Linux-libre 5.4.49-gnu
[librecmc/linux-libre.git] / drivers / nvdimm / bus.c
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
4  */
5 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
6 #include <linux/libnvdimm.h>
7 #include <linux/sched/mm.h>
8 #include <linux/vmalloc.h>
9 #include <linux/uaccess.h>
10 #include <linux/module.h>
11 #include <linux/blkdev.h>
12 #include <linux/fcntl.h>
13 #include <linux/async.h>
14 #include <linux/genhd.h>
15 #include <linux/ndctl.h>
16 #include <linux/sched.h>
17 #include <linux/slab.h>
18 #include <linux/cpu.h>
19 #include <linux/fs.h>
20 #include <linux/io.h>
21 #include <linux/mm.h>
22 #include <linux/nd.h>
23 #include "nd-core.h"
24 #include "nd.h"
25 #include "pfn.h"
26
27 int nvdimm_major;
28 static int nvdimm_bus_major;
29 struct class *nd_class;
30 static DEFINE_IDA(nd_ida);
31
32 static int to_nd_device_type(struct device *dev)
33 {
34         if (is_nvdimm(dev))
35                 return ND_DEVICE_DIMM;
36         else if (is_memory(dev))
37                 return ND_DEVICE_REGION_PMEM;
38         else if (is_nd_blk(dev))
39                 return ND_DEVICE_REGION_BLK;
40         else if (is_nd_dax(dev))
41                 return ND_DEVICE_DAX_PMEM;
42         else if (is_nd_region(dev->parent))
43                 return nd_region_to_nstype(to_nd_region(dev->parent));
44
45         return 0;
46 }
47
48 static int nvdimm_bus_uevent(struct device *dev, struct kobj_uevent_env *env)
49 {
50         return add_uevent_var(env, "MODALIAS=" ND_DEVICE_MODALIAS_FMT,
51                         to_nd_device_type(dev));
52 }
53
54 static struct module *to_bus_provider(struct device *dev)
55 {
56         /* pin bus providers while regions are enabled */
57         if (is_nd_region(dev)) {
58                 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
59
60                 return nvdimm_bus->nd_desc->module;
61         }
62         return NULL;
63 }
64
65 static void nvdimm_bus_probe_start(struct nvdimm_bus *nvdimm_bus)
66 {
67         nvdimm_bus_lock(&nvdimm_bus->dev);
68         nvdimm_bus->probe_active++;
69         nvdimm_bus_unlock(&nvdimm_bus->dev);
70 }
71
72 static void nvdimm_bus_probe_end(struct nvdimm_bus *nvdimm_bus)
73 {
74         nvdimm_bus_lock(&nvdimm_bus->dev);
75         if (--nvdimm_bus->probe_active == 0)
76                 wake_up(&nvdimm_bus->wait);
77         nvdimm_bus_unlock(&nvdimm_bus->dev);
78 }
79
80 static int nvdimm_bus_probe(struct device *dev)
81 {
82         struct nd_device_driver *nd_drv = to_nd_device_driver(dev->driver);
83         struct module *provider = to_bus_provider(dev);
84         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
85         int rc;
86
87         if (!try_module_get(provider))
88                 return -ENXIO;
89
90         dev_dbg(&nvdimm_bus->dev, "START: %s.probe(%s)\n",
91                         dev->driver->name, dev_name(dev));
92
93         nvdimm_bus_probe_start(nvdimm_bus);
94         debug_nvdimm_lock(dev);
95         rc = nd_drv->probe(dev);
96         debug_nvdimm_unlock(dev);
97
98         if ((rc == 0 || rc == -EOPNOTSUPP) &&
99                         dev->parent && is_nd_region(dev->parent))
100                 nd_region_advance_seeds(to_nd_region(dev->parent), dev);
101         nvdimm_bus_probe_end(nvdimm_bus);
102
103         dev_dbg(&nvdimm_bus->dev, "END: %s.probe(%s) = %d\n", dev->driver->name,
104                         dev_name(dev), rc);
105
106         if (rc != 0)
107                 module_put(provider);
108         return rc;
109 }
110
111 static int nvdimm_bus_remove(struct device *dev)
112 {
113         struct nd_device_driver *nd_drv = to_nd_device_driver(dev->driver);
114         struct module *provider = to_bus_provider(dev);
115         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
116         int rc = 0;
117
118         if (nd_drv->remove) {
119                 debug_nvdimm_lock(dev);
120                 rc = nd_drv->remove(dev);
121                 debug_nvdimm_unlock(dev);
122         }
123
124         dev_dbg(&nvdimm_bus->dev, "%s.remove(%s) = %d\n", dev->driver->name,
125                         dev_name(dev), rc);
126         module_put(provider);
127         return rc;
128 }
129
130 static void nvdimm_bus_shutdown(struct device *dev)
131 {
132         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
133         struct nd_device_driver *nd_drv = NULL;
134
135         if (dev->driver)
136                 nd_drv = to_nd_device_driver(dev->driver);
137
138         if (nd_drv && nd_drv->shutdown) {
139                 nd_drv->shutdown(dev);
140                 dev_dbg(&nvdimm_bus->dev, "%s.shutdown(%s)\n",
141                                 dev->driver->name, dev_name(dev));
142         }
143 }
144
145 void nd_device_notify(struct device *dev, enum nvdimm_event event)
146 {
147         nd_device_lock(dev);
148         if (dev->driver) {
149                 struct nd_device_driver *nd_drv;
150
151                 nd_drv = to_nd_device_driver(dev->driver);
152                 if (nd_drv->notify)
153                         nd_drv->notify(dev, event);
154         }
155         nd_device_unlock(dev);
156 }
157 EXPORT_SYMBOL(nd_device_notify);
158
159 void nvdimm_region_notify(struct nd_region *nd_region, enum nvdimm_event event)
160 {
161         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
162
163         if (!nvdimm_bus)
164                 return;
165
166         /* caller is responsible for holding a reference on the device */
167         nd_device_notify(&nd_region->dev, event);
168 }
169 EXPORT_SYMBOL_GPL(nvdimm_region_notify);
170
171 struct clear_badblocks_context {
172         resource_size_t phys, cleared;
173 };
174
175 static int nvdimm_clear_badblocks_region(struct device *dev, void *data)
176 {
177         struct clear_badblocks_context *ctx = data;
178         struct nd_region *nd_region;
179         resource_size_t ndr_end;
180         sector_t sector;
