drivers/base/memory.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Memory subsystem support
   4  *
   5  * Written by Matt Tolentino <matthew.e.tolentino@intel.com>
   6  *            Dave Hansen <haveblue@us.ibm.com>
   7  *
   8  * This file provides the necessary infrastructure to represent
   9  * a SPARSEMEM-memory-model system's physical memory in /sysfs.
  10  * All arch-independent code that assumes MEMORY_HOTPLUG requires
  11  * SPARSEMEM should be contained here, or in mm/memory_hotplug.c.
  12  */
  13
  14 #include <linux/module.h>
  15 #include <linux/init.h>
  16 #include <linux/topology.h>
  17 #include <linux/capability.h>
  18 #include <linux/device.h>
  19 #include <linux/memory.h>
  20 #include <linux/memory_hotplug.h>
  21 #include <linux/mm.h>
  22 #include <linux/stat.h>
  23 #include <linux/slab.h>
  24 #include <linux/xarray.h>
  25
  26 #include <linux/atomic.h>
  27 #include <linux/uaccess.h>
  28
  29 #define MEMORY_CLASS_NAME       "memory"
  30
  31 static const char *const online_type_to_str[] = {
  32         [MMOP_OFFLINE] = "offline",
  33         [MMOP_ONLINE] = "online",
  34         [MMOP_ONLINE_KERNEL] = "online_kernel",
  35         [MMOP_ONLINE_MOVABLE] = "online_movable",
  36 };
  37
  38 int memhp_online_type_from_str(const char *str)
  39 {
  40         int i;
  41
  42         for (i = 0; i < ARRAY_SIZE(online_type_to_str); i++) {
  43                 if (sysfs_streq(str, online_type_to_str[i]))
  44                         return i;
  45         }
  46         return -EINVAL;
  47 }
  48
  49 #define to_memory_block(dev) container_of(dev, struct memory_block, dev)
  50
  51 static int sections_per_block;
  52
  53 static inline unsigned long base_memory_block_id(unsigned long section_nr)
  54 {
  55         return section_nr / sections_per_block;
  56 }
  57
  58 static inline unsigned long pfn_to_block_id(unsigned long pfn)
  59 {
  60         return base_memory_block_id(pfn_to_section_nr(pfn));
  61 }
  62
  63 static inline unsigned long phys_to_block_id(unsigned long phys)
  64 {
  65         return pfn_to_block_id(PFN_DOWN(phys));
  66 }
  67
  68 static int memory_subsys_online(struct device *dev);
  69 static int memory_subsys_offline(struct device *dev);
  70
  71 static struct bus_type memory_subsys = {
  72         .name = MEMORY_CLASS_NAME,
  73         .dev_name = MEMORY_CLASS_NAME,
  74         .online = memory_subsys_online,
  75         .offline = memory_subsys_offline,
  76 };
  77
  78 /*
  79  * Memory blocks are cached in a local radix tree to avoid
  80  * a costly linear search for the corresponding device on
  81  * the subsystem bus.
  82  */
  83 static DEFINE_XARRAY(memory_blocks);
  84
  85 static BLOCKING_NOTIFIER_HEAD(memory_chain);
  86
  87 int register_memory_notifier(struct notifier_block *nb)
  88 {
  89         return blocking_notifier_chain_register(&memory_chain, nb);
  90 }
  91 EXPORT_SYMBOL(register_memory_notifier);
  92
  93 void unregister_memory_notifier(struct notifier_block *nb)
  94 {
  95         blocking_notifier_chain_unregister(&memory_chain, nb);
  96 }
  97 EXPORT_SYMBOL(unregister_memory_notifier);
  98
  99 static void memory_block_release(struct device *dev)
 100 {
 101         struct memory_block *mem = to_memory_block(dev);
 102
 103         kfree(mem);
 104 }
 105
 106 unsigned long __weak memory_block_size_bytes(void)
 107 {
 108         return MIN_MEMORY_BLOCK_SIZE;
 109 }
 110 EXPORT_SYMBOL_GPL(memory_block_size_bytes);
 111
 112 /*
 113  * Show the first physical section index (number) of this memory block.
