drivers/base/memory.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Memory subsystem support
   4  *
   5  * Written by Matt Tolentino <matthew.e.tolentino@intel.com>
   6  *            Dave Hansen <haveblue@us.ibm.com>
   7  *
   8  * This file provides the necessary infrastructure to represent
   9  * a SPARSEMEM-memory-model system's physical memory in /sysfs.
  10  * All arch-independent code that assumes MEMORY_HOTPLUG requires
  11  * SPARSEMEM should be contained here, or in mm/memory_hotplug.c.
  12  */
  13
  14 #include <linux/module.h>
  15 #include <linux/init.h>
  16 #include <linux/topology.h>
  17 #include <linux/capability.h>
  18 #include <linux/device.h>
  19 #include <linux/memory.h>
  20 #include <linux/memory_hotplug.h>
  21 #include <linux/mm.h>
  22 #include <linux/mutex.h>
  23 #include <linux/stat.h>
  24 #include <linux/slab.h>
  25
  26 #include <linux/atomic.h>
  27 #include <linux/uaccess.h>
  28
  29 static DEFINE_MUTEX(mem_sysfs_mutex);
  30
  31 #define MEMORY_CLASS_NAME       "memory"
  32
  33 #define to_memory_block(dev) container_of(dev, struct memory_block, dev)
  34
  35 static int sections_per_block;
  36
  37 static inline int base_memory_block_id(int section_nr)
  38 {
  39         return section_nr / sections_per_block;
  40 }
  41
  42 static int memory_subsys_online(struct device *dev);
  43 static int memory_subsys_offline(struct device *dev);
  44
  45 static struct bus_type memory_subsys = {
  46         .name = MEMORY_CLASS_NAME,
  47         .dev_name = MEMORY_CLASS_NAME,
  48         .online = memory_subsys_online,
  49         .offline = memory_subsys_offline,
  50 };
  51
  52 static BLOCKING_NOTIFIER_HEAD(memory_chain);
  53
  54 int register_memory_notifier(struct notifier_block *nb)
  55 {
  56         return blocking_notifier_chain_register(&memory_chain, nb);
  57 }
  58 EXPORT_SYMBOL(register_memory_notifier);
  59
  60 void unregister_memory_notifier(struct notifier_block *nb)
  61 {
  62         blocking_notifier_chain_unregister(&memory_chain, nb);
  63 }
  64 EXPORT_SYMBOL(unregister_memory_notifier);
  65
  66 static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain);
  67
  68 int register_memory_isolate_notifier(struct notifier_block *nb)
  69 {
  70         return atomic_notifier_chain_register(&memory_isolate_chain, nb);
  71 }
  72 EXPORT_SYMBOL(register_memory_isolate_notifier);
  73
  74 void unregister_memory_isolate_notifier(struct notifier_block *nb)
  75 {
  76         atomic_notifier_chain_unregister(&memory_isolate_chain, nb);
  77 }
  78 EXPORT_SYMBOL(unregister_memory_isolate_notifier);
  79
  80 static void memory_block_release(struct device *dev)
  81 {
  82         struct memory_block *mem = to_memory_block(dev);
  83
  84         kfree(mem);
  85 }
  86
  87 unsigned long __weak memory_block_size_bytes(void)
  88 {
  89         return MIN_MEMORY_BLOCK_SIZE;
  90 }
  91 EXPORT_SYMBOL_GPL(memory_block_size_bytes);
  92
  93 static unsigned long get_memory_block_size(void)
  94 {
  95         unsigned long block_sz;
  96
  97         block_sz = memory_block_size_bytes();
  98
  99         /* Validate blk_sz is a power of 2 and not less than section size */
 100         if ((block_sz & (block_sz - 1)) || (block_sz < MIN_MEMORY_BLOCK_SIZE)) {
 101                 WARN_ON(1);
 102                 block_sz = MIN_MEMORY_BLOCK_SIZE;
 103         }
 104
 105         return block_sz;
 106 }
 107
 108 /*
 109  * use this as the physical section index that this memsection
 110  * uses.
