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 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         /* Can't offline block with non-present sections */
 307         if (mem->section_count != sections_per_block)
 308                 return -EINVAL;
 309
 310         return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
 311 }
 312
 313 static ssize_t
 314 store_mem_state(struct device *dev,
 315                 struct device_attribute *attr, const char *buf, size_t count)
 316 {
 317         struct memory_block *mem = to_memory_block(dev);
 318         int ret, online_type;
 319
 320         ret = lock_device_hotplug_sysfs();
 321         if (ret)
 322                 return ret;
 323
 324         if (sysfs_streq(buf, "online_kernel"))
 325                 online_type = MMOP_ONLINE_KERNEL;
 326         else if (sysfs_streq(buf, "online_movable"))
 327                 online_type = MMOP_ONLINE_MOVABLE;
 328         else if (sysfs_streq(buf, "online"))
 329                 online_type = MMOP_ONLINE_KEEP;
 330         else if (sysfs_streq(buf, "offline"))
 331                 online_type = MMOP_OFFLINE;
 332         else {
 333                 ret = -EINVAL;
 334                 goto err;
 335         }
 336
 337         /*
 338          * Memory hotplug needs to hold mem_hotplug_begin() for probe to find
 339          * the correct memory block to online before doing device_online(dev),
 340          * which will take dev->mutex.  Take the lock early to prevent an
 341          * inversion, memory_subsys_online() callbacks will be implemented by
 342          * assuming it's already protected.
 343          */
 344         mem_hotplug_begin();
 345
 346         switch (online_type) {
 347         case MMOP_ONLINE_KERNEL:
 348         case MMOP_ONLINE_MOVABLE:
 349         case MMOP_ONLINE_KEEP:
 350                 mem->online_type = online_type;
 351                 ret = device_online(&mem->dev);
 352                 break;
 353         case MMOP_OFFLINE:
 354                 ret = device_offline(&mem->dev);
 355                 break;
 356         default:
 357                 ret = -EINVAL; /* should never happen */
 358         }
 359
 360         mem_hotplug_done();
 361 err:
 362         unlock_device_hotplug();
 363
 364         if (ret)
 365                 return ret;
 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 show_phys_device(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 ssize_t show_valid_zones(struct device *dev,
 387                                 struct device_attribute *attr, char *buf)
 388 {
 389         struct memory_block *mem = to_memory_block(dev);
 390         unsigned long start_pfn, end_pfn;
 391         unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
 392         struct page *first_page;
 393         struct zone *zone;
 394         int zone_shift = 0;
 395
 396         start_pfn = section_nr_to_pfn(mem->start_section_nr);
 397         end_pfn = start_pfn + nr_pages;
 398         first_page = pfn_to_page(start_pfn);
 399
 400         /* The block contains more than one zone can not be offlined. */
 401         if (!test_pages_in_a_zone(start_pfn, end_pfn))
 402                 return sprintf(buf, "none\n");
 403
 404         zone = page_zone(first_page);
 405
 406         /* MMOP_ONLINE_KEEP */
 407         sprintf(buf, "%s", zone->name);
 408
 409         /* MMOP_ONLINE_KERNEL */
 410         zone_shift = zone_can_shift(start_pfn, nr_pages, ZONE_NORMAL);
 411         if (zone_shift) {
 412                 strcat(buf, " ");
 413                 strcat(buf, (zone + zone_shift)->name);
 414         }
 415
 416         /* MMOP_ONLINE_MOVABLE */
 417         zone_shift = zone_can_shift(start_pfn, nr_pages, ZONE_MOVABLE);
 418         if (zone_shift) {
 419                 strcat(buf, " ");
 420                 strcat(buf, (zone + zone_shift)->name);
 421         }
 422
 423         strcat(buf, "\n");
 424
 425         return strlen(buf);
 426 }
 427 static DEVICE_ATTR(valid_zones, 0444, show_valid_zones, NULL);
 428 #endif
 429
 430 static DEVICE_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL);
 431 static DEVICE_ATTR(state, 0644, show_mem_state, store_mem_state);
 432 static DEVICE_ATTR(phys_device, 0444, show_phys_device, NULL);
 433 static DEVICE_ATTR(removable, 0444, show_mem_removable, NULL);
 434
 435 /*
 436  * Block size attribute stuff
 437  */
 438 static ssize_t
 439 print_block_size(struct device *dev, struct device_attribute *attr,
 440                  char *buf)
 441 {
 442         return sprintf(buf, "%lx\n", get_memory_block_size());
 443 }
 444
 445 static DEVICE_ATTR(block_size_bytes, 0444, print_block_size, NULL);
 446
 447 /*
 448  * Memory auto online policy.
