1 | // SPDX-License-Identifier: GPL-2.0 |
2 | /* Copyright(c) 2017-2018 Intel Corporation. All rights reserved. */ |
3 | #include <linux/memremap.h> |
4 | #include <linux/device.h> |
5 | #include <linux/mutex.h> |
6 | #include <linux/list.h> |
7 | #include <linux/slab.h> |
8 | #include <linux/dax.h> |
9 | #include <linux/io.h> |
10 | #include "dax-private.h" |
11 | #include "bus.h" |
12 | |
13 | static DEFINE_MUTEX(dax_bus_lock); |
14 | |
15 | #define DAX_NAME_LEN 30 |
16 | struct dax_id { |
17 | struct list_head list; |
18 | char dev_name[DAX_NAME_LEN]; |
19 | }; |
20 | |
21 | static int dax_bus_uevent(const struct device *dev, struct kobj_uevent_env *env) |
22 | { |
23 | /* |
24 | * We only ever expect to handle device-dax instances, i.e. the |
25 | * @type argument to MODULE_ALIAS_DAX_DEVICE() is always zero |
26 | */ |
27 | return add_uevent_var(env, format: "MODALIAS=" DAX_DEVICE_MODALIAS_FMT, 0); |
28 | } |
29 | |
30 | static struct dax_device_driver *to_dax_drv(struct device_driver *drv) |
31 | { |
32 | return container_of(drv, struct dax_device_driver, drv); |
33 | } |
34 | |
35 | static struct dax_id *__dax_match_id(struct dax_device_driver *dax_drv, |
36 | const char *dev_name) |
37 | { |
38 | struct dax_id *dax_id; |
39 | |
40 | lockdep_assert_held(&dax_bus_lock); |
41 | |
42 | list_for_each_entry(dax_id, &dax_drv->ids, list) |
43 | if (sysfs_streq(s1: dax_id->dev_name, s2: dev_name)) |
44 | return dax_id; |
45 | return NULL; |
46 | } |
47 | |
48 | static int dax_match_id(struct dax_device_driver *dax_drv, struct device *dev) |
49 | { |
50 | int match; |
51 | |
52 | mutex_lock(&dax_bus_lock); |
53 | match = !!__dax_match_id(dax_drv, dev_name: dev_name(dev)); |
54 | mutex_unlock(lock: &dax_bus_lock); |
55 | |
56 | return match; |
57 | } |
58 | |
59 | static int dax_match_type(struct dax_device_driver *dax_drv, struct device *dev) |
60 | { |
61 | enum dax_driver_type type = DAXDRV_DEVICE_TYPE; |
62 | struct dev_dax *dev_dax = to_dev_dax(dev); |
63 | |
64 | if (dev_dax->region->res.flags & IORESOURCE_DAX_KMEM) |
65 | type = DAXDRV_KMEM_TYPE; |
66 | |
67 | if (dax_drv->type == type) |
68 | return 1; |
69 | |
70 | /* default to device mode if dax_kmem is disabled */ |
71 | if (dax_drv->type == DAXDRV_DEVICE_TYPE && |
72 | !IS_ENABLED(CONFIG_DEV_DAX_KMEM)) |
73 | return 1; |
74 | |
75 | return 0; |
76 | } |
77 | |
78 | enum id_action { |
79 | ID_REMOVE, |
80 | ID_ADD, |
81 | }; |
82 | |
83 | static ssize_t do_id_store(struct device_driver *drv, const char *buf, |
84 | size_t count, enum id_action action) |
85 | { |
86 | struct dax_device_driver *dax_drv = to_dax_drv(drv); |
87 | unsigned int region_id, id; |
88 | char devname[DAX_NAME_LEN]; |
89 | struct dax_id *dax_id; |
90 | ssize_t rc = count; |
91 | int fields; |
92 | |
93 | fields = sscanf(buf, "dax%d.%d" , ®ion_id, &id); |
94 | if (fields != 2) |
95 | return -EINVAL; |
96 | sprintf(buf: devname, fmt: "dax%d.%d" , region_id, id); |
97 | if (!sysfs_streq(s1: buf, s2: devname)) |
98 | return -EINVAL; |
99 | |
100 | mutex_lock(&dax_bus_lock); |
101 | dax_id = __dax_match_id(dax_drv, dev_name: buf); |
102 | if (!dax_id) { |
103 | if (action == ID_ADD) { |
104 | dax_id = kzalloc(size: sizeof(*dax_id), GFP_KERNEL); |
105 | if (dax_id) { |
106 | strscpy(p: dax_id->dev_name, q: buf, DAX_NAME_LEN); |
107 | list_add(new: &dax_id->list, head: &dax_drv->ids); |
108 | } else |
109 | rc = -ENOMEM; |
110 | } |
111 | } else if (action == ID_REMOVE) { |
112 | list_del(entry: &dax_id->list); |
113 | kfree(objp: dax_id); |
114 | } |
115 | mutex_unlock(lock: &dax_bus_lock); |
116 | |
117 | if (rc < 0) |
118 | return rc; |
119 | if (action == ID_ADD) |
120 | rc = driver_attach(drv); |
121 | if (rc) |
122 | return rc; |
123 | return count; |
124 | } |
125 | |
126 | static ssize_t new_id_store(struct device_driver *drv, const char *buf, |
127 | size_t count) |
128 | { |
129 | return do_id_store(drv, buf, count, action: ID_ADD); |
130 | } |
131 | static DRIVER_ATTR_WO(new_id); |
132 | |
133 | static ssize_t remove_id_store(struct device_driver *drv, const char *buf, |
134 | size_t count) |
135 | { |
136 | return do_id_store(drv, buf, count, action: ID_REMOVE); |
137 | } |
138 | static DRIVER_ATTR_WO(remove_id); |
139 | |
140 | static struct attribute *dax_drv_attrs[] = { |
141 | &driver_attr_new_id.attr, |
142 | &driver_attr_remove_id.attr, |
143 | NULL, |
144 | }; |
145 | ATTRIBUTE_GROUPS(dax_drv); |
146 | |
147 | static int dax_bus_match(struct device *dev, struct device_driver *drv); |
148 | |
149 | /* |
150 | * Static dax regions are regions created by an external subsystem |
151 | * nvdimm where a single range is assigned. Its boundaries are by the external |
152 | * subsystem and are usually limited to one physical memory range. For example, |
153 | * for PMEM it is usually defined by NVDIMM Namespace boundaries (i.e. a |
154 | * single contiguous range) |
155 | * |
156 | * On dynamic dax regions, the assigned region can be partitioned by dax core |
157 | * into multiple subdivisions. A subdivision is represented into one |
158 | * /dev/daxN.M device composed by one or more potentially discontiguous ranges. |
159 | * |
160 | * When allocating a dax region, drivers must set whether it's static |
161 | * (IORESOURCE_DAX_STATIC). On static dax devices, the @pgmap is pre-assigned |
162 | * to dax core when calling devm_create_dev_dax(), whereas in dynamic dax |
163 | * devices it is NULL but afterwards allocated by dax core on device ->probe(). |
164 | * Care is needed to make sure that dynamic dax devices are torn down with a |
165 | * cleared @pgmap field (see kill_dev_dax()). |
166 | */ |
167 | static bool is_static(struct dax_region *dax_region) |
168 | { |
169 | return (dax_region->res.flags & IORESOURCE_DAX_STATIC) != 0; |
170 | } |
171 | |
172 | bool static_dev_dax(struct dev_dax *dev_dax) |
173 | { |
174 | return is_static(dax_region: dev_dax->region); |
175 | } |
176 | EXPORT_SYMBOL_GPL(static_dev_dax); |
177 | |
178 | static u64 dev_dax_size(struct dev_dax *dev_dax) |
