1
2
3
4
5
6
7
8
9
10
11
12
13#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
14#include <linux/libnvdimm.h>
15#include <linux/sched/mm.h>
16#include <linux/vmalloc.h>
17#include <linux/uaccess.h>
18#include <linux/module.h>
19#include <linux/blkdev.h>
20#include <linux/fcntl.h>
21#include <linux/async.h>
22#include <linux/genhd.h>
23#include <linux/ndctl.h>
24#include <linux/sched.h>
25#include <linux/slab.h>
26#include <linux/cpu.h>
27#include <linux/fs.h>
28#include <linux/io.h>
29#include <linux/mm.h>
30#include <linux/nd.h>
31#include "nd-core.h"
32#include "nd.h"
33#include "pfn.h"
34
35int nvdimm_major;
36static int nvdimm_bus_major;
37static struct class *nd_class;
38static DEFINE_IDA(nd_ida);
39
40static int to_nd_device_type(struct device *dev)
41{
42 if (is_nvdimm(dev))
43 return ND_DEVICE_DIMM;
44 else if (is_memory(dev))
45 return ND_DEVICE_REGION_PMEM;
46 else if (is_nd_blk(dev))
47 return ND_DEVICE_REGION_BLK;
48 else if (is_nd_dax(dev))
49 return ND_DEVICE_DAX_PMEM;
50 else if (is_nd_region(dev->parent))
51 return nd_region_to_nstype(to_nd_region(dev->parent));
52
53 return 0;
54}
55
56static int nvdimm_bus_uevent(struct device *dev, struct kobj_uevent_env *env)
57{
58 return add_uevent_var(env, "MODALIAS=" ND_DEVICE_MODALIAS_FMT,
59 to_nd_device_type(dev));
60}
61
62static struct module *to_bus_provider(struct device *dev)
63{
64
65 if (is_nd_region(dev)) {
66 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
67
68 return nvdimm_bus->nd_desc->module;
69 }
70 return NULL;
71}
72
73static void nvdimm_bus_probe_start(struct nvdimm_bus *nvdimm_bus)
74{
75 nvdimm_bus_lock(&nvdimm_bus->dev);
76 nvdimm_bus->probe_active++;
77 nvdimm_bus_unlock(&nvdimm_bus->dev);
78}
79
80static void nvdimm_bus_probe_end(struct nvdimm_bus *nvdimm_bus)
81{
82 nvdimm_bus_lock(&nvdimm_bus->dev);
83 if (--nvdimm_bus->probe_active == 0)
84 wake_up(&nvdimm_bus->wait);
85 nvdimm_bus_unlock(&nvdimm_bus->dev);
86}
87
88static int nvdimm_bus_probe(struct device *dev)
89{
90 struct nd_device_driver *nd_drv = to_nd_device_driver(dev->driver);
91 struct module *provider = to_bus_provider(dev);
92 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
93 int rc;
94
95 if (!try_module_get(provider))
96 return -ENXIO;
97
98 dev_dbg(&nvdimm_bus->dev, "START: %s.probe(%s)\n",
99 dev->driver->name, dev_name(dev));
100
101 nvdimm_bus_probe_start(nvdimm_bus);
102 rc = nd_drv->probe(dev);
103 if ((rc == 0 || rc == -EOPNOTSUPP) &&
104 dev->parent && is_nd_region(dev->parent))
105 nd_region_advance_seeds(to_nd_region(dev->parent), dev);
106 nvdimm_bus_probe_end(nvdimm_bus);
107
108 dev_dbg(&nvdimm_bus->dev, "END: %s.probe(%s) = %d\n", dev->driver->name,
109 dev_name(dev), rc);
110
111 if (rc != 0)
112 module_put(provider);
113 return rc;
114}
115
116static int nvdimm_bus_remove(struct device *dev)
117{
118 struct nd_device_driver *nd_drv = to_nd_device_driver(dev->driver);
119 struct module *provider = to_bus_provider(dev);
120 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
121 int rc = 0;
122
123 if (nd_drv->remove)
124 rc = nd_drv->remove(dev);
125
126 dev_dbg(&nvdimm_bus->dev, "%s.remove(%s) = %d\n", dev->driver->name,
127 dev_name(dev), rc);
128 module_put(provider);
129 return rc;
130}
131
