mm/memory_hotplug.c

   1 /*
   2  *  linux/mm/memory_hotplug.c
   3  *
   4  *  Copyright (C)
   5  */
   6
   7 #include <linux/stddef.h>
   8 #include <linux/mm.h>
   9 #include <linux/swap.h>
  10 #include <linux/interrupt.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/compiler.h>
  13 #include <linux/export.h>
  14 #include <linux/pagevec.h>
  15 #include <linux/writeback.h>
  16 #include <linux/slab.h>
  17 #include <linux/sysctl.h>
  18 #include <linux/cpu.h>
  19 #include <linux/memory.h>
  20 #include <linux/memory_hotplug.h>
  21 #include <linux/highmem.h>
  22 #include <linux/vmalloc.h>
  23 #include <linux/ioport.h>
  24 #include <linux/delay.h>
  25 #include <linux/migrate.h>
  26 #include <linux/page-isolation.h>
  27 #include <linux/pfn.h>
  28 #include <linux/suspend.h>
  29 #include <linux/mm_inline.h>
  30 #include <linux/firmware-map.h>
  31 #include <linux/stop_machine.h>
  32 #include <linux/hugetlb.h>
  33 #include <linux/memblock.h>
  34 #include <linux/bootmem.h>
  35
  36 #include <asm/tlbflush.h>
  37
  38 #include "internal.h"
  39
  40 /*
  41  * online_page_callback contains pointer to current page onlining function.
  42  * Initially it is generic_online_page(). If it is required it could be
  43  * changed by calling set_online_page_callback() for callback registration
  44  * and restore_online_page_callback() for generic callback restore.
  45  */
  46
  47 static void generic_online_page(struct page *page);
  48
  49 static online_page_callback_t online_page_callback = generic_online_page;
  50 static DEFINE_MUTEX(online_page_callback_lock);
  51
  52 /* The same as the cpu_hotplug lock, but for memory hotplug. */
  53 static struct {
  54         struct task_struct *active_writer;
  55         struct mutex lock; /* Synchronizes accesses to refcount, */
  56         /*
  57          * Also blocks the new readers during
  58          * an ongoing mem hotplug operation.
  59          */
  60         int refcount;
  61
  62 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  63         struct lockdep_map dep_map;
  64 #endif
  65 } mem_hotplug = {
  66         .active_writer = NULL,
  67         .lock = __MUTEX_INITIALIZER(mem_hotplug.lock),
  68         .refcount = 0,
  69 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  70         .dep_map = {.name = "mem_hotplug.lock" },
  71 #endif
  72 };
  73
  74 /* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */
  75 #define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map)
  76 #define memhp_lock_acquire()      lock_map_acquire(&mem_hotplug.dep_map)
  77 #define memhp_lock_release()      lock_map_release(&mem_hotplug.dep_map)
  78
  79 void get_online_mems(void)
  80 {
  81         might_sleep();
  82         if (mem_hotplug.active_writer == current)
  83                 return;
  84         memhp_lock_acquire_read();
  85         mutex_lock(&mem_hotplug.lock);
  86         mem_hotplug.refcount++;
  87         mutex_unlock(&mem_hotplug.lock);
  88
  89 }
  90
  91 void put_online_mems(void)
  92 {
  93         if (mem_hotplug.active_writer == current)
  94                 return;
  95         mutex_lock(&mem_hotplug.lock);
  96
  97         if (WARN_ON(!mem_hotplug.refcount))
  98                 mem_hotplug.refcount++; /* try to fix things up */
  99
 100         if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer))
 101                 wake_up_process(mem_hotplug.active_writer);
 102         mutex_unlock(&mem_hotplug.lock);
 103         memhp_lock_release();
 104
 105 }
 106
 107 void mem_hotplug_begin(void)
 108 {
 109         mem_hotplug.active_writer = current;
 110
 111         memhp_lock_acquire();
 112         for (;;) {
 113                 mutex_lock(&mem_hotplug.lock);
 114                 if (likely(!mem_hotplug.refcount))
 115                         break;
 116                 __set_current_state(TASK_UNINTERRUPTIBLE);
 117                 mutex_unlock(&mem_hotplug.lock);
 118                 schedule();
 119         }
 120 }
 121
 122 void mem_hotplug_done(void)
 123 {
 124         mem_hotplug.active_writer = NULL;
 125         mutex_unlock(&mem_hotplug.lock);
 126         memhp_lock_release();
 127 }
 128
 129 /* add this memory to iomem resource */
 130 static struct resource *register_memory_resource(u64 start, u64 size)
 131 {
 132         struct resource *res;
 133         res = kzalloc(sizeof(struct resource), GFP_KERNEL);
 134         BUG_ON(!res);
 135
 136         res->name = "System RAM";
 137         res->start = start;
 138         res->end = start + size - 1;
 139         res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
 140         if (request_resource(&iomem_resource, res) < 0) {
 141                 pr_debug("System RAM resource %pR cannot be added\n", res);
 142                 kfree(res);
 143                 res = NULL;
 144         }
 145         return res;
 146 }
 147
 148 static void release_memory_resource(struct resource *res)
 149 {
 150         if (!res)
 151                 return;
