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