
188 .match = dax_bus_match,
189 .probe = dax_bus_probe,
190 .remove = dax_bus_remove,
191 .drv_groups = dax_drv_groups,
194 static int dax_bus_match(struct device *dev, struct device_driver *drv)
196 struct dax_device_driver *dax_drv = to_dax_drv(drv);
199 * All but the 'device-dax' driver, which has 'match_always'
200 * set, requires an exact id match.
202 if (dax_drv->match_always)
203 return 1;
205 return dax_match_id(dax_drv, dev);
209 * Rely on the fact that drvdata is set before the attributes are
210 * registered, and that the attributes are unregistered before drvdata
211 * is cleared to assume that drvdata is always valid.
213 static ssize_t id_show(struct device *dev,
214 struct device_attribute *attr, char *buf)
216 struct dax_region *dax_region = dev_get_drvdata(dev);
218 return sprintf(buf, "%d\n", dax_region->id);
220 static DEVICE_ATTR_RO(id);
222 static ssize_t region_size_show(struct device *dev,
223 struct device_attribute *attr, char *buf)
225 struct dax_region *dax_region = dev_get_drvdata(dev);
227 return sprintf(buf, "%llu\n", (unsigned long long)
228 resource_size(&dax_region->res));
230 static struct device_attribute dev_attr_region_size = __ATTR(size, 0444,
231 region_size_show, NULL);
233 static ssize_t region_align_show(struct device *dev,
234 struct device_attribute *attr, char *buf)
236 struct dax_region *dax_region = dev_get_drvdata(dev);
238 return sprintf(buf, "%u\n", dax_region->align);
240 static struct device_attribute dev_attr_region_align =
241 __ATTR(align, 0400, region_align_show, NULL);
243 #define for_each_dax_region_resource(dax_region, res) \
244 for (res = (dax_region)->res.child; res; res = res->sibling)
246 static unsigned long long dax_region_avail_size(struct dax_region *dax_region)
248 resource_size_t size = resource_size(&dax_region->res);
249 struct resource *res;
251 device_lock_assert(dax_region->dev);
253 for_each_dax_region_resource(dax_region, res)
254 size -= resource_size(res);
255 return size;
258 static ssize_t available_size_show(struct device *dev,
259 struct device_attribute *attr, char *buf)
261 struct dax_region *dax_region = dev_get_drvdata(dev);
262 unsigned long long size;
264 device_lock(dev);
265 size = dax_region_avail_size(dax_region);
266 device_unlock(dev);
268 return sprintf(buf, "%llu\n", size);
270 static DEVICE_ATTR_RO(available_size);
272 static ssize_t seed_show(struct device *dev,
273 struct device_attribute *attr, char *buf)
275 struct dax_region *dax_region = dev_get_drvdata(dev);
276 struct device *seed;
277 ssize_t rc;
279 if (is_static(dax_region))
280 return -EINVAL;
282 device_lock(dev);
283 seed = dax_region->seed;
284 rc = sprintf(buf, "%s\n", seed ? dev_name(seed) : "");
285 device_unlock(dev);
287 return rc;
289 static DEVICE_ATTR_RO(seed);
291 static ssize_t create_show(struct device *dev,
292 struct device_attribute *attr, char *buf)
294 struct dax_region *dax_region = dev_get_drvdata(dev);
295 struct device *youngest;
296 ssize_t rc;
298 if (is_static(dax_region))
299 return -EINVAL;
301 device_lock(dev);
302 youngest = dax_region->youngest;
303 rc = sprintf(buf, "%s\n", youngest ? dev_name(youngest) : "");
304 device_unlock(dev);
306 return rc;
309 static ssize_t create_store(struct device *dev, struct device_attribute *attr,
310 const char *buf, size_t len)
312 struct dax_region *dax_region = dev_get_drvdata(dev);
313 unsigned long long avail;
314 ssize_t rc;
315 int val;
317 if (is_static(dax_region))
318 return -EINVAL;
320 rc = kstrtoint(buf, 0, &val);
321 if (rc)
322 return rc;
323 if (val != 1)
324 return -EINVAL;
326 device_lock(dev);
327 avail = dax_region_avail_size(dax_region);
328 if (avail == 0)
329 rc = -ENOSPC;
330 else {
331 struct dev_dax_data data = {
332 .dax_region = dax_region,
333 .size = 0,
334 .id = -1,
336 struct dev_dax *dev_dax = devm_create_dev_dax(&data);
338 if (IS_ERR(dev_dax))
339 rc = PTR_ERR(dev_dax);
340 else {
342 * In support of crafting multiple new devices
343 * simultaneously multiple seeds can be created,
344 * but only the first one that has not been
345 * successfully bound is tracked as the region
346 * seed.