 152         release_resource(res);
 153         kfree(res);
 154         return;
 155 }
 156
 157 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
 158 void get_page_bootmem(unsigned long info,  struct page *page,
 159                       unsigned long type)
 160 {
 161         page->lru.next = (struct list_head *) type;
 162         SetPagePrivate(page);
 163         set_page_private(page, info);
 164         atomic_inc(&page->_count);
 165 }
 166
 167 void put_page_bootmem(struct page *page)
 168 {
 169         unsigned long type;
 170
 171         type = (unsigned long) page->lru.next;
 172         BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
 173                type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
 174
 175         if (atomic_dec_return(&page->_count) == 1) {
 176                 ClearPagePrivate(page);
 177                 set_page_private(page, 0);
 178                 INIT_LIST_HEAD(&page->lru);
 179                 free_reserved_page(page);
 180         }
 181 }
 182
 183 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
 184 #ifndef CONFIG_SPARSEMEM_VMEMMAP
 185 static void register_page_bootmem_info_section(unsigned long start_pfn)
 186 {
 187         unsigned long *usemap, mapsize, section_nr, i;
 188         struct mem_section *ms;
 189         struct page *page, *memmap;
 190
 191         section_nr = pfn_to_section_nr(start_pfn);
 192         ms = __nr_to_section(section_nr);
 193
 194         /* Get section's memmap address */
 195         memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
 196
 197         /*
 198          * Get page for the memmap's phys address
 199          * XXX: need more consideration for sparse_vmemmap...
 200          */
 201         page = virt_to_page(memmap);
 202         mapsize = sizeof(struct page) * PAGES_PER_SECTION;
 203         mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
 204
 205         /* remember memmap's page */
 206         for (i = 0; i < mapsize; i++, page++)
 207                 get_page_bootmem(section_nr, page, SECTION_INFO);
 208
 209         usemap = __nr_to_section(section_nr)->pageblock_flags;
 210         page = virt_to_page(usemap);
 211
 212         mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
 213
 214         for (i = 0; i < mapsize; i++, page++)
 215                 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
 216
 217 }
 218 #else /* CONFIG_SPARSEMEM_VMEMMAP */
 219 static void register_page_bootmem_info_section(unsigned long start_pfn)
 220 {
 221         unsigned long *usemap, mapsize, section_nr, i;
 222         struct mem_section *ms;
 223         struct page *page, *memmap;
 224
 225         if (!pfn_valid(start_pfn))
 226                 return;
 227
 228         section_nr = pfn_to_section_nr(start_pfn);
 229         ms = __nr_to_section(section_nr);
 230
 231         memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
 232
 233         register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
 234
 235         usemap = __nr_to_section(section_nr)->pageblock_flags;
 236         page = virt_to_page(usemap);
 237
 238         mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
 239
 240         for (i = 0; i < mapsize; i++, page++)
 241                 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
 242 }
 243 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
 244
 245 void register_page_bootmem_info_node(struct pglist_data *pgdat)
 246 {
 247         unsigned long i, pfn, end_pfn, nr_pages;
 248         int node = pgdat->node_id;
 249         struct page *page;
 250         struct zone *zone;
 251
 252         nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
 253         page = virt_to_page(pgdat);
 254
 255         for (i = 0; i < nr_pages; i++, page++)
 256                 get_page_bootmem(node, page, NODE_INFO);
 257
 258         zone = &pgdat->node_zones[0];
 259         for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
 260                 if (zone_is_initialized(zone)) {
 261                         nr_pages = zone->wait_table_hash_nr_entries
 262                                 * sizeof(wait_queue_head_t);
 263                         nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
 264                         page = virt_to_page(zone->wait_table);
 265
 266                         for (i = 0; i < nr_pages; i++, page++)
 267                                 get_page_bootmem(node, page, NODE_INFO);
 268                 }
 269         }
 270
 271         pfn = pgdat->node_start_pfn;
 272         end_pfn = pgdat_end_pfn(pgdat);
 273
 274         /* register section info */
 275         for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
 276                 /*
 277                  * Some platforms can assign the same pfn to multiple nodes - on
 278                  * node0 as well as nodeN.  To avoid registering a pfn against
 279                  * multiple nodes we check that this pfn does not already
 280                  * reside in some other nodes.