179 | { |
180 | u64 size = 0; |
181 | int i; |
182 | |
183 | device_lock_assert(dev: &dev_dax->dev); |
184 | |
185 | for (i = 0; i < dev_dax->nr_range; i++) |
186 | size += range_len(range: &dev_dax->ranges[i].range); |
187 | |
188 | return size; |
189 | } |
190 | |
191 | static int dax_bus_probe(struct device *dev) |
192 | { |
193 | struct dax_device_driver *dax_drv = to_dax_drv(drv: dev->driver); |
194 | struct dev_dax *dev_dax = to_dev_dax(dev); |
195 | struct dax_region *dax_region = dev_dax->region; |
196 | int rc; |
197 | |
198 | if (dev_dax_size(dev_dax) == 0 || dev_dax->id < 0) |
199 | return -ENXIO; |
200 | |
201 | rc = dax_drv->probe(dev_dax); |
202 | |
203 | if (rc || is_static(dax_region)) |
204 | return rc; |
205 | |
206 | /* |
207 | * Track new seed creation only after successful probe of the |
208 | * previous seed. |
209 | */ |
210 | if (dax_region->seed == dev) |
211 | dax_region->seed = NULL; |
212 | |
213 | return 0; |
214 | } |
215 | |
216 | static void dax_bus_remove(struct device *dev) |
217 | { |
218 | struct dax_device_driver *dax_drv = to_dax_drv(drv: dev->driver); |
219 | struct dev_dax *dev_dax = to_dev_dax(dev); |
220 | |
221 | if (dax_drv->remove) |
222 | dax_drv->remove(dev_dax); |
223 | } |
224 | |
225 | static struct bus_type dax_bus_type = { |
226 | .name = "dax" , |
227 | .uevent = dax_bus_uevent, |
228 | .match = dax_bus_match, |
229 | .probe = dax_bus_probe, |
230 | .remove = dax_bus_remove, |
231 | .drv_groups = dax_drv_groups, |
232 | }; |
233 | |
234 | static int dax_bus_match(struct device *dev, struct device_driver *drv) |
235 | { |
236 | struct dax_device_driver *dax_drv = to_dax_drv(drv); |
237 | |
238 | if (dax_match_id(dax_drv, dev)) |
239 | return 1; |
240 | return dax_match_type(dax_drv, dev); |
241 | } |
242 | |
243 | /* |
244 | * Rely on the fact that drvdata is set before the attributes are |
245 | * registered, and that the attributes are unregistered before drvdata |
246 | * is cleared to assume that drvdata is always valid. |
247 | */ |
248 | static ssize_t id_show(struct device *dev, |
249 | struct device_attribute *attr, char *buf) |
250 | { |
251 | struct dax_region *dax_region = dev_get_drvdata(dev); |
252 | |
253 | return sprintf(buf, fmt: "%d\n" , dax_region->id); |
254 | } |
255 | static DEVICE_ATTR_RO(id); |
256 | |
257 | static ssize_t region_size_show(struct device *dev, |
258 | struct device_attribute *attr, char *buf) |
259 | { |
260 | struct dax_region *dax_region = dev_get_drvdata(dev); |
261 | |
262 | return sprintf(buf, fmt: "%llu\n" , (unsigned long long) |
263 | resource_size(res: &dax_region->res)); |
264 | } |
265 | static struct device_attribute dev_attr_region_size = __ATTR(size, 0444, |
266 | region_size_show, NULL); |
267 | |
268 | static ssize_t region_align_show(struct device *dev, |
269 | struct device_attribute *attr, char *buf) |
270 | { |
271 | struct dax_region *dax_region = dev_get_drvdata(dev); |
272 | |
273 | return sprintf(buf, fmt: "%u\n" , dax_region->align); |
274 | } |
275 | static struct device_attribute dev_attr_region_align = |
276 | __ATTR(align, 0400, region_align_show, NULL); |
277 | |
278 | #define for_each_dax_region_resource(dax_region, res) \ |
279 | for (res = (dax_region)->res.child; res; res = res->sibling) |
280 | |
281 | static unsigned long long dax_region_avail_size(struct dax_region *dax_region) |
282 | { |
283 | resource_size_t size = resource_size(res: &dax_region->res); |
284 | struct resource *res; |
285 | |
286 | device_lock_assert(dev: dax_region->dev); |
287 | |
288 | for_each_dax_region_resource(dax_region, res) |
289 | size -= resource_size(res); |
290 | return size; |
291 | } |
292 | |
293 | static ssize_t available_size_show(struct device *dev, |
294 | struct device_attribute *attr, char *buf) |
295 | { |
296 | struct dax_region *dax_region = dev_get_drvdata(dev); |
297 | unsigned long long size; |
298 | |
299 | device_lock(dev); |
300 | size = dax_region_avail_size(dax_region); |
301 | device_unlock(dev); |
302 | |
303 | return sprintf(buf, fmt: "%llu\n" , size); |
304 | } |
305 | static DEVICE_ATTR_RO(available_size); |
306 | |
307 | static ssize_t seed_show(struct device *dev, |
308 | struct device_attribute *attr, char *buf) |
309 | { |
310 | struct dax_region *dax_region = dev_get_drvdata(dev); |
311 | struct device *seed; |
312 | ssize_t rc; |
313 | |
314 | if (is_static(dax_region)) |
315 | return -EINVAL; |
316 | |
317 | device_lock(dev); |
318 | seed = dax_region->seed; |
319 | rc = sprintf(buf, fmt: "%s\n" , seed ? dev_name(dev: seed) : "" ); |
320 | device_unlock(dev); |
321 | |
322 | return rc; |
323 | } |
324 | static DEVICE_ATTR_RO(seed); |
325 | |
326 | static ssize_t create_show(struct device *dev, |
327 | struct device_attribute *attr, char *buf) |
328 | { |
329 | struct dax_region *dax_region = dev_get_drvdata(dev); |
330 | struct device *youngest; |
331 | ssize_t rc; |
332 | |
333 | if (is_static(dax_region)) |
334 | return -EINVAL; |
335 | |
336 | device_lock(dev); |
337 | youngest = dax_region->youngest; |
338 | rc = sprintf(buf, fmt: "%s\n" , youngest ? dev_name(dev: youngest) : "" ); |
339 | device_unlock(dev); |
340 | |
341 | return rc; |
342 | } |
343 | |
344 | static ssize_t create_store(struct device *dev, struct device_attribute *attr, |
345 | const char *buf, size_t len) |
346 | { |
347 | struct dax_region *dax_region = dev_get_drvdata(dev); |
348 | unsigned long long avail; |
349 | ssize_t rc; |
350 | int val; |
351 | |
352 | if (is_static(dax_region)) |
353 | return -EINVAL; |
354 | |
355 | rc = kstrtoint(s: buf, base: 0, res: &val); |
356 | if (rc) |
357 | return rc; |
358 | if (val != 1) |
359 | return -EINVAL; |
360 | |
361 | device_lock(dev); |
362 | avail = dax_region_avail_size(dax_region); |
363 | if (avail == 0) |
364 | rc = -ENOSPC; |
365 | else { |
366 | struct dev_dax_data data = { |
367 | .dax_region = dax_region, |
368 | .size = 0, |
369 | .id = -1, |
370 | }; |
371 | struct dev_dax *dev_dax = devm_create_dev_dax(data: &data); |
