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 <linux/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         if (mem->state != MEM_ONLINE)
 132                 goto out;
 133
 134         for (i = 0; i < sections_per_block; i++) {
 135                 if (!present_section_nr(mem->start_section_nr + i))
 136                         continue;
 137                 pfn = section_nr_to_pfn(mem->start_section_nr + i);
 138                 ret &= is_mem_section_removable(pfn, PAGES_PER_SECTION);
 139         }
 140
 141 out:
 142         return sprintf(buf, "%d\n", ret);
 143 }
 144
 145 /*
 146  * online, offline, going offline, etc.
 147  */
 148 static ssize_t show_mem_state(struct device *dev,
 149                         struct device_attribute *attr, char *buf)
 150 {
 151         struct memory_block *mem = to_memory_block(dev);
 152         ssize_t len = 0;
 153
 154         /*
 155          * We can probably put these states in a nice little array
 156          * so that they're not open-coded
 157          */
 158         switch (mem->state) {
 159         case MEM_ONLINE:
 160                 len = sprintf(buf, "online\n");
 161                 break;
 162         case MEM_OFFLINE:
 163                 len = sprintf(buf, "offline\n");
 164                 break;
 165         case MEM_GOING_OFFLINE:
 166                 len = sprintf(buf, "going-offline\n");
 167                 break;
 168         default:
 169                 len = sprintf(buf, "ERROR-UNKNOWN-%ld\n",
 170                                 mem->state);
 171                 WARN_ON(1);
 172                 break;
 173         }
 174
 175         return len;
 176 }
 177
 178 int memory_notify(unsigned long val, void *v)
 179 {
 180         return blocking_notifier_call_chain(&memory_chain, val, v);
 181 }
 182
 183 int memory_isolate_notify(unsigned long val, void *v)
 184 {
 185         return atomic_notifier_call_chain(&memory_isolate_chain, val, v);
 186 }
 187
 188 /*
 189  * The probe routines leave the pages reserved, just as the bootmem code does.
 190  * Make sure they're still that way.
 191  */
 192 static bool pages_correctly_reserved(unsigned long start_pfn)
 193 {
 194         int i, j;
 195         struct page *page;
 196         unsigned long pfn = start_pfn;
 197
 198         /*
 199          * memmap between sections is not contiguous except with
 200          * SPARSEMEM_VMEMMAP. We lookup the page once per section
 201          * and assume memmap is contiguous within each section
 202          */
 203         for (i = 0; i < sections_per_block; i++, pfn += PAGES_PER_SECTION) {
 204                 if (WARN_ON_ONCE(!pfn_valid(pfn)))
 205                         return false;
 206                 page = pfn_to_page(pfn);
 207
 208                 for (j = 0; j < PAGES_PER_SECTION; j++) {
 209                         if (PageReserved(page + j))
 210                                 continue;
 211
 212                         printk(KERN_WARNING "section number %ld page number %d "
 213                                 "not reserved, was it already online?\n",
 214                                 pfn_to_section_nr(pfn), j);
 215
 216                         return false;
 217                 }
 218         }
 219
 220         return true;
 221 }
 222
 223 /*
 224  * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
 225  * OK to have direct references to sparsemem variables in here.
 226  * Must already be protected by mem_hotplug_begin().
