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
  25 #include <linux/atomic.h>
  26 #include <linux/uaccess.h>
  27
  28 #define MEMORY_CLASS_NAME       "memory"
  29
  30 static const char *const online_type_to_str[] = {
  31         [MMOP_OFFLINE] = "offline",
  32         [MMOP_ONLINE] = "online",
  33         [MMOP_ONLINE_KERNEL] = "online_kernel",
  34         [MMOP_ONLINE_MOVABLE] = "online_movable",
  35 };
  36
  37 int memhp_online_type_from_str(const char *str)
  38 {
  39         int i;
  40
  41         for (i = 0; i < ARRAY_SIZE(online_type_to_str); i++) {
  42                 if (sysfs_streq(str, online_type_to_str[i]))
  43                         return i;
  44         }
  45         return -EINVAL;
  46 }
  47
  48 #define to_memory_block(dev) container_of(dev, struct memory_block, dev)
  49
  50 static int sections_per_block;
  51
  52 static inline unsigned long base_memory_block_id(unsigned long section_nr)
  53 {
  54         return section_nr / sections_per_block;
  55 }
  56
  57 static inline unsigned long pfn_to_block_id(unsigned long pfn)
  58 {
  59         return base_memory_block_id(pfn_to_section_nr(pfn));
  60 }
  61
  62 static inline unsigned long phys_to_block_id(unsigned long phys)
  63 {
  64         return pfn_to_block_id(PFN_DOWN(phys));
  65 }
  66
  67 static int memory_subsys_online(struct device *dev);
  68 static int memory_subsys_offline(struct device *dev);
  69
  70 static struct bus_type memory_subsys = {
  71         .name = MEMORY_CLASS_NAME,
  72         .dev_name = MEMORY_CLASS_NAME,
  73         .online = memory_subsys_online,
  74         .offline = memory_subsys_offline,
  75 };
  76
  77 static BLOCKING_NOTIFIER_HEAD(memory_chain);
  78
  79 int register_memory_notifier(struct notifier_block *nb)
  80 {
  81         return blocking_notifier_chain_register(&memory_chain, nb);
  82 }
  83 EXPORT_SYMBOL(register_memory_notifier);
  84
  85 void unregister_memory_notifier(struct notifier_block *nb)
  86 {
  87         blocking_notifier_chain_unregister(&memory_chain, nb);
  88 }
  89 EXPORT_SYMBOL(unregister_memory_notifier);
  90
  91 static void memory_block_release(struct device *dev)
  92 {
  93         struct memory_block *mem = to_memory_block(dev);
  94
  95         kfree(mem);
  96 }
  97
  98 unsigned long __weak memory_block_size_bytes(void)
  99 {
 100         return MIN_MEMORY_BLOCK_SIZE;
 101 }
 102 EXPORT_SYMBOL_GPL(memory_block_size_bytes);
 103
 104 /*
 105  * Show the first physical section index (number) of this memory block.
 106  */
 107 static ssize_t phys_index_show(struct device *dev,
 108                                struct device_attribute *attr, char *buf)
 109 {
 110         struct memory_block *mem = to_memory_block(dev);
 111         unsigned long phys_index;
 112
 113         phys_index = mem->start_section_nr / sections_per_block;
 114         return sprintf(buf, "%08lx\n", phys_index);
 115 }
 116
 117 /*
 118  * Legacy interface that we cannot remove. Always indicate "removable"
 119  * with CONFIG_MEMORY_HOTREMOVE - bad heuristic.
 120  */
 121 static ssize_t removable_show(struct device *dev, struct device_attribute *attr,
 122                               char *buf)
 123 {
 124         return sprintf(buf, "%d\n", (int)IS_ENABLED(CONFIG_MEMORY_HOTREMOVE));
 125 }
 126
 127 /*
 128  * online, offline, going offline, etc.
