Loading...
Note: File does not exist in v3.1.
1/*
2 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 */
13#include <linux/libnvdimm.h>
14#include <linux/badblocks.h>
15#include <linux/export.h>
16#include <linux/module.h>
17#include <linux/blkdev.h>
18#include <linux/device.h>
19#include <linux/ctype.h>
20#include <linux/ndctl.h>
21#include <linux/mutex.h>
22#include <linux/slab.h>
23#include <linux/io.h>
24#include "nd-core.h"
25#include "nd.h"
26
27LIST_HEAD(nvdimm_bus_list);
28DEFINE_MUTEX(nvdimm_bus_list_mutex);
29
30void nvdimm_bus_lock(struct device *dev)
31{
32 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
33
34 if (!nvdimm_bus)
35 return;
36 mutex_lock(&nvdimm_bus->reconfig_mutex);
37}
38EXPORT_SYMBOL(nvdimm_bus_lock);
39
40void nvdimm_bus_unlock(struct device *dev)
41{
42 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
43
44 if (!nvdimm_bus)
45 return;
46 mutex_unlock(&nvdimm_bus->reconfig_mutex);
47}
48EXPORT_SYMBOL(nvdimm_bus_unlock);
49
50bool is_nvdimm_bus_locked(struct device *dev)
51{
52 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
53
54 if (!nvdimm_bus)
55 return false;
56 return mutex_is_locked(&nvdimm_bus->reconfig_mutex);
57}
58EXPORT_SYMBOL(is_nvdimm_bus_locked);
59
60struct nvdimm_map {
61 struct nvdimm_bus *nvdimm_bus;
62 struct list_head list;
63 resource_size_t offset;
64 unsigned long flags;
65 size_t size;
66 union {
67 void *mem;
68 void __iomem *iomem;
69 };
70 struct kref kref;
71};
72
73static struct nvdimm_map *find_nvdimm_map(struct device *dev,
74 resource_size_t offset)
75{
76 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
77 struct nvdimm_map *nvdimm_map;
78
79 list_for_each_entry(nvdimm_map, &nvdimm_bus->mapping_list, list)
80 if (nvdimm_map->offset == offset)
81 return nvdimm_map;
82 return NULL;
83}
84
85static struct nvdimm_map *alloc_nvdimm_map(struct device *dev,
86 resource_size_t offset, size_t size, unsigned long flags)
87{
88 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
89 struct nvdimm_map *nvdimm_map;
90
91 nvdimm_map = kzalloc(sizeof(*nvdimm_map), GFP_KERNEL);
92 if (!nvdimm_map)
93 return NULL;
94
95 INIT_LIST_HEAD(&nvdimm_map->list);
96 nvdimm_map->nvdimm_bus = nvdimm_bus;
97 nvdimm_map->offset = offset;
98 nvdimm_map->flags = flags;
99 nvdimm_map->size = size;
100 kref_init(&nvdimm_map->kref);
101
102 if (!request_mem_region(offset, size, dev_name(&nvdimm_bus->dev))) {
103 dev_err(&nvdimm_bus->dev, "failed to request %pa + %zd for %s\n",
104 &offset, size, dev_name(dev));
105 goto err_request_region;
106 }
107
108 if (flags)
109 nvdimm_map->mem = memremap(offset, size, flags);
110 else
111 nvdimm_map->iomem = ioremap(offset, size);
112
113 if (!nvdimm_map->mem)
114 goto err_map;
115
116 dev_WARN_ONCE(dev, !is_nvdimm_bus_locked(dev), "%s: bus unlocked!",
117 __func__);
118 list_add(&nvdimm_map->list, &nvdimm_bus->mapping_list);
119
120 return nvdimm_map;
121
122 err_map:
123 release_mem_region(offset, size);
124 err_request_region:
125 kfree(nvdimm_map);
126 return NULL;
127}
128
129static void nvdimm_map_release(struct kref *kref)
130{
131 struct nvdimm_bus *nvdimm_bus;
132 struct nvdimm_map *nvdimm_map;
133