 227  */
 228 static int
 229 memory_block_action(unsigned long phys_index, unsigned long action, int online_type)
 230 {
 231         unsigned long start_pfn;
 232         unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
 233         int ret;
 234
 235         start_pfn = section_nr_to_pfn(phys_index);
 236
 237         switch (action) {
 238         case MEM_ONLINE:
 239                 if (!pages_correctly_reserved(start_pfn))
 240                         return -EBUSY;
 241
 242                 ret = online_pages(start_pfn, nr_pages, online_type);
 243                 break;
 244         case MEM_OFFLINE:
 245                 ret = offline_pages(start_pfn, nr_pages);
 246                 break;
 247         default:
 248                 WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: "
 249                      "%ld\n", __func__, phys_index, action, action);
 250                 ret = -EINVAL;
 251         }
 252
 253         return ret;
 254 }
 255
 256 static int memory_block_change_state(struct memory_block *mem,
 257                 unsigned long to_state, unsigned long from_state_req)
 258 {
 259         int ret = 0;
 260
 261         if (mem->state != from_state_req)
 262                 return -EINVAL;
 263
 264         if (to_state == MEM_OFFLINE)
 265                 mem->state = MEM_GOING_OFFLINE;
 266
 267         ret = memory_block_action(mem->start_section_nr, to_state,
 268                                 mem->online_type);
 269
 270         mem->state = ret ? from_state_req : to_state;
 271
 272         return ret;
 273 }
 274
 275 /* The device lock serializes operations on memory_subsys_[online|offline] */
 276 static int memory_subsys_online(struct device *dev)
 277 {
 278         struct memory_block *mem = to_memory_block(dev);
 279         int ret;
 280
 281         if (mem->state == MEM_ONLINE)
 282                 return 0;
 283
 284         /*
 285          * If we are called from store_mem_state(), online_type will be
 286          * set >= 0 Otherwise we were called from the device online
 287          * attribute and need to set the online_type.
 288          */
 289         if (mem->online_type < 0)
 290                 mem->online_type = MMOP_ONLINE_KEEP;
 291
 292         /* Already under protection of mem_hotplug_begin() */
 293         ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
 294
 295         /* clear online_type */
 296         mem->online_type = -1;
 297
 298         return ret;
 299 }
 300
 301 static int memory_subsys_offline(struct device *dev)
 302 {
 303         struct memory_block *mem = to_memory_block(dev);
 304
 305         if (mem->state == MEM_OFFLINE)
 306                 return 0;
 307
 308         /* Can't offline block with non-present sections */
 309         if (mem->section_count != sections_per_block)
 310                 return -EINVAL;
 311
 312         return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
 313 }
 314
 315 static ssize_t
 316 store_mem_state(struct device *dev,
 317                 struct device_attribute *attr, const char *buf, size_t count)
 318 {
 319         struct memory_block *mem = to_memory_block(dev);
 320         int ret, online_type;
 321
 322         ret = lock_device_hotplug_sysfs();
 323         if (ret)
 324                 return ret;
 325
 326         if (sysfs_streq(buf, "online_kernel"))
 327                 online_type = MMOP_ONLINE_KERNEL;
 328         else if (sysfs_streq(buf, "online_movable"))
 329                 online_type = MMOP_ONLINE_MOVABLE;
 330         else if (sysfs_streq(buf, "online"))
 331                 online_type = MMOP_ONLINE_KEEP;
 332         else if (sysfs_streq(buf, "offline"))
 333                 online_type = MMOP_OFFLINE;
 334         else {
 335                 ret = -EINVAL;
 336                 goto err;
 337         }
 338
 339         /*
 340          * Memory hotplug needs to hold mem_hotplug_begin() for probe to find
 341          * the correct memory block to online before doing device_online(dev),
 342          * which will take dev->mutex.  Take the lock early to prevent an
 343          * inversion, memory_subsys_online() callbacks will be implemented by
 344          * assuming it's already protected.
 345          */
 346         mem_hotplug_begin();
 347
 348         switch (online_type) {
 349         case MMOP_ONLINE_KERNEL:
 350         case MMOP_ONLINE_MOVABLE:
 351         case MMOP_ONLINE_KEEP:
 352                 mem->online_type = online_type;
 353                 ret = device_online(&mem->dev);
 354                 break;
 355         case MMOP_OFFLINE:
 356                 ret = device_offline(&mem->dev);
 357                 break;
 358         default:
 359                 ret = -EINVAL; /* should never happen */
 360         }
 361
 362         mem_hotplug_done();
 363 err:
 364         unlock_device_hotplug();
 365
 366         if (ret < 0)
 367                 return ret;
 368         if (ret)
 369                 return -EINVAL;
 370
 371         return count;
 372 }
 373
 374 /*
 375  * phys_device is a bad name for this.  What I really want
 376  * is a way to differentiate between memory ranges that
 377  * are part of physical devices that constitute
 378  * a complete removable unit or fru.