132static void nvdimm_bus_shutdown(struct device *dev)
133{
134 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
135 struct nd_device_driver *nd_drv = NULL;
136
137 if (dev->driver)
138 nd_drv = to_nd_device_driver(dev->driver);
139
140 if (nd_drv && nd_drv->shutdown) {
141 nd_drv->shutdown(dev);
142 dev_dbg(&nvdimm_bus->dev, "%s.shutdown(%s)\n",
143 dev->driver->name, dev_name(dev));
144 }
145}
146
147void nd_device_notify(struct device *dev, enum nvdimm_event event)
148{
149 device_lock(dev);
150 if (dev->driver) {
151 struct nd_device_driver *nd_drv;
152
153 nd_drv = to_nd_device_driver(dev->driver);
154 if (nd_drv->notify)
155 nd_drv->notify(dev, event);
156 }
157 device_unlock(dev);
158}
159EXPORT_SYMBOL(nd_device_notify);
160
161void nvdimm_region_notify(struct nd_region *nd_region, enum nvdimm_event event)
162{
163 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
164
165 if (!nvdimm_bus)
166 return;
167
168
169 nd_device_notify(&nd_region->dev, event);
170}
171EXPORT_SYMBOL_GPL(nvdimm_region_notify);
172
173struct clear_badblocks_context {
174 resource_size_t phys, cleared;
175};
176
177static int nvdimm_clear_badblocks_region(struct device *dev, void *data)
178{
179 struct clear_badblocks_context *ctx = data;
180 struct nd_region *nd_region;
181 resource_size_t ndr_end;
182 sector_t sector;
183
184
185 if (!is_nd_pmem(dev))
186 return 0;
187
188 nd_region = to_nd_region(dev);
189 ndr_end = nd_region->ndr_start + nd_region->ndr_size - 1;
190
191
192 if (ctx->phys < nd_region->ndr_start
193 || (ctx->phys + ctx->cleared) > ndr_end)
194 return 0;
195
196 sector = (ctx->phys - nd_region->ndr_start) / 512;
197 badblocks_clear(&nd_region->bb, sector, ctx->cleared / 512);
198
199 if (nd_region->bb_state)
200 sysfs_notify_dirent(nd_region->bb_state);
201
202 return 0;
203}
204
205static void nvdimm_clear_badblocks_regions(struct nvdimm_bus *nvdimm_bus,
206 phys_addr_t phys, u64 cleared)
207{
208 struct clear_badblocks_context ctx = {
209 .phys = phys,
210 .cleared = cleared,
211 };
212
213 device_for_each_child(&nvdimm_bus->dev, &ctx,
214 nvdimm_clear_badblocks_region);
215}
216
217static void nvdimm_account_cleared_poison(struct nvdimm_bus *nvdimm_bus,
218 phys_addr_t phys, u64 cleared)
219{
220 if (cleared > 0)
221 badrange_forget(&nvdimm_bus->badrange, phys, cleared);
222
223 if (cleared > 0 && cleared / 512)
224 nvdimm_clear_badblocks_regions(nvdimm_bus, phys, cleared);
225}
226
227long nvdimm_clear_poison(struct device *dev, phys_addr_t phys,
228 unsigned int len)
229{
230 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
231 struct nvdimm_bus_descriptor *nd_desc;
232 struct nd_cmd_clear_error clear_err;
233 struct nd_cmd_ars_cap ars_cap;
234 u32 clear_err_unit, mask;
235 unsigned int noio_flag;
236 int cmd_rc, rc;
237
238 if (!nvdimm_bus)
239 return -ENXIO;
240
241 nd_desc = nvdimm_bus->nd_desc;
242
243
244
245
246 if (!nd_desc->ndctl)
247 return len;
248
249 memset(&ars_cap, 0, sizeof(ars_cap));
250 ars_cap.address = phys;
251 ars_cap.length = len;
252 noio_flag = memalloc_noio_save();
253 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_ARS_CAP, &ars_cap,
254 sizeof(ars_cap), &cmd_rc);
255 memalloc_noio_restore(noio_flag);
256 if (rc < 0)
257 return rc;
258 if (cmd_rc < 0)
259 return cmd_rc;