 129  */
 130 static ssize_t state_show(struct device *dev, struct device_attribute *attr,
 131                           char *buf)
 132 {
 133         struct memory_block *mem = to_memory_block(dev);
 134         ssize_t len = 0;
 135
 136         /*
 137          * We can probably put these states in a nice little array
 138          * so that they're not open-coded
 139          */
 140         switch (mem->state) {
 141         case MEM_ONLINE:
 142                 len = sprintf(buf, "online\n");
 143                 break;
 144         case MEM_OFFLINE:
 145                 len = sprintf(buf, "offline\n");
 146                 break;
 147         case MEM_GOING_OFFLINE:
 148                 len = sprintf(buf, "going-offline\n");
 149                 break;
 150         default:
 151                 len = sprintf(buf, "ERROR-UNKNOWN-%ld\n",
 152                                 mem->state);
 153                 WARN_ON(1);
 154                 break;
 155         }
 156
 157         return len;
 158 }
 159
 160 int memory_notify(unsigned long val, void *v)
 161 {
 162         return blocking_notifier_call_chain(&memory_chain, val, v);
 163 }
 164
 165 /*
 166  * MEMORY_HOTPLUG depends on SPARSEMEM in mm/Kconfig, so it is
 167  * OK to have direct references to sparsemem variables in here.
 168  */
 169 static int
 170 memory_block_action(unsigned long start_section_nr, unsigned long action,
 171                     int online_type, int nid)
 172 {
 173         unsigned long start_pfn;
 174         unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
 175         int ret;
 176
 177         start_pfn = section_nr_to_pfn(start_section_nr);
 178
 179         switch (action) {
 180         case MEM_ONLINE:
 181                 ret = online_pages(start_pfn, nr_pages, online_type, nid);
 182                 break;
 183         case MEM_OFFLINE:
 184                 ret = offline_pages(start_pfn, nr_pages);
 185                 break;
 186         default:
 187                 WARN(1, KERN_WARNING "%s(%ld, %ld) unknown action: "
 188                      "%ld\n", __func__, start_section_nr, action, action);
 189                 ret = -EINVAL;
 190         }
 191
 192         return ret;
 193 }
 194
 195 static int memory_block_change_state(struct memory_block *mem,
 196                 unsigned long to_state, unsigned long from_state_req)
 197 {
 198         int ret = 0;
 199
 200         if (mem->state != from_state_req)
 201                 return -EINVAL;
 202
 203         if (to_state == MEM_OFFLINE)
 204                 mem->state = MEM_GOING_OFFLINE;
 205
 206         ret = memory_block_action(mem->start_section_nr, to_state,
 207                                   mem->online_type, mem->nid);
 208
 209         mem->state = ret ? from_state_req : to_state;
 210
 211         return ret;
 212 }
 213
 214 /* The device lock serializes operations on memory_subsys_[online|offline] */
 215 static int memory_subsys_online(struct device *dev)
 216 {
 217         struct memory_block *mem = to_memory_block(dev);
 218         int ret;
 219
 220         if (mem->state == MEM_ONLINE)
 221                 return 0;
 222
 223         /*
 224          * When called via device_online() without configuring the online_type,
 225          * we want to default to MMOP_ONLINE.