 111  */
 112
 113 static ssize_t phys_index_show(struct device *dev,
 114                                struct device_attribute *attr, char *buf)
 115 {
 116         struct memory_block *mem = to_memory_block(dev);
 117         unsigned long phys_index;
 118
 119         phys_index = mem->start_section_nr / sections_per_block;
 120         return sprintf(buf, "%08lx\n", phys_index);
 121 }
 122
 123 /*
 124  * Show whether the section of memory is likely to be hot-removable
 125  */
 126 static ssize_t removable_show(struct device *dev, struct device_attribute *attr,
 127                               char *buf)
 128 {
 129         unsigned long i, pfn;
 130         int ret = 1;
 131         struct memory_block *mem = to_memory_block(dev);
 132
 133         if (mem->state != MEM_ONLINE)
 134                 goto out;
 135
 136         for (i = 0; i < sections_per_block; i++) {
 137                 if (!present_section_nr(mem->start_section_nr + i))
 138                         continue;
 139                 pfn = section_nr_to_pfn(mem->start_section_nr + i);
 140                 ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION);
 141         }
 142
 143 out:
 144         return sprintf(buf, "%d\n", ret);
 145 }
 146
 147 /*
 148  * online, offline, going offline, etc.
 149  */
 150 static ssize_t state_show(struct device *dev, struct device_attribute *attr,
 151                           char *buf)
 152 {
 153         struct memory_block *mem = to_memory_block(dev);
 154         ssize_t len = 0;
 155
 156         /*
 157          * We can probably put these states in a nice little array
 158          * so that they're not open-coded
 159          */
 160         switch (mem->state) {
 161         case MEM_ONLINE:
 162                 len = sprintf(buf, "online\n");
 163                 break;
 164         case MEM_OFFLINE:
 165                 len = sprintf(buf, "offline\n");
 166                 break;
 167         case MEM_GOING_OFFLINE:
 168                 len = sprintf(buf, "going-offline\n");
 169                 break;
 170         default:
 171                 len = sprintf(buf, "ERROR-UNKNOWN-%ld\n",
 172                                 mem->state);
 173                 WARN_ON(1);
 174                 break;
 175         }
 176
 177         return len;
 178 }
 179
 180 int memory_notify(unsigned long val, void *v)
 181 {
 182         return blocking_notifier_call_chain(&memory_chain, val, v);
 183 }
 184
 185 int memory_isolate_notify(unsigned long val, void *v)
 186 {
 187         return atomic_notifier_call_chain(&memory_isolate_chain, val, v);
 188 }
 189
 190 /*
 191  * The probe routines leave the pages uninitialized, just as the bootmem code
 192  * does. Make sure we do not access them, but instead use only information from
 193  * within sections.
 194  */
 195 static bool pages_correctly_probed(unsigned long start_pfn)
 196 {
 197         unsigned long section_nr = pfn_to_section_nr(start_pfn);
 198         unsigned long section_nr_end = section_nr + sections_per_block;
 199         unsigned long pfn = start_pfn;
 200
 201         /*
 202          * memmap between sections is not contiguous except with
 203          * SPARSEMEM_VMEMMAP. We lookup the page once per section
 204          * and assume memmap is contiguous within each section
 205          */
 206         for (; section_nr < section_nr_end; section_nr++) {
 207                 if (WARN_ON_ONCE(!pfn_valid(pfn)))
 208                         return false;
 209
 210                 if (!present_section_nr(section_nr)) {
 211                         pr_warn("section %ld pfn[%lx, %lx) not present\n",
 212                                 section_nr, pfn, pfn + PAGES_PER_SECTION);
 213                         return false;
 214                 } else if (!valid_section_nr(section_nr)) {
 215                         pr_warn("section %ld pfn[%lx, %lx) no valid memmap\n",
 216                                 section_nr, pfn, pfn + PAGES_PER_SECTION);
 217                         return false;
 218                 } else if (online_section_nr(section_nr)) {
 219                         pr_warn("section %ld pfn[%lx, %lx) is already online\n",
 220                                 section_nr, pfn, pfn + PAGES_PER_SECTION);
 221                         return false;
 222                 }
 223                 pfn += PAGES_PER_SECTION;
 224         }
 225
 226         return true;
 227 }
 228
 229 /*
 230  * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
 231  * OK to have direct references to sparsemem variables in here.