348 if (!dax_region->seed)
349 dax_region->seed = &dev_dax->dev;
350 dax_region->youngest = &dev_dax->dev;
351 rc = len;
354 device_unlock(dev);
356 return rc;
358 static DEVICE_ATTR_RW(create);
360 void kill_dev_dax(struct dev_dax *dev_dax)
362 struct dax_device *dax_dev = dev_dax->dax_dev;
363 struct inode *inode = dax_inode(dax_dev);
365 kill_dax(dax_dev);
366 unmap_mapping_range(inode->i_mapping, 0, 0, 1);
368 EXPORT_SYMBOL_GPL(kill_dev_dax);
370 static void trim_dev_dax_range(struct dev_dax *dev_dax)
372 int i = dev_dax->nr_range - 1;
373 struct range *range = &dev_dax->ranges[i].range;
374 struct dax_region *dax_region = dev_dax->region;
376 device_lock_assert(dax_region->dev);
377 dev_dbg(&dev_dax->dev, "delete range[%d]: %#llx:%#llx\n", i,
378 (unsigned long long)range->start,
379 (unsigned long long)range->end);
381 __release_region(&dax_region->res, range->start, range_len(range));
382 if (--dev_dax->nr_range == 0) {
383 kfree(dev_dax->ranges);
384 dev_dax->ranges = NULL;
388 static void free_dev_dax_ranges(struct dev_dax *dev_dax)
390 while (dev_dax->nr_range)
391 trim_dev_dax_range(dev_dax);
394 static void unregister_dev_dax(void *dev)
396 struct dev_dax *dev_dax = to_dev_dax(dev);
398 dev_dbg(dev, "%s\n", __func__);
400 kill_dev_dax(dev_dax);
401 free_dev_dax_ranges(dev_dax);
402 device_del(dev);
403 put_device(dev);
406 /* a return value >= 0 indicates this invocation invalidated the id */
407 static int __free_dev_dax_id(struct dev_dax *dev_dax)
409 struct dax_region *dax_region = dev_dax->region;
410 struct device *dev = &dev_dax->dev;
411 int rc = dev_dax->id;
413 device_lock_assert(dev);
415 if (is_static(dax_region) || dev_dax->id < 0)
416 return -1;
417 ida_free(&dax_region->ida, dev_dax->id);
418 dev_dax->id = -1;
419 return rc;
422 static int free_dev_dax_id(struct dev_dax *dev_dax)
424 struct device *dev = &dev_dax->dev;
425 int rc;
427 device_lock(dev);
428 rc = __free_dev_dax_id(dev_dax);
429 device_unlock(dev);
430 return rc;
433 static ssize_t delete_store(struct device *dev, struct device_attribute *attr,
434 const char *buf, size_t len)
436 struct dax_region *dax_region = dev_get_drvdata(dev);
437 struct dev_dax *dev_dax;
438 struct device *victim;
439 bool do_del = false;
440 int rc;
442 if (is_static(dax_region))
443 return -EINVAL;
445 victim = device_find_child_by_name(dax_region->dev, buf);
446 if (!victim)
447 return -ENXIO;
449 device_lock(dev);
450 device_lock(victim);
451 dev_dax = to_dev_dax(victim);
452 if (victim->driver || dev_dax_size(dev_dax))
453 rc = -EBUSY;
454 else {
456 * Invalidate the device so it does not become active
457 * again, but always preserve device-id-0 so that
458 * /sys/bus/dax/ is guaranteed to be populated while any
459 * dax_region is registered.
461 if (dev_dax->id > 0) {
462 do_del = __free_dev_dax_id(dev_dax) >= 0;
463 rc = len;
464 if (dax_region->seed == victim)
465 dax_region->seed = NULL;
466 if (dax_region->youngest == victim)
467 dax_region->youngest = NULL;
468 } else
469 rc = -EBUSY;
471 device_unlock(victim);
473 /* won the race to invalidate the device, clean it up */
474 if (do_del)
475 devm_release_action(dev, unregister_dev_dax, victim);
476 device_unlock(dev);
477 put_device(victim);
479 return rc;
481 static DEVICE_ATTR_WO(delete);
483 static umode_t dax_region_visible(struct kobject *kobj, struct attribute *a,
484 int n)
486 struct device *dev = container_of(kobj, struct device, kobj);
487 struct dax_region *dax_region = dev_get_drvdata(dev);
489 if (is_static(dax_region))
490 if (a == &dev_attr_available_size.attr
491 || a == &dev_attr_create.attr
492 || a == &dev_attr_seed.attr
493 || a == &dev_attr_delete.attr)
494 return 0;
495 return a->mode;
498 static struct attribute *dax_region_attributes[] = {
499 &dev_attr_available_size.attr,
500 &dev_attr_region_size.attr,
501 &dev_attr_region_align.attr,
502 &dev_attr_create.attr,
503 &dev_attr_seed.attr,
504 &dev_attr_delete.attr,
505 &dev_attr_id.attr,
506 NULL,
509 static const struct attribute_group dax_region_attribute_group = {
510 .name = "dax_region",
511 .attrs = dax_region_attributes,
512 .is_visible = dax_region_visible,
515 static const struct attribute_group *dax_region_attribute_groups[] = {
516 &dax_region_attribute_group,
517 NULL,
520 static void dax_region_free(struct kref *kref)
522 struct dax_region *dax_region;
524 dax_region = container_of(kref, struct dax_region, kref);
525 kfree(dax_region);
528 void dax_region_put(struct dax_region *dax_region)
530 kref_put(&dax_region->kref, dax_region_free);
532 EXPORT_SYMBOL_GPL(dax_region_put);
534 static void dax_region_unregister(void *region)
536 struct dax_region *dax_region = region;
538 sysfs_remove_groups(&dax_region->dev->kobj,
539 dax_region_attribute_groups);
540 dax_region_put(dax_region);
543 struct dax_region *alloc_dax_region(struct device *parent, int region_id,
544 struct range *range, int target_node, unsigned int align,
545 unsigned long flags)
547 struct dax_region *dax_region;
550 * The DAX core assumes that it can store its private data in
551 * parent->driver_data. This WARN is a reminder / safeguard for
552 * developers of device-dax drivers.