 226          */
 227         if (mem->online_type == MMOP_OFFLINE)
 228                 mem->online_type = MMOP_ONLINE;
 229
 230         ret = memory_block_change_state(mem, MEM_ONLINE, MEM_OFFLINE);
 231         mem->online_type = MMOP_OFFLINE;
 232
 233         return ret;
 234 }
 235
 236 static int memory_subsys_offline(struct device *dev)
 237 {
 238         struct memory_block *mem = to_memory_block(dev);
 239
 240         if (mem->state == MEM_OFFLINE)
 241                 return 0;
 242
 243         return memory_block_change_state(mem, MEM_OFFLINE, MEM_ONLINE);
 244 }
 245
 246 static ssize_t state_store(struct device *dev, struct device_attribute *attr,
 247                            const char *buf, size_t count)
 248 {
 249         const int online_type = memhp_online_type_from_str(buf);
 250         struct memory_block *mem = to_memory_block(dev);
 251         int ret;
 252
 253         if (online_type < 0)
 254                 return -EINVAL;
 255
 256         ret = lock_device_hotplug_sysfs();
 257         if (ret)
 258                 return ret;
 259
 260         switch (online_type) {
 261         case MMOP_ONLINE_KERNEL:
 262         case MMOP_ONLINE_MOVABLE:
 263         case MMOP_ONLINE:
 264                 /* mem->online_type is protected by device_hotplug_lock */
 265                 mem->online_type = online_type;
 266                 ret = device_online(&mem->dev);
 267                 break;
 268         case MMOP_OFFLINE:
 269                 ret = device_offline(&mem->dev);
 270                 break;
 271         default:
 272                 ret = -EINVAL; /* should never happen */
 273         }
 274
 275         unlock_device_hotplug();
 276
 277         if (ret < 0)
 278                 return ret;
 279         if (ret)
 280                 return -EINVAL;
 281
 282         return count;
 283 }
 284
 285 /*
 286  * phys_device is a bad name for this.  What I really want
 287  * is a way to differentiate between memory ranges that
 288  * are part of physical devices that constitute
 289  * a complete removable unit or fru.
 290  * i.e. do these ranges belong to the same physical device,
 291  * s.t. if I offline all of these sections I can then
 292  * remove the physical device?
 293  */
 294 static ssize_t phys_device_show(struct device *dev,
 295                                 struct device_attribute *attr, char *buf)
 296 {
 297         struct memory_block *mem = to_memory_block(dev);
 298         return sprintf(buf, "%d\n", mem->phys_device);
 299 }
 300
 301 #ifdef CONFIG_MEMORY_HOTREMOVE
 302 static void print_allowed_zone(char *buf, int nid, unsigned long start_pfn,
 303                 unsigned long nr_pages, int online_type,
 304                 struct zone *default_zone)
 305 {
 306         struct zone *zone;
 307
 308         zone = zone_for_pfn_range(online_type, nid, start_pfn, nr_pages);
 309         if (zone != default_zone) {
 310                 strcat(buf, " ");
 311                 strcat(buf, zone->name);
 312         }
 313 }
 314
 315 static ssize_t valid_zones_show(struct device *dev,
 316                                 struct device_attribute *attr, char *buf)
 317 {
 318         struct memory_block *mem = to_memory_block(dev);
 319         unsigned long start_pfn = section_nr_to_pfn(mem->start_section_nr);
 320         unsigned long nr_pages = PAGES_PER_SECTION * sections_per_block;
 321         struct zone *default_zone;
 322         int nid;
 323
 324         /*
 325          * Check the existing zone. Make sure that we do that only on the
 326          * online nodes otherwise the page_zone is not reliable
 327          */
 328         if (mem->state == MEM_ONLINE) {
 329                 /*
 330                  * The block contains more than one zone can not be offlined.
 331                  * This can happen e.g. for ZONE_DMA and ZONE_DMA32
 332                  */
 333                 default_zone = test_pages_in_a_zone(start_pfn,
 334                                                     start_pfn + nr_pages);
 335                 if (!default_zone)
 336                         return sprintf(buf, "none\n");
 337                 strcat(buf, default_zone->name);
 338                 goto out;
 339         }
 340
 341         nid = mem->nid;
 342         default_zone = zone_for_pfn_range(MMOP_ONLINE, nid, start_pfn,
 343                                           nr_pages);
 344         strcat(buf, default_zone->name);
 345
 346         print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_KERNEL,
 347                         default_zone);
 348         print_allowed_zone(buf, nid, start_pfn, nr_pages, MMOP_ONLINE_MOVABLE,
 349                         default_zone);
 350 out:
 351         strcat(buf, "\n");
 352
 353         return strlen(buf);
 354 }
 355 static DEVICE_ATTR_RO(valid_zones);
 356 #endif
 357
 358 static DEVICE_ATTR_RO(phys_index);
 359 static DEVICE_ATTR_RW(state);
 360 static DEVICE_ATTR_RO(phys_device);
 361 static DEVICE_ATTR_RO(removable);
 362
 363 /*
 364  * Show the memory block size (shared by all memory blocks).