134 nvdimm_map = container_of(kref, struct nvdimm_map, kref);
135 nvdimm_bus = nvdimm_map->nvdimm_bus;
136
137 dev_dbg(&nvdimm_bus->dev, "%s: %pa\n", __func__, &nvdimm_map->offset);
138 list_del(&nvdimm_map->list);
139 if (nvdimm_map->flags)
140 memunmap(nvdimm_map->mem);
141 else
142 iounmap(nvdimm_map->iomem);
143 release_mem_region(nvdimm_map->offset, nvdimm_map->size);
144 kfree(nvdimm_map);
145}
146
147static void nvdimm_map_put(void *data)
148{
149 struct nvdimm_map *nvdimm_map = data;
150 struct nvdimm_bus *nvdimm_bus = nvdimm_map->nvdimm_bus;
151
152 nvdimm_bus_lock(&nvdimm_bus->dev);
153 kref_put(&nvdimm_map->kref, nvdimm_map_release);
154 nvdimm_bus_unlock(&nvdimm_bus->dev);
155}
156
157/**
158 * devm_nvdimm_memremap - map a resource that is shared across regions
159 * @dev: device that will own a reference to the shared mapping
160 * @offset: physical base address of the mapping
161 * @size: mapping size
162 * @flags: memremap flags, or, if zero, perform an ioremap instead
163 */
164void *devm_nvdimm_memremap(struct device *dev, resource_size_t offset,
165 size_t size, unsigned long flags)
166{
167 struct nvdimm_map *nvdimm_map;
168
169 nvdimm_bus_lock(dev);
170 nvdimm_map = find_nvdimm_map(dev, offset);
171 if (!nvdimm_map)
172 nvdimm_map = alloc_nvdimm_map(dev, offset, size, flags);
173 else
174 kref_get(&nvdimm_map->kref);
175 nvdimm_bus_unlock(dev);
176
177 if (!nvdimm_map)
178 return NULL;
179
180 if (devm_add_action_or_reset(dev, nvdimm_map_put, nvdimm_map))
181 return NULL;
182
183 return nvdimm_map->mem;
184}
185EXPORT_SYMBOL_GPL(devm_nvdimm_memremap);
186
187u64 nd_fletcher64(void *addr, size_t len, bool le)
188{
189 u32 *buf = addr;
190 u32 lo32 = 0;
191 u64 hi32 = 0;
192 int i;
193
194 for (i = 0; i < len / sizeof(u32); i++) {
195 lo32 += le ? le32_to_cpu((__le32) buf[i]) : buf[i];
196 hi32 += lo32;
197 }
198
199 return hi32 << 32 | lo32;
200}
201EXPORT_SYMBOL_GPL(nd_fletcher64);
202
203struct nvdimm_bus_descriptor *to_nd_desc(struct nvdimm_bus *nvdimm_bus)
204{
205 /* struct nvdimm_bus definition is private to libnvdimm */
206 return nvdimm_bus->nd_desc;
207}
208EXPORT_SYMBOL_GPL(to_nd_desc);
209
210struct device *to_nvdimm_bus_dev(struct nvdimm_bus *nvdimm_bus)
211{
212 /* struct nvdimm_bus definition is private to libnvdimm */
213 return &nvdimm_bus->dev;
214}
215EXPORT_SYMBOL_GPL(to_nvdimm_bus_dev);
216
217static bool is_uuid_sep(char sep)
218{
219 if (sep == '\n' || sep == '-' || sep == ':' || sep == '\0')
220 return true;
221 return false;
222}
223
224static int nd_uuid_parse(struct device *dev, u8 *uuid_out, const char *buf,
225 size_t len)
226{
227 const char *str = buf;
228 u8 uuid[16];
229 int i;
230
231 for (i = 0; i < 16; i++) {
232 if (!isxdigit(str[0]) || !isxdigit(str[1])) {
233 dev_dbg(dev, "%s: pos: %d buf[%zd]: %c buf[%zd]: %c\n",
234 __func__, i, str - buf, str[0],
235 str + 1 - buf, str[1]);
236 return -EINVAL;
237 }
238
239 uuid[i] = (hex_to_bin(str[0]) << 4) | hex_to_bin(str[1]);
240 str += 2;
241 if (is_uuid_sep(*str))
242 str++;
243 }
244
245 memcpy(uuid_out, uuid, sizeof(uuid));
246 return 0;
247}
248
249/**
250 * nd_uuid_store: common implementation for writing 'uuid' sysfs attributes