 281                  */
 282                 if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node))
 283                         register_page_bootmem_info_section(pfn);
 284         }
 285 }
 286 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
 287
 288 static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn,
 289                                      unsigned long end_pfn)
 290 {
 291         unsigned long old_zone_end_pfn;
 292
 293         zone_span_writelock(zone);
 294
 295         old_zone_end_pfn = zone_end_pfn(zone);
 296         if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
 297                 zone->zone_start_pfn = start_pfn;
 298
 299         zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
 300                                 zone->zone_start_pfn;
 301
 302         zone_span_writeunlock(zone);
 303 }
 304
 305 static void resize_zone(struct zone *zone, unsigned long start_pfn,
 306                 unsigned long end_pfn)
 307 {
 308         zone_span_writelock(zone);
 309
 310         if (end_pfn - start_pfn) {
 311                 zone->zone_start_pfn = start_pfn;
 312                 zone->spanned_pages = end_pfn - start_pfn;
 313         } else {
 314                 /*
 315                  * make it consist as free_area_init_core(),
 316                  * if spanned_pages = 0, then keep start_pfn = 0
 317                  */
 318                 zone->zone_start_pfn = 0;
 319                 zone->spanned_pages = 0;
 320         }
 321
 322         zone_span_writeunlock(zone);
 323 }
 324
 325 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
 326                 unsigned long end_pfn)
 327 {
 328         enum zone_type zid = zone_idx(zone);
 329         int nid = zone->zone_pgdat->node_id;
 330         unsigned long pfn;
 331
 332         for (pfn = start_pfn; pfn < end_pfn; pfn++)
 333                 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
 334 }
 335
 336 /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
 337  * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */
 338 static int __ref ensure_zone_is_initialized(struct zone *zone,
 339                         unsigned long start_pfn, unsigned long num_pages)
 340 {
 341         if (!zone_is_initialized(zone))
 342                 return init_currently_empty_zone(zone, start_pfn, num_pages,
 343                                                  MEMMAP_HOTPLUG);
 344         return 0;
 345 }
 346
 347 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
 348                 unsigned long start_pfn, unsigned long end_pfn)
 349 {
 350         int ret;
 351         unsigned long flags;
 352         unsigned long z1_start_pfn;
 353
 354         ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
 355         if (ret)
 356                 return ret;
 357
 358         pgdat_resize_lock(z1->zone_pgdat, &flags);
 359
 360         /* can't move pfns which are higher than @z2 */
 361         if (end_pfn > zone_end_pfn(z2))
 362                 goto out_fail;
 363         /* the move out part must be at the left most of @z2 */
 364         if (start_pfn > z2->zone_start_pfn)
 365                 goto out_fail;
 366         /* must included/overlap */
 367         if (end_pfn <= z2->zone_start_pfn)
 368                 goto out_fail;
 369
 370         /* use start_pfn for z1's start_pfn if z1 is empty */
 371         if (!zone_is_empty(z1))
 372                 z1_start_pfn = z1->zone_start_pfn;
 373         else
 374                 z1_start_pfn = start_pfn;
 375
 376         resize_zone(z1, z1_start_pfn, end_pfn);
 377         resize_zone(z2, end_pfn, zone_end_pfn(z2));
 378
 379         pgdat_resize_unlock(z1->zone_pgdat, &flags);
 380
 381         fix_zone_id(z1, start_pfn, end_pfn);
 382
 383         return 0;
 384 out_fail:
 385         pgdat_resize_unlock(z1->zone_pgdat, &flags);
 386         return -1;
 387 }
 388
 389 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
 390                 unsigned long start_pfn, unsigned long end_pfn)
 391 {
 392         int ret;
 393         unsigned long flags;
 394         unsigned long z2_end_pfn;
 395
 396         ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
 397         if (ret)
 398                 return ret;
 399
 400         pgdat_resize_lock(z1->zone_pgdat, &flags);
 401
 402         /* can't move pfns which are lower than @z1 */
 403         if (z1->zone_start_pfn > start_pfn)
 404                 goto out_fail;
 405         /* the move out part mast at the right most of @z1 */
 406         if (zone_end_pfn(z1) >  end_pfn)
 407                 goto out_fail;
 408         /* must included/overlap */
 409         if (start_pfn >= zone_end_pfn(z1))
 410                 goto out_fail;
 411
 412         /* use end_pfn for z2's end_pfn if z2 is empty */
 413         if (!zone_is_empty(z2))
 414                 z2_end_pfn = zone_end_pfn(z2);
 415         else
 416                 z2_end_pfn = end_pfn;
 417
 418         resize_zone(z1, z1->zone_start_pfn, start_pfn);
 419         resize_zone(z2, start_pfn, z2_end_pfn);
 420
 421         pgdat_resize_unlock(z1->zone_pgdat, &flags);
 422
 423         fix_zone_id(z2, start_pfn, end_pfn);
 424
 425         return 0;
 426 out_fail:
 427         pgdat_resize_unlock(z1->zone_pgdat, &flags);
 428         return -1;
 429 }
 430
 431 static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
 432                                       unsigned long end_pfn)
 433 {
 434         unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
 435
 436         if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
 437                 pgdat->node_start_pfn = start_pfn;
 438