181
182         /* make sure device is a region */
183         if (!is_memory(dev))
184                 return 0;
185
186         nd_region = to_nd_region(dev);
187         ndr_end = nd_region->ndr_start + nd_region->ndr_size - 1;
188
189         /* make sure we are in the region */
190         if (ctx->phys < nd_region->ndr_start
191                         || (ctx->phys + ctx->cleared) > ndr_end)
192                 return 0;
193
194         sector = (ctx->phys - nd_region->ndr_start) / 512;
195         badblocks_clear(&nd_region->bb, sector, ctx->cleared / 512);
196
197         if (nd_region->bb_state)
198                 sysfs_notify_dirent(nd_region->bb_state);
199
200         return 0;
201 }
202
203 static void nvdimm_clear_badblocks_regions(struct nvdimm_bus *nvdimm_bus,
204                 phys_addr_t phys, u64 cleared)
205 {
206         struct clear_badblocks_context ctx = {
207                 .phys = phys,
208                 .cleared = cleared,
209         };
210
211         device_for_each_child(&nvdimm_bus->dev, &ctx,
212                         nvdimm_clear_badblocks_region);
213 }
214
215 static void nvdimm_account_cleared_poison(struct nvdimm_bus *nvdimm_bus,
216                 phys_addr_t phys, u64 cleared)
217 {
218         if (cleared > 0)
219                 badrange_forget(&nvdimm_bus->badrange, phys, cleared);
220
221         if (cleared > 0 && cleared / 512)
222                 nvdimm_clear_badblocks_regions(nvdimm_bus, phys, cleared);
223 }
224
225 long nvdimm_clear_poison(struct device *dev, phys_addr_t phys,
226                 unsigned int len)
227 {
228         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
229         struct nvdimm_bus_descriptor *nd_desc;
230         struct nd_cmd_clear_error clear_err;
231         struct nd_cmd_ars_cap ars_cap;
232         u32 clear_err_unit, mask;
233         unsigned int noio_flag;
234         int cmd_rc, rc;
235
236         if (!nvdimm_bus)
237                 return -ENXIO;
238
239         nd_desc = nvdimm_bus->nd_desc;
240         /*
241          * if ndctl does not exist, it's PMEM_LEGACY and
242          * we want to just pretend everything is handled.
243          */
244         if (!nd_desc->ndctl)
245                 return len;
246
247         memset(&ars_cap, 0, sizeof(ars_cap));
248         ars_cap.address = phys;
249         ars_cap.length = len;
250         noio_flag = memalloc_noio_save();
251         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, &ars_cap,
252                         sizeof(ars_cap), &cmd_rc);
253         memalloc_noio_restore(noio_flag);
254         if (rc < 0)
255                 return rc;
256         if (cmd_rc < 0)
257                 return cmd_rc;
258         clear_err_unit = ars_cap.clear_err_unit;
259         if (!clear_err_unit || !is_power_of_2(clear_err_unit))
260                 return -ENXIO;
261
262         mask = clear_err_unit - 1;
263         if ((phys | len) & mask)
264                 return -ENXIO;
265         memset(&clear_err, 0, sizeof(clear_err));
266         clear_err.address = phys;
267         clear_err.length = len;
268         noio_flag = memalloc_noio_save();
269         rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_CLEAR_ERROR, &clear_err,
270                         sizeof(clear_err), &cmd_rc);
271         memalloc_noio_restore(noio_flag);
272         if (rc < 0)
273                 return rc;
274         if (cmd_rc < 0)
275                 return cmd_rc;
276
277         nvdimm_account_cleared_poison(nvdimm_bus, phys, clear_err.cleared);
278
279         return clear_err.cleared;
280 }
281 EXPORT_SYMBOL_GPL(nvdimm_clear_poison);
282
283 static int nvdimm_bus_match(struct device *dev, struct device_driver *drv);
284
285 static struct bus_type nvdimm_bus_type = {
286         .name = "nd",
287         .uevent = nvdimm_bus_uevent,
288         .match = nvdimm_bus_match,
289         .probe = nvdimm_bus_probe,
290         .remove = nvdimm_bus_remove,
291         .shutdown = nvdimm_bus_shutdown,
292 };
293
294 static void nvdimm_bus_release(struct device *dev)
295 {
296         struct nvdimm_bus *nvdimm_bus;
297
298         nvdimm_bus = container_of(dev, struct nvdimm_bus, dev);
299         ida_simple_remove(&nd_ida, nvdimm_bus->id);
300         kfree(nvdimm_bus);
301 }
302
303 bool is_nvdimm_bus(struct device *dev)
304 {
305         return dev->release == nvdimm_bus_release;
306 }
307
308 struct nvdimm_bus *walk_to_nvdimm_bus(struct device *nd_dev)
309 {
310         struct device *dev;
311
312         for (dev = nd_dev; dev; dev = dev->parent)
313                 if (is_nvdimm_bus(dev))
314                         break;
315         dev_WARN_ONCE(nd_dev, !dev, "invalid dev, not on nd bus\n");
316         if (dev)
317                 return to_nvdimm_bus(dev);
318         return NULL;
319 }
320
321 struct nvdimm_bus *to_nvdimm_bus(struct device *dev)
322 {
323         struct nvdimm_bus *nvdimm_bus;
324
325         nvdimm_bus = container_of(dev, struct nvdimm_bus, dev);
326         WARN_ON(!is_nvdimm_bus(dev));
327         return nvdimm_bus;
328 }
329 EXPORT_SYMBOL_GPL(to_nvdimm_bus);
330
331 struct nvdimm_bus *nvdimm_to_bus(struct nvdimm *nvdimm)
332 {
333         return to_nvdimm_bus(nvdimm->dev.parent);
334 }
335 EXPORT_SYMBOL_GPL(nvdimm_to_bus);
336
337 struct nvdimm_bus *nvdimm_bus_register(struct device *parent,
338                 struct nvdimm_bus_descriptor *nd_desc)
339 {
340         struct nvdimm_bus *nvdimm_bus;
341         int rc;
342
343         nvdimm_bus = kzalloc(sizeof(*nvdimm_bus), GFP_KERNEL);
344         if (!nvdimm_bus)
345                 return NULL;
346         INIT_LIST_HEAD(&nvdimm_bus->list);
347         INIT_LIST_HEAD(&nvdimm_bus->mapping_list);
348         init_waitqueue_head(&nvdimm_bus->wait);
349         nvdimm_bus->id = ida_simple_get(&nd_ida, 0, 0, GFP_KERNEL);
350         if (nvdimm_bus->id < 0) {
351                 kfree(nvdimm_bus);