554 if (dev_get_drvdata(parent)) {
555 dev_WARN(parent, "dax core failed to setup private data\n");
556 return NULL;
559 if (!IS_ALIGNED(range->start, align)
560 || !IS_ALIGNED(range_len(range), align))
561 return NULL;
563 dax_region = kzalloc(sizeof(*dax_region), GFP_KERNEL);
564 if (!dax_region)
565 return NULL;
567 dev_set_drvdata(parent, dax_region);
568 kref_init(&dax_region->kref);
569 dax_region->id = region_id;
570 dax_region->align = align;
571 dax_region->dev = parent;
572 dax_region->target_node = target_node;
573 ida_init(&dax_region->ida);
574 dax_region->res = (struct resource) {
575 .start = range->start,
576 .end = range->end,
577 .flags = IORESOURCE_MEM | flags,
580 if (sysfs_create_groups(&parent->kobj, dax_region_attribute_groups)) {
581 kfree(dax_region);
582 return NULL;
585 kref_get(&dax_region->kref);
586 if (devm_add_action_or_reset(parent, dax_region_unregister, dax_region))
587 return NULL;
588 return dax_region;
590 EXPORT_SYMBOL_GPL(alloc_dax_region);
592 static void dax_mapping_release(struct device *dev)
594 struct dax_mapping *mapping = to_dax_mapping(dev);
595 struct dev_dax *dev_dax = to_dev_dax(dev->parent);
597 ida_free(&dev_dax->ida, mapping->id);
598 kfree(mapping);
601 static void unregister_dax_mapping(void *data)
603 struct device *dev = data;
604 struct dax_mapping *mapping = to_dax_mapping(dev);
605 struct dev_dax *dev_dax = to_dev_dax(dev->parent);
606 struct dax_region *dax_region = dev_dax->region;
608 dev_dbg(dev, "%s\n", __func__);
610 device_lock_assert(dax_region->dev);
612 dev_dax->ranges[mapping->range_id].mapping = NULL;
613 mapping->range_id = -1;
615 device_del(dev);
616 put_device(dev);
619 static struct dev_dax_range *get_dax_range(struct device *dev)
621 struct dax_mapping *mapping = to_dax_mapping(dev);
622 struct dev_dax *dev_dax = to_dev_dax(dev->parent);
623 struct dax_region *dax_region = dev_dax->region;
625 device_lock(dax_region->dev);
626 if (mapping->range_id < 0) {
627 device_unlock(dax_region->dev);
628 return NULL;
631 return &dev_dax->ranges[mapping->range_id];
634 static void put_dax_range(struct dev_dax_range *dax_range)
636 struct dax_mapping *mapping = dax_range->mapping;
637 struct dev_dax *dev_dax = to_dev_dax(mapping->dev.parent);
638 struct dax_region *dax_region = dev_dax->region;
640 device_unlock(dax_region->dev);
643 static ssize_t start_show(struct device *dev,
644 struct device_attribute *attr, char *buf)
646 struct dev_dax_range *dax_range;
647 ssize_t rc;
649 dax_range = get_dax_range(dev);
650 if (!dax_range)
651 return -ENXIO;
652 rc = sprintf(buf, "%#llx\n", dax_range->range.start);
653 put_dax_range(dax_range);
655 return rc;
657 static DEVICE_ATTR(start, 0400, start_show, NULL);
659 static ssize_t end_show(struct device *dev,
660 struct device_attribute *attr, char *buf)
662 struct dev_dax_range *dax_range;
663 ssize_t rc;
665 dax_range = get_dax_range(dev);
666 if (!dax_range)
667 return -ENXIO;
668 rc = sprintf(buf, "%#llx\n", dax_range->range.end);
669 put_dax_range(dax_range);
671 return rc;
673 static DEVICE_ATTR(end, 0400, end_show, NULL);
675 static ssize_t pgoff_show(struct device *dev,
676 struct device_attribute *attr, char *buf)
678 struct dev_dax_range *dax_range;
679 ssize_t rc;
681 dax_range = get_dax_range(dev);
682 if (!dax_range)
683 return -ENXIO;
684 rc = sprintf(buf, "%#lx\n", dax_range->pgoff);
685 put_dax_range(dax_range);
687 return rc;
689 static DEVICE_ATTR(page_offset, 0400, pgoff_show, NULL);
691 static struct attribute *dax_mapping_attributes[] = {
692 &dev_attr_start.attr,
693 &dev_attr_end.attr,
694 &dev_attr_page_offset.attr,
695 NULL,
698 static const struct attribute_group dax_mapping_attribute_group = {
699 .attrs = dax_mapping_attributes,
702 static const struct attribute_group *dax_mapping_attribute_groups[] = {
703 &dax_mapping_attribute_group,
704 NULL,
707 static struct device_type dax_mapping_type = {
708 .release = dax_mapping_release,
709 .groups = dax_mapping_attribute_groups,
712 static int devm_register_dax_mapping(struct dev_dax *dev_dax, int range_id)