 439         pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
 440                                         pgdat->node_start_pfn;
 441 }
 442
 443 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
 444 {
 445         struct pglist_data *pgdat = zone->zone_pgdat;
 446         int nr_pages = PAGES_PER_SECTION;
 447         int nid = pgdat->node_id;
 448         int zone_type;
 449         unsigned long flags;
 450         int ret;
 451
 452         zone_type = zone - pgdat->node_zones;
 453         ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
 454         if (ret)
 455                 return ret;
 456
 457         pgdat_resize_lock(zone->zone_pgdat, &flags);
 458         grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
 459         grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
 460                         phys_start_pfn + nr_pages);
 461         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 462         memmap_init_zone(nr_pages, nid, zone_type,
 463                          phys_start_pfn, MEMMAP_HOTPLUG);
 464         return 0;
 465 }
 466
 467 static int __meminit __add_section(int nid, struct zone *zone,
 468                                         unsigned long phys_start_pfn)
 469 {
 470         int ret;
 471
 472         if (pfn_valid(phys_start_pfn))
 473                 return -EEXIST;
 474
 475         ret = sparse_add_one_section(zone, phys_start_pfn);
 476
 477         if (ret < 0)
 478                 return ret;
 479
 480         ret = __add_zone(zone, phys_start_pfn);
 481
 482         if (ret < 0)
 483                 return ret;
 484
 485         return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
 486 }
 487
 488 /*
 489  * Reasonably generic function for adding memory.  It is
 490  * expected that archs that support memory hotplug will
 491  * call this function after deciding the zone to which to
 492  * add the new pages.
 493  */
 494 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
 495                         unsigned long nr_pages)
 496 {
 497         unsigned long i;
 498         int err = 0;
 499         int start_sec, end_sec;
 500         /* during initialize mem_map, align hot-added range to section */
 501         start_sec = pfn_to_section_nr(phys_start_pfn);
 502         end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
 503
 504         for (i = start_sec; i <= end_sec; i++) {
 505                 err = __add_section(nid, zone, section_nr_to_pfn(i));
 506
 507                 /*
 508                  * EEXIST is finally dealt with by ioresource collision
 509                  * check. see add_memory() => register_memory_resource()
 510                  * Warning will be printed if there is collision.
 511                  */
 512                 if (err && (err != -EEXIST))
 513                         break;
 514                 err = 0;
 515         }
 516         vmemmap_populate_print_last();
 517
 518         return err;
 519 }
 520 EXPORT_SYMBOL_GPL(__add_pages);
 521
 522 #ifdef CONFIG_MEMORY_HOTREMOVE
 523 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
 524 static int find_smallest_section_pfn(int nid, struct zone *zone,
 525                                      unsigned long start_pfn,
 526                                      unsigned long end_pfn)
 527 {
 528         struct mem_section *ms;
 529
 530         for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
 531                 ms = __pfn_to_section(start_pfn);
 532
 533                 if (unlikely(!valid_section(ms)))
 534                         continue;
 535
 536                 if (unlikely(pfn_to_nid(start_pfn) != nid))
 537                         continue;
 538
 539                 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
 540                         continue;
 541
 542                 return start_pfn;
 543         }
 544
 545         return 0;
 546 }
 547
 548 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
 549 static int find_biggest_section_pfn(int nid, struct zone *zone,
 550                                     unsigned long start_pfn,
 551                                     unsigned long end_pfn)
 552 {
 553         struct mem_section *ms;
 554         unsigned long pfn;
 555
 556         /* pfn is the end pfn of a memory section. */
 557         pfn = end_pfn - 1;
 558         for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
 559                 ms = __pfn_to_section(pfn);
 560
 561                 if (unlikely(!valid_section(ms)))
 562                         continue;
 563
 564                 if (unlikely(pfn_to_nid(pfn) != nid))
 565                         continue;
 566
 567                 if (zone && zone != page_zone(pfn_to_page(pfn)))
 568                         continue;
 569
 570                 return pfn;
 571         }
 572
 573         return 0;
 574 }
 575
 576 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
 577                              unsigned long end_pfn)
 578 {
 579         unsigned long zone_start_pfn = zone->zone_start_pfn;
 580         unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
 581         unsigned long zone_end_pfn = z;
 582         unsigned long pfn;
 583         struct mem_section *ms;
 584         int nid = zone_to_nid(zone);
 585
 586         zone_span_writelock(zone);
 587         if (zone_start_pfn == start_pfn) {
 588                 /*
 589                  * If the section is smallest section in the zone, it need
 590                  * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
 591                  * In this case, we find second smallest valid mem_section
 592                  * for shrinking zone.
