drivers/nvdimm/core.c

   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
  27 LIST_HEAD(nvdimm_bus_list);
  28 DEFINE_MUTEX(nvdimm_bus_list_mutex);
  29
  30 void 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 }
  38 EXPORT_SYMBOL(nvdimm_bus_lock);
  39
  40 void 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 }
  48 EXPORT_SYMBOL(nvdimm_bus_unlock);
  49
  50 bool 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 }
  58 EXPORT_SYMBOL(is_nvdimm_bus_locked);
  59
  60 struct 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
  73 static 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
  85 static 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                 goto err_request_region;
 104
 105         if (flags)
 106                 nvdimm_map->mem = memremap(offset, size, flags);
 107         else
 108                 nvdimm_map->iomem = ioremap(offset, size);
 109
 110         if (!nvdimm_map->mem)
 111                 goto err_map;
 112
 113         dev_WARN_ONCE(dev, !is_nvdimm_bus_locked(dev), "%s: bus unlocked!",
 114                         __func__);
 115         list_add(&nvdimm_map->list, &nvdimm_bus->mapping_list);
 116
 117         return nvdimm_map;
 118
 119  err_map:
 120         release_mem_region(offset, size);
 121  err_request_region:
 122         kfree(nvdimm_map);
 123         return NULL;
 124 }
 125
 126 static void nvdimm_map_release(struct kref *kref)
 127 {
 128         struct nvdimm_bus *nvdimm_bus;
 129         struct nvdimm_map *nvdimm_map;
 130
 131         nvdimm_map = container_of(kref, struct nvdimm_map, kref);
 132         nvdimm_bus = nvdimm_map->nvdimm_bus;
 133
 134         dev_dbg(&nvdimm_bus->dev, "%s: %pa\n", __func__, &nvdimm_map->offset);
 135         list_del(&nvdimm_map->list);
 136         if (nvdimm_map->flags)
 137                 memunmap(nvdimm_map->mem);
 138         else
 139                 iounmap(nvdimm_map->iomem);
 140         release_mem_region(nvdimm_map->offset, nvdimm_map->size);
 141         kfree(nvdimm_map);
 142 }
 143
 144 static void nvdimm_map_put(void *data)
 145 {
 146         struct nvdimm_map *nvdimm_map = data;
 147         struct nvdimm_bus *nvdimm_bus = nvdimm_map->nvdimm_bus;
 148
 149         nvdimm_bus_lock(&nvdimm_bus->dev);
 150         kref_put(&nvdimm_map->kref, nvdimm_map_release);
 151         nvdimm_bus_unlock(&nvdimm_bus->dev);
 152 }
 153
 154 /**
 155  * devm_nvdimm_memremap - map a resource that is shared across regions
 156  * @dev: device that will own a reference to the shared mapping
 157  * @offset: physical base address of the mapping
 158  * @size: mapping size
 159  * @flags: memremap flags, or, if zero, perform an ioremap instead
 160  */
 161 void *devm_nvdimm_memremap(struct device *dev, resource_size_t offset,
 162                 size_t size, unsigned long flags)
 163 {
 164         struct nvdimm_map *nvdimm_map;
 165
 166         nvdimm_bus_lock(dev);
 167         nvdimm_map = find_nvdimm_map(dev, offset);
 168         if (!nvdimm_map)
 169                 nvdimm_map = alloc_nvdimm_map(dev, offset, size, flags);
 170         else
 171                 kref_get(&nvdimm_map->kref);
 172         nvdimm_bus_unlock(dev);
 173
 174         if (devm_add_action_or_reset(dev, nvdimm_map_put, nvdimm_map))
 175                 return NULL;
 176
 177         return nvdimm_map->mem;
 178 }
 179 EXPORT_SYMBOL_GPL(devm_nvdimm_memremap);
 180
 181 u64 nd_fletcher64(void *addr, size_t len, bool le)
 182 {
 183         u32 *buf = addr;
 184         u32 lo32 = 0;
 185         u64 hi32 = 0;
 186         int i;
 187
 188         for (i = 0; i < len / sizeof(u32); i++) {
 189                 lo32 += le ? le32_to_cpu((__le32) buf[i]) : buf[i];
 190                 hi32 += lo32;
 191         }
 192
 193         return hi32 << 32 | lo32;
 194 }
 195 EXPORT_SYMBOL_GPL(nd_fletcher64);
 196
 197 struct nvdimm_bus_descriptor *to_nd_desc(struct nvdimm_bus *nvdimm_bus)
 198 {
 199         /* struct nvdimm_bus definition is private to libnvdimm */
 200         return nvdimm_bus->nd_desc;
 201 }
 202 EXPORT_SYMBOL_GPL(to_nd_desc);
 203
 204 static bool is_uuid_sep(char sep)
 205 {
 206         if (sep == '\n' || sep == '-' || sep == ':' || sep == '\0')
 207                 return true;
 208         return false;
 209 }
 210