 232  */
 233 static int
 234 memory_block_action(unsigned long start_section_nr, unsigned long action,
 235                     int online_type)
 236 {
 237         unsigned long start_pfn;
 238         unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
 239         int ret;
 240
 241         start_pfn = section_nr_to_pfn(start_section_nr);
 242
 243         switch (action) {
 244         case MEM_ONLINE:
 245                 if (!pages_correctly_probed(start_pfn))
 246                         return -EBUSY;
 247
 248                 ret = online_pages(start_pfn, nr_pages, online_type);
 249                 break;
 250         case MEM_OFFLINE:
 251                 ret = offline_pages(start_pfn, nr_pages);
 252                 break;
 253         default:
 254                 WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: "
 255                      "%ld\n", __func__, start_section_nr, action, action);
 256                 ret = -EINVAL;
 257         }
 258
 259         return ret;
 260 }
 261
 262 static int memory_block_change_state(struct memory_block *mem,
 263                 unsigned long to_state, unsigned long from_state_req)
 264 {
 265         int ret = 0;
 266
 267         if (mem->state != from_state_req)
 268                 return -EINVAL;
 269
 270         if (to_state == MEM_OFFLINE)
 271                 mem->state = MEM_GOING_OFFLINE;
 272
 273         ret = memory_block_action(mem->start_section_nr, to_state,
 274                                 mem->online_type);
 275
 276         mem->state = ret ? from_state_req : to_state;
 277
 278         return ret;
 279 }
 280
 281 /* The device lock serializes operations on memory_subsys_[online|offline] */
 282 static int memory_subsys_online(struct device *dev)
 283 {
 284         struct memory_block *mem = to_memory_block(dev);
 285         int ret;
 286
 287         if (mem->state == MEM_ONLINE)
 288                 return 0;
 289
 290         /*
 291          * If we are called from state_store(), online_type will be
 292          * set >= 0 Otherwise we were called from the device online
 293          * attribute and need to set the online_type.
 294          */
 295         if (mem->online_type < 0)
 296                 mem->online_type = MMOP_ONLINE_KEEP;
 297
 298         ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
 299
 300         /* clear online_type */
 301         mem->online_type = -1;
 302
 303         return ret;
 304 }
 305
 306 static int memory_subsys_offline(struct device *dev)
 307 {
 308         struct memory_block *mem = to_memory_block(dev);
 309
 310         if (mem->state == MEM_OFFLINE)
 311                 return 0;
 312
 313         /* Can't offline block with non-present sections */
 314         if (mem->section_count != sections_per_block)
 315                 return -EINVAL;
 316
 317         return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
 318 }
 319
 320 static ssize_t state_store(struct device *dev, struct device_attribute *attr,
 321                            const char *buf, size_t count)
 322 {
 323         struct memory_block *mem = to_memory_block(dev);
 324         int ret, online_type;
 325
 326         ret = lock_device_hotplug_sysfs();
 327         if (ret)
 328                 return ret;
 329
 330         if (sysfs_streq(buf, "online_kernel"))
 331                 online_type = MMOP_ONLINE_KERNEL;
 332         else if (sysfs_streq(buf, "online_movable"))
 333                 online_type = MMOP_ONLINE_MOVABLE;
 334         else if (sysfs_streq(buf, "online"))
 335                 online_type = MMOP_ONLINE_KEEP;
 336         else if (sysfs_streq(buf, "offline"))
 337                 online_type = MMOP_OFFLINE;
 338         else {
 339                 ret = -EINVAL;
 340                 goto err;
 341         }
 342
 343         switch (online_type) {
 344         case MMOP_ONLINE_KERNEL:
 345         case MMOP_ONLINE_MOVABLE:
 346         case MMOP_ONLINE_KEEP:
 347                 /* mem->online_type is protected by device_hotplug_lock */
 348                 mem->online_type = online_type;
 349                 ret = device_online(&mem->dev);
 350                 break;
 351         case MMOP_OFFLINE:
 352                 ret = device_offline(&mem->dev);
 353                 break;
 354         default:
 355                 ret = -EINVAL; /* should never happen */
 356         }
 357
 358 err:
 359         unlock_device_hotplug();
 360
 361         if (ret < 0)
 362                 return ret;
 363         if (ret)
 364                 return -EINVAL;
 365
 366         return count;
 367 }
 368
 369 /*
 370  * phys_device is a bad name for this.  What I really want
 371  * is a way to differentiate between memory ranges that
 372  * are part of physical devices that constitute
 373  * a complete removable unit or fru.