 114  */
 115 static ssize_t phys_index_show(struct device *dev,
 116                                struct device_attribute *attr, char *buf)
 117 {
 118         struct memory_block *mem = to_memory_block(dev);
 119         unsigned long phys_index;
 120
 121         phys_index = mem->start_section_nr / sections_per_block;
 122         return sprintf(buf, "%08lx\n", phys_index);
 123 }
 124
 125 /*
 126  * Legacy interface that we cannot remove. Always indicate "removable"
 127  * with CONFIG_MEMORY_HOTREMOVE - bad heuristic.
 128  */
 129 static ssize_t removable_show(struct device *dev, struct device_attribute *attr,
 130                               char *buf)
 131 {
 132         return sprintf(buf, "%d\n", (int)IS_ENABLED(CONFIG_MEMORY_HOTREMOVE));
 133 }
 134
 135 /*
 136  * online, offline, going offline, etc.
 137  */
 138 static ssize_t state_show(struct device *dev, struct device_attribute *attr,
 139                           char *buf)
 140 {
 141         struct memory_block *mem = to_memory_block(dev);
 142         ssize_t len = 0;
 143
 144         /*
 145          * We can probably put these states in a nice little array
 146          * so that they're not open-coded
 147          */
 148         switch (mem->state) {
 149         case MEM_ONLINE:
 150                 len = sprintf(buf, "online\n");
 151                 break;
 152         case MEM_OFFLINE:
 153                 len = sprintf(buf, "offline\n");
 154                 break;
 155         case MEM_GOING_OFFLINE:
 156                 len = sprintf(buf, "going-offline\n");
 157                 break;
 158         default:
 159                 len = sprintf(buf, "ERROR-UNKNOWN-%ld\n",
 160                                 mem->state);
 161                 WARN_ON(1);
 162                 break;
 163         }
 164
 165         return len;
 166 }
 167
 168 int memory_notify(unsigned long val, void *v)
 169 {
 170         return blocking_notifier_call_chain(&memory_chain, val, v);
 171 }
 172
 173 /*
 174  * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
 175  * OK to have direct references to sparsemem variables in here.
 176  */
 177 static int
 178 memory_block_action(unsigned long start_section_nr, unsigned long action,
 179                     int online_type, int nid)
 180 {
 181         unsigned long start_pfn;
 182         unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
 183         int ret;
 184
 185         start_pfn = section_nr_to_pfn(start_section_nr);
 186
 187         switch (action) {
 188         case MEM_ONLINE:
 189                 ret = online_pages(start_pfn, nr_pages, online_type, nid);
 190                 break;
 191         case MEM_OFFLINE:
 192                 ret = offline_pages(start_pfn, nr_pages);
 193                 break;
 194         default:
 195                 WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: "
 196                      "%ld\n", __func__, start_section_nr, action, action);
 197                 ret = -EINVAL;
 198         }
 199
 200         return ret;
 201 }
 202
 203 static int memory_block_change_state(struct memory_block *mem,
 204                 unsigned long to_state, unsigned long from_state_req)
 205 {
 206         int ret = 0;
 207
 208         if (mem->state != from_state_req)
 209                 return -EINVAL;
 210
 211         if (to_state == MEM_OFFLINE)
 212                 mem->state = MEM_GOING_OFFLINE;
 213
 214         ret = memory_block_action(mem->start_section_nr, to_state,
 215                                   mem->online_type, mem->nid);
 216
 217         mem->state = ret ? from_state_req : to_state;
 218
 219         return ret;
 220 }
 221
 222 /* The device lock serializes operations on memory_subsys_[online|offline] */
 223 static int memory_subsys_online(struct device *dev)
 224 {
 225         struct memory_block *mem = to_memory_block(dev);
 226         int ret;
 227
 228         if (mem->state == MEM_ONLINE)
 229                 return 0;
 230
 231         /*
 232          * When called via device_online() without configuring the online_type,
 233          * we want to default to MMOP_ONLINE.