260 clear_err_unit = ars_cap.clear_err_unit;
261 if (!clear_err_unit || !is_power_of_2(clear_err_unit))
262 return -ENXIO;
263
264 mask = clear_err_unit - 1;
265 if ((phys | len) & mask)
266 return -ENXIO;
267 memset(&clear_err, 0, sizeof(clear_err));
268 clear_err.address = phys;
269 clear_err.length = len;
270 noio_flag = memalloc_noio_save();
271 rc = nd_desc->ndctl(nd_desc, NULL, ND_CMD_CLEAR_ERROR, &clear_err,
272 sizeof(clear_err), &cmd_rc);
273 memalloc_noio_restore(noio_flag);
274 if (rc < 0)
275 return rc;
276 if (cmd_rc < 0)
277 return cmd_rc;
278
279 nvdimm_account_cleared_poison(nvdimm_bus, phys, clear_err.cleared);
280
281 return clear_err.cleared;
282}
283EXPORT_SYMBOL_GPL(nvdimm_clear_poison);
284
285static int nvdimm_bus_match(struct device *dev, struct device_driver *drv);
286
287static struct bus_type nvdimm_bus_type = {
288 .name = "nd",
289 .uevent = nvdimm_bus_uevent,
290 .match = nvdimm_bus_match,
291 .probe = nvdimm_bus_probe,
292 .remove = nvdimm_bus_remove,
293 .shutdown = nvdimm_bus_shutdown,
294};
295
296static void nvdimm_bus_release(struct device *dev)
297{
298 struct nvdimm_bus *nvdimm_bus;
299
300 nvdimm_bus = container_of(dev, struct nvdimm_bus, dev);
301 ida_simple_remove(&nd_ida, nvdimm_bus->id);
302 kfree(nvdimm_bus);
303}
304
305static bool is_nvdimm_bus(struct device *dev)
306{
307 return dev->release == nvdimm_bus_release;
308}
309
310struct nvdimm_bus *walk_to_nvdimm_bus(struct device *nd_dev)
311{
312 struct device *dev;
313
314 for (dev = nd_dev; dev; dev = dev->parent)
315 if (is_nvdimm_bus(dev))
316 break;
317 dev_WARN_ONCE(nd_dev, !dev, "invalid dev, not on nd bus\n");
318 if (dev)
319 return to_nvdimm_bus(dev);
320 return NULL;
321}
322
323struct nvdimm_bus *to_nvdimm_bus(struct device *dev)
324{
325 struct nvdimm_bus *nvdimm_bus;
326
327 nvdimm_bus = container_of(dev, struct nvdimm_bus, dev);
328 WARN_ON(!is_nvdimm_bus(dev));
329 return nvdimm_bus;
330}
331EXPORT_SYMBOL_GPL(to_nvdimm_bus);
332
333struct nvdimm_bus *nvdimm_to_bus(struct nvdimm *nvdimm)
334{
335 return to_nvdimm_bus(nvdimm->dev.parent);
336}
337EXPORT_SYMBOL_GPL(nvdimm_to_bus);
338
339struct nvdimm_bus *nvdimm_bus_register(struct device *parent,
340 struct nvdimm_bus_descriptor *nd_desc)
341{
342 struct nvdimm_bus *nvdimm_bus;
343 int rc;
344
345 nvdimm_bus = kzalloc(sizeof(*nvdimm_bus), GFP_KERNEL);
346 if (!nvdimm_bus)
347 return NULL;
348 INIT_LIST_HEAD(&nvdimm_bus->list);
349 INIT_LIST_HEAD(&nvdimm_bus->mapping_list);
350 init_waitqueue_head(&nvdimm_bus->wait);
351 nvdimm_bus->id = ida_simple_get(&nd_ida, 0, 0, GFP_KERNEL);
352 if (nvdimm_bus->id < 0) {
353 kfree(nvdimm_bus);
354 return NULL;
355 }
356 mutex_init(&nvdimm_bus->reconfig_mutex);
357 badrange_init(&nvdimm_bus->badrange);
358 nvdimm_bus->nd_desc = nd_desc;
359 nvdimm_bus->dev.parent = parent;
360 nvdimm_bus->dev.release = nvdimm_bus_release;
361 nvdimm_bus->dev.groups = nd_desc->attr_groups;
362 nvdimm_bus->dev.bus = &nvdimm_bus_type;
363 nvdimm_bus->dev.of_node = nd_desc->of_node;
364 dev_set_name(&nvdimm_bus->dev, "ndbus%d", nvdimm_bus->id);
365 rc = device_register(&nvdimm_bus->dev);
366 if (rc) {
367 dev_dbg(&nvdimm_bus->dev, "registration failed: %d\n", rc);
368 goto err;
369 }
370
371 return nvdimm_bus;
372 err:
373 put_device(&nvdimm_bus->dev);
