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