251 * @dev: container device for the uuid property
252 * @uuid_out: uuid buffer to replace
253 * @buf: raw sysfs buffer to parse
254 *
255 * Enforce that uuids can only be changed while the device is disabled
256 * (driver detached)
257 * LOCKING: expects device_lock() is held on entry
258 */
259int nd_uuid_store(struct device *dev, u8 **uuid_out, const char *buf,
260 size_t len)
261{
262 u8 uuid[16];
263 int rc;
264
265 if (dev->driver)
266 return -EBUSY;
267
268 rc = nd_uuid_parse(dev, uuid, buf, len);
269 if (rc)
270 return rc;
271
272 kfree(*uuid_out);
273 *uuid_out = kmemdup(uuid, sizeof(uuid), GFP_KERNEL);
274 if (!(*uuid_out))
275 return -ENOMEM;
276
277 return 0;
278}
279
280ssize_t nd_sector_size_show(unsigned long current_lbasize,
281 const unsigned long *supported, char *buf)
282{
283 ssize_t len = 0;
284 int i;
285
286 for (i = 0; supported[i]; i++)
287 if (current_lbasize == supported[i])
288 len += sprintf(buf + len, "[%ld] ", supported[i]);
289 else
290 len += sprintf(buf + len, "%ld ", supported[i]);
291 len += sprintf(buf + len, "\n");
292 return len;
293}
294
295ssize_t nd_sector_size_store(struct device *dev, const char *buf,
296 unsigned long *current_lbasize, const unsigned long *supported)
297{
298 unsigned long lbasize;
299 int rc, i;
300
301 if (dev->driver)
302 return -EBUSY;
303
304 rc = kstrtoul(buf, 0, &lbasize);
305 if (rc)
306 return rc;
307
308 for (i = 0; supported[i]; i++)
309 if (lbasize == supported[i])
310 break;
311
312 if (supported[i]) {
313 *current_lbasize = lbasize;
314 return 0;
315 } else {
316 return -EINVAL;
317 }
318}
319
320static ssize_t commands_show(struct device *dev,
321 struct device_attribute *attr, char *buf)
322{
323 int cmd, len = 0;
324 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
325 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
326
327 for_each_set_bit(cmd, &nd_desc->cmd_mask, BITS_PER_LONG)
328 len += sprintf(buf + len, "%s ", nvdimm_bus_cmd_name(cmd));
329 len += sprintf(buf + len, "\n");
330 return len;
331}
332static DEVICE_ATTR_RO(commands);
333
334static const char *nvdimm_bus_provider(struct nvdimm_bus *nvdimm_bus)
335{
336 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
337 struct device *parent = nvdimm_bus->dev.parent;
338
339 if (nd_desc->provider_name)
340 return nd_desc->provider_name;
341 else if (parent)
342 return dev_name(parent);
343 else
344 return "unknown";
345}
346
347static ssize_t provider_show(struct device *dev,
348 struct device_attribute *attr, char *buf)
349{
350 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
351
352 return sprintf(buf, "%s\n", nvdimm_bus_provider(nvdimm_bus));
353}
354static DEVICE_ATTR_RO(provider);
355
356static int flush_namespaces(struct device *dev, void *data)
357{
358 device_lock(dev);
359 device_unlock(dev);
360 return 0;
361}
362
363static int flush_regions_dimms(struct device *dev, void *data)
364{
365 device_lock(dev);
366 device_unlock(dev);
367 device_for_each_child(dev, NULL, flush_namespaces);
368 return 0;
369}
370
371static ssize_t wait_probe_show(struct device *dev,
372 struct device_attribute *attr, char *buf)
373{
374 struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