714 struct dax_region *dax_region = dev_dax->region;
715 struct dax_mapping *mapping;
716 struct device *dev;
717 int rc;
719 device_lock_assert(dax_region->dev);
721 if (dev_WARN_ONCE(&dev_dax->dev, !dax_region->dev->driver,
722 "region disabled\n"))
723 return -ENXIO;
725 mapping = kzalloc(sizeof(*mapping), GFP_KERNEL);
726 if (!mapping)
727 return -ENOMEM;
728 mapping->range_id = range_id;
729 mapping->id = ida_alloc(&dev_dax->ida, GFP_KERNEL);
730 if (mapping->id < 0) {
731 kfree(mapping);
732 return -ENOMEM;
734 dev_dax->ranges[range_id].mapping = mapping;
735 dev = &mapping->dev;
736 device_initialize(dev);
737 dev->parent = &dev_dax->dev;
738 dev->type = &dax_mapping_type;
739 dev_set_name(dev, "mapping%d", mapping->id);
740 rc = device_add(dev);
741 if (rc) {
742 put_device(dev);
743 return rc;
746 rc = devm_add_action_or_reset(dax_region->dev, unregister_dax_mapping,
747 dev);
748 if (rc)
749 return rc;
750 return 0;
753 static int alloc_dev_dax_range(struct dev_dax *dev_dax, u64 start,
754 resource_size_t size)
756 struct dax_region *dax_region = dev_dax->region;
757 struct resource *res = &dax_region->res;
758 struct device *dev = &dev_dax->dev;
759 struct dev_dax_range *ranges;
760 unsigned long pgoff = 0;
761 struct resource *alloc;
762 int i, rc;
764 device_lock_assert(dax_region->dev);
766 /* handle the seed alloc special case */
767 if (!size) {
768 if (dev_WARN_ONCE(dev, dev_dax->nr_range,
769 "0-size allocation must be first\n"))
770 return -EBUSY;
771 /* nr_range == 0 is elsewhere special cased as 0-size device */
772 return 0;
775 alloc = __request_region(res, start, size, dev_name(dev), 0);
776 if (!alloc)
777 return -ENOMEM;
779 ranges = krealloc(dev_dax->ranges, sizeof(*ranges)
780 * (dev_dax->nr_range + 1), GFP_KERNEL);
781 if (!ranges) {
782 __release_region(res, alloc->start, resource_size(alloc));
783 return -ENOMEM;
786 for (i = 0; i < dev_dax->nr_range; i++)
787 pgoff += PHYS_PFN(range_len(&ranges[i].range));
788 dev_dax->ranges = ranges;
789 ranges[dev_dax->nr_range++] = (struct dev_dax_range) {
790 .pgoff = pgoff,
791 .range = {
792 .start = alloc->start,
793 .end = alloc->end,
797 dev_dbg(dev, "alloc range[%d]: %pa:%pa\n", dev_dax->nr_range - 1,
798 &alloc->start, &alloc->end);
800 * A dev_dax instance must be registered before mapping device
801 * children can be added. Defer to devm_create_dev_dax() to add
802 * the initial mapping device.
804 if (!device_is_registered(&dev_dax->dev))
805 return 0;
807 rc = devm_register_dax_mapping(dev_dax, dev_dax->nr_range - 1);
808 if (rc)
809 trim_dev_dax_range(dev_dax);
811 return rc;
814 static int adjust_dev_dax_range(struct dev_dax *dev_dax, struct resource *res, resource_size_t size)
816 int last_range = dev_dax->nr_range - 1;
817 struct dev_dax_range *dax_range = &dev_dax->ranges[last_range];
818 struct dax_region *dax_region = dev_dax->region;
819 bool is_shrink = resource_size(res) > size;
820 struct range *range = &dax_range->range;
821 struct device *dev = &dev_dax->dev;
822 int rc;
824 device_lock_assert(dax_region->dev);
826 if (dev_WARN_ONCE(dev, !size, "deletion is handled by dev_dax_shrink\n"))
827 return -EINVAL;
829 rc = adjust_resource(res, range->start, size);
830 if (rc)
831 return rc;
833 *range = (struct range) {
834 .start = range->start,
835 .end = range->start + size - 1,
838 dev_dbg(dev, "%s range[%d]: %#llx:%#llx\n", is_shrink ? "shrink" : "extend",
839 last_range, (unsigned long long) range->start,
840 (unsigned long long) range->end);
842 return 0;
845 static ssize_t size_show(struct device *dev,
846 struct device_attribute *attr, char *buf)
848 struct dev_dax *dev_dax = to_dev_dax(dev);
849 unsigned long long size;
851 device_lock(dev);
852 size = dev_dax_size(dev_dax);
853 device_unlock(dev);
855 return sprintf(buf, "%llu\n", size);
858 static bool alloc_is_aligned(struct dev_dax *dev_dax, resource_size_t size)
861 * The minimum mapping granularity for a device instance is a
862 * single subsection, unless the arch says otherwise.