 593                  */
 594                 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
 595                                                 zone_end_pfn);
 596                 if (pfn) {
 597                         zone->zone_start_pfn = pfn;
 598                         zone->spanned_pages = zone_end_pfn - pfn;
 599                 }
 600         } else if (zone_end_pfn == end_pfn) {
 601                 /*
 602                  * If the section is biggest section in the zone, it need
 603                  * shrink zone->spanned_pages.
 604                  * In this case, we find second biggest valid mem_section for
 605                  * shrinking zone.
 606                  */
 607                 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
 608                                                start_pfn);
 609                 if (pfn)
 610                         zone->spanned_pages = pfn - zone_start_pfn + 1;
 611         }
 612
 613         /*
 614          * The section is not biggest or smallest mem_section in the zone, it
 615          * only creates a hole in the zone. So in this case, we need not
 616          * change the zone. But perhaps, the zone has only hole data. Thus
 617          * it check the zone has only hole or not.
 618          */
 619         pfn = zone_start_pfn;
 620         for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
 621                 ms = __pfn_to_section(pfn);
 622
 623                 if (unlikely(!valid_section(ms)))
 624                         continue;
 625
 626                 if (page_zone(pfn_to_page(pfn)) != zone)
 627                         continue;
 628
 629                  /* If the section is current section, it continues the loop */
 630                 if (start_pfn == pfn)
 631                         continue;
 632
 633                 /* If we find valid section, we have nothing to do */
 634                 zone_span_writeunlock(zone);
 635                 return;
 636         }
 637
 638         /* The zone has no valid section */
 639         zone->zone_start_pfn = 0;
 640         zone->spanned_pages = 0;
 641         zone_span_writeunlock(zone);
 642 }
 643
 644 static void shrink_pgdat_span(struct pglist_data *pgdat,
 645                               unsigned long start_pfn, unsigned long end_pfn)
 646 {
 647         unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
 648         unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
 649         unsigned long pgdat_end_pfn = p;
 650         unsigned long pfn;
 651         struct mem_section *ms;
 652         int nid = pgdat->node_id;
 653
 654         if (pgdat_start_pfn == start_pfn) {
 655                 /*
 656                  * If the section is smallest section in the pgdat, it need
 657                  * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
 658                  * In this case, we find second smallest valid mem_section
 659                  * for shrinking zone.
 660                  */
 661                 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
 662                                                 pgdat_end_pfn);
 663                 if (pfn) {
 664                         pgdat->node_start_pfn = pfn;
 665                         pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
 666                 }
 667         } else if (pgdat_end_pfn == end_pfn) {
 668                 /*
 669                  * If the section is biggest section in the pgdat, it need
 670                  * shrink pgdat->node_spanned_pages.
 671                  * In this case, we find second biggest valid mem_section for
 672                  * shrinking zone.
 673                  */
 674                 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
 675                                                start_pfn);
 676                 if (pfn)
 677                         pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
 678         }
 679
 680         /*
 681          * If the section is not biggest or smallest mem_section in the pgdat,
 682          * it only creates a hole in the pgdat. So in this case, we need not
 683          * change the pgdat.
 684          * But perhaps, the pgdat has only hole data. Thus it check the pgdat
 685          * has only hole or not.