372 | |
373 | if (IS_ERR(ptr: dev_dax)) |
374 | rc = PTR_ERR(ptr: dev_dax); |
375 | else { |
376 | /* |
377 | * In support of crafting multiple new devices |
378 | * simultaneously multiple seeds can be created, |
379 | * but only the first one that has not been |
380 | * successfully bound is tracked as the region |
381 | * seed. |
382 | */ |
383 | if (!dax_region->seed) |
384 | dax_region->seed = &dev_dax->dev; |
385 | dax_region->youngest = &dev_dax->dev; |
386 | rc = len; |
387 | } |
388 | } |
389 | device_unlock(dev); |
390 | |
391 | return rc; |
392 | } |
393 | static DEVICE_ATTR_RW(create); |
394 | |
395 | void kill_dev_dax(struct dev_dax *dev_dax) |
396 | { |
397 | struct dax_device *dax_dev = dev_dax->dax_dev; |
398 | struct inode *inode = dax_inode(dax_dev); |
399 | |
400 | kill_dax(dax_dev); |
401 | unmap_mapping_range(mapping: inode->i_mapping, holebegin: 0, holelen: 0, even_cows: 1); |
402 | |
403 | /* |
404 | * Dynamic dax region have the pgmap allocated via dev_kzalloc() |
405 | * and thus freed by devm. Clear the pgmap to not have stale pgmap |
406 | * ranges on probe() from previous reconfigurations of region devices. |
407 | */ |
408 | if (!static_dev_dax(dev_dax)) |
409 | dev_dax->pgmap = NULL; |
410 | } |
411 | EXPORT_SYMBOL_GPL(kill_dev_dax); |
412 | |
413 | static void trim_dev_dax_range(struct dev_dax *dev_dax) |
414 | { |
415 | int i = dev_dax->nr_range - 1; |
416 | struct range *range = &dev_dax->ranges[i].range; |
417 | struct dax_region *dax_region = dev_dax->region; |
418 | |
419 | device_lock_assert(dev: dax_region->dev); |
420 | dev_dbg(&dev_dax->dev, "delete range[%d]: %#llx:%#llx\n" , i, |
421 | (unsigned long long)range->start, |
422 | (unsigned long long)range->end); |
423 | |
424 | __release_region(&dax_region->res, range->start, range_len(range)); |
425 | if (--dev_dax->nr_range == 0) { |
426 | kfree(objp: dev_dax->ranges); |
427 | dev_dax->ranges = NULL; |
428 | } |
429 | } |
430 | |
431 | static void free_dev_dax_ranges(struct dev_dax *dev_dax) |
432 | { |
433 | while (dev_dax->nr_range) |
434 | trim_dev_dax_range(dev_dax); |
435 | } |
436 | |
437 | static void unregister_dev_dax(void *dev) |
438 | { |
439 | struct dev_dax *dev_dax = to_dev_dax(dev); |
440 | |
441 | dev_dbg(dev, "%s\n" , __func__); |
442 | |
443 | kill_dev_dax(dev_dax); |
444 | device_del(dev); |
445 | free_dev_dax_ranges(dev_dax); |
446 | put_device(dev); |
447 | } |
448 | |
449 | static void dax_region_free(struct kref *kref) |
450 | { |
451 | struct dax_region *dax_region; |
452 | |
453 | dax_region = container_of(kref, struct dax_region, kref); |
454 | kfree(objp: dax_region); |
455 | } |
456 | |
457 | static void dax_region_put(struct dax_region *dax_region) |
458 | { |
459 | kref_put(kref: &dax_region->kref, release: dax_region_free); |
460 | } |
461 | |
462 | /* a return value >= 0 indicates this invocation invalidated the id */ |
463 | static int __free_dev_dax_id(struct dev_dax *dev_dax) |
464 | { |
465 | struct device *dev = &dev_dax->dev; |
466 | struct dax_region *dax_region; |
467 | int rc = dev_dax->id; |
468 | |
469 | device_lock_assert(dev); |
470 | |
471 | if (!dev_dax->dyn_id || dev_dax->id < 0) |
472 | return -1; |
473 | dax_region = dev_dax->region; |
474 | ida_free(&dax_region->ida, id: dev_dax->id); |
475 | dax_region_put(dax_region); |
476 | dev_dax->id = -1; |
477 | return rc; |
478 | } |
479 | |
480 | static int free_dev_dax_id(struct dev_dax *dev_dax) |
481 | { |
482 | struct device *dev = &dev_dax->dev; |
483 | int rc; |
484 | |
485 | device_lock(dev); |
486 | rc = __free_dev_dax_id(dev_dax); |
487 | device_unlock(dev); |
488 | return rc; |
489 | } |
490 | |
491 | static int alloc_dev_dax_id(struct dev_dax *dev_dax) |
492 | { |
493 | struct dax_region *dax_region = dev_dax->region; |
494 | int id; |
495 | |
496 | id = ida_alloc(ida: &dax_region->ida, GFP_KERNEL); |
497 | if (id < 0) |
498 | return id; |
499 | kref_get(kref: &dax_region->kref); |
500 | dev_dax->dyn_id = true; |
501 | dev_dax->id = id; |
502 | return id; |
503 | } |
504 | |
505 | static ssize_t delete_store(struct device *dev, struct device_attribute *attr, |
506 | const char *buf, size_t len) |
507 | { |
508 | struct dax_region *dax_region = dev_get_drvdata(dev); |
509 | struct dev_dax *dev_dax; |
510 | struct device *victim; |
511 | bool do_del = false; |
512 | int rc; |
513 | |
514 | if (is_static(dax_region)) |
515 | return -EINVAL; |
516 | |
517 | victim = device_find_child_by_name(parent: dax_region->dev, name: buf); |
518 | if (!victim) |
519 | return -ENXIO; |
520 | |
521 | device_lock(dev); |
522 | device_lock(dev: victim); |
523 | dev_dax = to_dev_dax(dev: victim); |
524 | if (victim->driver || dev_dax_size(dev_dax)) |
525 | rc = -EBUSY; |
526 | else { |
527 | /* |
528 | * Invalidate the device so it does not become active |
529 | * again, but always preserve device-id-0 so that |
530 | * /sys/bus/dax/ is guaranteed to be populated while any |
531 | * dax_region is registered. |
532 | */ |
533 | if (dev_dax->id > 0) { |
534 | do_del = __free_dev_dax_id(dev_dax) >= 0; |
535 | rc = len; |
536 | if (dax_region->seed == victim) |
537 | dax_region->seed = NULL; |
538 | if (dax_region->youngest == victim) |
539 | dax_region->youngest = NULL; |
540 | } else |
541 | rc = -EBUSY; |
542 | } |
543 | device_unlock(dev: victim); |
544 | |
545 | /* won the race to invalidate the device, clean it up */ |
546 | if (do_del) |
547 | devm_release_action(dev, action: unregister_dev_dax, data: victim); |
548 | device_unlock(dev); |
549 | put_device(dev: victim); |
550 | |
551 | return rc; |
552 | } |
553 | static DEVICE_ATTR_WO(delete); |
554 | |
555 | static umode_t dax_region_visible(struct kobject *kobj, struct attribute *a, |
556 | int n) |
557 | { |
558 | struct device *dev = container_of(kobj, struct device, kobj); |
559 | struct dax_region *dax_region = dev_get_drvdata(dev); |
560 | |
561 | if (is_static(dax_region)) |
562 | if (a == &dev_attr_available_size.attr |