352                 return NULL;
353         }
354         mutex_init(&nvdimm_bus->reconfig_mutex);
355         badrange_init(&nvdimm_bus->badrange);
356         nvdimm_bus->nd_desc = nd_desc;
357         nvdimm_bus->dev.parent = parent;
358         nvdimm_bus->dev.release = nvdimm_bus_release;
359         nvdimm_bus->dev.groups = nd_desc->attr_groups;
360         nvdimm_bus->dev.bus = &nvdimm_bus_type;
361         nvdimm_bus->dev.of_node = nd_desc->of_node;
362         dev_set_name(&nvdimm_bus->dev, "ndbus%d", nvdimm_bus->id);
363         rc = device_register(&nvdimm_bus->dev);
364         if (rc) {
365                 dev_dbg(&nvdimm_bus->dev, "registration failed: %d\n", rc);
366                 goto err;
367         }
368
369         return nvdimm_bus;
370  err:
371         put_device(&nvdimm_bus->dev);
372         return NULL;
373 }
374 EXPORT_SYMBOL_GPL(nvdimm_bus_register);
375
376 void nvdimm_bus_unregister(struct nvdimm_bus *nvdimm_bus)
377 {
378         if (!nvdimm_bus)
379                 return;
380         device_unregister(&nvdimm_bus->dev);
381 }
382 EXPORT_SYMBOL_GPL(nvdimm_bus_unregister);
383
384 static int child_unregister(struct device *dev, void *data)
385 {
386         /*
387          * the singular ndctl class device per bus needs to be
388          * "device_destroy"ed, so skip it here
389          *
390          * i.e. remove classless children
391          */
392         if (dev->class)
393                 return 0;
394
395         if (is_nvdimm(dev)) {
396                 struct nvdimm *nvdimm = to_nvdimm(dev);
397                 bool dev_put = false;
398
399                 /* We are shutting down. Make state frozen artificially. */
400                 nvdimm_bus_lock(dev);
401                 set_bit(NVDIMM_SECURITY_FROZEN, &nvdimm->sec.flags);
402                 if (test_and_clear_bit(NDD_WORK_PENDING, &nvdimm->flags))
403                         dev_put = true;
404                 nvdimm_bus_unlock(dev);
405                 cancel_delayed_work_sync(&nvdimm->dwork);
406                 if (dev_put)
407                         put_device(dev);
408         }
409         nd_device_unregister(dev, ND_SYNC);
410
411         return 0;
412 }
413
414 static void free_badrange_list(struct list_head *badrange_list)
415 {
416         struct badrange_entry *bre, *next;
417
418         list_for_each_entry_safe(bre, next, badrange_list, list) {
419                 list_del(&bre->list);
420                 kfree(bre);
421         }
422         list_del_init(badrange_list);
423 }
424
425 static int nd_bus_remove(struct device *dev)
426 {
427         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
428
429         mutex_lock(&nvdimm_bus_list_mutex);
430         list_del_init(&nvdimm_bus->list);
431         mutex_unlock(&nvdimm_bus_list_mutex);
432
433         wait_event(nvdimm_bus->wait,
434                         atomic_read(&nvdimm_bus->ioctl_active) == 0);
435
436         nd_synchronize();
437         device_for_each_child(&nvdimm_bus->dev, NULL, child_unregister);
438
439         spin_lock(&nvdimm_bus->badrange.lock);
440         free_badrange_list(&nvdimm_bus->badrange.list);
441         spin_unlock(&nvdimm_bus->badrange.lock);
442
443         nvdimm_bus_destroy_ndctl(nvdimm_bus);
444
445         return 0;
446 }
447
448 static int nd_bus_probe(struct device *dev)
449 {
450         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
451         int rc;
452
453         rc = nvdimm_bus_create_ndctl(nvdimm_bus);
454         if (rc)
455                 return rc;
456
457         mutex_lock(&nvdimm_bus_list_mutex);
458         list_add_tail(&nvdimm_bus->list, &nvdimm_bus_list);
459         mutex_unlock(&nvdimm_bus_list_mutex);
460
461         /* enable bus provider attributes to look up their local context */
462         dev_set_drvdata(dev, nvdimm_bus->nd_desc);
463
464         return 0;
465 }
466
467 static struct nd_device_driver nd_bus_driver = {
468         .probe = nd_bus_probe,
469         .remove = nd_bus_remove,
470         .drv = {
471                 .name = "nd_bus",
472                 .suppress_bind_attrs = true,
473                 .bus = &nvdimm_bus_type,
474                 .owner = THIS_MODULE,
475                 .mod_name = KBUILD_MODNAME,
476         },
477 };
478
479 static int nvdimm_bus_match(struct device *dev, struct device_driver *drv)
480 {
481         struct nd_device_driver *nd_drv = to_nd_device_driver(drv);
482
483         if (is_nvdimm_bus(dev) && nd_drv == &nd_bus_driver)
484                 return true;
485
486         return !!test_bit(to_nd_device_type(dev), &nd_drv->type);
487 }
488
489 static ASYNC_DOMAIN_EXCLUSIVE(nd_async_domain);
490
491 void nd_synchronize(void)
492 {
493         async_synchronize_full_domain(&nd_async_domain);
494 }
495 EXPORT_SYMBOL_GPL(nd_synchronize);
496
497 static void nd_async_device_register(void *d, async_cookie_t cookie)
498 {
499         struct device *dev = d;
500
501         if (device_add(dev) != 0) {
502                 dev_err(dev, "%s: failed\n", __func__);
503                 put_device(dev);
504         }
505         put_device(dev);
506         if (dev->parent)
507                 put_device(dev->parent);
508 }
509
510 static void nd_async_device_unregister(void *d, async_cookie_t cookie)
511 {
512         struct device *dev = d;
513
514         /* flush bus operations before delete */
515         nvdimm_bus_lock(dev);
516         nvdimm_bus_unlock(dev);
517
518         device_unregister(dev);
519         put_device(dev);
520 }
521
522 void __nd_device_register(struct device *dev)
523 {
524         if (!dev)
525                 return;
526
527         /*
528          * Ensure that region devices always have their NUMA node set as
529          * early as possible. This way we are able to make certain that
530          * any memory associated with the creation and the creation
531          * itself of the region is associated with the correct node.