 365  */
 366 static ssize_t block_size_bytes_show(struct device *dev,
 367                                      struct device_attribute *attr, char *buf)
 368 {
 369         return sprintf(buf, "%lx\n", memory_block_size_bytes());
 370 }
 371
 372 static DEVICE_ATTR_RO(block_size_bytes);
 373
 374 /*
 375  * Memory auto online policy.
 376  */
 377
 378 static ssize_t auto_online_blocks_show(struct device *dev,
 379                                        struct device_attribute *attr, char *buf)
 380 {
 381         return sprintf(buf, "%s\n",
 382                        online_type_to_str[memhp_default_online_type]);
 383 }
 384
 385 static ssize_t auto_online_blocks_store(struct device *dev,
 386                                         struct device_attribute *attr,
 387                                         const char *buf, size_t count)
 388 {
 389         const int online_type = memhp_online_type_from_str(buf);
 390
 391         if (online_type < 0)
 392                 return -EINVAL;
 393
 394         memhp_default_online_type = online_type;
 395         return count;
 396 }
 397
 398 static DEVICE_ATTR_RW(auto_online_blocks);
 399
 400 /*
 401  * Some architectures will have custom drivers to do this, and
 402  * will not need to do it from userspace.  The fake hot-add code
 403  * as well as ppc64 will do all of their discovery in userspace
 404  * and will require this interface.
 405  */
 406 #ifdef CONFIG_ARCH_MEMORY_PROBE
 407 static ssize_t probe_store(struct device *dev, struct device_attribute *attr,
 408                            const char *buf, size_t count)
 409 {
 410         u64 phys_addr;
 411         int nid, ret;
 412         unsigned long pages_per_block = PAGES_PER_SECTION * sections_per_block;
 413
 414         ret = kstrtoull(buf, 0, &phys_addr);
 415         if (ret)
 416                 return ret;
 417
 418         if (phys_addr & ((pages_per_block << PAGE_SHIFT) - 1))
 419                 return -EINVAL;
 420
 421         ret = lock_device_hotplug_sysfs();
 422         if (ret)
 423                 return ret;
 424
 425         nid = memory_add_physaddr_to_nid(phys_addr);
 426         ret = __add_memory(nid, phys_addr,
 427                            MIN_MEMORY_BLOCK_SIZE * sections_per_block);
 428
 429         if (ret)
 430                 goto out;
 431
 432         ret = count;
 433 out:
 434         unlock_device_hotplug();
 435         return ret;
 436 }
 437
 438 static DEVICE_ATTR_WO(probe);
 439 #endif
 440
 441 #ifdef CONFIG_MEMORY_FAILURE
 442 /*
 443  * Support for offlining pages of memory
 444  */
 445
 446 /* Soft offline a page */
 447 static ssize_t soft_offline_page_store(struct device *dev,
 448                                        struct device_attribute *attr,
 449                                        const char *buf, size_t count)
 450 {
 451         int ret;
 452         u64 pfn;
 453         if (!capable(CAP_SYS_ADMIN))
 454                 return -EPERM;
 455         if (kstrtoull(buf, 0, &pfn) < 0)
 456                 return -EINVAL;
 457         pfn >>= PAGE_SHIFT;
 458         ret = soft_offline_page(pfn, 0);
 459         return ret == 0 ? count : ret;
 460 }
 461
 462 /* Forcibly offline a page, including killing processes. */
 463 static ssize_t hard_offline_page_store(struct device *dev,
 464                                        struct device_attribute *attr,
 465                                        const char *buf, size_t count)
 466 {
 467         int ret;
 468         u64 pfn;
 469         if (!capable(CAP_SYS_ADMIN))
 470                 return -EPERM;
 471         if (kstrtoull(buf, 0, &pfn) < 0)
 472                 return -EINVAL;
 473         pfn >>= PAGE_SHIFT;
 474         ret = memory_failure(pfn, 0);
 475         return ret ? ret : count;
 476 }
 477
 478 static DEVICE_ATTR_WO(soft_offline_page);
 479 static DEVICE_ATTR_WO(hard_offline_page);
 480 #endif
 481
 482 /*
 483  * Note that phys_device is optional.  It is here to allow for
 484  * differentiation between which *physical* devices each
 485  * section belongs to...