374 return NULL;
375}
376EXPORT_SYMBOL_GPL(nvdimm_bus_register);
377
378void nvdimm_bus_unregister(struct nvdimm_bus *nvdimm_bus)
379{
380 if (!nvdimm_bus)
381 return;
382 device_unregister(&nvdimm_bus->dev);
383}
384EXPORT_SYMBOL_GPL(nvdimm_bus_unregister);
385
386static int child_unregister(struct device *dev, void *data)
387{
388
389
390
391
392
393
394 if (dev->class)
395 return 0;
396
397 if (is_nvdimm(dev)) {
398 struct nvdimm *nvdimm = to_nvdimm(dev);
399 bool dev_put = false;
400
401
402 nvdimm_bus_lock(dev);
403 set_bit(NVDIMM_SECURITY_FROZEN, &nvdimm->sec.flags);
404 if (test_and_clear_bit(NDD_WORK_PENDING, &nvdimm->flags))
405 dev_put = true;
406 nvdimm_bus_unlock(dev);
407 cancel_delayed_work_sync(&nvdimm->dwork);
408 if (dev_put)
409 put_device(dev);
410 }
411 nd_device_unregister(dev, ND_SYNC);
412
413 return 0;
414}
415
416static void free_badrange_list(struct list_head *badrange_list)
417{
418 struct badrange_entry *bre, *next;
419
420 list_for_each_entry_safe(bre, next, badrange_list, list) {
421 list_del(&bre->list);
422 kfree(bre);
423 }
424 list_del_init(badrange_list);
425}
426
427static int nd_bus_remove(struct device *dev)
428{
429 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
430
431 mutex_lock(&nvdimm_bus_list_mutex);
432 list_del_init(&nvdimm_bus->list);
433 mutex_unlock(&nvdimm_bus_list_mutex);
434
435 wait_event(nvdimm_bus->wait,
436 atomic_read(&nvdimm_bus->ioctl_active) == 0);
437
438 nd_synchronize();
439 device_for_each_child(&nvdimm_bus->dev, NULL, child_unregister);
440
441 spin_lock(&nvdimm_bus->badrange.lock);
442 free_badrange_list(&nvdimm_bus->badrange.list);
443 spin_unlock(&nvdimm_bus->badrange.lock);
444
445 nvdimm_bus_destroy_ndctl(nvdimm_bus);
446
447 return 0;
448}
449
450static int nd_bus_probe(struct device *dev)
451{
452 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
453 int rc;
454
455 rc = nvdimm_bus_create_ndctl(nvdimm_bus);
456 if (rc)
457 return rc;
458
459 mutex_lock(&nvdimm_bus_list_mutex);
460 list_add_tail(&nvdimm_bus->list, &nvdimm_bus_list);
461 mutex_unlock(&nvdimm_bus_list_mutex);
462
463
464 dev_set_drvdata(dev, nvdimm_bus->nd_desc);
465
466 return 0;
467}
468
469static struct nd_device_driver nd_bus_driver = {
470 .probe = nd_bus_probe,
471 .remove = nd_bus_remove,
472 .drv = {
473 .name = "nd_bus",
474 .suppress_bind_attrs = true,
475 .bus = &nvdimm_bus_type,
476 .owner = THIS_MODULE,
477 .mod_name = KBUILD_MODNAME,
478 },
479};
480
481static int nvdimm_bus_match(struct device *dev, struct device_driver *drv)
482{
483 struct nd_device_driver *nd_drv = to_nd_device_driver(drv);
484
485 if (is_nvdimm_bus(dev) && nd_drv == &nd_bus_driver)
486 return true;
487
488 return !!test_bit(to_nd_device_type(dev), &nd_drv->type);
489}
490
491static ASYNC_DOMAIN_EXCLUSIVE(nd_async_domain);
492
493void nd_synchronize(void)
494{
495 async_synchronize_full_domain(&nd_async_domain);
496}
497EXPORT_SYMBOL_GPL(nd_synchronize);
498
499static void nd_async_device_register(void *d, async_cookie_t cookie)
500{
501 struct device *dev = d;
502
503 if (device_add(dev) != 0) {
504 dev_err(dev, "%s: failed\n", __func__);
505 put_device(dev);
506 }
507 put_device(dev);
508 if (dev->parent)
509 put_device(dev->parent);
510}
511
