drivers/base/memory.c

   1 /*
   2  * Memory subsystem support
   3  *
   4  * Written by Matt Tolentino <matthew.e.tolentino@intel.com>
   5  *            Dave Hansen <haveblue@us.ibm.com>
   6  *
   7  * This file provides the necessary infrastructure to represent
   8  * a SPARSEMEM-memory-model system's physical memory in /sysfs.
   9  * All arch-independent code that assumes MEMORY_HOTPLUG requires
  10  * SPARSEMEM should be contained here, or in mm/memory_hotplug.c.
  11  */
  12
  13 #include <linux/module.h>
  14 #include <linux/init.h>
  15 #include <linux/topology.h>
  16 #include <linux/capability.h>
  17 #include <linux/device.h>
  18 #include <linux/memory.h>
  19 #include <linux/memory_hotplug.h>
  20 #include <linux/mm.h>
  21 #include <linux/mutex.h>
  22 #include <linux/stat.h>
  23 #include <linux/slab.h>
  24
  25 #include <linux/atomic.h>
  26 #include <asm/uaccess.h>
  27
  28 static DEFINE_MUTEX(mem_sysfs_mutex);
  29
  30 #define MEMORY_CLASS_NAME       "memory"
  31
  32 #define to_memory_block(dev) container_of(dev, struct memory_block, dev)
  33
  34 static int sections_per_block;
  35
  36 static inline int base_memory_block_id(int section_nr)
  37 {
  38         return section_nr / sections_per_block;
  39 }
  40
  41 static int memory_subsys_online(struct device *dev);
  42 static int memory_subsys_offline(struct device *dev);
  43
  44 static struct bus_type memory_subsys = {
  45         .name = MEMORY_CLASS_NAME,
  46         .dev_name = MEMORY_CLASS_NAME,
  47         .online = memory_subsys_online,
  48         .offline = memory_subsys_offline,
  49 };
  50
  51 static BLOCKING_NOTIFIER_HEAD(memory_chain);
  52
  53 int register_memory_notifier(struct notifier_block *nb)
  54 {
  55         return blocking_notifier_chain_register(&memory_chain, nb);
  56 }
  57 EXPORT_SYMBOL(register_memory_notifier);
  58
  59 void unregister_memory_notifier(struct notifier_block *nb)
  60 {
  61         blocking_notifier_chain_unregister(&memory_chain, nb);
  62 }
  63 EXPORT_SYMBOL(unregister_memory_notifier);
  64
  65 static ATOMIC_NOTIFIER_HEAD(memory_isolate_chain);
  66
  67 int register_memory_isolate_notifier(struct notifier_block *nb)
  68 {
  69         return atomic_notifier_chain_register(&memory_isolate_chain, nb);
  70 }
  71 EXPORT_SYMBOL(register_memory_isolate_notifier);
  72
  73 void unregister_memory_isolate_notifier(struct notifier_block *nb)
  74 {
  75         atomic_notifier_chain_unregister(&memory_isolate_chain, nb);
  76 }
  77 EXPORT_SYMBOL(unregister_memory_isolate_notifier);
  78
  79 static void memory_block_release(struct device *dev)
  80 {
  81         struct memory_block *mem = to_memory_block(dev);
  82
  83         kfree(mem);
  84 }
  85
  86 unsigned long __weak memory_block_size_bytes(void)
  87 {
  88         return MIN_MEMORY_BLOCK_SIZE;
  89 }
  90
  91 static unsigned long get_memory_block_size(void)
  92 {
  93         unsigned long block_sz;
  94
  95         block_sz = memory_block_size_bytes();
  96
  97         /* Validate blk_sz is a power of 2 and not less than section size */
  98         if ((block_sz & (block_sz - 1)) || (block_sz < MIN_MEMORY_BLOCK_SIZE)) {
  99                 WARN_ON(1);
 100                 block_sz = MIN_MEMORY_BLOCK_SIZE;
 101         }
 102
 103         return block_sz;
 104 }
 105
 106 /*
 107  * use this as the physical section index that this memsection
 108  * uses.