 686          */
 687         pfn = pgdat_start_pfn;
 688         for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
 689                 ms = __pfn_to_section(pfn);
 690
 691                 if (unlikely(!valid_section(ms)))
 692                         continue;
 693
 694                 if (pfn_to_nid(pfn) != nid)
 695                         continue;
 696
 697                  /* If the section is current section, it continues the loop */
 698                 if (start_pfn == pfn)
 699                         continue;
 700
 701                 /* If we find valid section, we have nothing to do */
 702                 return;
 703         }
 704
 705         /* The pgdat has no valid section */
 706         pgdat->node_start_pfn = 0;
 707         pgdat->node_spanned_pages = 0;
 708 }
 709
 710 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
 711 {
 712         struct pglist_data *pgdat = zone->zone_pgdat;
 713         int nr_pages = PAGES_PER_SECTION;
 714         int zone_type;
 715         unsigned long flags;
 716
 717         zone_type = zone - pgdat->node_zones;
 718
 719         pgdat_resize_lock(zone->zone_pgdat, &flags);
 720         shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
 721         shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
 722         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 723 }
 724
 725 static int __remove_section(struct zone *zone, struct mem_section *ms)
 726 {
 727         unsigned long start_pfn;
 728         int scn_nr;
 729         int ret = -EINVAL;
 730
 731         if (!valid_section(ms))
 732                 return ret;
 733
 734         ret = unregister_memory_section(ms);
 735         if (ret)
 736                 return ret;
 737
 738         scn_nr = __section_nr(ms);
 739         start_pfn = section_nr_to_pfn(scn_nr);
 740         __remove_zone(zone, start_pfn);
 741
 742         sparse_remove_one_section(zone, ms);
 743         return 0;
 744 }
 745
 746 /**
 747  * __remove_pages() - remove sections of pages from a zone
 748  * @zone: zone from which pages need to be removed
 749  * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
 750  * @nr_pages: number of pages to remove (must be multiple of section size)
 751  *
 752  * Generic helper function to remove section mappings and sysfs entries
 753  * for the section of the memory we are removing. Caller needs to make
 754  * sure that pages are marked reserved and zones are adjust properly by
 755  * calling offline_pages().
 756  */
 757 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
 758                  unsigned long nr_pages)
 759 {
 760         unsigned long i;
 761         int sections_to_remove;
 762         resource_size_t start, size;
 763         int ret = 0;
 764
 765         /*
 766          * We can only remove entire sections
 767          */
 768         BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
 769         BUG_ON(nr_pages % PAGES_PER_SECTION);
 770
 771         start = phys_start_pfn << PAGE_SHIFT;
 772         size = nr_pages * PAGE_SIZE;
 773         ret = release_mem_region_adjustable(&iomem_resource, start, size);
 774         if (ret) {
 775                 resource_size_t endres = start + size - 1;
 776
 777                 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
 778                                 &start, &endres, ret);
 779         }
 780
 781         sections_to_remove = nr_pages / PAGES_PER_SECTION;
 782         for (i = 0; i < sections_to_remove; i++) {
 783                 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
 784                 ret = __remove_section(zone, __pfn_to_section(pfn));
 785                 if (ret)
 786                         break;
 787         }
 788         return ret;
 789 }
 790 EXPORT_SYMBOL_GPL(__remove_pages);
 791 #endif /* CONFIG_MEMORY_HOTREMOVE */
 792
 793 int set_online_page_callback(online_page_callback_t callback)
 794 {
 795         int rc = -EINVAL;
 796
 797         get_online_mems();
 798         mutex_lock(&online_page_callback_lock);
 799
 800         if (online_page_callback == generic_online_page) {
 801                 online_page_callback = callback;
 802                 rc = 0;
 803         }
 804
 805         mutex_unlock(&online_page_callback_lock);
 806         put_online_mems();
 807
 808         return rc;
 809 }
 810 EXPORT_SYMBOL_GPL(set_online_page_callback);
 811
 812 int restore_online_page_callback(online_page_callback_t callback)
 813 {
 814         int rc = -EINVAL;
 815
 816         get_online_mems();
 817         mutex_lock(&online_page_callback_lock);
 818
 819         if (online_page_callback == callback) {
 820                 online_page_callback = generic_online_page;
 821                 rc = 0;
 822         }
 823
 824         mutex_unlock(&online_page_callback_lock);
 825         put_online_mems();
 826
 827         return rc;
 828 }
 829 EXPORT_SYMBOL_GPL(restore_online_page_callback);
 830
 831 void __online_page_set_limits(struct page *page)
 832 {
 833 }
 834 EXPORT_SYMBOL_GPL(__online_page_set_limits);
 835
 836 void __online_page_increment_counters(struct page *page)
 837 {
 838         adjust_managed_page_count(page, 1);
 839 }
 840 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
 841
 842 void __online_page_free(struct page *page)
 843 {
 844         __free_reserved_page(page);
 845 }
 846 EXPORT_SYMBOL_GPL(__online_page_free);
 847
 848 static void generic_online_page(struct page *page)
 849 {
 850         __online_page_set_limits(page);
 851         __online_page_increment_counters(page);
 852         __online_page_free(page);
 853 }
 854
 855 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
 856                         void *arg)
 857 {
 858         unsigned long i;
 859         unsigned long onlined_pages = *(unsigned long *)arg;
 860         struct page *page;
 861         if (PageReserved(pfn_to_page(start_pfn)))
 862                 for (i = 0; i < nr_pages; i++) {
 863                         page = pfn_to_page(start_pfn + i);
 864                         (*online_page_callback)(page);
 865                         onlined_pages++;
 866                 }
 867         *(unsigned long *)arg = onlined_pages;
 868         return 0;
 869 }
 870
 871 #ifdef CONFIG_MOVABLE_NODE
 872 /*
 873  * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
 874  * normal memory.