864 return IS_ALIGNED(size, max_t(unsigned long, dev_dax->align, memremap_compat_align()));
867 static int dev_dax_shrink(struct dev_dax *dev_dax, resource_size_t size)
869 resource_size_t to_shrink = dev_dax_size(dev_dax) - size;
870 struct dax_region *dax_region = dev_dax->region;
871 struct device *dev = &dev_dax->dev;
872 int i;
874 for (i = dev_dax->nr_range - 1; i >= 0; i--) {
875 struct range *range = &dev_dax->ranges[i].range;
876 struct dax_mapping *mapping = dev_dax->ranges[i].mapping;
877 struct resource *adjust = NULL, *res;
878 resource_size_t shrink;
880 shrink = min_t(u64, to_shrink, range_len(range));
881 if (shrink >= range_len(range)) {
882 devm_release_action(dax_region->dev,
883 unregister_dax_mapping, &mapping->dev);
884 trim_dev_dax_range(dev_dax);
885 to_shrink -= shrink;
886 if (!to_shrink)
887 break;
888 continue;
891 for_each_dax_region_resource(dax_region, res)
892 if (strcmp(res->name, dev_name(dev)) == 0
893 && res->start == range->start) {
894 adjust = res;
895 break;
898 if (dev_WARN_ONCE(dev, !adjust || i != dev_dax->nr_range - 1,
899 "failed to find matching resource\n"))
900 return -ENXIO;
901 return adjust_dev_dax_range(dev_dax, adjust, range_len(range)
902 - shrink);
904 return 0;
908 * Only allow adjustments that preserve the relative pgoff of existing
909 * allocations. I.e. the dev_dax->ranges array is ordered by increasing pgoff.
911 static bool adjust_ok(struct dev_dax *dev_dax, struct resource *res)
913 struct dev_dax_range *last;
914 int i;
916 if (dev_dax->nr_range == 0)
917 return false;
918 if (strcmp(res->name, dev_name(&dev_dax->dev)) != 0)
919 return false;
920 last = &dev_dax->ranges[dev_dax->nr_range - 1];
921 if (last->range.start != res->start || last->range.end != res->end)
922 return false;
923 for (i = 0; i < dev_dax->nr_range - 1; i++) {
924 struct dev_dax_range *dax_range = &dev_dax->ranges[i];
926 if (dax_range->pgoff > last->pgoff)
927 return false;
930 return true;
933 static ssize_t dev_dax_resize(struct dax_region *dax_region,
934 struct dev_dax *dev_dax, resource_size_t size)
936 resource_size_t avail = dax_region_avail_size(dax_region), to_alloc;
937 resource_size_t dev_size = dev_dax_size(dev_dax);
938 struct resource *region_res = &dax_region->res;
939 struct device *dev = &dev_dax->dev;
940 struct resource *res, *first;
941 resource_size_t alloc = 0;
942 int rc;
944 if (dev->driver)
945 return -EBUSY;
946 if (size == dev_size)
947 return 0;
948 if (size > dev_size && size - dev_size > avail)
949 return -ENOSPC;
950 if (size < dev_size)
951 return dev_dax_shrink(dev_dax, size);
953 to_alloc = size - dev_size;
954 if (dev_WARN_ONCE(dev, !alloc_is_aligned(dev_dax, to_alloc),
955 "resize of %pa misaligned\n", &to_alloc))
956 return -ENXIO;
959 * Expand the device into the unused portion of the region. This
960 * may involve adjusting the end of an existing resource, or
961 * allocating a new resource.