532          */
533         if (is_nd_region(dev))
534                 set_dev_node(dev, to_nd_region(dev)->numa_node);
535
536         dev->bus = &nvdimm_bus_type;
537         if (dev->parent) {
538                 get_device(dev->parent);
539                 if (dev_to_node(dev) == NUMA_NO_NODE)
540                         set_dev_node(dev, dev_to_node(dev->parent));
541         }
542         get_device(dev);
543
544         async_schedule_dev_domain(nd_async_device_register, dev,
545                                   &nd_async_domain);
546 }
547
548 void nd_device_register(struct device *dev)
549 {
550         device_initialize(dev);
551         __nd_device_register(dev);
552 }
553 EXPORT_SYMBOL(nd_device_register);
554
555 void nd_device_unregister(struct device *dev, enum nd_async_mode mode)
556 {
557         bool killed;
558
559         switch (mode) {
560         case ND_ASYNC:
561                 /*
562                  * In the async case this is being triggered with the
563                  * device lock held and the unregistration work needs to
564                  * be moved out of line iff this is thread has won the
565                  * race to schedule the deletion.
566                  */
567                 if (!kill_device(dev))
568                         return;
569
570                 get_device(dev);
571                 async_schedule_domain(nd_async_device_unregister, dev,
572                                 &nd_async_domain);
573                 break;
574         case ND_SYNC:
575                 /*
576                  * In the sync case the device is being unregistered due
577                  * to a state change of the parent. Claim the kill state
578                  * to synchronize against other unregistration requests,
579                  * or otherwise let the async path handle it if the
580                  * unregistration was already queued.
581                  */
582                 nd_device_lock(dev);
583                 killed = kill_device(dev);
584                 nd_device_unlock(dev);
585
586                 if (!killed)
587                         return;
588
589                 nd_synchronize();
590                 device_unregister(dev);
591                 break;
592         }
593 }
594 EXPORT_SYMBOL(nd_device_unregister);
595
596 /**
597  * __nd_driver_register() - register a region or a namespace driver
598  * @nd_drv: driver to register
599  * @owner: automatically set by nd_driver_register() macro
600  * @mod_name: automatically set by nd_driver_register() macro
601  */
602 int __nd_driver_register(struct nd_device_driver *nd_drv, struct module *owner,
603                 const char *mod_name)
604 {
605         struct device_driver *drv = &nd_drv->drv;
606
607         if (!nd_drv->type) {
608                 pr_debug("driver type bitmask not set (%ps)\n",
609                                 __builtin_return_address(0));
610                 return -EINVAL;
611         }
612
613         if (!nd_drv->probe) {
614                 pr_debug("%s ->probe() must be specified\n", mod_name);
615                 return -EINVAL;
616         }
617
618         drv->bus = &nvdimm_bus_type;
619         drv->owner = owner;
620         drv->mod_name = mod_name;
621
622         return driver_register(drv);
623 }
624 EXPORT_SYMBOL(__nd_driver_register);
625
626 int nvdimm_revalidate_disk(struct gendisk *disk)
627 {
628         struct device *dev = disk_to_dev(disk)->parent;
629         struct nd_region *nd_region = to_nd_region(dev->parent);
630         int disk_ro = get_disk_ro(disk);
631
632         /*
633          * Upgrade to read-only if the region is read-only preserve as
634          * read-only if the disk is already read-only.