 379  * i.e. do these ranges belong to the same physical device,
 380  * s.t. if I offline all of these sections I can then
 381  * remove the physical device?
 382  */
 383 static ssize_t show_phys_device(struct device *dev,
 384                                 struct device_attribute *attr, char *buf)
 385 {
 386         struct memory_block *mem = to_memory_block(dev);
 387         return sprintf(buf, "%d\n", mem->phys_device);
 388 }
 389
 390 #ifdef CONFIG_MEMORY_HOTREMOVE
 391 static ssize_t show_valid_zones(struct device *dev,
 392                                 struct device_attribute *attr, char *buf)
 393 {
 394         struct memory_block *mem = to_memory_block(dev);
 395         unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr);
 396         unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
 397         unsigned long valid_start_pfn, valid_end_pfn;
 398         bool append = false;
 399         int nid;
 400
 401         /*
 402          * The block contains more than one zone can not be offlined.
 403          * This can happen e.g. for ZONE_DMA and ZONE_DMA32
 404          */
 405         if (!test_pages_in_a_zone(start_pfn, start_pfn + nr_pages, &valid_start_pfn, &valid_end_pfn))
 406                 return sprintf(buf, "none\n");
 407
 408         start_pfn = valid_start_pfn;
 409         nr_pages = valid_end_pfn - start_pfn;
 410
 411         /*
 412          * Check the existing zone. Make sure that we do that only on the
 413          * online nodes otherwise the page_zone is not reliable
 414          */
 415         if (mem->state == MEM_ONLINE) {
 416                 strcat(buf, page_zone(pfn_to_page(start_pfn))->name);
 417                 goto out;
 418         }
 419
 420         nid = pfn_to_nid(start_pfn);
 421         if (allow_online_pfn_range(nid, start_pfn, nr_pages, MMOP_ONLINE_KERNEL)) {
 422                 strcat(buf, default_zone_for_pfn(nid, start_pfn, nr_pages)->name);
 423                 append = true;
 424         }
 425
 426         if (allow_online_pfn_range(nid, start_pfn, nr_pages, MMOP_ONLINE_MOVABLE)) {
 427                 if (append)
 428                         strcat(buf, " ");
 429                 strcat(buf, NODE_DATA(nid)->node_zones[ZONE_MOVABLE].name);
 430         }
 431 out:
 432         strcat(buf, "\n");
 433
 434         return strlen(buf);
 435 }
 436 static DEVICE_ATTR(valid_zones, 0444, show_valid_zones, NULL);
 437 #endif
 438
 439 static DEVICE_ATTR(phys_index, 0444, show_mem_start_phys_index, NULL);
 440 static DEVICE_ATTR(state, 0644, show_mem_state, store_mem_state);
 441 static DEVICE_ATTR(phys_device, 0444, show_phys_device, NULL);
 442 static DEVICE_ATTR(removable, 0444, show_mem_removable, NULL);
 443
 444 /*
 445  * Block size attribute stuff
 446  */
 447 static ssize_t
 448 print_block_size(struct device *dev, struct device_attribute *attr,
 449                  char *buf)
 450 {
 451         return sprintf(buf, "%lx\n", get_memory_block_size());
 452 }
 453
 454 static DEVICE_ATTR(block_size_bytes, 0444, print_block_size, NULL);
 455
 456 /*
 457  * Memory auto online policy.
