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