563 | || a == &dev_attr_create.attr |
564 | || a == &dev_attr_seed.attr |
565 | || a == &dev_attr_delete.attr) |
566 | return 0; |
567 | return a->mode; |
568 | } |
569 | |
570 | static struct attribute *dax_region_attributes[] = { |
571 | &dev_attr_available_size.attr, |
572 | &dev_attr_region_size.attr, |
573 | &dev_attr_region_align.attr, |
574 | &dev_attr_create.attr, |
575 | &dev_attr_seed.attr, |
576 | &dev_attr_delete.attr, |
577 | &dev_attr_id.attr, |
578 | NULL, |
579 | }; |
580 | |
581 | static const struct attribute_group dax_region_attribute_group = { |
582 | .name = "dax_region" , |
583 | .attrs = dax_region_attributes, |
584 | .is_visible = dax_region_visible, |
585 | }; |
586 | |
587 | static const struct attribute_group *dax_region_attribute_groups[] = { |
588 | &dax_region_attribute_group, |
589 | NULL, |
590 | }; |
591 | |
592 | static void dax_region_unregister(void *region) |
593 | { |
594 | struct dax_region *dax_region = region; |
595 | |
596 | sysfs_remove_groups(kobj: &dax_region->dev->kobj, |
597 | groups: dax_region_attribute_groups); |
598 | dax_region_put(dax_region); |
599 | } |
600 | |
601 | struct dax_region *alloc_dax_region(struct device *parent, int region_id, |
602 | struct range *range, int target_node, unsigned int align, |
603 | unsigned long flags) |
604 | { |
605 | struct dax_region *dax_region; |
606 | |
607 | /* |
608 | * The DAX core assumes that it can store its private data in |
609 | * parent->driver_data. This WARN is a reminder / safeguard for |
610 | * developers of device-dax drivers. |
611 | */ |
612 | if (dev_get_drvdata(dev: parent)) { |
613 | dev_WARN(parent, "dax core failed to setup private data\n" ); |
614 | return NULL; |
615 | } |
616 | |
617 | if (!IS_ALIGNED(range->start, align) |
618 | || !IS_ALIGNED(range_len(range), align)) |
619 | return NULL; |
620 | |
621 | dax_region = kzalloc(size: sizeof(*dax_region), GFP_KERNEL); |
622 | if (!dax_region) |
623 | return NULL; |
624 | |
625 | dev_set_drvdata(dev: parent, data: dax_region); |
626 | kref_init(kref: &dax_region->kref); |
627 | dax_region->id = region_id; |
628 | dax_region->align = align; |
629 | dax_region->dev = parent; |
630 | dax_region->target_node = target_node; |
631 | ida_init(ida: &dax_region->ida); |
632 | dax_region->res = (struct resource) { |
633 | .start = range->start, |
634 | .end = range->end, |
635 | .flags = IORESOURCE_MEM | flags, |
636 | }; |
637 | |
638 | if (sysfs_create_groups(kobj: &parent->kobj, groups: dax_region_attribute_groups)) { |
639 | kfree(objp: dax_region); |
640 | return NULL; |
641 | } |
642 | |
643 | if (devm_add_action_or_reset(parent, dax_region_unregister, dax_region)) |
644 | return NULL; |
645 | return dax_region; |
646 | } |
647 | EXPORT_SYMBOL_GPL(alloc_dax_region); |
648 | |
649 | static void dax_mapping_release(struct device *dev) |
650 | { |
651 | struct dax_mapping *mapping = to_dax_mapping(dev); |
652 | struct device *parent = dev->parent; |
653 | struct dev_dax *dev_dax = to_dev_dax(dev: parent); |
654 | |
655 | ida_free(&dev_dax->ida, id: mapping->id); |
656 | kfree(objp: mapping); |
657 | put_device(dev: parent); |
658 | } |
659 | |
660 | static void unregister_dax_mapping(void *data) |
661 | { |
662 | struct device *dev = data; |
663 | struct dax_mapping *mapping = to_dax_mapping(dev); |
664 | struct dev_dax *dev_dax = to_dev_dax(dev: dev->parent); |
665 | struct dax_region *dax_region = dev_dax->region; |
666 | |
667 | dev_dbg(dev, "%s\n" , __func__); |
668 | |
669 | device_lock_assert(dev: dax_region->dev); |
670 | |
671 | dev_dax->ranges[mapping->range_id].mapping = NULL; |
672 | mapping->range_id = -1; |
673 | |
674 | device_unregister(dev); |
675 | } |
676 | |
677 | static struct dev_dax_range *get_dax_range(struct device *dev) |
678 | { |
679 | struct dax_mapping *mapping = to_dax_mapping(dev); |
680 | struct dev_dax *dev_dax = to_dev_dax(dev: dev->parent); |
681 | struct dax_region *dax_region = dev_dax->region; |
682 | |
683 | device_lock(dev: dax_region->dev); |
684 | if (mapping->range_id < 0) { |
685 | device_unlock(dev: dax_region->dev); |
686 | return NULL; |
687 | } |
688 | |
689 | return &dev_dax->ranges[mapping->range_id]; |
690 | } |
691 | |
692 | static void put_dax_range(struct dev_dax_range *dax_range) |
693 | { |
694 | struct dax_mapping *mapping = dax_range->mapping; |
695 | struct dev_dax *dev_dax = to_dev_dax(dev: mapping->dev.parent); |
696 | struct dax_region *dax_region = dev_dax->region; |
697 | |
698 | device_unlock(dev: dax_region->dev); |
699 | } |
700 | |
701 | static ssize_t start_show(struct device *dev, |
702 | struct device_attribute *attr, char *buf) |
703 | { |
704 | struct dev_dax_range *dax_range; |
705 | ssize_t rc; |
706 | |
707 | dax_range = get_dax_range(dev); |
708 | if (!dax_range) |
709 | return -ENXIO; |
710 | rc = sprintf(buf, fmt: "%#llx\n" , dax_range->range.start); |
711 | put_dax_range(dax_range); |
712 | |
713 | return rc; |
714 | } |
715 | static DEVICE_ATTR(start, 0400, start_show, NULL); |
716 | |
717 | static ssize_t end_show(struct device *dev, |
718 | struct device_attribute *attr, char *buf) |
719 | { |
720 | struct dev_dax_range *dax_range; |
721 | ssize_t rc; |
722 | |
723 | dax_range = get_dax_range(dev); |
724 | if (!dax_range) |
725 | return -ENXIO; |
726 | rc = sprintf(buf, fmt: "%#llx\n" , dax_range->range.end); |
727 | put_dax_range(dax_range); |
728 | |
729 | return rc; |
730 | } |
731 | static DEVICE_ATTR(end, 0400, end_show, NULL); |
732 | |
733 | static ssize_t pgoff_show(struct device *dev, |
734 | struct device_attribute *attr, char *buf) |
735 | { |
736 | struct dev_dax_range *dax_range; |
737 | ssize_t rc; |
738 | |
739 | dax_range = get_dax_range(dev); |
740 | if (!dax_range) |
741 | return -ENXIO; |
742 | rc = sprintf(buf, fmt: "%#lx\n" , dax_range->pgoff); |
743 | put_dax_range(dax_range); |
744 | |
745 | return rc; |
746 | } |
747 | static DEVICE_ATTR(page_offset, 0400, pgoff_show, NULL); |
748 | |
749 | static struct attribute *dax_mapping_attributes[] = { |