 449  */
 450
 451 static ssize_t
 452 show_auto_online_blocks(struct device *dev, struct device_attribute *attr,
 453                         char *buf)
 454 {
 455         if (memhp_auto_online)
 456                 return sprintf(buf, "online\n");
 457         else
 458                 return sprintf(buf, "offline\n");
 459 }
 460
 461 static ssize_t
 462 store_auto_online_blocks(struct device *dev, struct device_attribute *attr,
 463                          const char *buf, size_t count)
 464 {
 465         if (sysfs_streq(buf, "online"))
 466                 memhp_auto_online = true;
 467         else if (sysfs_streq(buf, "offline"))
 468                 memhp_auto_online = false;
 469         else
 470                 return -EINVAL;
 471
 472         return count;
 473 }
 474
 475 static DEVICE_ATTR(auto_online_blocks, 0644, show_auto_online_blocks,
 476                    store_auto_online_blocks);
 477
 478 /*
 479  * Some architectures will have custom drivers to do this, and
 480  * will not need to do it from userspace.  The fake hot-add code
 481  * as well as ppc64 will do all of their discovery in userspace
 482  * and will require this interface.
 483  */
 484 #ifdef CONFIG_ARCH_MEMORY_PROBE
 485 static ssize_t
 486 memory_probe_store(struct device *dev, struct device_attribute *attr,
 487                    const char *buf, size_t count)
 488 {
 489         u64 phys_addr;
 490         int nid, ret;
 491         unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block;
 492
 493         ret = kstrtoull(buf, 0, &phys_addr);
 494         if (ret)
 495                 return ret;
 496
 497         if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1))
 498                 return -EINVAL;
 499
 500         nid = memory_add_physaddr_to_nid(phys_addr);
 501         ret = add_memory(nid, phys_addr,
 502                          MIN_MEMORY_BLOCK_SIZE * sections_per_block);
 503
 504         if (ret)
 505                 goto out;
 506
 507         ret = count;
 508 out:
 509         return ret;
 510 }
 511
 512 static DEVICE_ATTR(probe, S_IWUSR, NULL, memory_probe_store);
 513 #endif
 514
 515 #ifdef CONFIG_MEMORY_FAILURE
 516 /*
 517  * Support for offlining pages of memory
 518  */
 519
 520 /* Soft offline a page */
 521 static ssize_t
 522 store_soft_offline_page(struct device *dev,
 523                         struct device_attribute *attr,
 524                         const char *buf, size_t count)
 525 {
 526         int ret;
 527         u64 pfn;
 528         if (!capable(CAP_SYS_ADMIN))
 529                 return -EPERM;
 530         if (kstrtoull(buf, 0, &pfn) < 0)
 531                 return -EINVAL;
 532         pfn >>= PAGE_SHIFT;
 533         if (!pfn_valid(pfn))
 534                 return -ENXIO;
 535         ret = soft_offline_page(pfn_to_page(pfn), 0);
 536         return ret == 0 ? count : ret;
 537 }
 538
 539 /* Forcibly offline a page, including killing processes. */
 540 static ssize_t
 541 store_hard_offline_page(struct device *dev,
 542                         struct device_attribute *attr,
 543                         const char *buf, size_t count)
 544 {
 545         int ret;
 546         u64 pfn;
 547         if (!capable(CAP_SYS_ADMIN))
 548                 return -EPERM;
 549         if (kstrtoull(buf, 0, &pfn) < 0)
 550                 return -EINVAL;
 551         pfn >>= PAGE_SHIFT;
 552         ret = memory_failure(pfn, 0, 0);
 553         return ret ? ret : count;
 554 }
 555
 556 static DEVICE_ATTR(soft_offline_page, S_IWUSR, NULL, store_soft_offline_page);
 557 static DEVICE_ATTR(hard_offline_page, S_IWUSR, NULL, store_hard_offline_page);
 558 #endif
 559
 560 /*
 561  * Note that phys_device is optional.  It is here to allow for
 562  * differentiation between which *physical* devices each
 563  * section belongs to...