 109  */
 110
 111 static ssize_t show_mem_start_phys_index(struct device *dev,
 112                         struct device_attribute *attr, char *buf)
 113 {
 114         struct memory_block *mem = to_memory_block(dev);
 115         unsigned long phys_index;
 116
 117         phys_index = mem->start_section_nr / sections_per_block;
 118         return sprintf(buf, "%08lx\n", phys_index);
 119 }
 120
 121 /*
 122  * Show whether the section of memory is likely to be hot-removable
 123  */
 124 static ssize_t show_mem_removable(struct device *dev,
 125                         struct device_attribute *attr, char *buf)
 126 {
 127         unsigned long i, pfn;
 128         int ret = 1;
 129         struct memory_block *mem = to_memory_block(dev);
 130
 131         for (i = 0; i < sections_per_block; i++) {
 132                 if (!present_section_nr(mem->start_section_nr + i))
 133                         continue;
 134                 pfn = section_nr_to_pfn(mem->start_section_nr + i);
 135                 ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION);
 136         }
 137
 138         return sprintf(buf, "%d\n", ret);
 139 }
 140
 141 /*
 142  * online, offline, going offline, etc.
 143  */
 144 static ssize_t show_mem_state(struct device *dev,
 145                         struct device_attribute *attr, char *buf)
 146 {
 147         struct memory_block *mem = to_memory_block(dev);
 148         ssize_t len = 0;
 149
 150         /*
 151          * We can probably put these states in a nice little array
 152          * so that they're not open-coded
 153          */
 154         switch (mem->state) {
 155         case MEM_ONLINE:
 156                 len = sprintf(buf, "online\n");
 157                 break;
 158         case MEM_OFFLINE:
 159                 len = sprintf(buf, "offline\n");
 160                 break;
 161         case MEM_GOING_OFFLINE:
 162                 len = sprintf(buf, "going-offline\n");
 163                 break;
 164         default:
 165                 len = sprintf(buf, "ERROR-UNKNOWN-%ld\n",
 166                                 mem->state);
 167                 WARN_ON(1);
 168                 break;
 169         }
 170
 171         return len;
 172 }
 173
 174 int memory_notify(unsigned long val, void *v)
 175 {
 176         return blocking_notifier_call_chain(&memory_chain, val, v);
 177 }
 178
 179 int memory_isolate_notify(unsigned long val, void *v)
 180 {
 181         return atomic_notifier_call_chain(&memory_isolate_chain, val, v);
 182 }
 183
 184 /*
 185  * The probe routines leave the pages reserved, just as the bootmem code does.
 186  * Make sure they're still that way.
 187  */
 188 static bool pages_correctly_reserved(unsigned long start_pfn)
 189 {
 190         int i, j;
 191         struct page *page;
 192         unsigned long pfn = start_pfn;
 193
 194         /*
 195          * memmap between sections is not contiguous except with
 196          * SPARSEMEM_VMEMMAP. We lookup the page once per section
 197          * and assume memmap is contiguous within each section
 198          */
 199         for (i = 0; i < sections_per_block; i++, pfn += PAGES_PER_SECTION) {
 200                 if (WARN_ON_ONCE(!pfn_valid(pfn)))
 201                         return false;
 202                 page = pfn_to_page(pfn);
 203
 204                 for (j = 0; j < PAGES_PER_SECTION; j++) {
 205                         if (PageReserved(page + j))
 206                                 continue;
 207
 208                         printk(KERN_WARNING "section number %ld page number %d "
 209                                 "not reserved, was it already online?\n",
 210                                 pfn_to_section_nr(pfn), j);
 211
 212                         return false;
 213                 }
 214         }
 215
 216         return true;
 217 }
 218
 219 /*
 220  * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
 221  * OK to have direct references to sparsemem variables in here.