512static void nd_async_device_unregister(void *d, async_cookie_t cookie)
513{
514 struct device *dev = d;
515
516
517 nvdimm_bus_lock(dev);
518 nvdimm_bus_unlock(dev);
519
520 device_unregister(dev);
521 put_device(dev);
522}
523
524void __nd_device_register(struct device *dev)
525{
526 if (!dev)
527 return;
528
529
530
531
532
533
534
535 if (is_nd_region(dev))
536 set_dev_node(dev, to_nd_region(dev)->numa_node);
537
538 dev->bus = &nvdimm_bus_type;
539 if (dev->parent) {
540 get_device(dev->parent);
541 if (dev_to_node(dev) == NUMA_NO_NODE)
542 set_dev_node(dev, dev_to_node(dev->parent));
543 }
544 get_device(dev);
545
546 async_schedule_dev_domain(nd_async_device_register, dev,
547 &nd_async_domain);
548}
549
550void nd_device_register(struct device *dev)
551{
552 device_initialize(dev);
553 __nd_device_register(dev);
554}
555EXPORT_SYMBOL(nd_device_register);
556
557void nd_device_unregister(struct device *dev, enum nd_async_mode mode)
558{
559 bool killed;
560
561 switch (mode) {
562 case ND_ASYNC:
563
564
565
566
567
568
569 if (!kill_device(dev))
570 return;
571
572 get_device(dev);
573 async_schedule_domain(nd_async_device_unregister, dev,
574 &nd_async_domain);
575 break;
576 case ND_SYNC:
577
578
579
580
581
582
583
584 device_lock(dev);
585 killed = kill_device(dev);
586 device_unlock(dev);
587
588 if (!killed)
589 return;
590
591 nd_synchronize();
592 device_unregister(dev);
593 break;
594 }
595}
596EXPORT_SYMBOL(nd_device_unregister);
597
598
599
600
601
602
603
604int __nd_driver_register(struct nd_device_driver *nd_drv, struct module *owner,
605 const char *mod_name)
606{
607 struct device_driver *drv = &nd_drv->drv;
608
609 if (!nd_drv->type) {
610 pr_debug("driver type bitmask not set (%pf)\n",
611 __builtin_return_address(0));
612 return -EINVAL;
613 }
614
615 if (!nd_drv->probe) {
616 pr_debug("%s ->probe() must be specified\n", mod_name);
617 return -EINVAL;
618 }
619
620 drv->bus = &nvdimm_bus_type;
621 drv->owner = owner;
622 drv->mod_name = mod_name;
623
624 return driver_register(drv);
625}
626EXPORT_SYMBOL(__nd_driver_register);
627
628int nvdimm_revalidate_disk(struct gendisk *disk)
629{
630 struct device *dev = disk_to_dev(disk)->parent;
631 struct nd_region *nd_region = to_nd_region(dev->parent);
632 int disk_ro = get_disk_ro(disk);
633
634
635
636
637
638 if (disk_ro || nd_region->ro == disk_ro)
639 return 0;
640
641 dev_info(dev, "%s read-only, marking %s read-only\n",
642 dev_name(&nd_region->dev), disk->disk_name);
643 set_disk_ro(disk, 1);
644
645 return 0;
646
647}
648EXPORT_SYMBOL(nvdimm_revalidate_disk);
649
650static ssize_t modalias_show(struct device *dev, struct device_attribute *attr,
651 char *buf)
652{
653 return sprintf(buf, ND_DEVICE_MODALIAS_FMT "\n",
654 to_nd_device_type(dev));
655}
656static DEVICE_ATTR_RO(modalias);
657
658static ssize_t devtype_show(struct device *dev, struct device_attribute *attr,
659 char *buf)
660{
661 return sprintf(buf, "%s\n", dev->type->name);
662}
663static DEVICE_ATTR_RO(devtype);
664
665static struct attribute *nd_device_attributes[] = {
666 &dev_attr_modalias.attr,
667 &dev_attr_devtype.attr,
668 NULL,
669};
670
671
672
673
674struct attribute_group nd_device_attribute_group = {
675 .attrs = nd_device_attributes,
676};
677EXPORT_SYMBOL_GPL(nd_device_attribute_group);