 211 static int nd_uuid_parse(struct device *dev, u8 *uuid_out, const char *buf,
 212                 size_t len)
 213 {
 214         const char *str = buf;
 215         u8 uuid[16];
 216         int i;
 217
 218         for (i = 0; i < 16; i++) {
 219                 if (!isxdigit(str[0]) || !isxdigit(str[1])) {
 220                         dev_dbg(dev, "%s: pos: %d buf[%zd]: %c buf[%zd]: %c\n",
 221                                         __func__, i, str - buf, str[0],
 222                                         str + 1 - buf, str[1]);
 223                         return -EINVAL;
 224                 }
 225
 226                 uuid[i] = (hex_to_bin(str[0]) << 4) | hex_to_bin(str[1]);
 227                 str += 2;
 228                 if (is_uuid_sep(*str))
 229                         str++;
 230         }
 231
 232         memcpy(uuid_out, uuid, sizeof(uuid));
 233         return 0;
 234 }
 235
 236 /**
 237  * nd_uuid_store: common implementation for writing 'uuid' sysfs attributes
 238  * @dev: container device for the uuid property
 239  * @uuid_out: uuid buffer to replace
 240  * @buf: raw sysfs buffer to parse
 241  *
 242  * Enforce that uuids can only be changed while the device is disabled
 243  * (driver detached)
 244  * LOCKING: expects device_lock() is held on entry
 245  */
 246 int nd_uuid_store(struct device *dev, u8 **uuid_out, const char *buf,
 247                 size_t len)
 248 {
 249         u8 uuid[16];
 250         int rc;
 251
 252         if (dev->driver)
 253                 return -EBUSY;
 254
 255         rc = nd_uuid_parse(dev, uuid, buf, len);
 256         if (rc)
 257                 return rc;
 258
 259         kfree(*uuid_out);
 260         *uuid_out = kmemdup(uuid, sizeof(uuid), GFP_KERNEL);
 261         if (!(*uuid_out))
 262                 return -ENOMEM;
 263
 264         return 0;
 265 }
 266
 267 ssize_t nd_sector_size_show(unsigned long current_lbasize,
 268                 const unsigned long *supported, char *buf)
 269 {
 270         ssize_t len = 0;
 271         int i;
 272
 273         for (i = 0; supported[i]; i++)
 274                 if (current_lbasize == supported[i])
 275                         len += sprintf(buf + len, "[%ld] ", supported[i]);
 276                 else
 277                         len += sprintf(buf + len, "%ld ", supported[i]);
 278         len += sprintf(buf + len, "\n");
 279         return len;
 280 }
 281
 282 ssize_t nd_sector_size_store(struct device *dev, const char *buf,
 283                 unsigned long *current_lbasize, const unsigned long *supported)
 284 {
 285         unsigned long lbasize;
 286         int rc, i;
 287
 288         if (dev->driver)
 289                 return -EBUSY;
 290
 291         rc = kstrtoul(buf, 0, &lbasize);
 292         if (rc)
 293                 return rc;
 294
 295         for (i = 0; supported[i]; i++)
 296                 if (lbasize == supported[i])
 297                         break;
 298
 299         if (supported[i]) {
 300                 *current_lbasize = lbasize;
 301                 return 0;
 302         } else {
 303                 return -EINVAL;
 304         }
 305 }
 306
 307 void __nd_iostat_start(struct bio *bio, unsigned long *start)
 308 {
 309         struct gendisk *disk = bio->bi_bdev->bd_disk;
 310         const int rw = bio_data_dir(bio);
 311         int cpu = part_stat_lock();
 312
 313         *start = jiffies;
 314         part_round_stats(cpu, &disk->part0);
 315         part_stat_inc(cpu, &disk->part0, ios[rw]);
 316         part_stat_add(cpu, &disk->part0, sectors[rw], bio_sectors(bio));
 317         part_inc_in_flight(&disk->part0, rw);
 318         part_stat_unlock();
 319 }
 320 EXPORT_SYMBOL(__nd_iostat_start);
 321
 322 void nd_iostat_end(struct bio *bio, unsigned long start)
 323 {
 324         struct gendisk *disk = bio->bi_bdev->bd_disk;
 325         unsigned long duration = jiffies - start;
 326         const int rw = bio_data_dir(bio);
 327         int cpu = part_stat_lock();
 328
 329         part_stat_add(cpu, &disk->part0, ticks[rw], duration);
 330         part_round_stats(cpu, &disk->part0);
 331         part_dec_in_flight(&disk->part0, rw);
 332         part_stat_unlock();
 333 }
 334 EXPORT_SYMBOL(nd_iostat_end);
 335
 336 static ssize_t commands_show(struct device *dev,