 875  */
 876 static bool can_online_high_movable(struct zone *zone)
 877 {
 878         return true;
 879 }
 880 #else /* CONFIG_MOVABLE_NODE */
 881 /* ensure every online node has NORMAL memory */
 882 static bool can_online_high_movable(struct zone *zone)
 883 {
 884         return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
 885 }
 886 #endif /* CONFIG_MOVABLE_NODE */
 887
 888 /* check which state of node_states will be changed when online memory */
 889 static void node_states_check_changes_online(unsigned long nr_pages,
 890         struct zone *zone, struct memory_notify *arg)
 891 {
 892         int nid = zone_to_nid(zone);
 893         enum zone_type zone_last = ZONE_NORMAL;
 894
 895         /*
 896          * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
 897          * contains nodes which have zones of 0...ZONE_NORMAL,
 898          * set zone_last to ZONE_NORMAL.
 899          *
 900          * If we don't have HIGHMEM nor movable node,
 901          * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
 902          * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
 903          */
 904         if (N_MEMORY == N_NORMAL_MEMORY)
 905                 zone_last = ZONE_MOVABLE;
 906
 907         /*
 908          * if the memory to be online is in a zone of 0...zone_last, and
 909          * the zones of 0...zone_last don't have memory before online, we will
 910          * need to set the node to node_states[N_NORMAL_MEMORY] after
 911          * the memory is online.
 912          */
 913         if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
 914                 arg->status_change_nid_normal = nid;
 915         else
 916                 arg->status_change_nid_normal = -1;
 917
 918 #ifdef CONFIG_HIGHMEM
 919         /*
 920          * If we have movable node, node_states[N_HIGH_MEMORY]
 921          * contains nodes which have zones of 0...ZONE_HIGHMEM,
 922          * set zone_last to ZONE_HIGHMEM.
 923          *
 924          * If we don't have movable node, node_states[N_NORMAL_MEMORY]
 925          * contains nodes which have zones of 0...ZONE_MOVABLE,
 926          * set zone_last to ZONE_MOVABLE.
 927          */
 928         zone_last = ZONE_HIGHMEM;
 929         if (N_MEMORY == N_HIGH_MEMORY)
 930                 zone_last = ZONE_MOVABLE;
 931
 932         if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
 933                 arg->status_change_nid_high = nid;
 934         else
 935                 arg->status_change_nid_high = -1;
 936 #else
 937         arg->status_change_nid_high = arg->status_change_nid_normal;
 938 #endif
 939
 940         /*
 941          * if the node don't have memory befor online, we will need to
 942          * set the node to node_states[N_MEMORY] after the memory
 943          * is online.
 944          */
 945         if (!node_state(nid, N_MEMORY))
 946                 arg->status_change_nid = nid;
 947         else
 948                 arg->status_change_nid = -1;
 949 }
 950
 951 static void node_states_set_node(int node, struct memory_notify *arg)
 952 {
 953         if (arg->status_change_nid_normal >= 0)
 954                 node_set_state(node, N_NORMAL_MEMORY);
 955
 956         if (arg->status_change_nid_high >= 0)
 957                 node_set_state(node, N_HIGH_MEMORY);
 958
 959         node_set_state(node, N_MEMORY);
 960 }
 961
 962
 963 /* Must be protected by mem_hotplug_begin() */
 964 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
 965 {
 966         unsigned long flags;
 967         unsigned long onlined_pages = 0;
 968         struct zone *zone;
 969         int need_zonelists_rebuild = 0;
 970         int nid;
 971         int ret;
 972         struct memory_notify arg;
 973
 974         /*
 975          * This doesn't need a lock to do pfn_to_page().
 976          * The section can't be removed here because of the
 977          * memory_block->state_mutex.