 374  * i.e. do these ranges belong to the same physical device,
 375  * s.t. if I offline all of these sections I can then
 376  * remove the physical device?
 377  */
 378 static ssize_t phys_device_show(struct device *dev,
 379                                 struct device_attribute *attr, char *buf)
 380 {
 381         struct memory_block *mem = to_memory_block(dev);
 382         return sprintf(buf, "%d\n", mem->phys_device);
 383 }
 384
 385 #ifdef CONFIG_MEMORY_HOTREMOVE
 386 static void print_allowed_zone(char *buf, int nid, unsigned long start_pfn,
 387                 unsigned long nr_pages, int online_type,
 388                 struct zone *default_zone)
 389 {
 390         struct zone *zone;
 391
 392         zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
 393         if (zone != default_zone) {
 394                 strcat(buf, " ");
 395                 strcat(buf, zone->name);
 396         }
 397 }
 398
 399 static ssize_t valid_zones_show(struct device *dev,
 400                                 struct device_attribute *attr, char *buf)
 401 {
 402         struct memory_block *mem = to_memory_block(dev);
 403         unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr);
 404         unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
 405         unsigned long valid_start_pfn, valid_end_pfn;
 406         struct zone *default_zone;
 407         int nid;
 408
 409         /*
 410          * Check the existing zone. Make sure that we do that only on the
 411          * online nodes otherwise the page_zone is not reliable
 412          */
 413         if (mem->state == MEM_ONLINE) {
 414                 /*
 415                  * The block contains more than one zone can not be offlined.
 416                  * This can happen e.g. for ZONE_DMA and ZONE_DMA32
 417                  */
 418                 if (!test_pages_in_a_zone(start_pfn, start_pfn + nr_pages,
 419                                           &valid_start_pfn, &valid_end_pfn))
 420                         return sprintf(buf, "none\n");
 421                 start_pfn = valid_start_pfn;
 422                 strcat(buf, page_zone(pfn_to_page(start_pfn))->name);
 423                 goto out;
 424         }
 425
 426         nid = mem->nid;
 427         default_zone = zone_for_pfn_range(MMOP_ONLINE_KEEP, nid, start_pfn, nr_pages);
 428         strcat(buf, default_zone->name);
 429
 430         print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_KERNEL,
 431                         default_zone);
 432         print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_MOVABLE,
 433                         default_zone);
 434 out:
 435         strcat(buf, "\n");
 436
 437         return strlen(buf);
 438 }
 439 static DEVICE_ATTR_RO(valid_zones);
 440 #endif
 441
 442 static DEVICE_ATTR_RO(phys_index);
 443 static DEVICE_ATTR_RW(state);
 444 static DEVICE_ATTR_RO(phys_device);
 445 static DEVICE_ATTR_RO(removable);
 446
 447 /*
 448  * Block size attribute stuff
 449  */
 450 static ssize_t block_size_bytes_show(struct device *dev,
 451                                      struct device_attribute *attr, char *buf)
 452 {
 453         return sprintf(buf, "%lx\n", get_memory_block_size());
 454 }
 455
 456 static DEVICE_ATTR_RO(block_size_bytes);
 457
 458 /*
 459  * Memory auto online policy.