635          */
636         if (disk_ro || nd_region->ro == disk_ro)
637                 return 0;
638
639         dev_info(dev, "%s read-only, marking %s read-only\n",
640                         dev_name(&nd_region->dev), disk->disk_name);
641         set_disk_ro(disk, 1);
642
643         return 0;
644
645 }
646 EXPORT_SYMBOL(nvdimm_revalidate_disk);
647
648 static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
649                 char *buf)
650 {
651         return sprintf(buf, ND_DEVICE_MODALIAS_FMT "\n",
652                         to_nd_device_type(dev));
653 }
654 static DEVICE_ATTR_RO(modalias);
655
656 static ssize_t devtype_show(struct device *dev, struct device_attribute *attr,
657                 char *buf)
658 {
659         return sprintf(buf, "%s\n", dev->type->name);
660 }
661 static DEVICE_ATTR_RO(devtype);
662
663 static struct attribute *nd_device_attributes[] = {
664         &dev_attr_modalias.attr,
665         &dev_attr_devtype.attr,
666         NULL,
667 };
668
669 /*
670  * nd_device_attribute_group - generic attributes for all devices on an nd bus
671  */
672 struct attribute_group nd_device_attribute_group = {
673         .attrs = nd_device_attributes,
674 };
675 EXPORT_SYMBOL_GPL(nd_device_attribute_group);
676
677 static ssize_t numa_node_show(struct device *dev,
678                 struct device_attribute *attr, char *buf)
679 {
680         return sprintf(buf, "%d\n", dev_to_node(dev));
681 }
682 static DEVICE_ATTR_RO(numa_node);
683
684 static struct attribute *nd_numa_attributes[] = {
685         &dev_attr_numa_node.attr,
686         NULL,
687 };
688
689 static umode_t nd_numa_attr_visible(struct kobject *kobj, struct attribute *a,
690                 int n)
691 {
692         if (!IS_ENABLED(CONFIG_NUMA))
693                 return 0;
694
695         return a->mode;
696 }
697
698 /*
699  * nd_numa_attribute_group - NUMA attributes for all devices on an nd bus
700  */
701 struct attribute_group nd_numa_attribute_group = {
702         .attrs = nd_numa_attributes,
703         .is_visible = nd_numa_attr_visible,
704 };
705 EXPORT_SYMBOL_GPL(nd_numa_attribute_group);
706
707 int nvdimm_bus_create_ndctl(struct nvdimm_bus *nvdimm_bus)
708 {
709         dev_t devt = MKDEV(nvdimm_bus_major, nvdimm_bus->id);
710         struct device *dev;
711
712         dev = device_create(nd_class, &nvdimm_bus->dev, devt, nvdimm_bus,
713                         "ndctl%d", nvdimm_bus->id);
714
715         if (IS_ERR(dev))
716                 dev_dbg(&nvdimm_bus->dev, "failed to register ndctl%d: %ld\n",
717                                 nvdimm_bus->id, PTR_ERR(dev));
718         return PTR_ERR_OR_ZERO(dev);
719 }
720
721 void nvdimm_bus_destroy_ndctl(struct nvdimm_bus *nvdimm_bus)
722 {
723         device_destroy(nd_class, MKDEV(nvdimm_bus_major, nvdimm_bus->id));
724 }
725
726 static const struct nd_cmd_desc __nd_cmd_dimm_descs[] = {
727         [ND_CMD_IMPLEMENTED] = { },
728         [ND_CMD_SMART] = {
729                 .out_num = 2,
730                 .out_sizes = { 4, 128, },
731         },
732         [ND_CMD_SMART_THRESHOLD] = {
733                 .out_num = 2,
734                 .out_sizes = { 4, 8, },
735         },
736         [ND_CMD_DIMM_FLAGS] = {
737                 .out_num = 2,
738                 .out_sizes = { 4, 4 },
739         },
740         [ND_CMD_GET_CONFIG_SIZE] = {
741                 .out_num = 3,
742                 .out_sizes = { 4, 4, 4, },
743         },
744         [ND_CMD_GET_CONFIG_DATA] = {
745                 .in_num = 2,
746                 .in_sizes = { 4, 4, },
747                 .out_num = 2,
748                 .out_sizes = { 4, UINT_MAX, },
749         },
750         [ND_CMD_SET_CONFIG_DATA] = {
751                 .in_num = 3,
752                 .in_sizes = { 4, 4, UINT_MAX, },
753                 .out_num = 1,
754                 .out_sizes = { 4, },
755         },
756         [ND_CMD_VENDOR] = {
757                 .in_num = 3,
758                 .in_sizes = { 4, 4, UINT_MAX, },
759                 .out_num = 3,
760                 .out_sizes = { 4, 4, UINT_MAX, },
761         },
762         [ND_CMD_CALL] = {
763                 .in_num = 2,
764                 .in_sizes = { sizeof(struct nd_cmd_pkg), UINT_MAX, },
765                 .out_num = 1,
766                 .out_sizes = { UINT_MAX, },
767         },
768 };
769
770 const struct nd_cmd_desc *nd_cmd_dimm_desc(int cmd)
771 {
772         if (cmd < ARRAY_SIZE(__nd_cmd_dimm_descs))
773                 return &__nd_cmd_dimm_descs[cmd];
774         return NULL;
775 }
776 EXPORT_SYMBOL_GPL(nd_cmd_dimm_desc);
777
778 static const struct nd_cmd_desc __nd_cmd_bus_descs[] = {
779         [ND_CMD_IMPLEMENTED] = { },
780         [ND_CMD_ARS_CAP] = {
781                 .in_num = 2,
782                 .in_sizes = { 8, 8, },
783                 .out_num = 4,
784                 .out_sizes = { 4, 4, 4, 4, },
785         },
786         [ND_CMD_ARS_START] = {
787                 .in_num = 5,
788                 .in_sizes = { 8, 8, 2, 1, 5, },
789                 .out_num = 2,
790                 .out_sizes = { 4, 4, },
791         },
792         [ND_CMD_ARS_STATUS] = {
793                 .out_num = 3,
794                 .out_sizes = { 4, 4, UINT_MAX, },
795         },
796         [ND_CMD_CLEAR_ERROR] = {
797                 .in_num = 2,
798                 .in_sizes = { 8, 8, },
799                 .out_num = 3,
800                 .out_sizes = { 4, 4, 8, },
801         },
802         [ND_CMD_CALL] = {
803                 .in_num = 2,
804                 .in_sizes = { sizeof(struct nd_cmd_pkg), UINT_MAX, },
805                 .out_num = 1,
806                 .out_sizes = { UINT_MAX, },
807         },
808 };
809
810 const struct nd_cmd_desc *nd_cmd_bus_desc(int cmd)
811 {
812         if (cmd < ARRAY_SIZE(__nd_cmd_bus_descs))
813                 return &__nd_cmd_bus_descs[cmd];
814         return NULL;
815 }
816 EXPORT_SYMBOL_GPL(nd_cmd_bus_desc);
817
818 u32 nd_cmd_in_size(struct nvdimm *nvdimm, int cmd,
819                 const struct nd_cmd_desc *desc, int idx, void *buf)
820 {
821         if (idx >= desc->in_num)
822                 return UINT_MAX;
823
824         if (desc->in_sizes[idx] < UINT_MAX)
825                 return desc->in_sizes[idx];
826
827         if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA && idx == 2) {
828                 struct nd_cmd_set_config_hdr *hdr = buf;
829
830                 return hdr->in_length;
831         } else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2) {
832                 struct nd_cmd_vendor_hdr *hdr = buf;
833
834                 return hdr->in_length;
835         } else if (cmd == ND_CMD_CALL) {
836                 struct nd_cmd_pkg *pkg = buf;
837
838                 return pkg->nd_size_in;
839         }
840
841         return UINT_MAX;
842 }
843 EXPORT_SYMBOL_GPL(nd_cmd_in_size);
844
845 u32 nd_cmd_out_size(struct nvdimm *nvdimm, int cmd,
846                 const struct nd_cmd_desc *desc, int idx, const u32 *in_field,
847                 const u32 *out_field, unsigned long remainder)
848 {
849         if (idx >= desc->out_num)
850                 return UINT_MAX;
851
852         if (desc->out_sizes[idx] < UINT_MAX)
853                 return desc->out_sizes[idx];
854
855         if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA && idx == 1)
856                 return in_field[1];
857         else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2)
858                 return out_field[1];
859         else if (!nvdimm && cmd == ND_CMD_ARS_STATUS && idx == 2) {
860                 /*
861                  * Per table 9-276 ARS Data in ACPI 6.1, out_field[1] is
862                  * "Size of Output Buffer in bytes, including this
863                  * field."