963 retry:
964 first = region_res->child;
965 if (!first)
966 return alloc_dev_dax_range(dev_dax, dax_region->res.start, to_alloc);
968 rc = -ENOSPC;
969 for (res = first; res; res = res->sibling) {
970 struct resource *next = res->sibling;
972 /* space at the beginning of the region */
973 if (res == first && res->start > dax_region->res.start) {
974 alloc = min(res->start - dax_region->res.start, to_alloc);
975 rc = alloc_dev_dax_range(dev_dax, dax_region->res.start, alloc);
976 break;
979 alloc = 0;
980 /* space between allocations */
981 if (next && next->start > res->end + 1)
982 alloc = min(next->start - (res->end + 1), to_alloc);
984 /* space at the end of the region */
985 if (!alloc && !next && res->end < region_res->end)
986 alloc = min(region_res->end - res->end, to_alloc);
988 if (!alloc)
989 continue;
991 if (adjust_ok(dev_dax, res)) {
992 rc = adjust_dev_dax_range(dev_dax, res, resource_size(res) + alloc);
993 break;
995 rc = alloc_dev_dax_range(dev_dax, res->end + 1, alloc);
996 break;
998 if (rc)
999 return rc;
1000 to_alloc -= alloc;
1001 if (to_alloc)
1002 goto retry;
1003 return 0;
1006 static ssize_t size_store(struct device *dev, struct device_attribute *attr,
1007 const char *buf, size_t len)
1009 ssize_t rc;
1010 unsigned long long val;
1011 struct dev_dax *dev_dax = to_dev_dax(dev);
1012 struct dax_region *dax_region = dev_dax->region;
1014 rc = kstrtoull(buf, 0, &val);
1015 if (rc)
1016 return rc;
1018 if (!alloc_is_aligned(dev_dax, val)) {
1019 dev_dbg(dev, "%s: size: %lld misaligned\n", __func__, val);
1020 return -EINVAL;
1023 device_lock(dax_region->dev);
1024 if (!dax_region->dev->driver) {
1025 device_unlock(dax_region->dev);
1026 return -ENXIO;
1028 device_lock(dev);
1029 rc = dev_dax_resize(dax_region, dev_dax, val);
1030 device_unlock(dev);
1031 device_unlock(dax_region->dev);
1033 return rc == 0 ? len : rc;
1035 static DEVICE_ATTR_RW(size);
1037 static ssize_t range_parse(const char *opt, size_t len, struct range *range)
1039 unsigned long long addr = 0;
1040 char *start, *end, *str;
1041 ssize_t rc = EINVAL;
1043 str = kstrdup(opt, GFP_KERNEL);
1044 if (!str)
1045 return rc;
1047 end = str;
1048 start = strsep(&end, "-");
1049 if (!start || !end)
1050 goto err;
1052 rc = kstrtoull(start, 16, &addr);
1053 if (rc)
1054 goto err;
1055 range->start = addr;
1057 rc = kstrtoull(end, 16, &addr);
1058 if (rc)
1059 goto err;
1060 range->end = addr;
1062 err:
1063 kfree(str);
1064 return rc;
1067 static ssize_t mapping_store(struct device *dev, struct device_attribute *attr,
1068 const char *buf, size_t len)
1070 struct dev_dax *dev_dax = to_dev_dax(dev);
1071 struct dax_region *dax_region = dev_dax->region;
1072 size_t to_alloc;
1073 struct range r;
1074 ssize_t rc;
1076 rc = range_parse(buf, len, &r);
1077 if (rc)
1078 return rc;
1080 rc = -ENXIO;
1081 device_lock(dax_region->dev);
1082 if (!dax_region->dev->driver) {
1083 device_unlock(dax_region->dev);
1084 return rc;
1086 device_lock(dev);
1088 to_alloc = range_len(&r);
1089 if (alloc_is_aligned(dev_dax, to_alloc))
1090 rc = alloc_dev_dax_range(dev_dax, r.start, to_alloc);
1091 device_unlock(dev);
1092 device_unlock(dax_region->dev);
1094 return rc == 0 ? len : rc;
1096 static DEVICE_ATTR_WO(mapping);
1098 static ssize_t align_show(struct device *dev,
1099 struct device_attribute *attr, char *buf)
1101 struct dev_dax *dev_dax = to_dev_dax(dev);
1103 return sprintf(buf, "%d\n", dev_dax->align);
1106 static ssize_t dev_dax_validate_align(struct dev_dax *dev_dax)
1108 struct device *dev = &dev_dax->dev;
1109 int i;
1111 for (i = 0; i < dev_dax->nr_range; i++) {
1112 size_t len = range_len(&dev_dax->ranges[i].range);
1114 if (!alloc_is_aligned(dev_dax, len)) {
1115 dev_dbg(dev, "%s: align %u invalid for range %d\n",
1116 __func__, dev_dax->align, i);
1117 return -EINVAL;
1121 return 0;
1124 static ssize_t align_store(struct device *dev, struct device_attribute *attr,
1125 const char *buf, size_t len)
1127 struct dev_dax *dev_dax = to_dev_dax(dev);
1128 struct dax_region *dax_region = dev_dax->region;
1129 unsigned long val, align_save;
1130 ssize_t rc;
1132 rc = kstrtoul(buf, 0, &val);
1133 if (rc)
1134 return -ENXIO;
1136 if (!dax_align_valid(val))
1137 return -EINVAL;
1139 device_lock(dax_region->dev);
1140 if (!dax_region->dev->driver) {
1141 device_unlock(dax_region->dev);
1142 return -ENXIO;
1145 device_lock(dev);
1146 if (dev->driver) {
1147 rc = -EBUSY;
1148 goto out_unlock;
1151 align_save = dev_dax->align;
1152 dev_dax->align = val;
1153 rc = dev_dax_validate_align(dev_dax);
1154 if (rc)
1155 dev_dax->align = align_save;
1156 out_unlock:
1157 device_unlock(dev);