 460  */
 461
 462 static ssize_t auto_online_blocks_show(struct device *dev,
 463                                        struct device_attribute *attr, char *buf)
 464 {
 465         if (memhp_auto_online)
 466                 return sprintf(buf, "online\n");
 467         else
 468                 return sprintf(buf, "offline\n");
 469 }
 470
 471 static ssize_t auto_online_blocks_store(struct device *dev,
 472                                         struct device_attribute *attr,
 473                                         const char *buf, size_t count)
 474 {
 475         if (sysfs_streq(buf, "online"))
 476                 memhp_auto_online = true;
 477         else if (sysfs_streq(buf, "offline"))
 478                 memhp_auto_online = false;
 479         else
 480                 return -EINVAL;
 481
 482         return count;
 483 }
 484
 485 static DEVICE_ATTR_RW(auto_online_blocks);
 486
 487 /*
 488  * Some architectures will have custom drivers to do this, and
 489  * will not need to do it from userspace.  The fake hot-add code
 490  * as well as ppc64 will do all of their discovery in userspace
 491  * and will require this interface.
 492  */
 493 #ifdef CONFIG_ARCH_MEMORY_PROBE
 494 static ssize_t probe_store(struct device *dev, struct device_attribute *attr,
 495                            const char *buf, size_t count)
 496 {
 497         u64 phys_addr;
 498         int nid, ret;
 499         unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block;
 500
 501         ret = kstrtoull(buf, 0, &phys_addr);
 502         if (ret)
 503                 return ret;
 504
 505         if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1))
 506                 return -EINVAL;
 507
 508         ret = lock_device_hotplug_sysfs();
 509         if (ret)
 510                 return ret;
 511
 512         nid = memory_add_physaddr_to_nid(phys_addr);
 513         ret = __add_memory(nid, phys_addr,
 514                            MIN_MEMORY_BLOCK_SIZE * sections_per_block);
 515
 516         if (ret)
 517                 goto out;
 518
 519         ret = count;
 520 out:
 521         unlock_device_hotplug();
 522         return ret;
 523 }
 524
 525 static DEVICE_ATTR_WO(probe);
 526 #endif
 527
 528 #ifdef CONFIG_MEMORY_FAILURE
 529 /*
 530  * Support for offlining pages of memory
 531  */
 532
 533 /* Soft offline a page */
 534 static ssize_t soft_offline_page_store(struct device *dev,
 535                                        struct device_attribute *attr,
 536                                        const char *buf, size_t count)
 537 {
 538         int ret;
 539         u64 pfn;
 540         if (!capable(CAP_SYS_ADMIN))
 541                 return -EPERM;
 542         if (kstrtoull(buf, 0, &pfn) < 0)
 543                 return -EINVAL;
 544         pfn >>= PAGE_SHIFT;
 545         if (!pfn_valid(pfn))
 546                 return -ENXIO;
 547         ret = soft_offline_page(pfn_to_page(pfn), 0);
 548         return ret == 0 ? count : ret;
 549 }
 550
 551 /* Forcibly offline a page, including killing processes. */
 552 static ssize_t hard_offline_page_store(struct device *dev,
 553                                        struct device_attribute *attr,
 554                                        const char *buf, size_t count)
 555 {
 556         int ret;
 557         u64 pfn;
 558         if (!capable(CAP_SYS_ADMIN))
 559                 return -EPERM;
 560         if (kstrtoull(buf, 0, &pfn) < 0)
 561                 return -EINVAL;
 562         pfn >>= PAGE_SHIFT;
 563         ret = memory_failure(pfn, 0);
 564         return ret ? ret : count;
 565 }
 566
 567 static DEVICE_ATTR_WO(soft_offline_page);
 568 static DEVICE_ATTR_WO(hard_offline_page);
 569 #endif
 570
 571 /*
 572  * Note that phys_device is optional.  It is here to allow for
 573  * differentiation between which *physical* devices each
 574  * section belongs to...
 575  */
 576 int __weak arch_get_memory_phys_device(unsigned long start_pfn)
 577 {
 578         return 0;
 579 }
 580
 581 /*
 582  * A reference for the returned object is held and the reference for the
 583  * hinted object is released.