375 struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
376 int rc;
377
378 if (nd_desc->flush_probe) {
379 rc = nd_desc->flush_probe(nd_desc);
380 if (rc)
381 return rc;
382 }
383 nd_synchronize();
384 device_for_each_child(dev, NULL, flush_regions_dimms);
385 return sprintf(buf, "1\n");
386}
387static DEVICE_ATTR_RO(wait_probe);
388
389static struct attribute *nvdimm_bus_attributes[] = {
390 &dev_attr_commands.attr,
391 &dev_attr_wait_probe.attr,
392 &dev_attr_provider.attr,
393 NULL,
394};
395
396struct attribute_group nvdimm_bus_attribute_group = {
397 .attrs = nvdimm_bus_attributes,
398};
399EXPORT_SYMBOL_GPL(nvdimm_bus_attribute_group);
400
401static void set_badblock(struct badblocks *bb, sector_t s, int num)
402{
403 dev_dbg(bb->dev, "Found a poison range (0x%llx, 0x%llx)\n",
404 (u64) s * 512, (u64) num * 512);
405 /* this isn't an error as the hardware will still throw an exception */
406 if (badblocks_set(bb, s, num, 1))
407 dev_info_once(bb->dev, "%s: failed for sector %llx\n",
408 __func__, (u64) s);
409}
410
411/**
412 * __add_badblock_range() - Convert a physical address range to bad sectors
413 * @bb: badblocks instance to populate
414 * @ns_offset: namespace offset where the error range begins (in bytes)
415 * @len: number of bytes of poison to be added
416 *
417 * This assumes that the range provided with (ns_offset, len) is within
418 * the bounds of physical addresses for this namespace, i.e. lies in the
419 * interval [ns_start, ns_start + ns_size)
420 */
421static void __add_badblock_range(struct badblocks *bb, u64 ns_offset, u64 len)
422{
423 const unsigned int sector_size = 512;
424 sector_t start_sector;
425 u64 num_sectors;
426 u32 rem;
427
428 start_sector = div_u64(ns_offset, sector_size);
429 num_sectors = div_u64_rem(len, sector_size, &rem);
430 if (rem)
431 num_sectors++;
432
433 if (unlikely(num_sectors > (u64)INT_MAX)) {
434 u64 remaining = num_sectors;
435 sector_t s = start_sector;
436
437 while (remaining) {
438 int done = min_t(u64, remaining, INT_MAX);
439
440 set_badblock(bb, s, done);
441 remaining -= done;
442 s += done;
443 }
444 } else
445 set_badblock(bb, start_sector, num_sectors);
446}
447
448static void badblocks_populate(struct list_head *poison_list,
449 struct badblocks *bb, const struct resource *res)
450{
451 struct nd_poison *pl;
452
453 if (list_empty(poison_list))
454 return;
455
456 list_for_each_entry(pl, poison_list, list) {
457 u64 pl_end = pl->start + pl->length - 1;
458
459 /* Discard intervals with no intersection */
460 if (pl_end < res->start)
461 continue;
462 if (pl->start > res->end)
463 continue;
464 /* Deal with any overlap after start of the namespace */
465 if (pl->start >= res->start) {
466 u64 start = pl->start;
467 u64 len;
468
469 if (pl_end <= res->end)
470 len = pl->length;
471 else
472 len = res->start + resource_size(res)
473 - pl->start;
474 __add_badblock_range(bb, start - res->start, len);
475 continue;
476 }
477 /* Deal with overlap for poison starting before the namespace */
478 if (pl->start < res->start) {
479 u64 len;
480
481 if (pl_end < res->end)
482 len = pl->start + pl->length - res->start;
483 else
484 len = resource_size(res);