 234          */
 235         if (mem->online_type == MMOP_OFFLINE)
 236                 mem->online_type = MMOP_ONLINE;
 237
 238         ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
 239         mem->online_type = MMOP_OFFLINE;
 240
 241         return ret;
 242 }
 243
 244 static int memory_subsys_offline(struct device *dev)
 245 {
 246         struct memory_block *mem = to_memory_block(dev);
 247
 248         if (mem->state == MEM_OFFLINE)
 249                 return 0;
 250
 251         return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
 252 }
 253
 254 static ssize_t state_store(struct device *dev, struct device_attribute *attr,
 255                            const char *buf, size_t count)
 256 {
 257         const int online_type = memhp_online_type_from_str(buf);
 258         struct memory_block *mem = to_memory_block(dev);
 259         int ret;
 260
 261         if (online_type < 0)
 262                 return -EINVAL;
 263
 264         ret = lock_device_hotplug_sysfs();
 265         if (ret)
 266                 return ret;
 267
 268         switch (online_type) {
 269         case MMOP_ONLINE_KERNEL:
 270         case MMOP_ONLINE_MOVABLE:
 271         case MMOP_ONLINE:
 272                 /* mem->online_type is protected by device_hotplug_lock */
 273                 mem->online_type = online_type;
 274                 ret = device_online(&mem->dev);
 275                 break;
 276         case MMOP_OFFLINE:
 277                 ret = device_offline(&mem->dev);
 278                 break;
 279         default:
 280                 ret = -EINVAL; /* should never happen */
 281         }
 282
 283         unlock_device_hotplug();
 284
 285         if (ret < 0)
 286                 return ret;
 287         if (ret)
 288                 return -EINVAL;
 289
 290         return count;
 291 }
 292
 293 /*
 294  * phys_device is a bad name for this.  What I really want
 295  * is a way to differentiate between memory ranges that
 296  * are part of physical devices that constitute
 297  * a complete removable unit or fru.
 298  * i.e. do these ranges belong to the same physical device,
 299  * s.t. if I offline all of these sections I can then
 300  * remove the physical device?
 301  */
 302 static ssize_t phys_device_show(struct device *dev,
 303                                 struct device_attribute *attr, char *buf)
 304 {
 305         struct memory_block *mem = to_memory_block(dev);
 306         return sprintf(buf, "%d\n", mem->phys_device);
 307 }
 308
 309 #ifdef CONFIG_MEMORY_HOTREMOVE
 310 static void print_allowed_zone(char *buf, int nid, unsigned long start_pfn,
 311                 unsigned long nr_pages, int online_type,
 312                 struct zone *default_zone)
 313 {
 314         struct zone *zone;
 315
 316         zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
 317         if (zone != default_zone) {
 318                 strcat(buf, " ");
 319                 strcat(buf, zone->name);
 320         }
 321 }
 322
 323 static ssize_t valid_zones_show(struct device *dev,
 324                                 struct device_attribute *attr, char *buf)
 325 {
 326         struct memory_block *mem = to_memory_block(dev);
 327         unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr);
 328         unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
 329         struct zone *default_zone;
 330         int nid;
 331
 332         /*
 333          * Check the existing zone. Make sure that we do that only on the
 334          * online nodes otherwise the page_zone is not reliable
 335          */
 336         if (mem->state == MEM_ONLINE) {
 337                 /*
 338                  * The block contains more than one zone can not be offlined.
 339                  * This can happen e.g. for ZONE_DMA and ZONE_DMA32
 340                  */
 341                 default_zone = test_pages_in_a_zone(start_pfn,
 342                                                     start_pfn + nr_pages);
 343                 if (!default_zone)
 344                         return sprintf(buf, "none\n");
 345                 strcat(buf, default_zone->name);
 346                 goto out;
 347         }
 348
 349         nid = mem->nid;
 350         default_zone = zone_for_pfn_range(MMOP_ONLINE, nid, start_pfn,
 351                                           nr_pages);
 352         strcat(buf, default_zone->name);
 353
 354         print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_KERNEL,
 355                         default_zone);
 356         print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_MOVABLE,
 357                         default_zone);
 358 out:
 359         strcat(buf, "\n");
 360
 361         return strlen(buf);
 362 }
 363 static DEVICE_ATTR_RO(valid_zones);
 364 #endif
 365
 366 static DEVICE_ATTR_RO(phys_index);
 367 static DEVICE_ATTR_RW(state);
 368 static DEVICE_ATTR_RO(phys_device);
 369 static DEVICE_ATTR_RO(removable);
 370
 371 /*
 372  * Show the memory block size (shared by all memory blocks).