750 | &dev_attr_start.attr, |
751 | &dev_attr_end.attr, |
752 | &dev_attr_page_offset.attr, |
753 | NULL, |
754 | }; |
755 | |
756 | static const struct attribute_group dax_mapping_attribute_group = { |
757 | .attrs = dax_mapping_attributes, |
758 | }; |
759 | |
760 | static const struct attribute_group *dax_mapping_attribute_groups[] = { |
761 | &dax_mapping_attribute_group, |
762 | NULL, |
763 | }; |
764 | |
765 | static struct device_type dax_mapping_type = { |
766 | .release = dax_mapping_release, |
767 | .groups = dax_mapping_attribute_groups, |
768 | }; |
769 | |
770 | static int devm_register_dax_mapping(struct dev_dax *dev_dax, int range_id) |
771 | { |
772 | struct dax_region *dax_region = dev_dax->region; |
773 | struct dax_mapping *mapping; |
774 | struct device *dev; |
775 | int rc; |
776 | |
777 | device_lock_assert(dev: dax_region->dev); |
778 | |
779 | if (dev_WARN_ONCE(&dev_dax->dev, !dax_region->dev->driver, |
780 | "region disabled\n" )) |
781 | return -ENXIO; |
782 | |
783 | mapping = kzalloc(size: sizeof(*mapping), GFP_KERNEL); |
784 | if (!mapping) |
785 | return -ENOMEM; |
786 | mapping->range_id = range_id; |
787 | mapping->id = ida_alloc(ida: &dev_dax->ida, GFP_KERNEL); |
788 | if (mapping->id < 0) { |
789 | kfree(objp: mapping); |
790 | return -ENOMEM; |
791 | } |
792 | dev_dax->ranges[range_id].mapping = mapping; |
793 | dev = &mapping->dev; |
794 | device_initialize(dev); |
795 | dev->parent = &dev_dax->dev; |
796 | get_device(dev: dev->parent); |
797 | dev->type = &dax_mapping_type; |
798 | dev_set_name(dev, name: "mapping%d" , mapping->id); |
799 | rc = device_add(dev); |
800 | if (rc) { |
801 | put_device(dev); |
802 | return rc; |
803 | } |
804 | |
805 | rc = devm_add_action_or_reset(dax_region->dev, unregister_dax_mapping, |
806 | dev); |
807 | if (rc) |
808 | return rc; |
809 | return 0; |
810 | } |
811 | |
812 | static int alloc_dev_dax_range(struct dev_dax *dev_dax, u64 start, |
813 | resource_size_t size) |
814 | { |
815 | struct dax_region *dax_region = dev_dax->region; |
816 | struct resource *res = &dax_region->res; |
817 | struct device *dev = &dev_dax->dev; |
818 | struct dev_dax_range *ranges; |
819 | unsigned long pgoff = 0; |
820 | struct resource *alloc; |
821 | int i, rc; |
822 | |
823 | device_lock_assert(dev: dax_region->dev); |
824 | |
825 | /* handle the seed alloc special case */ |
826 | if (!size) { |
827 | if (dev_WARN_ONCE(dev, dev_dax->nr_range, |
828 | "0-size allocation must be first\n" )) |
829 | return -EBUSY; |
830 | /* nr_range == 0 is elsewhere special cased as 0-size device */ |
831 | return 0; |
832 | } |
833 | |
834 | alloc = __request_region(res, start, n: size, name: dev_name(dev), flags: 0); |
835 | if (!alloc) |
836 | return -ENOMEM; |
837 | |
838 | ranges = krealloc(objp: dev_dax->ranges, new_size: sizeof(*ranges) |
839 | * (dev_dax->nr_range + 1), GFP_KERNEL); |
840 | if (!ranges) { |
841 | __release_region(res, alloc->start, resource_size(res: alloc)); |
842 | return -ENOMEM; |
843 | } |
844 | |
845 | for (i = 0; i < dev_dax->nr_range; i++) |
846 | pgoff += PHYS_PFN(range_len(&ranges[i].range)); |
847 | dev_dax->ranges = ranges; |
848 | ranges[dev_dax->nr_range++] = (struct dev_dax_range) { |
849 | .pgoff = pgoff, |
850 | .range = { |
851 | .start = alloc->start, |
852 | .end = alloc->end, |
853 | }, |
854 | }; |
855 | |
856 | dev_dbg(dev, "alloc range[%d]: %pa:%pa\n" , dev_dax->nr_range - 1, |
857 | &alloc->start, &alloc->end); |
858 | /* |
859 | * A dev_dax instance must be registered before mapping device |
860 | * children can be added. Defer to devm_create_dev_dax() to add |
861 | * the initial mapping device. |
862 | */ |
863 | if (!device_is_registered(dev: &dev_dax->dev)) |
864 | return 0; |
865 | |
866 | rc = devm_register_dax_mapping(dev_dax, range_id: dev_dax->nr_range - 1); |
867 | if (rc) |
868 | trim_dev_dax_range(dev_dax); |
869 | |
870 | return rc; |
871 | } |
872 | |
873 | static int adjust_dev_dax_range(struct dev_dax *dev_dax, struct resource *res, resource_size_t size) |
874 | { |
875 | int last_range = dev_dax->nr_range - 1; |
876 | struct dev_dax_range *dax_range = &dev_dax->ranges[last_range]; |
877 | struct dax_region *dax_region = dev_dax->region; |
878 | bool is_shrink = resource_size(res) > size; |
879 | struct range *range = &dax_range->range; |
880 | struct device *dev = &dev_dax->dev; |
881 | int rc; |
882 | |
883 | device_lock_assert(dev: dax_region->dev); |
884 | |
885 | if (dev_WARN_ONCE(dev, !size, "deletion is handled by dev_dax_shrink\n" )) |
886 | return -EINVAL; |
887 | |
888 | rc = adjust_resource(res, start: range->start, size); |
889 | if (rc) |
890 | return rc; |
891 | |
892 | *range = (struct range) { |
893 | .start = range->start, |
894 | .end = range->start + size - 1, |
895 | }; |
896 | |
897 | dev_dbg(dev, "%s range[%d]: %#llx:%#llx\n" , is_shrink ? "shrink" : "extend" , |
898 | last_range, (unsigned long long) range->start, |
899 | (unsigned long long) range->end); |
900 | |
901 | return 0; |
902 | } |
903 | |
904 | static ssize_t size_show(struct device *dev, |
905 | struct device_attribute *attr, char *buf) |
906 | { |
907 | struct dev_dax *dev_dax = to_dev_dax(dev); |
908 | unsigned long long size; |
909 | |
910 | device_lock(dev); |
911 | size = dev_dax_size(dev_dax); |
912 | device_unlock(dev); |
913 | |
914 | return sprintf(buf, fmt: "%llu\n" , size); |
915 | } |
916 | |
917 | static bool alloc_is_aligned(struct dev_dax *dev_dax, resource_size_t size) |
918 | { |
919 | /* |
920 | * The minimum mapping granularity for a device instance is a |
921 | * single subsection, unless the arch says otherwise. |
922 | */ |
923 | return IS_ALIGNED(size, max_t(unsigned long, dev_dax->align, memremap_compat_align())); |
924 | } |
925 | |
926 | static int dev_dax_shrink(struct dev_dax *dev_dax, resource_size_t size) |
927 | { |
928 | resource_size_t to_shrink = dev_dax_size(dev_dax) - size; |
929 | struct dax_region *dax_region = dev_dax->region; |