485 __add_badblock_range(bb, 0, len);
486 }
487 }
488}
489
490/**
491 * nvdimm_badblocks_populate() - Convert a list of poison ranges to badblocks
492 * @region: parent region of the range to interrogate
493 * @bb: badblocks instance to populate
494 * @res: resource range to consider
495 *
496 * The poison list generated during bus initialization may contain
497 * multiple, possibly overlapping physical address ranges. Compare each
498 * of these ranges to the resource range currently being initialized,
499 * and add badblocks entries for all matching sub-ranges
500 */
501void nvdimm_badblocks_populate(struct nd_region *nd_region,
502 struct badblocks *bb, const struct resource *res)
503{
504 struct nvdimm_bus *nvdimm_bus;
505 struct list_head *poison_list;
506
507 if (!is_nd_pmem(&nd_region->dev)) {
508 dev_WARN_ONCE(&nd_region->dev, 1,
509 "%s only valid for pmem regions\n", __func__);
510 return;
511 }
512 nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
513 poison_list = &nvdimm_bus->poison_list;
514
515 nvdimm_bus_lock(&nvdimm_bus->dev);
516 badblocks_populate(poison_list, bb, res);
517 nvdimm_bus_unlock(&nvdimm_bus->dev);
518}
519EXPORT_SYMBOL_GPL(nvdimm_badblocks_populate);
520
521static int add_poison(struct nvdimm_bus *nvdimm_bus, u64 addr, u64 length,
522 gfp_t flags)
523{
524 struct nd_poison *pl;
525
526 pl = kzalloc(sizeof(*pl), flags);
527 if (!pl)
528 return -ENOMEM;
529
530 pl->start = addr;
531 pl->length = length;
532 list_add_tail(&pl->list, &nvdimm_bus->poison_list);
533
534 return 0;
535}
536
537static int bus_add_poison(struct nvdimm_bus *nvdimm_bus, u64 addr, u64 length)
538{
539 struct nd_poison *pl;
540
541 if (list_empty(&nvdimm_bus->poison_list))
542 return add_poison(nvdimm_bus, addr, length, GFP_KERNEL);
543
544 /*
545 * There is a chance this is a duplicate, check for those first.
546 * This will be the common case as ARS_STATUS returns all known
547 * errors in the SPA space, and we can't query it per region
548 */
549 list_for_each_entry(pl, &nvdimm_bus->poison_list, list)
550 if (pl->start == addr) {
551 /* If length has changed, update this list entry */
552 if (pl->length != length)
553 pl->length = length;
554 return 0;
555 }
556
557 /*
558 * If not a duplicate or a simple length update, add the entry as is,
559 * as any overlapping ranges will get resolved when the list is consumed
560 * and converted to badblocks
561 */
562 return add_poison(nvdimm_bus, addr, length, GFP_KERNEL);
563}
564
565int nvdimm_bus_add_poison(struct nvdimm_bus *nvdimm_bus, u64 addr, u64 length)
566{
567 int rc;
568
569 nvdimm_bus_lock(&nvdimm_bus->dev);
570 rc = bus_add_poison(nvdimm_bus, addr, length);
571 nvdimm_bus_unlock(&nvdimm_bus->dev);
572
573 return rc;
574}
575EXPORT_SYMBOL_GPL(nvdimm_bus_add_poison);
576
577void nvdimm_clear_from_poison_list(struct nvdimm_bus *nvdimm_bus,
578 phys_addr_t start, unsigned int len)
579{
580 struct list_head *poison_list = &nvdimm_bus->poison_list;
581 u64 clr_end = start + len - 1;
582 struct nd_poison *pl, *next;
583
584 nvdimm_bus_lock(&nvdimm_bus->dev);
585 WARN_ON_ONCE(list_empty(poison_list));
586
587 /*
588 * [start, clr_end] is the poison interval being cleared.