 222  * Must already be protected by mem_hotplug_begin().
 223  */
 224 static int
 225 memory_block_action(unsigned long phys_index, unsigned long action, int online_type)
 226 {
 227         unsigned long start_pfn;
 228         unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
 229         struct page *first_page;
 230         int ret;
 231
 232         start_pfn = section_nr_to_pfn(phys_index);
 233         first_page = pfn_to_page(start_pfn);
 234
 235         switch (action) {
 236         case MEM_ONLINE:
 237                 if (!pages_correctly_reserved(start_pfn))
 238                         return -EBUSY;
 239
 240                 ret = online_pages(start_pfn, nr_pages, online_type);
 241                 break;
 242         case MEM_OFFLINE:
 243                 ret = offline_pages(start_pfn, nr_pages);
 244                 break;
 245         default:
 246                 WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: "
 247                      "%ld\n", __func__, phys_index, action, action);
 248                 ret = -EINVAL;
 249         }
 250
 251         return ret;
 252 }
 253
 254 static int memory_block_change_state(struct memory_block *mem,
 255                 unsigned long to_state, unsigned long from_state_req)
 256 {
 257         int ret = 0;
 258
 259         if (mem->state != from_state_req)
 260                 return -EINVAL;
 261
 262         if (to_state == MEM_OFFLINE)
 263                 mem->state = MEM_GOING_OFFLINE;
 264
 265         ret = memory_block_action(mem->start_section_nr, to_state,
 266                                 mem->online_type);
 267
 268         mem->state = ret ? from_state_req : to_state;
 269
 270         return ret;
 271 }
 272
 273 /* The device lock serializes operations on memory_subsys_[online|offline] */
 274 static int memory_subsys_online(struct device *dev)
 275 {
 276         struct memory_block *mem = to_memory_block(dev);
 277         int ret;
 278
 279         if (mem->state == MEM_ONLINE)
 280                 return 0;
 281
 282         /*
 283          * If we are called from store_mem_state(), online_type will be
 284          * set >= 0 Otherwise we were called from the device online
 285          * attribute and need to set the online_type.
 286          */
 287         if (mem->online_type < 0)
 288                 mem->online_type = MMOP_ONLINE_KEEP;
 289
 290         /* Already under protection of mem_hotplug_begin() */
 291         ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
 292
 293         /* clear online_type */
 294         mem->online_type = -1;
 295
 296         return ret;
 297 }
 298
 299 static int memory_subsys_offline(struct device *dev)
 300 {
 301         struct memory_block *mem = to_memory_block(dev);
 302
 303         if (mem->state == MEM_OFFLINE)
 304                 return 0;
 305
 306         return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
 307 }
 308
 309 static ssize_t
 310 store_mem_state(struct device *dev,
 311                 struct device_attribute *attr, const char *buf, size_t count)
 312 {
 313         struct memory_block *mem = to_memory_block(dev);
 314         int ret, online_type;
 315
 316         ret = lock_device_hotplug_sysfs();
 317         if (ret)
 318                 return ret;
 319
 320         if (sysfs_streq(buf, "online_kernel"))
 321                 online_type = MMOP_ONLINE_KERNEL;
 322         else if (sysfs_streq(buf, "online_movable"))
 323                 online_type = MMOP_ONLINE_MOVABLE;
 324         else if (sysfs_streq(buf, "online"))
 325                 online_type = MMOP_ONLINE_KEEP;
 326         else if (sysfs_streq(buf, "offline"))
 327                 online_type = MMOP_OFFLINE;
 328         else {
 329                 ret = -EINVAL;
 330                 goto err;
 331         }
 332
 333         /*
 334          * Memory hotplug needs to hold mem_hotplug_begin() for probe to find
 335          * the correct memory block to online before doing device_online(dev),
 336          * which will take dev->mutex.  Take the lock early to prevent an
 337          * inversion, memory_subsys_online() callbacks will be implemented by
 338          * assuming it's already protected.