864                  */
865                 if (out_field[1] < 4)
866                         return 0;
867                 /*
868                  * ACPI 6.1 is ambiguous if 'status' is included in the
869                  * output size. If we encounter an output size that
870                  * overshoots the remainder by 4 bytes, assume it was
871                  * including 'status'.
872                  */
873                 if (out_field[1] - 4 == remainder)
874                         return remainder;
875                 return out_field[1] - 8;
876         } else if (cmd == ND_CMD_CALL) {
877                 struct nd_cmd_pkg *pkg = (struct nd_cmd_pkg *) in_field;
878
879                 return pkg->nd_size_out;
880         }
881
882
883         return UINT_MAX;
884 }
885 EXPORT_SYMBOL_GPL(nd_cmd_out_size);
886
887 void wait_nvdimm_bus_probe_idle(struct device *dev)
888 {
889         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
890
891         do {
892                 if (nvdimm_bus->probe_active == 0)
893                         break;
894                 nvdimm_bus_unlock(dev);
895                 nd_device_unlock(dev);
896                 wait_event(nvdimm_bus->wait,
897                                 nvdimm_bus->probe_active == 0);
898                 nd_device_lock(dev);
899                 nvdimm_bus_lock(dev);
900         } while (true);
901 }
902
903 static int nd_pmem_forget_poison_check(struct device *dev, void *data)
904 {
905         struct nd_cmd_clear_error *clear_err =
906                 (struct nd_cmd_clear_error *)data;
907         struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL;
908         struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL;
909         struct nd_dax *nd_dax = is_nd_dax(dev) ? to_nd_dax(dev) : NULL;
910         struct nd_namespace_common *ndns = NULL;
911         struct nd_namespace_io *nsio;
912         resource_size_t offset = 0, end_trunc = 0, start, end, pstart, pend;
913
914         if (nd_dax || !dev->driver)
915                 return 0;
916
917         start = clear_err->address;
918         end = clear_err->address + clear_err->cleared - 1;
919
920         if (nd_btt || nd_pfn || nd_dax) {
921                 if (nd_btt)
922                         ndns = nd_btt->ndns;
923                 else if (nd_pfn)
924                         ndns = nd_pfn->ndns;
925                 else if (nd_dax)
926                         ndns = nd_dax->nd_pfn.ndns;
927
928                 if (!ndns)
929                         return 0;
930         } else
931                 ndns = to_ndns(dev);
932
933         nsio = to_nd_namespace_io(&ndns->dev);
934         pstart = nsio->res.start + offset;
935         pend = nsio->res.end - end_trunc;
936
937         if ((pstart >= start) && (pend <= end))
938                 return -EBUSY;
939
940         return 0;
941
942 }
943
944 static int nd_ns_forget_poison_check(struct device *dev, void *data)
945 {
946         return device_for_each_child(dev, data, nd_pmem_forget_poison_check);
947 }
948
949 /* set_config requires an idle interleave set */
950 static int nd_cmd_clear_to_send(struct nvdimm_bus *nvdimm_bus,
951                 struct nvdimm *nvdimm, unsigned int cmd, void *data)
952 {
953         struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
954
955         /* ask the bus provider if it would like to block this request */
956         if (nd_desc->clear_to_send) {
957                 int rc = nd_desc->clear_to_send(nd_desc, nvdimm, cmd, data);
958
959                 if (rc)
960                         return rc;
961         }
962
963         /* require clear error to go through the pmem driver */
964         if (!nvdimm && cmd == ND_CMD_CLEAR_ERROR)
965                 return device_for_each_child(&nvdimm_bus->dev, data,
966                                 nd_ns_forget_poison_check);
967
968         if (!nvdimm || cmd != ND_CMD_SET_CONFIG_DATA)
969                 return 0;
970
971         /* prevent label manipulation while the kernel owns label updates */
972         wait_nvdimm_bus_probe_idle(&nvdimm_bus->dev);
973         if (atomic_read(&nvdimm->busy))
974                 return -EBUSY;
975         return 0;
976 }
977
978 static int __nd_ioctl(struct nvdimm_bus *nvdimm_bus, struct nvdimm *nvdimm,
979                 int read_only, unsigned int ioctl_cmd, unsigned long arg)
980 {
981         struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
982         const struct nd_cmd_desc *desc = NULL;
983         unsigned int cmd = _IOC_NR(ioctl_cmd);
984         struct device *dev = &nvdimm_bus->dev;
985         void __user *p = (void __user *) arg;
986         char *out_env = NULL, *in_env = NULL;
987         const char *cmd_name, *dimm_name;
988         u32 in_len = 0, out_len = 0;
989         unsigned int func = cmd;
990         unsigned long cmd_mask;
991         struct nd_cmd_pkg pkg;
992         int rc, i, cmd_rc;
993         void *buf = NULL;
994         u64 buf_len = 0;
995
996         if (nvdimm) {
997                 desc = nd_cmd_dimm_desc(cmd);
998                 cmd_name = nvdimm_cmd_name(cmd);
999                 cmd_mask = nvdimm->cmd_mask;