1158 device_unlock(dax_region->dev);
1159 return rc == 0 ? len : rc;
1161 static DEVICE_ATTR_RW(align);
1163 static int dev_dax_target_node(struct dev_dax *dev_dax)
1165 struct dax_region *dax_region = dev_dax->region;
1167 return dax_region->target_node;
1170 static ssize_t target_node_show(struct device *dev,
1171 struct device_attribute *attr, char *buf)
1173 struct dev_dax *dev_dax = to_dev_dax(dev);
1175 return sprintf(buf, "%d\n", dev_dax_target_node(dev_dax));
1177 static DEVICE_ATTR_RO(target_node);
1179 static ssize_t resource_show(struct device *dev,
1180 struct device_attribute *attr, char *buf)
1182 struct dev_dax *dev_dax = to_dev_dax(dev);
1183 struct dax_region *dax_region = dev_dax->region;
1184 unsigned long long start;
1186 if (dev_dax->nr_range < 1)
1187 start = dax_region->res.start;
1188 else
1189 start = dev_dax->ranges[0].range.start;
1191 return sprintf(buf, "%#llx\n", start);
1193 static DEVICE_ATTR(resource, 0400, resource_show, NULL);
1195 static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
1196 char *buf)
1199 * We only ever expect to handle device-dax instances, i.e. the
1200 * @type argument to MODULE_ALIAS_DAX_DEVICE() is always zero
1202 return sprintf(buf, DAX_DEVICE_MODALIAS_FMT "\n", 0);
1204 static DEVICE_ATTR_RO(modalias);
1206 static ssize_t numa_node_show(struct device *dev,
1207 struct device_attribute *attr, char *buf)
1209 return sprintf(buf, "%d\n", dev_to_node(dev));
1211 static DEVICE_ATTR_RO(numa_node);
1213 static umode_t dev_dax_visible(struct kobject *kobj, struct attribute *a, int n)
1215 struct device *dev = container_of(kobj, struct device, kobj);
1216 struct dev_dax *dev_dax = to_dev_dax(dev);
1217 struct dax_region *dax_region = dev_dax->region;
1219 if (a == &dev_attr_target_node.attr && dev_dax_target_node(dev_dax) < 0)
1220 return 0;
1221 if (a == &dev_attr_numa_node.attr && !IS_ENABLED(CONFIG_NUMA))
1222 return 0;
1223 if (a == &dev_attr_mapping.attr && is_static(dax_region))
1224 return 0;
1225 if ((a == &dev_attr_align.attr ||
1226 a == &dev_attr_size.attr) && is_static(dax_region))
1227 return 0444;
1228 return a->mode;
1231 static struct attribute *dev_dax_attributes[] = {
1232 &dev_attr_modalias.attr,
1233 &dev_attr_size.attr,
1234 &dev_attr_mapping.attr,
1235 &dev_attr_target_node.attr,
1236 &dev_attr_align.attr,
1237 &dev_attr_resource.attr,
1238 &dev_attr_numa_node.attr,
1239 NULL,
1242 static const struct attribute_group dev_dax_attribute_group = {
1243 .attrs = dev_dax_attributes,
1244 .is_visible = dev_dax_visible,
1247 static const struct attribute_group *dax_attribute_groups[] = {
1248 &dev_dax_attribute_group,
1249 NULL,
1252 static void dev_dax_release(struct device *dev)
1254 struct dev_dax *dev_dax = to_dev_dax(dev);
1255 struct dax_region *dax_region = dev_dax->region;
1256 struct dax_device *dax_dev = dev_dax->dax_dev;
1258 put_dax(dax_dev);
1259 free_dev_dax_id(dev_dax);
1260 dax_region_put(dax_region);
1261 kfree(dev_dax->pgmap);
1262 kfree(dev_dax);
1265 static const struct device_type dev_dax_type = {
1266 .release = dev_dax_release,
1267 .groups = dax_attribute_groups,
1270 struct dev_dax *devm_create_dev_dax(struct dev_dax_data *data)
1272 struct dax_region *dax_region = data->dax_region;
1273 struct device *parent = dax_region->dev;
1274 struct dax_device *dax_dev;
1275 struct dev_dax *dev_dax;
1276 struct inode *inode;
1277 struct device *dev;
1278 int rc;
1280 dev_dax = kzalloc(sizeof(*dev_dax), GFP_KERNEL);
1281 if (!dev_dax)
1282 return ERR_PTR(-ENOMEM);
1284 if (is_static(dax_region)) {
1285 if (dev_WARN_ONCE(parent, data->id < 0,
1286 "dynamic id specified to static region\n")) {
1287 rc = -EINVAL;
1288 goto err_id;
1291 dev_dax->id = data->id;
1292 } else {
1293 if (dev_WARN_ONCE(parent, data->id >= 0,
1294 "static id specified to dynamic region\n")) {
1295 rc = -EINVAL;
1296 goto err_id;
1299 rc = ida_alloc(&dax_region->ida, GFP_KERNEL);
1300 if (rc < 0)
1301 goto err_id;
1302 dev_dax->id = rc;
1305 dev_dax->region = dax_region;
1306 dev = &dev_dax->dev;
1307 device_initialize(dev);
1308 dev_set_name(dev, "dax%d.%d", dax_region->id, dev_dax->id);
1310 rc = alloc_dev_dax_range(dev_dax, dax_region->res.start, data->size);
1311 if (rc)
1312 goto err_range;
1314 if (data->pgmap) {
1315 dev_WARN_ONCE(parent, !is_static(dax_region),
1316 "custom dev_pagemap requires a static dax_region\n");
1318 dev_dax->pgmap = kmemdup(data->pgmap,
1319 sizeof(struct dev_pagemap), GFP_KERNEL);
1320 if (!dev_dax->pgmap) {
1321 rc = -ENOMEM;
1322 goto err_pgmap;
1327 * No 'host' or dax_operations since there is no access to this
1328 * device outside of mmap of the resulting character device.