589 * [pl->start, pl_end] is the poison_list entry we're comparing
590 * the above interval against. The poison list entry may need
591 * to be modified (update either start or length), deleted, or
592 * split into two based on the overlap characteristics
593 */
594
595 list_for_each_entry_safe(pl, next, poison_list, list) {
596 u64 pl_end = pl->start + pl->length - 1;
597
598 /* Skip intervals with no intersection */
599 if (pl_end < start)
600 continue;
601 if (pl->start > clr_end)
602 continue;
603 /* Delete completely overlapped poison entries */
604 if ((pl->start >= start) && (pl_end <= clr_end)) {
605 list_del(&pl->list);
606 kfree(pl);
607 continue;
608 }
609 /* Adjust start point of partially cleared entries */
610 if ((start <= pl->start) && (clr_end > pl->start)) {
611 pl->length -= clr_end - pl->start + 1;
612 pl->start = clr_end + 1;
613 continue;
614 }
615 /* Adjust pl->length for partial clearing at the tail end */
616 if ((pl->start < start) && (pl_end <= clr_end)) {
617 /* pl->start remains the same */
618 pl->length = start - pl->start;
619 continue;
620 }
621 /*
622 * If clearing in the middle of an entry, we split it into
623 * two by modifying the current entry to represent one half of
624 * the split, and adding a new entry for the second half.
625 */
626 if ((pl->start < start) && (pl_end > clr_end)) {
627 u64 new_start = clr_end + 1;
628 u64 new_len = pl_end - new_start + 1;
629
630 /* Add new entry covering the right half */
631 add_poison(nvdimm_bus, new_start, new_len, GFP_NOIO);
632 /* Adjust this entry to cover the left half */
633 pl->length = start - pl->start;
634 continue;
635 }
636 }
637 nvdimm_bus_unlock(&nvdimm_bus->dev);
638}
639EXPORT_SYMBOL_GPL(nvdimm_clear_from_poison_list);
640
641#ifdef CONFIG_BLK_DEV_INTEGRITY
642int nd_integrity_init(struct gendisk *disk, unsigned long meta_size)
643{
644 struct blk_integrity bi;
645
646 if (meta_size == 0)
647 return 0;
648
649 memset(&bi, 0, sizeof(bi));
650
651 bi.tuple_size = meta_size;
652 bi.tag_size = meta_size;
653
654 blk_integrity_register(disk, &bi);
655 blk_queue_max_integrity_segments(disk->queue, 1);
656
657 return 0;
658}
659EXPORT_SYMBOL(nd_integrity_init);
660
661#else /* CONFIG_BLK_DEV_INTEGRITY */
662int nd_integrity_init(struct gendisk *disk, unsigned long meta_size)
663{
664 return 0;
665}
666EXPORT_SYMBOL(nd_integrity_init);
667
668#endif
669
670static __init int libnvdimm_init(void)
671{
672 int rc;
673
674 rc = nvdimm_bus_init();
675 if (rc)
676 return rc;
677 rc = nvdimm_init();
678 if (rc)
679 goto err_dimm;
680 rc = nd_region_init();
681 if (rc)
682 goto err_region;
683 return 0;
684 err_region:
685 nvdimm_exit();
686 err_dimm:
687 nvdimm_bus_exit();
688 return rc;
689}
690
691static __exit void libnvdimm_exit(void)
692{
693 WARN_ON(!list_empty(&nvdimm_bus_list));
694 nd_region_exit();
695 nvdimm_exit();
696 nvdimm_bus_exit();
697 nd_region_devs_exit();
698 nvdimm_devs_exit();
699}
700
701MODULE_LICENSE("GPL v2");
702MODULE_AUTHOR("Intel Corporation");
703subsys_initcall(libnvdimm_init);
704module_exit(libnvdimm_exit);