930 | struct device *dev = &dev_dax->dev; |
931 | int i; |
932 | |
933 | for (i = dev_dax->nr_range - 1; i >= 0; i--) { |
934 | struct range *range = &dev_dax->ranges[i].range; |
935 | struct dax_mapping *mapping = dev_dax->ranges[i].mapping; |
936 | struct resource *adjust = NULL, *res; |
937 | resource_size_t shrink; |
938 | |
939 | shrink = min_t(u64, to_shrink, range_len(range)); |
940 | if (shrink >= range_len(range)) { |
941 | devm_release_action(dev: dax_region->dev, |
942 | action: unregister_dax_mapping, data: &mapping->dev); |
943 | trim_dev_dax_range(dev_dax); |
944 | to_shrink -= shrink; |
945 | if (!to_shrink) |
946 | break; |
947 | continue; |
948 | } |
949 | |
950 | for_each_dax_region_resource(dax_region, res) |
951 | if (strcmp(res->name, dev_name(dev)) == 0 |
952 | && res->start == range->start) { |
953 | adjust = res; |
954 | break; |
955 | } |
956 | |
957 | if (dev_WARN_ONCE(dev, !adjust || i != dev_dax->nr_range - 1, |
958 | "failed to find matching resource\n" )) |
959 | return -ENXIO; |
960 | return adjust_dev_dax_range(dev_dax, res: adjust, size: range_len(range) |
961 | - shrink); |
962 | } |
963 | return 0; |
964 | } |
965 | |
966 | /* |
967 | * Only allow adjustments that preserve the relative pgoff of existing |
968 | * allocations. I.e. the dev_dax->ranges array is ordered by increasing pgoff. |
969 | */ |
970 | static bool adjust_ok(struct dev_dax *dev_dax, struct resource *res) |
971 | { |
972 | struct dev_dax_range *last; |
973 | int i; |
974 | |
975 | if (dev_dax->nr_range == 0) |
976 | return false; |
977 | if (strcmp(res->name, dev_name(dev: &dev_dax->dev)) != 0) |
978 | return false; |
979 | last = &dev_dax->ranges[dev_dax->nr_range - 1]; |
980 | if (last->range.start != res->start || last->range.end != res->end) |
981 | return false; |
982 | for (i = 0; i < dev_dax->nr_range - 1; i++) { |
983 | struct dev_dax_range *dax_range = &dev_dax->ranges[i]; |
984 | |
985 | if (dax_range->pgoff > last->pgoff) |
986 | return false; |
987 | } |
988 | |
989 | return true; |
990 | } |
991 | |
992 | static ssize_t dev_dax_resize(struct dax_region *dax_region, |
993 | struct dev_dax *dev_dax, resource_size_t size) |
994 | { |
995 | resource_size_t avail = dax_region_avail_size(dax_region), to_alloc; |
996 | resource_size_t dev_size = dev_dax_size(dev_dax); |
997 | struct resource *region_res = &dax_region->res; |
998 | struct device *dev = &dev_dax->dev; |
999 | struct resource *res, *first; |
1000 | resource_size_t alloc = 0; |
1001 | int rc; |
1002 | |
1003 | if (dev->driver) |
1004 | return -EBUSY; |
1005 | if (size == dev_size) |
1006 | return 0; |
1007 | if (size > dev_size && size - dev_size > avail) |
1008 | return -ENOSPC; |
1009 | if (size < dev_size) |
1010 | return dev_dax_shrink(dev_dax, size); |
1011 | |
1012 | to_alloc = size - dev_size; |
1013 | if (dev_WARN_ONCE(dev, !alloc_is_aligned(dev_dax, to_alloc), |
1014 | "resize of %pa misaligned\n" , &to_alloc)) |
1015 | return -ENXIO; |
1016 | |
1017 | /* |
1018 | * Expand the device into the unused portion of the region. This |
1019 | * may involve adjusting the end of an existing resource, or |
1020 | * allocating a new resource. |
1021 | */ |
1022 | retry: |
1023 | first = region_res->child; |
1024 | if (!first) |
1025 | return alloc_dev_dax_range(dev_dax, start: dax_region->res.start, size: to_alloc); |
1026 | |
1027 | rc = -ENOSPC; |
1028 | for (res = first; res; res = res->sibling) { |
1029 | struct resource *next = res->sibling; |
1030 | |
1031 | /* space at the beginning of the region */ |
1032 | if (res == first && res->start > dax_region->res.start) { |
1033 | alloc = min(res->start - dax_region->res.start, to_alloc); |
1034 | rc = alloc_dev_dax_range(dev_dax, start: dax_region->res.start, size: alloc); |
1035 | break; |
1036 | } |
1037 | |
1038 | alloc = 0; |
1039 | /* space between allocations */ |
1040 | if (next && next->start > res->end + 1) |
1041 | alloc = min(next->start - (res->end + 1), to_alloc); |
1042 | |
1043 | /* space at the end of the region */ |
1044 | if (!alloc && !next && res->end < region_res->end) |
1045 | alloc = min(region_res->end - res->end, to_alloc); |
1046 | |
1047 | if (!alloc) |
1048 | continue; |
1049 | |
1050 | if (adjust_ok(dev_dax, res)) { |
1051 | rc = adjust_dev_dax_range(dev_dax, res, size: resource_size(res) + alloc); |
1052 | break; |
1053 | } |
1054 | rc = alloc_dev_dax_range(dev_dax, start: res->end + 1, size: alloc); |
1055 | break; |
1056 | } |
1057 | if (rc) |
1058 | return rc; |
1059 | to_alloc -= alloc; |
1060 | if (to_alloc) |
1061 | goto retry; |
1062 | return 0; |
1063 | } |
1064 | |
1065 | static ssize_t size_store(struct device *dev, struct device_attribute *attr, |
1066 | const char *buf, size_t len) |
1067 | { |
1068 | ssize_t rc; |
1069 | unsigned long long val; |
1070 | struct dev_dax *dev_dax = to_dev_dax(dev); |
1071 | struct dax_region *dax_region = dev_dax->region; |
1072 | |
1073 | rc = kstrtoull(s: buf, base: 0, res: &val); |
1074 | if (rc) |
1075 | return rc; |
1076 | |
1077 | if (!alloc_is_aligned(dev_dax, size: val)) { |
1078 | dev_dbg(dev, "%s: size: %lld misaligned\n" , __func__, val); |
1079 | return -EINVAL; |
1080 | } |
1081 | |
1082 | device_lock(dev: dax_region->dev); |
1083 | if (!dax_region->dev->driver) { |
1084 | device_unlock(dev: dax_region->dev); |
1085 | return -ENXIO; |
1086 | } |
1087 | device_lock(dev); |
1088 | rc = dev_dax_resize(dax_region, dev_dax, size: val); |
1089 | device_unlock(dev); |
1090 | device_unlock(dev: dax_region->dev); |
1091 | |
1092 | return rc == 0 ? len : rc; |
1093 | } |
1094 | static DEVICE_ATTR_RW(size); |
1095 | |
1096 | static ssize_t range_parse(const char *opt, size_t len, struct range *range) |
1097 | { |
1098 | unsigned long long addr = 0; |
1099 | char *start, *end, *str; |
1100 | ssize_t rc = -EINVAL; |
1101 | |
1102 | str = kstrdup(s: opt, GFP_KERNEL); |
1103 | if (!str) |
1104 | return rc; |
1105 | |
1106 | end = str; |
1107 | start = strsep(&end, "-" ); |
1108 | if (!start || !end) |
1109 | goto err; |