 339          */
 340         mem_hotplug_begin();
 341
 342         switch (online_type) {
 343         case MMOP_ONLINE_KERNEL:
 344         case MMOP_ONLINE_MOVABLE:
 345         case MMOP_ONLINE_KEEP:
 346                 mem->online_type = online_type;
 347                 ret = device_online(&mem->dev);
 348                 break;
 349         case MMOP_OFFLINE:
 350                 ret = device_offline(&mem->dev);
 351                 break;
 352         default:
 353                 ret = -EINVAL; /* should never happen */
 354         }
 355
 356         mem_hotplug_done();
 357 err:
 358         unlock_device_hotplug();
 359
 360         if (ret)
 361                 return ret;
 362         return count;
 363 }
 364
 365 /*
 366  * phys_device is a bad name for this.  What I really want
 367  * is a way to differentiate between memory ranges that
 368  * are part of physical devices that constitute
 369  * a complete removable unit or fru.
 370  * i.e. do these ranges belong to the same physical device,
 371  * s.t. if I offline all of these sections I can then
 372  * remove the physical device?
 373  */
 374 static ssize_t show_phys_device(struct device *dev,
 375                                 struct device_attribute *attr, char *buf)
 376 {
 377         struct memory_block *mem = to_memory_block(dev);
 378         return sprintf(buf, "%d\n", mem->phys_device);
 379 }
 380
 381 #ifdef CONFIG_MEMORY_HOTREMOVE
 382 static ssize_t show_valid_zones(struct device *dev,
 383                                 struct device_attribute *attr, char *buf)
 384 {
 385         struct memory_block *mem = to_memory_block(dev);
 386         unsigned long start_pfn, end_pfn;
 387         unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
 388         struct page *first_page;
 389         struct zone *zone;
 390
 391         start_pfn = section_nr_to_pfn(mem->start_section_nr);
 392         end_pfn = start_pfn + nr_pages;
 393         first_page = pfn_to_page(start_pfn);
 394
 395         /* The block contains more than one zone can not be offlined. */
 396         if (!test_pages_in_a_zone(start_pfn, end_pfn))
 397                 return sprintf(buf, "none\n");
 398
 399         zone = page_zone(first_page);
 400
 401         if (zone_idx(zone) == ZONE_MOVABLE - 1) {
 402                 /*The mem block is the last memoryblock of this zone.*/
 403                 if (end_pfn == zone_end_pfn(zone))
 404                         return sprintf(buf, "%s %s\n",
 405                                         zone->name, (zone + 1)->name);
 406         }
 407
 408         if (zone_idx(zone) == ZONE_MOVABLE) {
 409                 /*The mem block is the first memoryblock of ZONE_MOVABLE.*/
 410                 if (start_pfn == zone->zone_start_pfn)
 411                         return sprintf(buf, "%s %s\n",
 412                                         zone->name, (zone - 1)->name);
 413         }
 414
 415         return sprintf(buf, "%s\n", zone->name);
 416 }
 417 static DEVICE_ATTR(valid_zones, 0444, show_valid_zones, NULL);
 418 #endif
 419
 420 static DEVICE_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL);
 421 static DEVICE_ATTR(state, 0644, show_mem_state, store_mem_state);
 422 static DEVICE_ATTR(phys_device, 0444, show_phys_device, NULL);
 423 static DEVICE_ATTR(removable, 0444, show_mem_removable, NULL);
 424
 425 /*
 426  * Block size attribute stuff
 427  */
 428 static ssize_t
 429 print_block_size(struct device *dev, struct device_attribute *attr,
 430                  char *buf)
 431 {
 432         return sprintf(buf, "%lx\n", get_memory_block_size());
 433 }
 434
 435 static DEVICE_ATTR(block_size_bytes, 0444, print_block_size, NULL);
 436
 437 /*
 438  * Some architectures will have custom drivers to do this, and
 439  * will not need to do it from userspace.  The fake hot-add code
 440  * as well as ppc64 will do all of their discovery in userspace
 441  * and will require this interface.