 373  */
 374 static ssize_t block_size_bytes_show(struct device *dev,
 375                                      struct device_attribute *attr, char *buf)
 376 {
 377         return sprintf(buf, "%lx\n", memory_block_size_bytes());
 378 }
 379
 380 static DEVICE_ATTR_RO(block_size_bytes);
 381
 382 /*
 383  * Memory auto online policy.
 384  */
 385
 386 static ssize_t auto_online_blocks_show(struct device *dev,
 387                                        struct device_attribute *attr, char *buf)
 388 {
 389         return sprintf(buf, "%s\n",
 390                        online_type_to_str[memhp_default_online_type]);
 391 }
 392
 393 static ssize_t auto_online_blocks_store(struct device *dev,
 394                                         struct device_attribute *attr,
 395                                         const char *buf, size_t count)
 396 {
 397         const int online_type = memhp_online_type_from_str(buf);
 398
 399         if (online_type < 0)
 400                 return -EINVAL;
 401
 402         memhp_default_online_type = online_type;
 403         return count;
 404 }
 405
 406 static DEVICE_ATTR_RW(auto_online_blocks);
 407
 408 /*
 409  * Some architectures will have custom drivers to do this, and
 410  * will not need to do it from userspace.  The fake hot-add code
 411  * as well as ppc64 will do all of their discovery in userspace
 412  * and will require this interface.
 413  */
 414 #ifdef CONFIG_ARCH_MEMORY_PROBE
 415 static ssize_t probe_store(struct device *dev, struct device_attribute *attr,
 416                            const char *buf, size_t count)
 417 {
 418         u64 phys_addr;
 419         int nid, ret;
 420         unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block;
 421
 422         ret = kstrtoull(buf, 0, &phys_addr);
 423         if (ret)
 424                 return ret;
 425
 426         if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1))
 427                 return -EINVAL;
 428
 429         ret = lock_device_hotplug_sysfs();
 430         if (ret)
 431                 return ret;
 432
 433         nid = memory_add_physaddr_to_nid(phys_addr);
 434         ret = __add_memory(nid, phys_addr,
 435                            MIN_MEMORY_BLOCK_SIZE * sections_per_block);
 436
 437         if (ret)
 438                 goto out;
 439
 440         ret = count;
 441 out:
 442         unlock_device_hotplug();
 443         return ret;
 444 }
 445
 446 static DEVICE_ATTR_WO(probe);
 447 #endif
 448
 449 #ifdef CONFIG_MEMORY_FAILURE
 450 /*
 451  * Support for offlining pages of memory
 452  */
 453
 454 /* Soft offline a page */
 455 static ssize_t soft_offline_page_store(struct device *dev,
 456                                        struct device_attribute *attr,
 457                                        const char *buf, size_t count)
 458 {
 459         int ret;
 460         u64 pfn;
 461         if (!capable(CAP_SYS_ADMIN))
 462                 return -EPERM;
 463         if (kstrtoull(buf, 0, &pfn) < 0)
 464                 return -EINVAL;
 465         pfn >>= PAGE_SHIFT;
 466         ret = soft_offline_page(pfn, 0);
 467         return ret == 0 ? count : ret;
 468 }
 469
 470 /* Forcibly offline a page, including killing processes. */
 471 static ssize_t hard_offline_page_store(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 (kstrtoull(buf, 0, &pfn) < 0)
 480                 return -EINVAL;
 481         pfn >>= PAGE_SHIFT;
 482         ret = memory_failure(pfn, 0);
 483         return ret ? ret : count;
 484 }
 485
 486 static DEVICE_ATTR_WO(soft_offline_page);
 487 static DEVICE_ATTR_WO(hard_offline_page);
 488 #endif
 489
 490 /*
 491  * Note that phys_device is optional.  It is here to allow for
 492  * differentiation between which *physical* devices each
 493  * section belongs to...