 564  */
 565 int __weak arch_get_memory_phys_device(unsigned long start_pfn)
 566 {
 567         return 0;
 568 }
 569
 570 /*
 571  * A reference for the returned object is held and the reference for the
 572  * hinted object is released.
 573  */
 574 struct memory_block *find_memory_block_hinted(struct mem_section *section,
 575                                               struct memory_block *hint)
 576 {
 577         int block_id = base_memory_block_id(__section_nr(section));
 578         struct device *hintdev = hint ? &hint->dev : NULL;
 579         struct device *dev;
 580
 581         dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev);
 582         if (hint)
 583                 put_device(&hint->dev);
 584         if (!dev)
 585                 return NULL;
 586         return to_memory_block(dev);
 587 }
 588
 589 /*
 590  * For now, we have a linear search to go find the appropriate
 591  * memory_block corresponding to a particular phys_index. If
 592  * this gets to be a real problem, we can always use a radix
 593  * tree or something here.
 594  *
 595  * This could be made generic for all device subsystems.
 596  */
 597 struct memory_block *find_memory_block(struct mem_section *section)
 598 {
 599         return find_memory_block_hinted(section, NULL);
 600 }
 601
 602 static struct attribute *memory_memblk_attrs[] = {
 603         &dev_attr_phys_index.attr,
 604         &dev_attr_state.attr,
 605         &dev_attr_phys_device.attr,
 606         &dev_attr_removable.attr,
 607 #ifdef CONFIG_MEMORY_HOTREMOVE
 608         &dev_attr_valid_zones.attr,
 609 #endif
 610         NULL
 611 };
 612
 613 static struct attribute_group memory_memblk_attr_group = {
 614         .attrs = memory_memblk_attrs,
 615 };
 616
 617 static const struct attribute_group *memory_memblk_attr_groups[] = {
 618         &memory_memblk_attr_group,
 619         NULL,
 620 };
 621
 622 /*
 623  * register_memory - Setup a sysfs device for a memory block
 624  */
 625 static
 626 int register_memory(struct memory_block *memory)
 627 {
 628         memory->dev.bus = &memory_subsys;
 629         memory->dev.id = memory->start_section_nr / sections_per_block;
 630         memory->dev.release = memory_block_release;
 631         memory->dev.groups = memory_memblk_attr_groups;
 632         memory->dev.offline = memory->state == MEM_OFFLINE;
 633
 634         return device_register(&memory->dev);
 635 }
 636
 637 static int init_memory_block(struct memory_block **memory,
 638                              struct mem_section *section, unsigned long state)
 639 {
 640         struct memory_block *mem;
 641         unsigned long start_pfn;
 642         int scn_nr;
 643         int ret = 0;
 644
 645         mem = kzalloc(sizeof(*mem), GFP_KERNEL);
 646         if (!mem)
 647                 return -ENOMEM;
 648
 649         scn_nr = __section_nr(section);
 650         mem->start_section_nr =
 651                         base_memory_block_id(scn_nr) * sections_per_block;
 652         mem->end_section_nr = mem->start_section_nr + sections_per_block - 1;
 653         mem->state = state;
 654         start_pfn = section_nr_to_pfn(mem->start_section_nr);
 655         mem->phys_device = arch_get_memory_phys_device(start_pfn);
 656
 657         ret = register_memory(mem);
 658
 659         *memory = mem;
 660         return ret;
 661 }
 662
 663 static int add_memory_block(int base_section_nr)
 664 {
 665         struct memory_block *mem;
 666         int i, ret, section_count = 0, section_nr;
 667
 668         for (i = base_section_nr;
 669              (i < base_section_nr + sections_per_block) && i < NR_MEM_SECTIONS;
 670              i++) {
 671                 if (!present_section_nr(i))
 672                         continue;
 673                 if (section_count == 0)
 674                         section_nr = i;
 675                 section_count++;
 676         }
 677
 678         if (section_count == 0)
 679                 return 0;
 680         ret = init_memory_block(&mem, __nr_to_section(section_nr), MEM_ONLINE);
 681         if (ret)
 682                 return ret;
 683         mem->section_count = section_count;
 684         return 0;
 685 }
 686
 687 static bool is_zone_device_section(struct mem_section *ms)
 688 {
 689         struct page *page;
 690
 691         page = sparse_decode_mem_map(ms->section_mem_map, __section_nr(ms));
 692         return is_zone_device_page(page);
 693 }
 694
 695 /*
 696  * need an interface for the VM to add new memory regions,
 697  * but without onlining it.