1330 dax_dev = alloc_dax(dev_dax, NULL, NULL, DAXDEV_F_SYNC);
1331 if (IS_ERR(dax_dev)) {
1332 rc = PTR_ERR(dax_dev);
1333 goto err_alloc_dax;
1336 /* a device_dax instance is dead while the driver is not attached */
1337 kill_dax(dax_dev);
1339 dev_dax->dax_dev = dax_dev;
1340 dev_dax->target_node = dax_region->target_node;
1341 dev_dax->align = dax_region->align;
1342 ida_init(&dev_dax->ida);
1343 kref_get(&dax_region->kref);
1345 inode = dax_inode(dax_dev);
1346 dev->devt = inode->i_rdev;
1347 if (data->subsys == DEV_DAX_BUS)
1348 dev->bus = &dax_bus_type;
1349 else
1350 dev->class = dax_class;
1351 dev->parent = parent;
1352 dev->type = &dev_dax_type;
1354 rc = device_add(dev);
1355 if (rc) {
1356 kill_dev_dax(dev_dax);
1357 put_device(dev);
1358 return ERR_PTR(rc);
1361 rc = devm_add_action_or_reset(dax_region->dev, unregister_dev_dax, dev);
1362 if (rc)
1363 return ERR_PTR(rc);
1365 /* register mapping device for the initial allocation range */
1366 if (dev_dax->nr_range && range_len(&dev_dax->ranges[0].range)) {
1367 rc = devm_register_dax_mapping(dev_dax, 0);
1368 if (rc)
1369 return ERR_PTR(rc);
1372 return dev_dax;
1374 err_alloc_dax:
1375 kfree(dev_dax->pgmap);
1376 err_pgmap:
1377 free_dev_dax_ranges(dev_dax);
1378 err_range:
1379 free_dev_dax_id(dev_dax);
1380 err_id:
1381 kfree(dev_dax);
1383 return ERR_PTR(rc);
1385 EXPORT_SYMBOL_GPL(devm_create_dev_dax);
1387 static int match_always_count;
1389 int __dax_driver_register(struct dax_device_driver *dax_drv,
1390 struct module *module, const char *mod_name)
1392 struct device_driver *drv = &dax_drv->drv;
1393 int rc = 0;
1395 INIT_LIST_HEAD(&dax_drv->ids);
1396 drv->owner = module;
1397 drv->name = mod_name;
1398 drv->mod_name = mod_name;
1399 drv->bus = &dax_bus_type;
1401 /* there can only be one default driver */
1402 mutex_lock(&dax_bus_lock);
1403 match_always_count += dax_drv->match_always;
1404 if (match_always_count > 1) {
1405 match_always_count--;
1406 WARN_ON(1);
1407 rc = -EINVAL;
1409 mutex_unlock(&dax_bus_lock);
1410 if (rc)
1411 return rc;
1412 return driver_register(drv);
1414 EXPORT_SYMBOL_GPL(__dax_driver_register);
1416 void dax_driver_unregister(struct dax_device_driver *dax_drv)
1418 struct device_driver *drv = &dax_drv->drv;
1419 struct dax_id *dax_id, *_id;
1421 mutex_lock(&dax_bus_lock);
1422 match_always_count -= dax_drv->match_always;
1423 list_for_each_entry_safe(dax_id, _id, &dax_drv->ids, list) {
1424 list_del(&dax_id->list);
1425 kfree(dax_id);
1427 mutex_unlock(&dax_bus_lock);
1428 driver_unregister(drv);
1430 EXPORT_SYMBOL_GPL(dax_driver_unregister);
1432 int __init dax_bus_init(void)
1434 int rc;
1436 if (IS_ENABLED(CONFIG_DEV_DAX_PMEM_COMPAT)) {
1437 dax_class = class_create(THIS_MODULE, "dax");
1438 if (IS_ERR(dax_class))
1439 return PTR_ERR(dax_class);
1442 rc = bus_register(&dax_bus_type);
1443 if (rc)
1444 class_destroy(dax_class);
1445 return rc;
1448 void __exit dax_bus_exit(void)
1450 bus_unregister(&dax_bus_type);
1451 class_destroy(dax_class);