 486  */
 487 int __weak arch_get_memory_phys_device(unsigned long start_pfn)
 488 {
 489         return 0;
 490 }
 491
 492 /* A reference for the returned memory block device is acquired. */
 493 static struct memory_block *find_memory_block_by_id(unsigned long block_id)
 494 {
 495         struct device *dev;
 496
 497         dev = subsys_find_device_by_id(&memory_subsys, block_id, NULL);
 498         return dev ? to_memory_block(dev) : NULL;
 499 }
 500
 501 /*
 502  * For now, we have a linear search to go find the appropriate
 503  * memory_block corresponding to a particular phys_index. If
 504  * this gets to be a real problem, we can always use a radix
 505  * tree or something here.
 506  *
 507  * This could be made generic for all device subsystems.
 508  */
 509 struct memory_block *find_memory_block(struct mem_section *section)
 510 {
 511         unsigned long block_id = base_memory_block_id(__section_nr(section));
 512
 513         return find_memory_block_by_id(block_id);
 514 }
 515
 516 static struct attribute *memory_memblk_attrs[] = {
 517         &dev_attr_phys_index.attr,
 518         &dev_attr_state.attr,
 519         &dev_attr_phys_device.attr,
 520         &dev_attr_removable.attr,
 521 #ifdef CONFIG_MEMORY_HOTREMOVE
 522         &dev_attr_valid_zones.attr,
 523 #endif
 524         NULL
 525 };
 526
 527 static struct attribute_group memory_memblk_attr_group = {
 528         .attrs = memory_memblk_attrs,
 529 };
 530
 531 static const struct attribute_group *memory_memblk_attr_groups[] = {
 532         &memory_memblk_attr_group,
 533         NULL,
 534 };
 535
 536 /*
 537  * register_memory - Setup a sysfs device for a memory block
 538  */
 539 static
 540 int register_memory(struct memory_block *memory)
 541 {
 542         int ret;
 543
 544         memory->dev.bus = &memory_subsys;
 545         memory->dev.id = memory->start_section_nr / sections_per_block;
 546         memory->dev.release = memory_block_release;
 547         memory->dev.groups = memory_memblk_attr_groups;
 548         memory->dev.offline = memory->state == MEM_OFFLINE;
 549
 550         ret = device_register(&memory->dev);
 551         if (ret)
 552                 put_device(&memory->dev);
 553
 554         return ret;
 555 }
 556
 557 static int init_memory_block(struct memory_block **memory,
 558                              unsigned long block_id, unsigned long state)
 559 {
 560         struct memory_block *mem;
 561         unsigned long start_pfn;
 562         int ret = 0;
 563
 564         mem = find_memory_block_by_id(block_id);
 565         if (mem) {
 566                 put_device(&mem->dev);
 567                 return -EEXIST;
 568         }
 569         mem = kzalloc(sizeof(*mem), GFP_KERNEL);
 570         if (!mem)
 571                 return -ENOMEM;
 572
 573         mem->start_section_nr = block_id * sections_per_block;
 574         mem->state = state;
 575         start_pfn = section_nr_to_pfn(mem->start_section_nr);
 576         mem->phys_device = arch_get_memory_phys_device(start_pfn);
 577         mem->nid = NUMA_NO_NODE;
 578
 579         ret = register_memory(mem);
 580
 581         *memory = mem;
 582         return ret;
 583 }
 584
 585 static int add_memory_block(unsigned long base_section_nr)
 586 {
 587         int section_count = 0;
 588         struct memory_block *mem;
 589         unsigned long nr;
 590
 591         for (nr = base_section_nr; nr < base_section_nr + sections_per_block;
 592              nr++)
 593                 if (present_section_nr(nr))
 594                         section_count++;
 595
 596         if (section_count == 0)
 597                 return 0;
 598         return init_memory_block(&mem, base_memory_block_id(base_section_nr),
 599                                  MEM_ONLINE);
 600 }
 601
 602 static void unregister_memory(struct memory_block *memory)
 603 {
 604         if (WARN_ON_ONCE(memory->dev.bus != &memory_subsys))
 605                 return;
 606
 607         /* drop the ref. we got via find_memory_block() */
 608         put_device(&memory->dev);
 609         device_unregister(&memory->dev);
 610 }
 611
 612 /*
 613  * Create memory block devices for the given memory area. Start and size
 614  * have to be aligned to memory block granularity. Memory block devices
 615  * will be initialized as offline.