1110 | |
1111 | rc = kstrtoull(s: start, base: 16, res: &addr); |
1112 | if (rc) |
1113 | goto err; |
1114 | range->start = addr; |
1115 | |
1116 | rc = kstrtoull(s: end, base: 16, res: &addr); |
1117 | if (rc) |
1118 | goto err; |
1119 | range->end = addr; |
1120 | |
1121 | err: |
1122 | kfree(objp: str); |
1123 | return rc; |
1124 | } |
1125 | |
1126 | static ssize_t mapping_store(struct device *dev, struct device_attribute *attr, |
1127 | const char *buf, size_t len) |
1128 | { |
1129 | struct dev_dax *dev_dax = to_dev_dax(dev); |
1130 | struct dax_region *dax_region = dev_dax->region; |
1131 | size_t to_alloc; |
1132 | struct range r; |
1133 | ssize_t rc; |
1134 | |
1135 | rc = range_parse(opt: buf, len, range: &r); |
1136 | if (rc) |
1137 | return rc; |
1138 | |
1139 | rc = -ENXIO; |
1140 | device_lock(dev: dax_region->dev); |
1141 | if (!dax_region->dev->driver) { |
1142 | device_unlock(dev: dax_region->dev); |
1143 | return rc; |
1144 | } |
1145 | device_lock(dev); |
1146 | |
1147 | to_alloc = range_len(range: &r); |
1148 | if (alloc_is_aligned(dev_dax, size: to_alloc)) |
1149 | rc = alloc_dev_dax_range(dev_dax, start: r.start, size: to_alloc); |
1150 | device_unlock(dev); |
1151 | device_unlock(dev: dax_region->dev); |
1152 | |
1153 | return rc == 0 ? len : rc; |
1154 | } |
1155 | static DEVICE_ATTR_WO(mapping); |
1156 | |
1157 | static ssize_t align_show(struct device *dev, |
1158 | struct device_attribute *attr, char *buf) |
1159 | { |
1160 | struct dev_dax *dev_dax = to_dev_dax(dev); |
1161 | |
1162 | return sprintf(buf, fmt: "%d\n" , dev_dax->align); |
1163 | } |
1164 | |
1165 | static ssize_t dev_dax_validate_align(struct dev_dax *dev_dax) |
1166 | { |
1167 | struct device *dev = &dev_dax->dev; |
1168 | int i; |
1169 | |
1170 | for (i = 0; i < dev_dax->nr_range; i++) { |
1171 | size_t len = range_len(range: &dev_dax->ranges[i].range); |
1172 | |
1173 | if (!alloc_is_aligned(dev_dax, size: len)) { |
1174 | dev_dbg(dev, "%s: align %u invalid for range %d\n" , |
1175 | __func__, dev_dax->align, i); |
1176 | return -EINVAL; |
1177 | } |
1178 | } |
1179 | |
1180 | return 0; |
1181 | } |
1182 | |
1183 | static ssize_t align_store(struct device *dev, struct device_attribute *attr, |
1184 | const char *buf, size_t len) |
1185 | { |
1186 | struct dev_dax *dev_dax = to_dev_dax(dev); |
1187 | struct dax_region *dax_region = dev_dax->region; |
1188 | unsigned long val, align_save; |
1189 | ssize_t rc; |
1190 | |
1191 | rc = kstrtoul(s: buf, base: 0, res: &val); |
1192 | if (rc) |
1193 | return -ENXIO; |
1194 | |
1195 | if (!dax_align_valid(align: val)) |
1196 | return -EINVAL; |
1197 | |
1198 | device_lock(dev: dax_region->dev); |
1199 | if (!dax_region->dev->driver) { |
1200 | device_unlock(dev: dax_region->dev); |
1201 | return -ENXIO; |
1202 | } |
1203 | |
1204 | device_lock(dev); |
1205 | if (dev->driver) { |
1206 | rc = -EBUSY; |
1207 | goto out_unlock; |
1208 | } |
1209 | |
1210 | align_save = dev_dax->align; |
1211 | dev_dax->align = val; |
1212 | rc = dev_dax_validate_align(dev_dax); |
1213 | if (rc) |
1214 | dev_dax->align = align_save; |
1215 | out_unlock: |
1216 | device_unlock(dev); |
1217 | device_unlock(dev: dax_region->dev); |
1218 | return rc == 0 ? len : rc; |
1219 | } |
1220 | static DEVICE_ATTR_RW(align); |
1221 | |
1222 | static int dev_dax_target_node(struct dev_dax *dev_dax) |
1223 | { |
1224 | struct dax_region *dax_region = dev_dax->region; |
1225 | |
1226 | return dax_region->target_node; |
1227 | } |
1228 | |
1229 | static ssize_t target_node_show(struct device *dev, |
1230 | struct device_attribute *attr, char *buf) |
1231 | { |
1232 | struct dev_dax *dev_dax = to_dev_dax(dev); |
1233 | |
1234 | return sprintf(buf, fmt: "%d\n" , dev_dax_target_node(dev_dax)); |
1235 | } |
1236 | static DEVICE_ATTR_RO(target_node); |
1237 | |
1238 | static ssize_t resource_show(struct device *dev, |
1239 | struct device_attribute *attr, char *buf) |
1240 | { |
1241 | struct dev_dax *dev_dax = to_dev_dax(dev); |
1242 | struct dax_region *dax_region = dev_dax->region; |
1243 | unsigned long long start; |
1244 | |
1245 | if (dev_dax->nr_range < 1) |
1246 | start = dax_region->res.start; |
1247 | else |
1248 | start = dev_dax->ranges[0].range.start; |
1249 | |
1250 | return sprintf(buf, fmt: "%#llx\n" , start); |
1251 | } |
1252 | static DEVICE_ATTR(resource, 0400, resource_show, NULL); |
1253 | |
1254 | static ssize_t modalias_show(struct device *dev, struct device_attribute *attr, |
1255 | char *buf) |
1256 | { |
1257 | /* |
1258 | * We only ever expect to handle device-dax instances, i.e. the |
1259 | * @type argument to MODULE_ALIAS_DAX_DEVICE() is always zero |
1260 | */ |
1261 | return sprintf(buf, DAX_DEVICE_MODALIAS_FMT "\n" , 0); |
1262 | } |
1263 | static DEVICE_ATTR_RO(modalias); |
1264 | |
1265 | static ssize_t numa_node_show(struct device *dev, |
1266 | struct device_attribute *attr, char *buf) |
1267 | { |
1268 | return sprintf(buf, fmt: "%d\n" , dev_to_node(dev)); |
1269 | } |
1270 | static DEVICE_ATTR_RO(numa_node); |
1271 | |
1272 | static umode_t dev_dax_visible(struct kobject *kobj, struct attribute *a, int n) |
1273 | { |
1274 | struct device *dev = container_of(kobj, struct device, kobj); |
1275 | struct dev_dax *dev_dax = to_dev_dax(dev); |
1276 | struct dax_region *dax_region = dev_dax->region; |
1277 | |
1278 | if (a == &dev_attr_target_node.attr && dev_dax_target_node(dev_dax) < 0) |
1279 | return 0; |
1280 | if (a == &dev_attr_numa_node.attr && !IS_ENABLED(CONFIG_NUMA)) |
1281 | return 0; |
1282 | if (a == &dev_attr_mapping.attr && is_static(dax_region)) |
1283 | return 0; |
1284 | if ((a == &dev_attr_align.attr || |
1285 | a == &dev_attr_size.attr) && is_static(dax_region)) |
1286 | return 0444; |
1287 | return a->mode; |
1288 | } |
1289 | |
1290 | static struct attribute *dev_dax_attributes[] = { |
1291 | &dev_attr_modalias.attr, |
1292 | &dev_attr_size.attr, |
1293 | &dev_attr_mapping.attr, |
1294 | &dev_attr_target_node.attr, |
1295 | &dev_attr_align.attr, |