 458  */
 459
 460 static ssize_t
 461 show_auto_online_blocks(struct device *dev, struct device_attribute *attr,
 462                         char *buf)
 463 {
 464         if (memhp_auto_online)
 465                 return sprintf(buf, "online\n");
 466         else
 467                 return sprintf(buf, "offline\n");
 468 }
 469
 470 static ssize_t
 471 store_auto_online_blocks(struct device *dev, struct device_attribute *attr,
 472                          const char *buf, size_t count)
 473 {
 474         if (sysfs_streq(buf, "online"))
 475                 memhp_auto_online = true;
 476         else if (sysfs_streq(buf, "offline"))
 477                 memhp_auto_online = false;
 478         else
 479                 return -EINVAL;
 480
 481         return count;
 482 }
 483
 484 static DEVICE_ATTR(auto_online_blocks, 0644, show_auto_online_blocks,
 485                    store_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
 495 memory_probe_store(struct device *dev, struct device_attribute *attr,
 496                    const char *buf, size_t count)
 497 {
 498         u64 phys_addr;
 499         int nid, ret;
 500         unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block;
 501
 502         ret = kstrtoull(buf, 0, &phys_addr);
 503         if (ret)
 504                 return ret;
 505
 506         if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1))
 507                 return -EINVAL;
 508
 509         nid = memory_add_physaddr_to_nid(phys_addr);
 510         ret = add_memory(nid, phys_addr,
 511                          MIN_MEMORY_BLOCK_SIZE * sections_per_block);
 512
 513         if (ret)
 514                 goto out;
 515
 516         ret = count;
 517 out:
 518         return ret;
 519 }
 520
 521 static DEVICE_ATTR(probe, S_IWUSR, NULL, memory_probe_store);
 522 #endif
 523
 524 #ifdef CONFIG_MEMORY_FAILURE
 525 /*
 526  * Support for offlining pages of memory
 527  */
 528
 529 /* Soft offline a page */
 530 static ssize_t
 531 store_soft_offline_page(struct device *dev,
 532                         struct device_attribute *attr,
 533                         const char *buf, size_t count)
 534 {
 535         int ret;
 536         u64 pfn;
 537         if (!capable(CAP_SYS_ADMIN))
 538                 return -EPERM;
 539         if (kstrtoull(buf, 0, &pfn) < 0)
 540                 return -EINVAL;
 541         pfn >>= PAGE_SHIFT;
 542         if (!pfn_valid(pfn))
 543                 return -ENXIO;
 544         ret = soft_offline_page(pfn_to_page(pfn), 0);
 545         return ret == 0 ? count : ret;
 546 }
 547
 548 /* Forcibly offline a page, including killing processes. */
 549 static ssize_t
 550 store_hard_offline_page(struct device *dev,
 551                         struct device_attribute *attr,
 552                         const char *buf, size_t count)
 553 {
 554         int ret;
 555         u64 pfn;
 556         if (!capable(CAP_SYS_ADMIN))
 557                 return -EPERM;
 558         if (kstrtoull(buf, 0, &pfn) < 0)
 559                 return -EINVAL;
 560         pfn >>= PAGE_SHIFT;
 561         ret = memory_failure(pfn, 0, 0);
 562         return ret ? ret : count;
 563 }
 564
 565 static DEVICE_ATTR(soft_offline_page, S_IWUSR, NULL, store_soft_offline_page);
 566 static DEVICE_ATTR(hard_offline_page, S_IWUSR, NULL, store_hard_offline_page);
 567 #endif
 568
 569 /*
 570  * Note that phys_device is optional.  It is here to allow for
 571  * differentiation between which *physical* devices each
 572  * section belongs to...
 573  */
 574 int __weak arch_get_memory_phys_device(unsigned long start_pfn)
 575 {
 576         return 0;
 577 }
 578
 579 /*
 580  * A reference for the returned object is held and the reference for the
 581  * hinted object is released.
 582  */
 583 struct memory_block *find_memory_block_hinted(struct mem_section *section,
 584                                               struct memory_block *hint)
 585 {
 586         int block_id = base_memory_block_id(__section_nr(section));
 587         struct device *hintdev = hint ? &hint->dev : NULL;
 588         struct device *dev;
 589
 590         dev = subsys_find_device_by_id(&memory_subsys, block_id, hintdev);
 591         if (hint)
 592                 put_device(&hint->dev);
 593         if (!dev)
 594                 return NULL;
 595         return to_memory_block(dev);
 596 }
 597
 598 /*
 599  * For now, we have a linear search to go find the appropriate
 600  * memory_block corresponding to a particular phys_index. If
 601  * this gets to be a real problem, we can always use a radix
 602  * tree or something here.