 442  */
 443 #ifdef CONFIG_ARCH_MEMORY_PROBE
 444 static ssize_t
 445 memory_probe_store(struct device *dev, struct device_attribute *attr,
 446                    const char *buf, size_t count)
 447 {
 448         u64 phys_addr;
 449         int nid;
 450         int i, ret;
 451         unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block;
 452
 453         ret = kstrtoull(buf, 0, &phys_addr);
 454         if (ret)
 455                 return ret;
 456
 457         if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1))
 458                 return -EINVAL;
 459
 460         for (i = 0; i < sections_per_block; i++) {
 461                 nid = memory_add_physaddr_to_nid(phys_addr);
 462                 ret = add_memory(nid, phys_addr,
 463                                  PAGES_PER_SECTION << PAGE_SHIFT);
 464                 if (ret)
 465                         goto out;
 466
 467                 phys_addr += MIN_MEMORY_BLOCK_SIZE;
 468         }
 469
 470         ret = count;
 471 out:
 472         return ret;
 473 }
 474
 475 static DEVICE_ATTR(probe, S_IWUSR, NULL, memory_probe_store);
 476 #endif
 477
 478 #ifdef CONFIG_MEMORY_FAILURE
 479 /*
 480  * Support for offlining pages of memory
 481  */
 482
 483 /* Soft offline a page */
 484 static ssize_t
 485 store_soft_offline_page(struct device *dev,
 486                         struct device_attribute *attr,
 487                         const char *buf, size_t count)
 488 {
 489         int ret;
 490         u64 pfn;
 491         if (!capable(CAP_SYS_ADMIN))
 492                 return -EPERM;
 493         if (kstrtoull(buf, 0, &pfn) < 0)
 494                 return -EINVAL;
 495         pfn >>= PAGE_SHIFT;
 496         if (!pfn_valid(pfn))
 497                 return -ENXIO;
 498         ret = soft_offline_page(pfn_to_page(pfn), 0);
 499         return ret == 0 ? count : ret;
 500 }
 501
 502 /* Forcibly offline a page, including killing processes. */
 503 static ssize_t
 504 store_hard_offline_page(struct device *dev,
 505                         struct device_attribute *attr,
 506                         const char *buf, size_t count)
 507 {
 508         int ret;
 509         u64 pfn;
 510         if (!capable(CAP_SYS_ADMIN))
 511                 return -EPERM;
 512         if (kstrtoull(buf, 0, &pfn) < 0)
 513                 return -EINVAL;
 514         pfn >>= PAGE_SHIFT;
 515         ret = memory_failure(pfn, 0, 0);
 516         return ret ? ret : count;
 517 }
 518
 519 static DEVICE_ATTR(soft_offline_page, S_IWUSR, NULL, store_soft_offline_page);
 520 static DEVICE_ATTR(hard_offline_page, S_IWUSR, NULL, store_hard_offline_page);
 521 #endif
 522
 523 /*
 524  * Note that phys_device is optional.  It is here to allow for
 525  * differentiation between which *physical* devices each
 526  * section belongs to...
 527  */
 528 int __weak arch_get_memory_phys_device(unsigned long start_pfn)
 529 {
 530         return 0;
 531 }
 532
 533 /*
 534  * A reference for the returned object is held and the reference for the
 535  * hinted object is released.
 536  */
 537 struct memory_block *find_memory_block_hinted(struct mem_section *section,
 538                                               struct memory_block *hint)
 539 {
 540         int block_id = base_memory_block_id(__section_nr(section));
 541         struct device *hintdev = hint ? &hint->dev : NULL;
 542         struct device *dev;
 543
 544         dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev);
 545         if (hint)
 546                 put_device(&hint->dev);
 547         if (!dev)
 548                 return NULL;
 549         return to_memory_block(dev);
 550 }
 551
 552 /*
 553  * For now, we have a linear search to go find the appropriate
 554  * memory_block corresponding to a particular phys_index. If
 555  * this gets to be a real problem, we can always use a radix
 556  * tree or something here.