678
679static ssize_t numa_node_show(struct device *dev,
680 struct device_attribute *attr, char *buf)
681{
682 return sprintf(buf, "%d\n", dev_to_node(dev));
683}
684static DEVICE_ATTR_RO(numa_node);
685
686static struct attribute *nd_numa_attributes[] = {
687 &dev_attr_numa_node.attr,
688 NULL,
689};
690
691static umode_t nd_numa_attr_visible(struct kobject *kobj, struct attribute *a,
692 int n)
693{
694 if (!IS_ENABLED(CONFIG_NUMA))
695 return 0;
696
697 return a->mode;
698}
699
700
701
702
703struct attribute_group nd_numa_attribute_group = {
704 .attrs = nd_numa_attributes,
705 .is_visible = nd_numa_attr_visible,
706};
707EXPORT_SYMBOL_GPL(nd_numa_attribute_group);
708
709int nvdimm_bus_create_ndctl(struct nvdimm_bus *nvdimm_bus)
710{
711 dev_t devt = MKDEV(nvdimm_bus_major, nvdimm_bus->id);
712 struct device *dev;
713
714 dev = device_create(nd_class, &nvdimm_bus->dev, devt, nvdimm_bus,
715 "ndctl%d", nvdimm_bus->id);
716
717 if (IS_ERR(dev))
718 dev_dbg(&nvdimm_bus->dev, "failed to register ndctl%d: %ld\n",
719 nvdimm_bus->id, PTR_ERR(dev));
720 return PTR_ERR_OR_ZERO(dev);
721}
722
723void nvdimm_bus_destroy_ndctl(struct nvdimm_bus *nvdimm_bus)
724{
725 device_destroy(nd_class, MKDEV(nvdimm_bus_major, nvdimm_bus->id));
726}
727
728static const struct nd_cmd_desc __nd_cmd_dimm_descs[] = {
729 [ND_CMD_IMPLEMENTED] = { },
730 [ND_CMD_SMART] = {
731 .out_num = 2,
732 .out_sizes = { 4, 128, },
733 },
734 [ND_CMD_SMART_THRESHOLD] = {
735 .out_num = 2,
736 .out_sizes = { 4, 8, },
737 },
738 [ND_CMD_DIMM_FLAGS] = {
739 .out_num = 2,
740 .out_sizes = { 4, 4 },
741 },
742 [ND_CMD_GET_CONFIG_SIZE] = {
743 .out_num = 3,
744 .out_sizes = { 4, 4, 4, },
745 },
746 [ND_CMD_GET_CONFIG_DATA] = {
747 .in_num = 2,
748 .in_sizes = { 4, 4, },
749 .out_num = 2,
750 .out_sizes = { 4, UINT_MAX, },
751 },
752 [ND_CMD_SET_CONFIG_DATA] = {
753 .in_num = 3,
754 .in_sizes = { 4, 4, UINT_MAX, },
755 .out_num = 1,
756 .out_sizes = { 4, },
757 },
758 [ND_CMD_VENDOR] = {
759 .in_num = 3,
760 .in_sizes = { 4, 4, UINT_MAX, },
761 .out_num = 3,
762 .out_sizes = { 4, 4, UINT_MAX, },
763 },
764 [ND_CMD_CALL] = {
765 .in_num = 2,
766 .in_sizes = { sizeof(struct nd_cmd_pkg), UINT_MAX, },
767 .out_num = 1,
768 .out_sizes = { UINT_MAX, },
769 },
770};
771
772const struct nd_cmd_desc *nd_cmd_dimm_desc(int cmd)
773{
774 if (cmd < ARRAY_SIZE(__nd_cmd_dimm_descs))
775 return &__nd_cmd_dimm_descs[cmd];
776 return NULL;
777}
778EXPORT_SYMBOL_GPL(nd_cmd_dimm_desc);
779
780static const struct nd_cmd_desc __nd_cmd_bus_descs[] = {
781 [ND_CMD_IMPLEMENTED] = { },
782 [ND_CMD_ARS_CAP] = {
783 .in_num = 2,
784 .in_sizes = { 8, 8, },
785 .out_num = 4,
786 .out_sizes = { 4, 4, 4, 4, },
787 },
788 [ND_CMD_ARS_START] = {
789 .in_num = 5,
790 .in_sizes = { 8, 8, 2, 1, 5, },
791 .out_num = 2,
792 .out_sizes = { 4, 4, },
793 },
794 [ND_CMD_ARS_STATUS] = {
795 .out_num = 3,
796 .out_sizes = { 4, 4, UINT_MAX, },
797 },
798 [ND_CMD_CLEAR_ERROR] = {
799 .in_num = 2,
800 .in_sizes = { 8, 8, },
801 .out_num = 3,
802 .out_sizes = { 4, 4, 8, },
803 },
804 [ND_CMD_CALL] = {
805 .in_num = 2,
806 .in_sizes = { sizeof(struct nd_cmd_pkg), UINT_MAX, },
807 .out_num = 1,
808 .out_sizes = { UINT_MAX, },
809 },
810};
811
812const struct nd_cmd_desc *nd_cmd_bus_desc(int cmd)