 698  */
 699 int register_new_memory(int nid, struct mem_section *section)
 700 {
 701         int ret = 0;
 702         struct memory_block *mem;
 703
 704         if (is_zone_device_section(section))
 705                 return 0;
 706
 707         mutex_lock(&mem_sysfs_mutex);
 708
 709         mem = find_memory_block(section);
 710         if (mem) {
 711                 mem->section_count++;
 712                 put_device(&mem->dev);
 713         } else {
 714                 ret = init_memory_block(&mem, section, MEM_OFFLINE);
 715                 if (ret)
 716                         goto out;
 717                 mem->section_count++;
 718         }
 719
 720         if (mem->section_count == sections_per_block)
 721                 ret = register_mem_sect_under_node(mem, nid);
 722 out:
 723         mutex_unlock(&mem_sysfs_mutex);
 724         return ret;
 725 }
 726
 727 #ifdef CONFIG_MEMORY_HOTREMOVE
 728 static void
 729 unregister_memory(struct memory_block *memory)
 730 {
 731         BUG_ON(memory->dev.bus != &memory_subsys);
 732
 733         /* drop the ref. we got in remove_memory_block() */
 734         put_device(&memory->dev);
 735         device_unregister(&memory->dev);
 736 }
 737
 738 static int remove_memory_section(unsigned long node_id,
 739                                struct mem_section *section, int phys_device)
 740 {
 741         struct memory_block *mem;
 742
 743         if (is_zone_device_section(section))
 744                 return 0;
 745
 746         mutex_lock(&mem_sysfs_mutex);
 747         mem = find_memory_block(section);
 748         unregister_mem_sect_under_nodes(mem, __section_nr(section));
 749
 750         mem->section_count--;
 751         if (mem->section_count == 0)
 752                 unregister_memory(mem);
 753         else
 754                 put_device(&mem->dev);
 755
 756         mutex_unlock(&mem_sysfs_mutex);
 757         return 0;
 758 }
 759
 760 int unregister_memory_section(struct mem_section *section)
 761 {
 762         if (!present_section(section))
 763                 return -EINVAL;
 764
 765         return remove_memory_section(0, section, 0);
 766 }
 767 #endif /* CONFIG_MEMORY_HOTREMOVE */
 768
 769 /* return true if the memory block is offlined, otherwise, return false */
 770 bool is_memblock_offlined(struct memory_block *mem)
 771 {
 772         return mem->state == MEM_OFFLINE;
 773 }
 774
 775 static struct attribute *memory_root_attrs[] = {
 776 #ifdef CONFIG_ARCH_MEMORY_PROBE
 777         &dev_attr_probe.attr,
 778 #endif
 779
 780 #ifdef CONFIG_MEMORY_FAILURE
 781         &dev_attr_soft_offline_page.attr,
 782         &dev_attr_hard_offline_page.attr,
 783 #endif
 784
 785         &dev_attr_block_size_bytes.attr,
 786         &dev_attr_auto_online_blocks.attr,
 787         NULL
 788 };
 789
 790 static struct attribute_group memory_root_attr_group = {
 791         .attrs = memory_root_attrs,
 792 };
 793
 794 static const struct attribute_group *memory_root_attr_groups[] = {
 795         &memory_root_attr_group,
 796         NULL,
 797 };
 798
 799 /*
 800  * Initialize the sysfs support for memory devices...
 801  */
 802 int __init memory_dev_init(void)
 803 {
 804         unsigned int i;
 805         int ret;
 806         int err;
 807         unsigned long block_sz;
 808
 809         ret = subsys_system_register(&memory_subsys, memory_root_attr_groups);
 810         if (ret)
 811                 goto out;
 812
 813         block_sz = get_memory_block_size();
 814         sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE;
 815
 816         /*
 817          * Create entries for memory sections that were found
 818          * during boot and have been initialized
 819          */
 820         mutex_lock(&mem_sysfs_mutex);
 821         for (i = 0; i < NR_MEM_SECTIONS; i += sections_per_block) {
 822                 err = add_memory_block(i);
 823                 if (!ret)
 824                         ret = err;
 825         }
 826         mutex_unlock(&mem_sysfs_mutex);
 827
 828 out:
 829         if (ret)
 830                 printk(KERN_ERR "%s() failed: %d\n", __func__, ret);
 831         return ret;
 832 }