 603  *
 604  * This could be made generic for all device subsystems.
 605  */
 606 struct memory_block *find_memory_block(struct mem_section *section)
 607 {
 608         return find_memory_block_hinted(section, NULL);
 609 }
 610
 611 static struct attribute *memory_memblk_attrs[] = {
 612         &dev_attr_phys_index.attr,
 613         &dev_attr_state.attr,
 614         &dev_attr_phys_device.attr,
 615         &dev_attr_removable.attr,
 616 #ifdef CONFIG_MEMORY_HOTREMOVE
 617         &dev_attr_valid_zones.attr,
 618 #endif
 619         NULL
 620 };
 621
 622 static struct attribute_group memory_memblk_attr_group = {
 623         .attrs = memory_memblk_attrs,
 624 };
 625
 626 static const struct attribute_group *memory_memblk_attr_groups[] = {
 627         &memory_memblk_attr_group,
 628         NULL,
 629 };
 630
 631 /*
 632  * register_memory - Setup a sysfs device for a memory block
 633  */
 634 static
 635 int register_memory(struct memory_block *memory)
 636 {
 637         memory->dev.bus = &memory_subsys;
 638         memory->dev.id = memory->start_section_nr / sections_per_block;
 639         memory->dev.release = memory_block_release;
 640         memory->dev.groups = memory_memblk_attr_groups;
 641         memory->dev.offline = memory->state == MEM_OFFLINE;
 642
 643         return device_register(&memory->dev);
 644 }
 645
 646 static int init_memory_block(struct memory_block **memory,
 647                              struct mem_section *section, unsigned long state)
 648 {
 649         struct memory_block *mem;
 650         unsigned long start_pfn;
 651         int scn_nr;
 652         int ret = 0;
 653
 654         mem = kzalloc(sizeof(*mem), GFP_KERNEL);
 655         if (!mem)
 656                 return -ENOMEM;
 657
 658         scn_nr = __section_nr(section);
 659         mem->start_section_nr =
 660                         base_memory_block_id(scn_nr) * sections_per_block;
 661         mem->end_section_nr = mem->start_section_nr + sections_per_block - 1;
 662         mem->state = state;
 663         start_pfn = section_nr_to_pfn(mem->start_section_nr);
 664         mem->phys_device = arch_get_memory_phys_device(start_pfn);
 665
 666         ret = register_memory(mem);
 667
 668         *memory = mem;
 669         return ret;
 670 }
 671
 672 static int add_memory_block(int base_section_nr)
 673 {
 674         struct memory_block *mem;
 675         int i, ret, section_count = 0, section_nr;
 676
 677         for (i = base_section_nr;
 678              (i < base_section_nr + sections_per_block) && i < NR_MEM_SECTIONS;
 679              i++) {
 680                 if (!present_section_nr(i))
 681                         continue;
 682                 if (section_count == 0)
 683                         section_nr = i;
 684                 section_count++;
 685         }
 686
 687         if (section_count == 0)
 688                 return 0;
 689         ret = init_memory_block(&mem, __nr_to_section(section_nr), MEM_ONLINE);
 690         if (ret)
 691                 return ret;
 692         mem->section_count = section_count;
 693         return 0;
 694 }
 695
 696 /*
 697  * need an interface for the VM to add new memory regions,
 698  * but without onlining it.