1000                 dimm_name = dev_name(&nvdimm->dev);
1001         } else {
1002                 desc = nd_cmd_bus_desc(cmd);
1003                 cmd_name = nvdimm_bus_cmd_name(cmd);
1004                 cmd_mask = nd_desc->cmd_mask;
1005                 dimm_name = "bus";
1006         }
1007
1008         if (cmd == ND_CMD_CALL) {
1009                 if (copy_from_user(&pkg, p, sizeof(pkg)))
1010                         return -EFAULT;
1011         }
1012
1013         if (!desc ||
1014             (desc->out_num + desc->in_num == 0) ||
1015             cmd > ND_CMD_CALL ||
1016             !test_bit(cmd, &cmd_mask))
1017                 return -ENOTTY;
1018
1019         /* fail write commands (when read-only) */
1020         if (read_only)
1021                 switch (cmd) {
1022                 case ND_CMD_VENDOR:
1023                 case ND_CMD_SET_CONFIG_DATA:
1024                 case ND_CMD_ARS_START:
1025                 case ND_CMD_CLEAR_ERROR:
1026                 case ND_CMD_CALL:
1027                         dev_dbg(dev, "'%s' command while read-only.\n",
1028                                         nvdimm ? nvdimm_cmd_name(cmd)
1029                                         : nvdimm_bus_cmd_name(cmd));
1030                         return -EPERM;
1031                 default:
1032                         break;
1033                 }
1034
1035         /* process an input envelope */
1036         in_env = kzalloc(ND_CMD_MAX_ENVELOPE, GFP_KERNEL);
1037         if (!in_env)
1038                 return -ENOMEM;
1039         for (i = 0; i < desc->in_num; i++) {
1040                 u32 in_size, copy;
1041
1042                 in_size = nd_cmd_in_size(nvdimm, cmd, desc, i, in_env);
1043                 if (in_size == UINT_MAX) {
1044                         dev_err(dev, "%s:%s unknown input size cmd: %s field: %d\n",
1045                                         __func__, dimm_name, cmd_name, i);
1046                         rc = -ENXIO;
1047                         goto out;
1048                 }
1049                 if (in_len < ND_CMD_MAX_ENVELOPE)
1050                         copy = min_t(u32, ND_CMD_MAX_ENVELOPE - in_len, in_size);
1051                 else
1052                         copy = 0;
1053                 if (copy && copy_from_user(&in_env[in_len], p + in_len, copy)) {
1054                         rc = -EFAULT;
1055                         goto out;
1056                 }
1057                 in_len += in_size;
1058         }
1059
1060         if (cmd == ND_CMD_CALL) {
1061                 func = pkg.nd_command;
1062                 dev_dbg(dev, "%s, idx: %llu, in: %u, out: %u, len %llu\n",
1063                                 dimm_name, pkg.nd_command,
1064                                 in_len, out_len, buf_len);
1065         }
1066
1067         /* process an output envelope */
1068         out_env = kzalloc(ND_CMD_MAX_ENVELOPE, GFP_KERNEL);
1069         if (!out_env) {
1070                 rc = -ENOMEM;
1071                 goto out;
1072         }
1073
1074         for (i = 0; i < desc->out_num; i++) {
1075                 u32 out_size = nd_cmd_out_size(nvdimm, cmd, desc, i,
1076                                 (u32 *) in_env, (u32 *) out_env, 0);
1077                 u32 copy;
1078
1079                 if (out_size == UINT_MAX) {
1080                         dev_dbg(dev, "%s unknown output size cmd: %s field: %d\n",
1081                                         dimm_name, cmd_name, i);
1082                         rc = -EFAULT;
1083                         goto out;
1084                 }
1085                 if (out_len < ND_CMD_MAX_ENVELOPE)
1086                         copy = min_t(u32, ND_CMD_MAX_ENVELOPE - out_len, out_size);
1087                 else
1088                         copy = 0;
1089                 if (copy && copy_from_user(&out_env[out_len],
1090                                         p + in_len + out_len, copy)) {
1091                         rc = -EFAULT;
1092                         goto out;
1093                 }
1094                 out_len += out_size;
1095         }
1096
1097         buf_len = (u64) out_len + (u64) in_len;
1098         if (buf_len > ND_IOCTL_MAX_BUFLEN) {
1099                 dev_dbg(dev, "%s cmd: %s buf_len: %llu > %d\n", dimm_name,
1100                                 cmd_name, buf_len, ND_IOCTL_MAX_BUFLEN);
1101                 rc = -EINVAL;
1102                 goto out;
1103         }
1104
1105         buf = vmalloc(buf_len);
1106         if (!buf) {
1107                 rc = -ENOMEM;
1108                 goto out;
1109         }
1110
1111         if (copy_from_user(buf, p, buf_len)) {
1112                 rc = -EFAULT;
1113                 goto out;
1114         }
1115
1116         nd_device_lock(dev);
1117         nvdimm_bus_lock(dev);
1118         rc = nd_cmd_clear_to_send(nvdimm_bus, nvdimm, func, buf);
1119         if (rc)
1120                 goto out_unlock;
1121
1122         rc = nd_desc->ndctl(nd_desc, nvdimm, cmd, buf, buf_len, &cmd_rc);
1123         if (rc < 0)
1124                 goto out_unlock;
1125
1126         if (!nvdimm && cmd == ND_CMD_CLEAR_ERROR && cmd_rc >= 0) {