 616  *
 617  * Called under device_hotplug_lock.
 618  */
 619 int create_memory_block_devices(unsigned long start, unsigned long size)
 620 {
 621         const unsigned long start_block_id = pfn_to_block_id(PFN_DOWN(start));
 622         unsigned long end_block_id = pfn_to_block_id(PFN_DOWN(start + size));
 623         struct memory_block *mem;
 624         unsigned long block_id;
 625         int ret = 0;
 626
 627         if (WARN_ON_ONCE(!IS_ALIGNED(start, memory_block_size_bytes()) ||
 628                          !IS_ALIGNED(size, memory_block_size_bytes())))
 629                 return -EINVAL;
 630
 631         for (block_id = start_block_id; block_id != end_block_id; block_id++) {
 632                 ret = init_memory_block(&mem, block_id, MEM_OFFLINE);
 633                 if (ret)
 634                         break;
 635         }
 636         if (ret) {
 637                 end_block_id = block_id;
 638                 for (block_id = start_block_id; block_id != end_block_id;
 639                      block_id++) {
 640                         mem = find_memory_block_by_id(block_id);
 641                         if (WARN_ON_ONCE(!mem))
 642                                 continue;
 643                         unregister_memory(mem);
 644                 }
 645         }
 646         return ret;
 647 }
 648
 649 /*
 650  * Remove memory block devices for the given memory area. Start and size
 651  * have to be aligned to memory block granularity. Memory block devices
 652  * have to be offline.
 653  *
 654  * Called under device_hotplug_lock.
 655  */
 656 void remove_memory_block_devices(unsigned long start, unsigned long size)
 657 {
 658         const unsigned long start_block_id = pfn_to_block_id(PFN_DOWN(start));
 659         const unsigned long end_block_id = pfn_to_block_id(PFN_DOWN(start + size));
 660         struct memory_block *mem;
 661         unsigned long block_id;
 662
 663         if (WARN_ON_ONCE(!IS_ALIGNED(start, memory_block_size_bytes()) ||
 664                          !IS_ALIGNED(size, memory_block_size_bytes())))
 665                 return;
 666
 667         for (block_id = start_block_id; block_id != end_block_id; block_id++) {
 668                 mem = find_memory_block_by_id(block_id);
 669                 if (WARN_ON_ONCE(!mem))
 670                         continue;
 671                 unregister_memory_block_under_nodes(mem);
 672                 unregister_memory(mem);
 673         }
 674 }
 675
 676 /* return true if the memory block is offlined, otherwise, return false */
 677 bool is_memblock_offlined(struct memory_block *mem)
 678 {
 679         return mem->state == MEM_OFFLINE;
 680 }
 681
 682 static struct attribute *memory_root_attrs[] = {
 683 #ifdef CONFIG_ARCH_MEMORY_PROBE
 684         &dev_attr_probe.attr,
 685 #endif
 686
 687 #ifdef CONFIG_MEMORY_FAILURE
 688         &dev_attr_soft_offline_page.attr,
 689         &dev_attr_hard_offline_page.attr,
 690 #endif
 691
 692         &dev_attr_block_size_bytes.attr,
 693         &dev_attr_auto_online_blocks.attr,
 694         NULL
 695 };
 696
 697 static struct attribute_group memory_root_attr_group = {
 698         .attrs = memory_root_attrs,
 699 };
 700
 701 static const struct attribute_group *memory_root_attr_groups[] = {
 702         &memory_root_attr_group,
 703         NULL,
 704 };
 705
 706 /*
 707  * Initialize the sysfs support for memory devices. At the time this function
 708  * is called, we cannot have concurrent creation/deletion of memory block
 709  * devices, the device_hotplug_lock is not needed.