 584  */
 585 struct memory_block *find_memory_block_hinted(struct mem_section *section,
 586                                               struct memory_block *hint)
 587 {
 588         int block_id = base_memory_block_id(__section_nr(section));
 589         struct device *hintdev = hint ? &hint->dev : NULL;
 590         struct device *dev;
 591
 592         dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev);
 593         if (hint)
 594                 put_device(&hint->dev);
 595         if (!dev)
 596                 return NULL;
 597         return to_memory_block(dev);
 598 }
 599
 600 /*
 601  * For now, we have a linear search to go find the appropriate
 602  * memory_block corresponding to a particular phys_index. If
 603  * this gets to be a real problem, we can always use a radix
 604  * tree or something here.
 605  *
 606  * This could be made generic for all device subsystems.
 607  */
 608 struct memory_block *find_memory_block(struct mem_section *section)
 609 {
 610         return find_memory_block_hinted(section, NULL);
 611 }
 612
 613 static struct attribute *memory_memblk_attrs[] = {
 614         &dev_attr_phys_index.attr,
 615         &dev_attr_state.attr,
 616         &dev_attr_phys_device.attr,
 617         &dev_attr_removable.attr,
 618 #ifdef CONFIG_MEMORY_HOTREMOVE
 619         &dev_attr_valid_zones.attr,
 620 #endif
 621         NULL
 622 };
 623
 624 static struct attribute_group memory_memblk_attr_group = {
 625         .attrs = memory_memblk_attrs,
 626 };
 627
 628 static const struct attribute_group *memory_memblk_attr_groups[] = {
 629         &memory_memblk_attr_group,
 630         NULL,
 631 };
 632
 633 /*
 634  * register_memory - Setup a sysfs device for a memory block
 635  */
 636 static
 637 int register_memory(struct memory_block *memory)
 638 {
 639         int ret;
 640
 641         memory->dev.bus = &memory_subsys;
 642         memory->dev.id = memory->start_section_nr / sections_per_block;
 643         memory->dev.release = memory_block_release;
 644         memory->dev.groups = memory_memblk_attr_groups;
 645         memory->dev.offline = memory->state == MEM_OFFLINE;
 646
 647         ret = device_register(&memory->dev);
 648         if (ret)
 649                 put_device(&memory->dev);
 650
 651         return ret;
 652 }
 653
 654 static int init_memory_block(struct memory_block **memory,
 655                              struct mem_section *section, unsigned long state)
 656 {
 657         struct memory_block *mem;
 658         unsigned long start_pfn;
 659         int scn_nr;
 660         int ret = 0;
 661
 662         mem = kzalloc(sizeof(*mem), GFP_KERNEL);
 663         if (!mem)
 664                 return -ENOMEM;
 665
 666         scn_nr = __section_nr(section);
 667         mem->start_section_nr =
 668                         base_memory_block_id(scn_nr) * sections_per_block;
 669         mem->end_section_nr = mem->start_section_nr + sections_per_block - 1;
 670         mem->state = state;
 671         start_pfn = section_nr_to_pfn(mem->start_section_nr);
 672         mem->phys_device = arch_get_memory_phys_device(start_pfn);
 673
 674         ret = register_memory(mem);
 675
 676         *memory = mem;
 677         return ret;
 678 }
 679
 680 static int add_memory_block(int base_section_nr)
 681 {
 682         struct memory_block *mem;
 683         int i, ret, section_count = 0, section_nr;
 684
 685         for (i = base_section_nr;
 686              i < base_section_nr + sections_per_block;
 687              i++) {
 688                 if (!present_section_nr(i))
 689                         continue;
 690                 if (section_count == 0)
 691                         section_nr = i;
 692                 section_count++;
 693         }
 694
 695         if (section_count == 0)
 696                 return 0;
 697         ret = init_memory_block(&mem, __nr_to_section(section_nr), MEM_ONLINE);
 698         if (ret)
 699                 return ret;
 700         mem->section_count = section_count;
 701         return 0;
 702 }
 703
 704 /*
 705  * need an interface for the VM to add new memory regions,
 706  * but without onlining it.