1127                 struct nd_cmd_clear_error *clear_err = buf;
1128
1129                 nvdimm_account_cleared_poison(nvdimm_bus, clear_err->address,
1130                                 clear_err->cleared);
1131         }
1132
1133         if (copy_to_user(p, buf, buf_len))
1134                 rc = -EFAULT;
1135
1136 out_unlock:
1137         nvdimm_bus_unlock(dev);
1138         nd_device_unlock(dev);
1139 out:
1140         kfree(in_env);
1141         kfree(out_env);
1142         vfree(buf);
1143         return rc;
1144 }
1145
1146 enum nd_ioctl_mode {
1147         BUS_IOCTL,
1148         DIMM_IOCTL,
1149 };
1150
1151 static int match_dimm(struct device *dev, void *data)
1152 {
1153         long id = (long) data;
1154
1155         if (is_nvdimm(dev)) {
1156                 struct nvdimm *nvdimm = to_nvdimm(dev);
1157
1158                 return nvdimm->id == id;
1159         }
1160
1161         return 0;
1162 }
1163
1164 static long nd_ioctl(struct file *file, unsigned int cmd, unsigned long arg,
1165                 enum nd_ioctl_mode mode)
1166
1167 {
1168         struct nvdimm_bus *nvdimm_bus, *found = NULL;
1169         long id = (long) file->private_data;
1170         struct nvdimm *nvdimm = NULL;
1171         int rc, ro;
1172
1173         ro = ((file->f_flags & O_ACCMODE) == O_RDONLY);
1174         mutex_lock(&nvdimm_bus_list_mutex);
1175         list_for_each_entry(nvdimm_bus, &nvdimm_bus_list, list) {
1176                 if (mode == DIMM_IOCTL) {
1177                         struct device *dev;
1178
1179                         dev = device_find_child(&nvdimm_bus->dev,
1180                                         file->private_data, match_dimm);
1181                         if (!dev)
1182                                 continue;
1183                         nvdimm = to_nvdimm(dev);
1184                         found = nvdimm_bus;
1185                 } else if (nvdimm_bus->id == id) {
1186                         found = nvdimm_bus;
1187                 }
1188
1189                 if (found) {
1190                         atomic_inc(&nvdimm_bus->ioctl_active);
1191                         break;
1192                 }
1193         }
1194         mutex_unlock(&nvdimm_bus_list_mutex);
1195
1196         if (!found)
1197                 return -ENXIO;
1198
1199         nvdimm_bus = found;
1200         rc = __nd_ioctl(nvdimm_bus, nvdimm, ro, cmd, arg);
1201
1202         if (nvdimm)
1203                 put_device(&nvdimm->dev);
1204         if (atomic_dec_and_test(&nvdimm_bus->ioctl_active))
1205                 wake_up(&nvdimm_bus->wait);
1206
1207         return rc;
1208 }
1209
1210 static long bus_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1211 {
1212         return nd_ioctl(file, cmd, arg, BUS_IOCTL);
1213 }
1214
1215 static long dimm_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1216 {
1217         return nd_ioctl(file, cmd, arg, DIMM_IOCTL);
1218 }
1219
1220 static int nd_open(struct inode *inode, struct file *file)
1221 {
1222         long minor = iminor(inode);
1223
1224         file->private_data = (void *) minor;
1225         return 0;
1226 }
1227
1228 static const struct file_operations nvdimm_bus_fops = {
1229         .owner = THIS_MODULE,
1230         .open = nd_open,
1231         .unlocked_ioctl = bus_ioctl,
1232         .compat_ioctl = bus_ioctl,
1233         .llseek = noop_llseek,
1234 };
1235
1236 static const struct file_operations nvdimm_fops = {
1237         .owner = THIS_MODULE,
1238         .open = nd_open,
1239         .unlocked_ioctl = dimm_ioctl,
1240         .compat_ioctl = dimm_ioctl,
1241         .llseek = noop_llseek,
1242 };
1243
1244 int __init nvdimm_bus_init(void)
1245 {
1246         int rc;
1247
1248         rc = bus_register(&nvdimm_bus_type);
1249         if (rc)
1250                 return rc;
1251
1252         rc = register_chrdev(0, "ndctl", &nvdimm_bus_fops);
1253         if (rc < 0)
1254                 goto err_bus_chrdev;
1255         nvdimm_bus_major = rc;
1256
1257         rc = register_chrdev(0, "dimmctl", &nvdimm_fops);
1258         if (rc < 0)
1259                 goto err_dimm_chrdev;
1260         nvdimm_major = rc;
1261
1262         nd_class = class_create(THIS_MODULE, "nd");
1263         if (IS_ERR(nd_class)) {
1264                 rc = PTR_ERR(nd_class);
1265                 goto err_class;
1266         }
1267
1268         rc = driver_register(&nd_bus_driver.drv);
1269         if (rc)
1270                 goto err_nd_bus;
1271
1272         return 0;
1273
1274  err_nd_bus:
1275         class_destroy(nd_class);
1276  err_class:
1277         unregister_chrdev(nvdimm_major, "dimmctl");
1278  err_dimm_chrdev:
1279         unregister_chrdev(nvdimm_bus_major, "ndctl");
1280  err_bus_chrdev:
1281         bus_unregister(&nvdimm_bus_type);
1282
1283         return rc;
1284 }
1285
1286 void nvdimm_bus_exit(void)
1287 {
1288         driver_unregister(&nd_bus_driver.drv);
1289         class_destroy(nd_class);
1290         unregister_chrdev(nvdimm_bus_major, "ndctl");
1291         unregister_chrdev(nvdimm_major, "dimmctl");
1292         bus_unregister(&nvdimm_bus_type);
1293         ida_destroy(&nd_ida);
1294 }