 557  *
 558  * This could be made generic for all device subsystems.
 559  */
 560 struct memory_block *find_memory_block(struct mem_section *section)
 561 {
 562         return find_memory_block_hinted(section, NULL);
 563 }
 564
 565 static struct attribute *memory_memblk_attrs[] = {
 566         &dev_attr_phys_index.attr,
 567         &dev_attr_state.attr,
 568         &dev_attr_phys_device.attr,
 569         &dev_attr_removable.attr,
 570 #ifdef CONFIG_MEMORY_HOTREMOVE
 571         &dev_attr_valid_zones.attr,
 572 #endif
 573         NULL
 574 };
 575
 576 static struct attribute_group memory_memblk_attr_group = {
 577         .attrs = memory_memblk_attrs,
 578 };
 579
 580 static const struct attribute_group *memory_memblk_attr_groups[] = {
 581         &memory_memblk_attr_group,
 582         NULL,
 583 };
 584
 585 /*
 586  * register_memory - Setup a sysfs device for a memory block
 587  */
 588 static
 589 int register_memory(struct memory_block *memory)
 590 {
 591         memory->dev.bus = &memory_subsys;
 592         memory->dev.id = memory->start_section_nr / sections_per_block;
 593         memory->dev.release = memory_block_release;
 594         memory->dev.groups = memory_memblk_attr_groups;
 595         memory->dev.offline = memory->state == MEM_OFFLINE;
 596
 597         return device_register(&memory->dev);
 598 }
 599
 600 static int init_memory_block(struct memory_block **memory,
 601                              struct mem_section *section, unsigned long state)
 602 {
 603         struct memory_block *mem;
 604         unsigned long start_pfn;
 605         int scn_nr;
 606         int ret = 0;
 607
 608         mem = kzalloc(sizeof(*mem), GFP_KERNEL);
 609         if (!mem)
 610                 return -ENOMEM;
 611
 612         scn_nr = __section_nr(section);
 613         mem->start_section_nr =
 614                         base_memory_block_id(scn_nr) * sections_per_block;
 615         mem->end_section_nr = mem->start_section_nr + sections_per_block - 1;
 616         mem->state = state;
 617         mem->section_count++;
 618         start_pfn = section_nr_to_pfn(mem->start_section_nr);
 619         mem->phys_device = arch_get_memory_phys_device(start_pfn);
 620
 621         ret = register_memory(mem);
 622
 623         *memory = mem;
 624         return ret;
 625 }
 626
 627 static int add_memory_block(int base_section_nr)
 628 {
 629         struct memory_block *mem;
 630         int i, ret, section_count = 0, section_nr;
 631
 632         for (i = base_section_nr;
 633              (i < base_section_nr + sections_per_block) && i < NR_MEM_SECTIONS;
 634              i++) {
 635                 if (!present_section_nr(i))
 636                         continue;
 637                 if (section_count == 0)
 638                         section_nr = i;
 639                 section_count++;
 640         }
 641
 642         if (section_count == 0)
 643                 return 0;
 644         ret = init_memory_block(&mem, __nr_to_section(section_nr), MEM_ONLINE);
 645         if (ret)
 646                 return ret;
 647         mem->section_count = section_count;
 648         return 0;
 649 }
 650
 651
 652 /*
 653  * need an interface for the VM to add new memory regions,
 654  * but without onlining it.