 710  */
 711 void __init memory_dev_init(void)
 712 {
 713         int ret;
 714         unsigned long block_sz, nr;
 715
 716         /* Validate the configured memory block size */
 717         block_sz = memory_block_size_bytes();
 718         if (!is_power_of_2(block_sz) || block_sz < MIN_MEMORY_BLOCK_SIZE)
 719                 panic("Memory block size not suitable: 0x%lx\n", block_sz);
 720         sections_per_block = block_sz / MIN_MEMORY_BLOCK_SIZE;
 721
 722         ret = subsys_system_register(&memory_subsys, memory_root_attr_groups);
 723         if (ret)
 724                 panic("%s() failed to register subsystem: %d\n", __func__, ret);
 725
 726         /*
 727          * Create entries for memory sections that were found
 728          * during boot and have been initialized
 729          */
 730         for (nr = 0; nr <= __highest_present_section_nr;
 731              nr += sections_per_block) {
 732                 ret = add_memory_block(nr);
 733                 if (ret)
 734                         panic("%s() failed to add memory block: %d\n", __func__,
 735                               ret);
 736         }
 737 }
 738
 739 /**
 740  * walk_memory_blocks - walk through all present memory blocks overlapped
 741  *                      by the range [start, start + size)
 742  *
 743  * @start: start address of the memory range
 744  * @size: size of the memory range
 745  * @arg: argument passed to func
 746  * @func: callback for each memory section walked
 747  *
 748  * This function walks through all present memory blocks overlapped by the
 749  * range [start, start + size), calling func on each memory block.
 750  *
 751  * In case func() returns an error, walking is aborted and the error is
 752  * returned.
 753  */
 754 int walk_memory_blocks(unsigned long start, unsigned long size,
 755                        void *arg, walk_memory_blocks_func_t func)
 756 {
 757         const unsigned long start_block_id = phys_to_block_id(start);
 758         const unsigned long end_block_id = phys_to_block_id(start + size - 1);
 759         struct memory_block *mem;
 760         unsigned long block_id;
 761         int ret = 0;
 762
 763         if (!size)
 764                 return 0;
 765
 766         for (block_id = start_block_id; block_id <= end_block_id; block_id++) {
 767                 mem = find_memory_block_by_id(block_id);
 768                 if (!mem)
 769                         continue;
 770
 771                 ret = func(mem, arg);
 772                 put_device(&mem->dev);
 773                 if (ret)
 774                         break;
 775         }
 776         return ret;
 777 }
 778
 779 struct for_each_memory_block_cb_data {
 780         walk_memory_blocks_func_t func;
 781         void *arg;
 782 };
 783
 784 static int for_each_memory_block_cb(struct device *dev, void *data)
 785 {
 786         struct memory_block *mem = to_memory_block(dev);
 787         struct for_each_memory_block_cb_data *cb_data = data;
 788
 789         return cb_data->func(mem, cb_data->arg);
 790 }
 791
 792 /**
 793  * for_each_memory_block - walk through all present memory blocks
 794  *
 795  * @arg: argument passed to func
 796  * @func: callback for each memory block walked
 797  *
 798  * This function walks through all present memory blocks, calling func on
 799  * each memory block.
 800  *
 801  * In case func() returns an error, walking is aborted and the error is
 802  * returned.
 803  */
 804 int for_each_memory_block(void *arg, walk_memory_blocks_func_t func)
 805 {
 806         struct for_each_memory_block_cb_data cb_data = {
 807                 .func = func,
 808                 .arg = arg,
 809         };
 810
 811         return bus_for_each_dev(&memory_subsys, NULL, &cb_data,
 812                                 for_each_memory_block_cb);
 813 }