 337                 struct device_attribute *attr, char *buf)
 338 {
 339         int cmd, len = 0;
 340         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
 341         struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
 342
 343         for_each_set_bit(cmd, &nd_desc->cmd_mask, BITS_PER_LONG)
 344                 len += sprintf(buf + len, "%s ", nvdimm_bus_cmd_name(cmd));
 345         len += sprintf(buf + len, "\n");
 346         return len;
 347 }
 348 static DEVICE_ATTR_RO(commands);
 349
 350 static const char *nvdimm_bus_provider(struct nvdimm_bus *nvdimm_bus)
 351 {
 352         struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
 353         struct device *parent = nvdimm_bus->dev.parent;
 354
 355         if (nd_desc->provider_name)
 356                 return nd_desc->provider_name;
 357         else if (parent)
 358                 return dev_name(parent);
 359         else
 360                 return "unknown";
 361 }
 362
 363 static ssize_t provider_show(struct device *dev,
 364                 struct device_attribute *attr, char *buf)
 365 {
 366         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
 367
 368         return sprintf(buf, "%s\n", nvdimm_bus_provider(nvdimm_bus));
 369 }
 370 static DEVICE_ATTR_RO(provider);
 371
 372 static int flush_namespaces(struct device *dev, void *data)
 373 {
 374         device_lock(dev);
 375         device_unlock(dev);
 376         return 0;
 377 }
 378
 379 static int flush_regions_dimms(struct device *dev, void *data)
 380 {
 381         device_lock(dev);
 382         device_unlock(dev);
 383         device_for_each_child(dev, NULL, flush_namespaces);
 384         return 0;
 385 }
 386
 387 static ssize_t wait_probe_show(struct device *dev,
 388                 struct device_attribute *attr, char *buf)
 389 {
 390         struct nvdimm_bus *nvdimm_bus = to_nvdimm_bus(dev);
 391         struct nvdimm_bus_descriptor *nd_desc = nvdimm_bus->nd_desc;
 392         int rc;
 393
 394         if (nd_desc->flush_probe) {
 395                 rc = nd_desc->flush_probe(nd_desc);
 396                 if (rc)
 397                         return rc;
 398         }
 399         nd_synchronize();
 400         device_for_each_child(dev, NULL, flush_regions_dimms);
 401         return sprintf(buf, "1\n");
 402 }
 403 static DEVICE_ATTR_RO(wait_probe);
 404
 405 static struct attribute *nvdimm_bus_attributes[] = {
 406         &dev_attr_commands.attr,
 407         &dev_attr_wait_probe.attr,
 408         &dev_attr_provider.attr,
 409         NULL,
 410 };
 411
 412 struct attribute_group nvdimm_bus_attribute_group = {
 413         .attrs = nvdimm_bus_attributes,
 414 };
 415 EXPORT_SYMBOL_GPL(nvdimm_bus_attribute_group);
 416
 417 static void set_badblock(struct badblocks *bb, sector_t s, int num)
 418 {
 419         dev_dbg(bb->dev, "Found a poison range (0x%llx, 0x%llx)\n",
 420                         (u64) s * 512, (u64) num * 512);
 421         /* this isn't an error as the hardware will still throw an exception */
 422         if (badblocks_set(bb, s, num, 1))
 423                 dev_info_once(bb->dev, "%s: failed for sector %llx\n",
 424                                 __func__, (u64) s);
 425 }
 426
 427 /**
 428  * __add_badblock_range() - Convert a physical address range to bad sectors
 429  * @bb:         badblocks instance to populate
 430  * @ns_offset:  namespace offset where the error range begins (in bytes)
 431  * @len:        number of bytes of poison to be added
 432  *
 433  * This assumes that the range provided with (ns_offset, len) is within
 434  * the bounds of physical addresses for this namespace, i.e. lies in the
 435  * interval [ns_start, ns_start + ns_size)
 436  */
 437 static void __add_badblock_range(struct badblocks *bb, u64 ns_offset, u64 len)
 438 {
 439         const unsigned int sector_size = 512;
 440         sector_t start_sector;
 441         u64 num_sectors;
 442         u32 rem;
 443
 444         start_sector = div_u64(ns_offset, sector_size);
 445         num_sectors = div_u64_rem(len, sector_size, &rem);
 446         if (rem)
 447                 num_sectors++;
 448
 449         if (unlikely(num_sectors > (u64)INT_MAX)) {
 450                 u64 remaining = num_sectors;
 451                 sector_t s = start_sector;