 655  */
 656 int register_new_memory(int nid, struct mem_section *section)
 657 {
 658         int ret = 0;
 659         struct memory_block *mem;
 660
 661         mutex_lock(&mem_sysfs_mutex);
 662
 663         mem = find_memory_block(section);
 664         if (mem) {
 665                 mem->section_count++;
 666                 put_device(&mem->dev);
 667         } else {
 668                 ret = init_memory_block(&mem, section, MEM_OFFLINE);
 669                 if (ret)
 670                         goto out;
 671         }
 672
 673         if (mem->section_count == sections_per_block)
 674                 ret = register_mem_sect_under_node(mem, nid);
 675 out:
 676         mutex_unlock(&mem_sysfs_mutex);
 677         return ret;
 678 }
 679
 680 #ifdef CONFIG_MEMORY_HOTREMOVE
 681 static void
 682 unregister_memory(struct memory_block *memory)
 683 {
 684         BUG_ON(memory->dev.bus != &memory_subsys);
 685
 686         /* drop the ref. we got in remove_memory_block() */
 687         put_device(&memory->dev);
 688         device_unregister(&memory->dev);
 689 }
 690
 691 static int remove_memory_block(unsigned long node_id,
 692                                struct mem_section *section, int phys_device)
 693 {
 694         struct memory_block *mem;
 695
 696         mutex_lock(&mem_sysfs_mutex);
 697         mem = find_memory_block(section);
 698         unregister_mem_sect_under_nodes(mem, __section_nr(section));
 699
 700         mem->section_count--;
 701         if (mem->section_count == 0)
 702                 unregister_memory(mem);
 703         else
 704                 put_device(&mem->dev);
 705
 706         mutex_unlock(&mem_sysfs_mutex);
 707         return 0;
 708 }
 709
 710 int unregister_memory_section(struct mem_section *section)
 711 {
 712         if (!present_section(section))
 713                 return -EINVAL;
 714
 715         return remove_memory_block(0, section, 0);
 716 }
 717 #endif /* CONFIG_MEMORY_HOTREMOVE */
 718
 719 /* return true if the memory block is offlined, otherwise, return false */
 720 bool is_memblock_offlined(struct memory_block *mem)
 721 {
 722         return mem->state == MEM_OFFLINE;
 723 }
 724
 725 static struct attribute *memory_root_attrs[] = {
 726 #ifdef CONFIG_ARCH_MEMORY_PROBE
 727         &dev_attr_probe.attr,
 728 #endif
 729
 730 #ifdef CONFIG_MEMORY_FAILURE
 731         &dev_attr_soft_offline_page.attr,
 732         &dev_attr_hard_offline_page.attr,
 733 #endif
 734
 735         &dev_attr_block_size_bytes.attr,
 736         NULL
 737 };
 738
 739 static struct attribute_group memory_root_attr_group = {
 740         .attrs = memory_root_attrs,
 741 };
 742
 743 static const struct attribute_group *memory_root_attr_groups[] = {
 744         &memory_root_attr_group,
 745         NULL,
 746 };
 747
 748 /*
 749  * Initialize the sysfs support for memory devices...
 750  */
 751 int __init memory_dev_init(void)
 752 {
 753         unsigned int i;
 754         int ret;
 755         int err;
 756         unsigned long block_sz;
 757
 758         ret = subsys_system_register(&memory_subsys, memory_root_attr_groups);
 759         if (ret)
 760                 goto out;
 761
 762         block_sz = get_memory_block_size();
 763         sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE;
 764
 765         /*
 766          * Create entries for memory sections that were found
 767          * during boot and have been initialized
 768          */
 769         mutex_lock(&mem_sysfs_mutex);
 770         for (i = 0; i < NR_MEM_SECTIONS; i += sections_per_block) {
 771                 err = add_memory_block(i);
 772                 if (!ret)
 773                         ret = err;
 774         }
 775         mutex_unlock(&mem_sysfs_mutex);
 776
 777 out:
 778         if (ret)
 779                 printk(KERN_ERR "%s() failed: %d\n", __func__, ret);
 780         return ret;
 781 }