 494  */
 495 int __weak arch_get_memory_phys_device(unsigned long start_pfn)
 496 {
 497         return 0;
 498 }
 499
 500 /*
 501  * A reference for the returned memory block device is acquired.
 502  *
 503  * Called under device_hotplug_lock.
 504  */
 505 static struct memory_block *find_memory_block_by_id(unsigned long block_id)
 506 {
 507         struct memory_block *mem;
 508
 509         mem = xa_load(&memory_blocks, block_id);
 510         if (mem)
 511                 get_device(&mem->dev);
 512         return mem;
 513 }
 514
 515 /*
 516  * Called under device_hotplug_lock.
 517  */
 518 struct memory_block *find_memory_block(struct mem_section *section)
 519 {
 520         unsigned long block_id = base_memory_block_id(__section_nr(section));
 521
 522         return find_memory_block_by_id(block_id);
 523 }
 524
 525 static struct attribute *memory_memblk_attrs[] = {
 526         &dev_attr_phys_index.attr,
 527         &dev_attr_state.attr,
 528         &dev_attr_phys_device.attr,
 529         &dev_attr_removable.attr,
 530 #ifdef CONFIG_MEMORY_HOTREMOVE
 531         &dev_attr_valid_zones.attr,
 532 #endif
 533         NULL
 534 };
 535
 536 static struct attribute_group memory_memblk_attr_group = {
 537         .attrs = memory_memblk_attrs,
 538 };
 539
 540 static const struct attribute_group *memory_memblk_attr_groups[] = {
 541         &memory_memblk_attr_group,
 542         NULL,
 543 };
 544
 545 /*
 546  * register_memory - Setup a sysfs device for a memory block
 547  */
 548 static
 549 int register_memory(struct memory_block *memory)
 550 {
 551         int ret;
 552
 553         memory->dev.bus = &memory_subsys;
 554         memory->dev.id = memory->start_section_nr / sections_per_block;
 555         memory->dev.release = memory_block_release;
 556         memory->dev.groups = memory_memblk_attr_groups;
 557         memory->dev.offline = memory->state == MEM_OFFLINE;
 558
 559         ret = device_register(&memory->dev);
 560         if (ret) {
 561                 put_device(&memory->dev);
 562                 return ret;
 563         }
 564         ret = xa_err(xa_store(&memory_blocks, memory->dev.id, memory,
 565                               GFP_KERNEL));
 566         if (ret) {
 567                 put_device(&memory->dev);
 568                 device_unregister(&memory->dev);
 569         }
 570         return ret;
 571 }
 572
 573 static int init_memory_block(struct memory_block **memory,
 574                              unsigned long block_id, unsigned long state)
 575 {
 576         struct memory_block *mem;
 577         unsigned long start_pfn;
 578         int ret = 0;
 579
 580         mem = find_memory_block_by_id(block_id);
 581         if (mem) {
 582                 put_device(&mem->dev);
 583                 return -EEXIST;
 584         }
 585         mem = kzalloc(sizeof(*mem), GFP_KERNEL);
 586         if (!mem)
 587                 return -ENOMEM;
 588
 589         mem->start_section_nr = block_id * sections_per_block;
 590         mem->state = state;
 591         start_pfn = section_nr_to_pfn(mem->start_section_nr);
 592         mem->phys_device = arch_get_memory_phys_device(start_pfn);
 593         mem->nid = NUMA_NO_NODE;
 594
 595         ret = register_memory(mem);
 596
 597         *memory = mem;
 598         return ret;
 599 }
 600
 601 static int add_memory_block(unsigned long base_section_nr)
 602 {
 603         int section_count = 0;
 604         struct memory_block *mem;
 605         unsigned long nr;
 606
 607         for (nr = base_section_nr; nr < base_section_nr + sections_per_block;
 608              nr++)
 609                 if (present_section_nr(nr))
 610                         section_count++;
 611
 612         if (section_count == 0)
 613                 return 0;
 614         return init_memory_block(&mem, base_memory_block_id(base_section_nr),
 615                                  MEM_ONLINE);
 616 }
 617
 618 static void unregister_memory(struct memory_block *memory)
 619 {
 620         if (WARN_ON_ONCE(memory->dev.bus != &memory_subsys))
 621                 return;
 622
 623         WARN_ON(xa_erase(&memory_blocks, memory->dev.id) == NULL);
 624
 625         /* drop the ref. we got via find_memory_block() */
 626         put_device(&memory->dev);
 627         device_unregister(&memory->dev);
 628 }
 629
 630 /*
 631  * Create memory block devices for the given memory area. Start and size
 632  * have to be aligned to memory block granularity. Memory block devices
 633  * will be initialized as offline.