 452
 453                 while (remaining) {
 454                         int done = min_t(u64, remaining, INT_MAX);
 455
 456                         set_badblock(bb, s, done);
 457                         remaining -= done;
 458                         s += done;
 459                 }
 460         } else
 461                 set_badblock(bb, start_sector, num_sectors);
 462 }
 463
 464 static void badblocks_populate(struct list_head *poison_list,
 465                 struct badblocks *bb, const struct resource *res)
 466 {
 467         struct nd_poison *pl;
 468
 469         if (list_empty(poison_list))
 470                 return;
 471
 472         list_for_each_entry(pl, poison_list, list) {
 473                 u64 pl_end = pl->start + pl->length - 1;
 474
 475                 /* Discard intervals with no intersection */
 476                 if (pl_end < res->start)
 477                         continue;
 478                 if (pl->start >  res->end)
 479                         continue;
 480                 /* Deal with any overlap after start of the namespace */
 481                 if (pl->start >= res->start) {
 482                         u64 start = pl->start;
 483                         u64 len;
 484
 485                         if (pl_end <= res->end)
 486                                 len = pl->length;
 487                         else
 488                                 len = res->start + resource_size(res)
 489                                         - pl->start;
 490                         __add_badblock_range(bb, start - res->start, len);
 491                         continue;
 492                 }
 493                 /* Deal with overlap for poison starting before the namespace */
 494                 if (pl->start < res->start) {
 495                         u64 len;
 496
 497                         if (pl_end < res->end)
 498                                 len = pl->start + pl->length - res->start;
 499                         else
 500                                 len = resource_size(res);
 501                         __add_badblock_range(bb, 0, len);
 502                 }
 503         }
 504 }
 505
 506 /**
 507  * nvdimm_badblocks_populate() - Convert a list of poison ranges to badblocks
 508  * @region: parent region of the range to interrogate
 509  * @bb: badblocks instance to populate
 510  * @res: resource range to consider
 511  *
 512  * The poison list generated during bus initialization may contain
 513  * multiple, possibly overlapping physical address ranges.  Compare each
 514  * of these ranges to the resource range currently being initialized,
 515  * and add badblocks entries for all matching sub-ranges
 516  */
 517 void nvdimm_badblocks_populate(struct nd_region *nd_region,
 518                 struct badblocks *bb, const struct resource *res)
 519 {
 520         struct nvdimm_bus *nvdimm_bus;
 521         struct list_head *poison_list;
 522
 523         if (!is_nd_pmem(&nd_region->dev)) {
 524                 dev_WARN_ONCE(&nd_region->dev, 1,
 525                                 "%s only valid for pmem regions\n", __func__);
 526                 return;
 527         }
 528         nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
 529         poison_list = &nvdimm_bus->poison_list;
 530
 531         nvdimm_bus_lock(&nvdimm_bus->dev);
 532         badblocks_populate(poison_list, bb, res);
 533         nvdimm_bus_unlock(&nvdimm_bus->dev);
 534 }
 535 EXPORT_SYMBOL_GPL(nvdimm_badblocks_populate);
 536
 537 static int add_poison(struct nvdimm_bus *nvdimm_bus, u64 addr, u64 length)
 538 {
 539         struct nd_poison *pl;
 540
 541         pl = kzalloc(sizeof(*pl), GFP_KERNEL);
 542         if (!pl)
 543                 return -ENOMEM;
 544
 545         pl->start = addr;
 546         pl->length = length;
 547         list_add_tail(&pl->list, &nvdimm_bus->poison_list);
 548
 549         return 0;
 550 }
 551
 552 static int bus_add_poison(struct nvdimm_bus *nvdimm_bus, u64 addr, u64 length)
 553 {
 554         struct nd_poison *pl;
 555
 556         if (list_empty(&nvdimm_bus->poison_list))
 557                 return add_poison(nvdimm_bus, addr, length);
 558
 559         /*
 560          * There is a chance this is a duplicate, check for those first.