813{
814 if (cmd < ARRAY_SIZE(__nd_cmd_bus_descs))
815 return &__nd_cmd_bus_descs[cmd];
816 return NULL;
817}
818EXPORT_SYMBOL_GPL(nd_cmd_bus_desc);
819
820u32 nd_cmd_in_size(struct nvdimm *nvdimm, int cmd,
821 const struct nd_cmd_desc *desc, int idx, void *buf)
822{
823 if (idx >= desc->in_num)
824 return UINT_MAX;
825
826 if (desc->in_sizes[idx] < UINT_MAX)
827 return desc->in_sizes[idx];
828
829 if (nvdimm && cmd == ND_CMD_SET_CONFIG_DATA && idx == 2) {
830 struct nd_cmd_set_config_hdr *hdr = buf;
831
832 return hdr->in_length;
833 } else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2) {
834 struct nd_cmd_vendor_hdr *hdr = buf;
835
836 return hdr->in_length;
837 } else if (cmd == ND_CMD_CALL) {
838 struct nd_cmd_pkg *pkg = buf;
839
840 return pkg->nd_size_in;
841 }
842
843 return UINT_MAX;
844}
845EXPORT_SYMBOL_GPL(nd_cmd_in_size);
846
847u32 nd_cmd_out_size(struct nvdimm *nvdimm, int cmd,
848 const struct nd_cmd_desc *desc, int idx, const u32 *in_field,
849 const u32 *out_field, unsigned long remainder)
850{
851 if (idx >= desc->out_num)
852 return UINT_MAX;
853
854 if (desc->out_sizes[idx] < UINT_MAX)
855 return desc->out_sizes[idx];
856
857 if (nvdimm && cmd == ND_CMD_GET_CONFIG_DATA && idx == 1)
858 return in_field[1];
859 else if (nvdimm && cmd == ND_CMD_VENDOR && idx == 2)
860 return out_field[1];
861 else if (!nvdimm && cmd == ND_CMD_ARS_STATUS && idx == 2) {
862
863
864
865
866
867 if (out_field[1] < 4)
868 return 0;
869
870
871
872
873
874
875 if (out_field[1] - 4 == remainder)
876 return remainder;
877 return out_field[1] - 8;
878 } else if (cmd == ND_CMD_CALL) {
879 struct nd_cmd_pkg *pkg = (struct nd_cmd_pkg *) in_field;
880
881 return pkg->nd_size_out;
882 }
883
884
885 return UINT_MAX;
886}
887EXPORT_SYMBOL_GPL(nd_cmd_out_size);
888
889void wait_nvdimm_bus_probe_idle(struct device *dev)
890{
891 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
892
893 do {
894 if (nvdimm_bus->probe_active == 0)
895 break;
896 nvdimm_bus_unlock(dev);
897 device_unlock(dev);
898 wait_event(nvdimm_bus->wait,
899 nvdimm_bus->probe_active == 0);
900 device_lock(dev);
901 nvdimm_bus_lock(dev);
902 } while (true);
903}
904
905static int nd_pmem_forget_poison_check(struct device *dev, void *data)
906{
907 struct nd_cmd_clear_error *clear_err =
908 (struct nd_cmd_clear_error *)data;
909 struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL;
910 struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL;
911 struct nd_dax *nd_dax = is_nd_dax(dev) ? to_nd_dax(dev) : NULL;
912 struct nd_namespace_common *ndns = NULL;
913 struct nd_namespace_io *nsio;
914 resource_size_t offset = 0, end_trunc = 0, start, end, pstart, pend;
915
916 if (nd_dax || !dev->driver)
917 return 0;
918
919 start = clear_err->address;
920 end = clear_err->address + clear_err->cleared - 1;
921
922 if (nd_btt || nd_pfn || nd_dax) {
923 if (nd_btt)
924 ndns = nd_btt->ndns;
925 else if (nd_pfn)
926 ndns = nd_pfn->ndns;
927 else if (nd_dax)
928 ndns = nd_dax->nd_pfn.ndns;
929
930 if (!ndns)
931 return 0;
932 } else
933 ndns = to_ndns(dev);
934
935 nsio = to_nd_namespace_io(&ndns->dev);
936 pstart = nsio->res.start + offset;
937 pend = nsio->res.end - end_trunc;
938
939 if ((pstart >= start) && (pend <= end))
940 return -EBUSY;