 634  *
 635  * Called under device_hotplug_lock.
 636  */
 637 int create_memory_block_devices(unsigned long start, unsigned long size)
 638 {
 639         const unsigned long start_block_id = pfn_to_block_id(PFN_DOWN(start));
 640         unsigned long end_block_id = pfn_to_block_id(PFN_DOWN(start + size));
 641         struct memory_block *mem;
 642         unsigned long block_id;
 643         int ret = 0;
 644
 645         if (WARN_ON_ONCE(!IS_ALIGNED(start, memory_block_size_bytes()) ||
 646                          !IS_ALIGNED(size, memory_block_size_bytes())))
 647                 return -EINVAL;
 648
 649         for (block_id = start_block_id; block_id != end_block_id; block_id++) {
 650                 ret = init_memory_block(&mem, block_id, MEM_OFFLINE);
 651                 if (ret)
 652                         break;
 653         }
 654         if (ret) {
 655                 end_block_id = block_id;
 656                 for (block_id = start_block_id; block_id != end_block_id;
 657                      block_id++) {
 658                         mem = find_memory_block_by_id(block_id);
 659                         if (WARN_ON_ONCE(!mem))
 660                                 continue;
 661                         unregister_memory(mem);
 662                 }
 663         }
 664         return ret;
 665 }
 666
 667 /*
 668  * Remove memory block devices for the given memory area. Start and size
 669  * have to be aligned to memory block granularity. Memory block devices
 670  * have to be offline.
 671  *
 672  * Called under device_hotplug_lock.
 673  */
 674 void remove_memory_block_devices(unsigned long start, unsigned long size)
 675 {
 676         const unsigned long start_block_id = pfn_to_block_id(PFN_DOWN(start));
 677         const unsigned long end_block_id = pfn_to_block_id(PFN_DOWN(start + size));
 678         struct memory_block *mem;
 679         unsigned long block_id;
 680
 681         if (WARN_ON_ONCE(!IS_ALIGNED(start, memory_block_size_bytes()) ||
 682                          !IS_ALIGNED(size, memory_block_size_bytes())))
 683                 return;
 684
 685         for (block_id = start_block_id; block_id != end_block_id; block_id++) {
 686                 mem = find_memory_block_by_id(block_id);
 687                 if (WARN_ON_ONCE(!mem))
 688                         continue;
 689                 unregister_memory_block_under_nodes(mem);
 690                 unregister_memory(mem);
 691         }
 692 }
 693
 694 /* return true if the memory block is offlined, otherwise, return false */
 695 bool is_memblock_offlined(struct memory_block *mem)
 696 {
 697         return mem->state == MEM_OFFLINE;
 698 }
 699
 700 static struct attribute *memory_root_attrs[] = {
 701 #ifdef CONFIG_ARCH_MEMORY_PROBE
 702         &dev_attr_probe.attr,
 703 #endif
 704
 705 #ifdef CONFIG_MEMORY_FAILURE
 706         &dev_attr_soft_offline_page.attr,
 707         &dev_attr_hard_offline_page.attr,
 708 #endif
 709
 710         &dev_attr_block_size_bytes.attr,
 711         &dev_attr_auto_online_blocks.attr,
 712         NULL
 713 };
 714
 715 static struct attribute_group memory_root_attr_group = {
 716         .attrs = memory_root_attrs,
 717 };
 718
 719 static const struct attribute_group *memory_root_attr_groups[] = {
 720         &memory_root_attr_group,
 721         NULL,
 722 };
 723
 724 /*
 725  * Initialize the sysfs support for memory devices. At the time this function
 726  * is called, we cannot have concurrent creation/deletion of memory block
 727  * devices, the device_hotplug_lock is not needed.