 707  */
 708 int hotplug_memory_register(int nid, struct mem_section *section)
 709 {
 710         int ret = 0;
 711         struct memory_block *mem;
 712
 713         mutex_lock(&mem_sysfs_mutex);
 714
 715         mem = find_memory_block(section);
 716         if (mem) {
 717                 mem->section_count++;
 718                 put_device(&mem->dev);
 719         } else {
 720                 ret = init_memory_block(&mem, section, MEM_OFFLINE);
 721                 if (ret)
 722                         goto out;
 723                 mem->section_count++;
 724         }
 725
 726 out:
 727         mutex_unlock(&mem_sysfs_mutex);
 728         return ret;
 729 }
 730
 731 #ifdef CONFIG_MEMORY_HOTREMOVE
 732 static void
 733 unregister_memory(struct memory_block *memory)
 734 {
 735         BUG_ON(memory->dev.bus != &memory_subsys);
 736
 737         /* drop the ref. we got via find_memory_block() */
 738         put_device(&memory->dev);
 739         device_unregister(&memory->dev);
 740 }
 741
 742 void unregister_memory_section(struct mem_section *section)
 743 {
 744         struct memory_block *mem;
 745
 746         if (WARN_ON_ONCE(!present_section(section)))
 747                 return;
 748
 749         mutex_lock(&mem_sysfs_mutex);
 750
 751         /*
 752          * Some users of the memory hotplug do not want/need memblock to
 753          * track all sections. Skip over those.
 754          */
 755         mem = find_memory_block(section);
 756         if (!mem)
 757                 goto out_unlock;
 758
 759         unregister_mem_sect_under_nodes(mem, __section_nr(section));
 760
 761         mem->section_count--;
 762         if (mem->section_count == 0)
 763                 unregister_memory(mem);
 764         else
 765                 put_device(&mem->dev);
 766
 767 out_unlock:
 768         mutex_unlock(&mem_sysfs_mutex);
 769 }
 770 #endif /* CONFIG_MEMORY_HOTREMOVE */
 771
 772 /* return true if the memory block is offlined, otherwise, return false */
 773 bool is_memblock_offlined(struct memory_block *mem)
 774 {
 775         return mem->state == MEM_OFFLINE;
 776 }
 777
 778 static struct attribute *memory_root_attrs[] = {
 779 #ifdef CONFIG_ARCH_MEMORY_PROBE
 780         &dev_attr_probe.attr,
 781 #endif
 782
 783 #ifdef CONFIG_MEMORY_FAILURE
 784         &dev_attr_soft_offline_page.attr,
 785         &dev_attr_hard_offline_page.attr,
 786 #endif
 787
 788         &dev_attr_block_size_bytes.attr,
 789         &dev_attr_auto_online_blocks.attr,
 790         NULL
 791 };
 792
 793 static struct attribute_group memory_root_attr_group = {
 794         .attrs = memory_root_attrs,
 795 };
 796
 797 static const struct attribute_group *memory_root_attr_groups[] = {
 798         &memory_root_attr_group,
 799         NULL,
 800 };
 801
 802 /*
 803  * Initialize the sysfs support for memory devices...
 804  */
 805 int __init memory_dev_init(void)
 806 {
 807         unsigned int i;
 808         int ret;
 809         int err;
 810         unsigned long block_sz;
 811
 812         ret = subsys_system_register(&memory_subsys, memory_root_attr_groups);
 813         if (ret)
 814                 goto out;
 815
 816         block_sz = get_memory_block_size();
 817         sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE;
 818
 819         /*
 820          * Create entries for memory sections that were found
 821          * during boot and have been initialized
 822          */
 823         mutex_lock(&mem_sysfs_mutex);
 824         for (i = 0; i <= __highest_present_section_nr;
 825                 i += sections_per_block) {
 826                 err = add_memory_block(i);
 827                 if (!ret)
 828                         ret = err;
 829         }
 830         mutex_unlock(&mem_sysfs_mutex);
 831
 832 out:
 833         if (ret)
 834                 printk(KERN_ERR "%s() failed: %d\n", __func__, ret);
 835         return ret;
 836 }