 699  */
 700 int register_new_memory(int nid, struct mem_section *section)
 701 {
 702         int ret = 0;
 703         struct memory_block *mem;
 704
 705         mutex_lock(&mem_sysfs_mutex);
 706
 707         mem = find_memory_block(section);
 708         if (mem) {
 709                 mem->section_count++;
 710                 put_device(&mem->dev);
 711         } else {
 712                 ret = init_memory_block(&mem, section, MEM_OFFLINE);
 713                 if (ret)
 714                         goto out;
 715                 mem->section_count++;
 716         }
 717
 718         if (mem->section_count == sections_per_block)
 719                 ret = register_mem_sect_under_node(mem, nid);
 720 out:
 721         mutex_unlock(&mem_sysfs_mutex);
 722         return ret;
 723 }
 724
 725 #ifdef CONFIG_MEMORY_HOTREMOVE
 726 static void
 727 unregister_memory(struct memory_block *memory)
 728 {
 729         BUG_ON(memory->dev.bus != &memory_subsys);
 730
 731         /* drop the ref. we got in remove_memory_block() */
 732         put_device(&memory->dev);
 733         device_unregister(&memory->dev);
 734 }
 735
 736 static int remove_memory_section(unsigned long node_id,
 737                                struct mem_section *section, int phys_device)
 738 {
 739         struct memory_block *mem;
 740
 741         mutex_lock(&mem_sysfs_mutex);
 742
 743         /*
 744          * Some users of the memory hotplug do not want/need memblock to
 745          * track all sections. Skip over those.
 746          */
 747         mem = find_memory_block(section);
 748         if (!mem)
 749                 goto out_unlock;
 750
 751         unregister_mem_sect_under_nodes(mem, __section_nr(section));
 752
 753         mem->section_count--;
 754         if (mem->section_count == 0)
 755                 unregister_memory(mem);
 756         else
 757                 put_device(&mem->dev);
 758
 759 out_unlock:
 760         mutex_unlock(&mem_sysfs_mutex);
 761         return 0;
 762 }
 763
 764 int unregister_memory_section(struct mem_section *section)
 765 {
 766         if (!present_section(section))
 767                 return -EINVAL;
 768
 769         return remove_memory_section(0, section, 0);
 770 }
 771 #endif /* CONFIG_MEMORY_HOTREMOVE */
 772
 773 /* return true if the memory block is offlined, otherwise, return false */
 774 bool is_memblock_offlined(struct memory_block *mem)
 775 {
 776         return mem->state == MEM_OFFLINE;
 777 }
 778
 779 static struct attribute *memory_root_attrs[] = {
 780 #ifdef CONFIG_ARCH_MEMORY_PROBE
 781         &dev_attr_probe.attr,
 782 #endif
 783
 784 #ifdef CONFIG_MEMORY_FAILURE
 785         &dev_attr_soft_offline_page.attr,
 786         &dev_attr_hard_offline_page.attr,
 787 #endif
 788
 789         &dev_attr_block_size_bytes.attr,
 790         &dev_attr_auto_online_blocks.attr,
 791         NULL
 792 };
 793
 794 static struct attribute_group memory_root_attr_group = {
 795         .attrs = memory_root_attrs,
 796 };
 797
 798 static const struct attribute_group *memory_root_attr_groups[] = {
 799         &memory_root_attr_group,
 800         NULL,
 801 };
 802
 803 /*
 804  * Initialize the sysfs support for memory devices...
 805  */
 806 int __init memory_dev_init(void)
 807 {
 808         unsigned int i;
 809         int ret;
 810         int err;
 811         unsigned long block_sz;
 812
 813         ret = subsys_system_register(&memory_subsys, memory_root_attr_groups);
 814         if (ret)
 815                 goto out;
 816
 817         block_sz = get_memory_block_size();
 818         sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE;
 819
 820         /*
 821          * Create entries for memory sections that were found
 822          * during boot and have been initialized
 823          */
 824         mutex_lock(&mem_sysfs_mutex);
 825         for (i = 0; i < NR_MEM_SECTIONS; i += sections_per_block) {
 826                 /* Don't iterate over sections we know are !present: */
 827                 if (i > __highest_present_section_nr)
 828                         break;
 829
 830                 err = add_memory_block(i);
 831                 if (!ret)
 832                         ret = err;
 833         }
 834         mutex_unlock(&mem_sysfs_mutex);
 835
 836 out:
 837         if (ret)
 838                 printk(KERN_ERR "%s() failed: %d\n", __func__, ret);
 839         return ret;
 840 }