 561          * This will be the common case as ARS_STATUS returns all known
 562          * errors in the SPA space, and we can't query it per region
 563          */
 564         list_for_each_entry(pl, &nvdimm_bus->poison_list, list)
 565                 if (pl->start == addr) {
 566                         /* If length has changed, update this list entry */
 567                         if (pl->length != length)
 568                                 pl->length = length;
 569                         return 0;
 570                 }
 571
 572         /*
 573          * If not a duplicate or a simple length update, add the entry as is,
 574          * as any overlapping ranges will get resolved when the list is consumed
 575          * and converted to badblocks
 576          */
 577         return add_poison(nvdimm_bus, addr, length);
 578 }
 579
 580 int nvdimm_bus_add_poison(struct nvdimm_bus *nvdimm_bus, u64 addr, u64 length)
 581 {
 582         int rc;
 583
 584         nvdimm_bus_lock(&nvdimm_bus->dev);
 585         rc = bus_add_poison(nvdimm_bus, addr, length);
 586         nvdimm_bus_unlock(&nvdimm_bus->dev);
 587
 588         return rc;
 589 }
 590 EXPORT_SYMBOL_GPL(nvdimm_bus_add_poison);
 591
 592 #ifdef CONFIG_BLK_DEV_INTEGRITY
 593 int nd_integrity_init(struct gendisk *disk, unsigned long meta_size)
 594 {
 595         struct blk_integrity bi;
 596
 597         if (meta_size == 0)
 598                 return 0;
 599
 600         memset(&bi, 0, sizeof(bi));
 601
 602         bi.tuple_size = meta_size;
 603         bi.tag_size = meta_size;
 604
 605         blk_integrity_register(disk, &bi);
 606         blk_queue_max_integrity_segments(disk->queue, 1);
 607
 608         return 0;
 609 }
 610 EXPORT_SYMBOL(nd_integrity_init);
 611
 612 #else /* CONFIG_BLK_DEV_INTEGRITY */
 613 int nd_integrity_init(struct gendisk *disk, unsigned long meta_size)
 614 {
 615         return 0;
 616 }
 617 EXPORT_SYMBOL(nd_integrity_init);
 618
 619 #endif
 620
 621 static __init int libnvdimm_init(void)
 622 {
 623         int rc;
 624
 625         rc = nvdimm_bus_init();
 626         if (rc)
 627                 return rc;
 628         rc = nvdimm_init();
 629         if (rc)
 630                 goto err_dimm;
 631         rc = nd_region_init();
 632         if (rc)
 633                 goto err_region;
 634         return 0;
 635  err_region:
 636         nvdimm_exit();
 637  err_dimm:
 638         nvdimm_bus_exit();
 639         return rc;
 640 }
 641
 642 static __exit void libnvdimm_exit(void)
 643 {
 644         WARN_ON(!list_empty(&nvdimm_bus_list));
 645         nd_region_exit();
 646         nvdimm_exit();
 647         nvdimm_bus_exit();
 648         nd_region_devs_exit();
 649         nvdimm_devs_exit();
 650 }
 651
 652 MODULE_LICENSE("GPL v2");
 653 MODULE_AUTHOR("Intel Corporation");
 654 subsys_initcall(libnvdimm_init);
 655 module_exit(libnvdimm_exit);