 728  */
 729 void __init memory_dev_init(void)
 730 {
 731         int ret;
 732         unsigned long block_sz, nr;
 733
 734         /* Validate the configured memory block size */
 735         block_sz = memory_block_size_bytes();
 736         if (!is_power_of_2(block_sz) || block_sz < MIN_MEMORY_BLOCK_SIZE)
 737                 panic("Memory block size not suitable: 0x%lx\n", block_sz);
 738         sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE;
 739
 740         ret = subsys_system_register(&memory_subsys, memory_root_attr_groups);
 741         if (ret)
 742                 panic("%s() failed to register subsystem: %d\n", __func__, ret);
 743
 744         /*
 745          * Create entries for memory sections that were found
 746          * during boot and have been initialized
 747          */
 748         for (nr = 0; nr <= __highest_present_section_nr;
 749              nr += sections_per_block) {
 750                 ret = add_memory_block(nr);
 751                 if (ret)
 752                         panic("%s() failed to add memory block: %d\n", __func__,
 753                               ret);
 754         }
 755 }
 756
 757 /**
 758  * walk_memory_blocks - walk through all present memory blocks overlapped
 759  *                      by the range [start, start + size)
 760  *
 761  * @start: start address of the memory range
 762  * @size: size of the memory range
 763  * @arg: argument passed to func
 764  * @func: callback for each memory section walked
 765  *
 766  * This function walks through all present memory blocks overlapped by the
 767  * range [start, start + size), calling func on each memory block.
 768  *
 769  * In case func() returns an error, walking is aborted and the error is
 770  * returned.
 771  *
 772  * Called under device_hotplug_lock.
 773  */
 774 int walk_memory_blocks(unsigned long start, unsigned long size,
 775                        void *arg, walk_memory_blocks_func_t func)
 776 {
 777         const unsigned long start_block_id = phys_to_block_id(start);
 778         const unsigned long end_block_id = phys_to_block_id(start + size - 1);
 779         struct memory_block *mem;
 780         unsigned long block_id;
 781         int ret = 0;
 782
 783         if (!size)
 784                 return 0;
 785
 786         for (block_id = start_block_id; block_id <= end_block_id; block_id++) {
 787                 mem = find_memory_block_by_id(block_id);
 788                 if (!mem)
 789                         continue;
 790
 791                 ret = func(mem, arg);
 792                 put_device(&mem->dev);
 793                 if (ret)
 794                         break;
 795         }
 796         return ret;
 797 }
 798
 799 struct for_each_memory_block_cb_data {
 800         walk_memory_blocks_func_t func;
 801         void *arg;
 802 };
 803
 804 static int for_each_memory_block_cb(struct device *dev, void *data)
 805 {
 806         struct memory_block *mem = to_memory_block(dev);
 807         struct for_each_memory_block_cb_data *cb_data = data;
 808
 809         return cb_data->func(mem, cb_data->arg);
 810 }
 811
 812 /**
 813  * for_each_memory_block - walk through all present memory blocks
 814  *
 815  * @arg: argument passed to func
 816  * @func: callback for each memory block walked
 817  *
 818  * This function walks through all present memory blocks, calling func on
 819  * each memory block.
 820  *
 821  * In case func() returns an error, walking is aborted and the error is
 822  * returned.
 823  */
 824 int for_each_memory_block(void *arg, walk_memory_blocks_func_t func)
 825 {
 826         struct for_each_memory_block_cb_data cb_data = {
 827                 .func = func,
 828                 .arg = arg,
 829         };
 830
 831         return bus_for_each_dev(&memory_subsys, NULL, &cb_data,
 832                                 for_each_memory_block_cb);
 833 }