941
942 return 0;
943
944}
945
946static int nd_ns_forget_poison_check(struct device *dev, void *data)
947{
948 return device_for_each_child(dev, data, nd_pmem_forget_poison_check);
949}
950
951
952static int nd_cmd_clear_to_send(struct nvdimm_bus *nvdimm_bus,
953 struct nvdimm *nvdimm, unsigned int cmd, void *data)
954{
955 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
956
957
958 if (nd_desc->clear_to_send) {
959 int rc = nd_desc->clear_to_send(nd_desc, nvdimm, cmd, data);
960
961 if (rc)
962 return rc;
963 }
964
965
966 if (!nvdimm && cmd == ND_CMD_CLEAR_ERROR)
967 return device_for_each_child(&nvdimm_bus->dev, data,
968 nd_ns_forget_poison_check);
969
970 if (!nvdimm || cmd != ND_CMD_SET_CONFIG_DATA)
971 return 0;
972
973
974 wait_nvdimm_bus_probe_idle(&nvdimm_bus->dev);
975 if (atomic_read(&nvdimm->busy))
976 return -EBUSY;
977 return 0;
978}
979
980static int __nd_ioctl(struct nvdimm_bus *nvdimm_bus, struct nvdimm *nvdimm,
981 int read_only, unsigned int ioctl_cmd, unsigned long arg)
982{
983 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
984 const struct nd_cmd_desc *desc = NULL;
985 unsigned int cmd = _IOC_NR(ioctl_cmd);
986 struct device *dev = &nvdimm_bus->dev;
987 void __user *p = (void __user *) arg;
988 char *out_env = NULL, *in_env = NULL;
989 const char *cmd_name, *dimm_name;
990 u32 in_len = 0, out_len = 0;
991 unsigned int func = cmd;
992 unsigned long cmd_mask;
993 struct nd_cmd_pkg pkg;
994 int rc, i, cmd_rc;
995 void *buf = NULL;
996 u64 buf_len = 0;
997
998 if (nvdimm) {
999 desc = nd_cmd_dimm_desc(cmd);
1000 cmd_name = nvdimm_cmd_name(cmd);
1001 cmd_mask = nvdimm->cmd_mask;
1002 dimm_name = dev_name(&nvdimm->dev);
1003 } else {
1004 desc = nd_cmd_bus_desc(cmd);
1005 cmd_name = nvdimm_bus_cmd_name(cmd);
1006 cmd_mask = nd_desc->cmd_mask;
1007 dimm_name = "bus";
1008 }
1009
1010 if (cmd == ND_CMD_CALL) {
1011 if (copy_from_user(&pkg, p, sizeof(pkg)))
1012 return -EFAULT;
1013 }
1014
1015 if (!desc || (desc->out_num + desc->in_num == 0) ||
1016 !test_bit(cmd, &cmd_mask))
1017 return -ENOTTY;
1018
1019
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
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
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 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
1136out_unlock:
1137 nvdimm_bus_unlock(dev);
1138 device_unlock(dev);
1139out:
1140 kfree(in_env);
1141 kfree(out_env);
1142 vfree(buf);
1143 return rc;
1144}
1145
1146enum nd_ioctl_mode {
1147 BUS_IOCTL,
1148 DIMM_IOCTL,
1149};
1150
1151static 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
1164static 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
1210static long bus_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1211{
1212 return nd_ioctl(file, cmd, arg, BUS_IOCTL);
1213}
1214
1215static long dimm_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
1216{
1217 return nd_ioctl(file, cmd, arg, DIMM_IOCTL);
1218}
1219
1220static 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
1228static 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
1236static 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
1244int __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
1286void 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}
1295