1296 | &dev_attr_resource.attr, |
1297 | &dev_attr_numa_node.attr, |
1298 | NULL, |
1299 | }; |
1300 | |
1301 | static const struct attribute_group dev_dax_attribute_group = { |
1302 | .attrs = dev_dax_attributes, |
1303 | .is_visible = dev_dax_visible, |
1304 | }; |
1305 | |
1306 | static const struct attribute_group *dax_attribute_groups[] = { |
1307 | &dev_dax_attribute_group, |
1308 | NULL, |
1309 | }; |
1310 | |
1311 | static void dev_dax_release(struct device *dev) |
1312 | { |
1313 | struct dev_dax *dev_dax = to_dev_dax(dev); |
1314 | struct dax_device *dax_dev = dev_dax->dax_dev; |
1315 | |
1316 | put_dax(dax_dev); |
1317 | free_dev_dax_id(dev_dax); |
1318 | kfree(objp: dev_dax->pgmap); |
1319 | kfree(objp: dev_dax); |
1320 | } |
1321 | |
1322 | static const struct device_type dev_dax_type = { |
1323 | .release = dev_dax_release, |
1324 | .groups = dax_attribute_groups, |
1325 | }; |
1326 | |
1327 | struct dev_dax *devm_create_dev_dax(struct dev_dax_data *data) |
1328 | { |
1329 | struct dax_region *dax_region = data->dax_region; |
1330 | struct device *parent = dax_region->dev; |
1331 | struct dax_device *dax_dev; |
1332 | struct dev_dax *dev_dax; |
1333 | struct inode *inode; |
1334 | struct device *dev; |
1335 | int rc; |
1336 | |
1337 | dev_dax = kzalloc(size: sizeof(*dev_dax), GFP_KERNEL); |
1338 | if (!dev_dax) |
1339 | return ERR_PTR(error: -ENOMEM); |
1340 | |
1341 | dev_dax->region = dax_region; |
1342 | if (is_static(dax_region)) { |
1343 | if (dev_WARN_ONCE(parent, data->id < 0, |
1344 | "dynamic id specified to static region\n" )) { |
1345 | rc = -EINVAL; |
1346 | goto err_id; |
1347 | } |
1348 | |
1349 | dev_dax->id = data->id; |
1350 | } else { |
1351 | if (dev_WARN_ONCE(parent, data->id >= 0, |
1352 | "static id specified to dynamic region\n" )) { |
1353 | rc = -EINVAL; |
1354 | goto err_id; |
1355 | } |
1356 | |
1357 | rc = alloc_dev_dax_id(dev_dax); |
1358 | if (rc < 0) |
1359 | goto err_id; |
1360 | } |
1361 | |
1362 | dev = &dev_dax->dev; |
1363 | device_initialize(dev); |
1364 | dev_set_name(dev, name: "dax%d.%d" , dax_region->id, dev_dax->id); |
1365 | |
1366 | rc = alloc_dev_dax_range(dev_dax, start: dax_region->res.start, size: data->size); |
1367 | if (rc) |
1368 | goto err_range; |
1369 | |
1370 | if (data->pgmap) { |
1371 | dev_WARN_ONCE(parent, !is_static(dax_region), |
1372 | "custom dev_pagemap requires a static dax_region\n" ); |
1373 | |
1374 | dev_dax->pgmap = kmemdup(p: data->pgmap, |
1375 | size: sizeof(struct dev_pagemap), GFP_KERNEL); |
1376 | if (!dev_dax->pgmap) { |
1377 | rc = -ENOMEM; |
1378 | goto err_pgmap; |
1379 | } |
1380 | } |
1381 | |
1382 | /* |
1383 | * No dax_operations since there is no access to this device outside of |
1384 | * mmap of the resulting character device. |
1385 | */ |
1386 | dax_dev = alloc_dax(private: dev_dax, NULL); |
1387 | if (IS_ERR(ptr: dax_dev)) { |
1388 | rc = PTR_ERR(ptr: dax_dev); |
1389 | goto err_alloc_dax; |
1390 | } |
1391 | set_dax_synchronous(dax_dev); |
1392 | set_dax_nocache(dax_dev); |
1393 | set_dax_nomc(dax_dev); |
1394 | |
1395 | /* a device_dax instance is dead while the driver is not attached */ |
1396 | kill_dax(dax_dev); |
1397 | |
1398 | dev_dax->dax_dev = dax_dev; |
1399 | dev_dax->target_node = dax_region->target_node; |
1400 | dev_dax->align = dax_region->align; |
1401 | ida_init(ida: &dev_dax->ida); |
1402 | |
1403 | inode = dax_inode(dax_dev); |
1404 | dev->devt = inode->i_rdev; |
1405 | dev->bus = &dax_bus_type; |
1406 | dev->parent = parent; |
1407 | dev->type = &dev_dax_type; |
1408 | |
1409 | rc = device_add(dev); |
1410 | if (rc) { |
1411 | kill_dev_dax(dev_dax); |
1412 | put_device(dev); |
1413 | return ERR_PTR(error: rc); |
1414 | } |
1415 | |
1416 | rc = devm_add_action_or_reset(dax_region->dev, unregister_dev_dax, dev); |
1417 | if (rc) |
1418 | return ERR_PTR(error: rc); |
1419 | |
1420 | /* register mapping device for the initial allocation range */ |
1421 | if (dev_dax->nr_range && range_len(range: &dev_dax->ranges[0].range)) { |
1422 | rc = devm_register_dax_mapping(dev_dax, range_id: 0); |
1423 | if (rc) |
1424 | return ERR_PTR(error: rc); |
1425 | } |
1426 | |
1427 | return dev_dax; |
1428 | |
1429 | err_alloc_dax: |
1430 | kfree(objp: dev_dax->pgmap); |
1431 | err_pgmap: |
1432 | free_dev_dax_ranges(dev_dax); |
1433 | err_range: |
1434 | free_dev_dax_id(dev_dax); |
1435 | err_id: |
1436 | kfree(objp: dev_dax); |
1437 | |
1438 | return ERR_PTR(error: rc); |
1439 | } |
1440 | EXPORT_SYMBOL_GPL(devm_create_dev_dax); |
1441 | |
1442 | int __dax_driver_register(struct dax_device_driver *dax_drv, |
1443 | struct module *module, const char *mod_name) |
1444 | { |
1445 | struct device_driver *drv = &dax_drv->drv; |
1446 | |
1447 | /* |
1448 | * dax_bus_probe() calls dax_drv->probe() unconditionally. |
1449 | * So better be safe than sorry and ensure it is provided. |
1450 | */ |
1451 | if (!dax_drv->probe) |
1452 | return -EINVAL; |
1453 | |
1454 | INIT_LIST_HEAD(list: &dax_drv->ids); |
1455 | drv->owner = module; |
1456 | drv->name = mod_name; |
1457 | drv->mod_name = mod_name; |
1458 | drv->bus = &dax_bus_type; |
1459 | |
1460 | return driver_register(drv); |
1461 | } |
1462 | EXPORT_SYMBOL_GPL(__dax_driver_register); |
1463 | |
1464 | void dax_driver_unregister(struct dax_device_driver *dax_drv) |
1465 | { |
1466 | struct device_driver *drv = &dax_drv->drv; |
1467 | struct dax_id *dax_id, *_id; |
1468 | |
1469 | mutex_lock(&dax_bus_lock); |
1470 | list_for_each_entry_safe(dax_id, _id, &dax_drv->ids, list) { |
1471 | list_del(entry: &dax_id->list); |
1472 | kfree(objp: dax_id); |
1473 | } |
1474 | mutex_unlock(lock: &dax_bus_lock); |
1475 | driver_unregister(drv); |
1476 | } |
1477 | EXPORT_SYMBOL_GPL(dax_driver_unregister); |
1478 | |
1479 | int __init dax_bus_init(void) |
1480 | { |
1481 | return bus_register(bus: &dax_bus_type); |
1482 | } |
1483 | |
1484 | void __exit dax_bus_exit(void) |
1485 | { |
1486 | bus_unregister(bus: &dax_bus_type); |
1487 | } |
1488 | |