 978          */
 979         zone = page_zone(pfn_to_page(pfn));
 980
 981         if ((zone_idx(zone) > ZONE_NORMAL ||
 982             online_type == MMOP_ONLINE_MOVABLE) &&
 983             !can_online_high_movable(zone))
 984                 return -EINVAL;
 985
 986         if (online_type == MMOP_ONLINE_KERNEL &&
 987             zone_idx(zone) == ZONE_MOVABLE) {
 988                 if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages))
 989                         return -EINVAL;
 990         }
 991         if (online_type == MMOP_ONLINE_MOVABLE &&
 992             zone_idx(zone) == ZONE_MOVABLE - 1) {
 993                 if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages))
 994                         return -EINVAL;
 995         }
 996
 997         /* Previous code may changed the zone of the pfn range */
 998         zone = page_zone(pfn_to_page(pfn));
 999
1000         arg.start_pfn = pfn;
1001         arg.nr_pages = nr_pages;
1002         node_states_check_changes_online(nr_pages, zone, &arg);
1003
1004         nid = pfn_to_nid(pfn);
1005
1006         ret = memory_notify(MEM_GOING_ONLINE, &arg);
1007         ret = notifier_to_errno(ret);
1008         if (ret) {
1009                 memory_notify(MEM_CANCEL_ONLINE, &arg);
1010                 return ret;
1011         }
1012         /*
1013          * If this zone is not populated, then it is not in zonelist.
1014          * This means the page allocator ignores this zone.
1015          * So, zonelist must be updated after online.
1016          */
1017         mutex_lock(&zonelists_mutex);
1018         if (!populated_zone(zone)) {
1019                 need_zonelists_rebuild = 1;
1020                 build_all_zonelists(NULL, zone);
1021         }
1022
1023         ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
1024                 online_pages_range);
1025         if (ret) {
1026                 if (need_zonelists_rebuild)
1027                         zone_pcp_reset(zone);
1028                 mutex_unlock(&zonelists_mutex);
1029                 printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n",
1030                        (unsigned long long) pfn << PAGE_SHIFT,
1031                        (((unsigned long long) pfn + nr_pages)
1032                             << PAGE_SHIFT) - 1);
1033                 memory_notify(MEM_CANCEL_ONLINE, &arg);
1034                 return ret;
1035         }
1036
1037         zone->present_pages += onlined_pages;
1038
1039         pgdat_resize_lock(zone->zone_pgdat, &flags);
1040         zone->zone_pgdat->node_present_pages += onlined_pages;
1041         pgdat_resize_unlock(zone->zone_pgdat, &flags);
1042
1043         if (onlined_pages) {
1044                 node_states_set_node(zone_to_nid(zone), &arg);
1045                 if (need_zonelists_rebuild)
1046                         build_all_zonelists(NULL, NULL);
1047                 else
1048                         zone_pcp_update(zone);
1049         }
1050
1051         mutex_unlock(&zonelists_mutex);
1052
1053         init_per_zone_wmark_min();
1054
1055         if (onlined_pages)
1056                 kswapd_run(zone_to_nid(zone));
1057
1058         vm_total_pages = nr_free_pagecache_pages();
1059
1060         writeback_set_ratelimit();
1061
1062         if (onlined_pages)
1063                 memory_notify(MEM_ONLINE, &arg);
1064         return 0;
1065 }
1066 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1067
1068 static void reset_node_present_pages(pg_data_t *pgdat)
1069 {
1070         struct zone *z;
1071
1072         for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1073                 z->present_pages = 0;
1074
1075         pgdat->node_present_pages = 0;
1076 }
1077
1078 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1079 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1080 {
1081         struct pglist_data *pgdat;
1082         unsigned long zones_size[MAX_NR_ZONES] = {0};
1083         unsigned long zholes_size[MAX_NR_ZONES] = {0};
1084         unsigned long start_pfn = PFN_DOWN(start);
1085
1086         pgdat = NODE_DATA(nid);
1087         if (!pgdat) {
1088                 pgdat = arch_alloc_nodedata(nid);
1089                 if (!pgdat)
1090                         return NULL;
1091
1092                 arch_refresh_nodedata(nid, pgdat);
1093         } else {
1094                 /* Reset the nr_zones and classzone_idx to 0 before reuse */
1095                 pgdat->nr_zones = 0;
1096                 pgdat->classzone_idx = 0;
1097         }
1098
1099         /* we can use NODE_DATA(nid) from here */
1100
1101         /* init node's zones as empty zones, we don't have any present pages.*/
1102         free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1103
1104         /*
1105          * The node we allocated has no zone fallback lists. For avoiding
1106          * to access not-initialized zonelist, build here.
1107          */
1108         mutex_lock(&zonelists_mutex);
1109         build_all_zonelists(pgdat, NULL);
1110         mutex_unlock(&zonelists_mutex);
1111
1112         /*
1113          * zone->managed_pages is set to an approximate value in
1114          * free_area_init_core(), which will cause
1115          * /sys/device/system/node/nodeX/meminfo has wrong data.
1116          * So reset it to 0 before any memory is onlined.
1117          */
1118         reset_node_managed_pages(pgdat);
1119
1120         /*
1121          * When memory is hot-added, all the memory is in offline state. So
1122          * clear all zones' present_pages because they will be updated in
1123          * online_pages() and offline_pages().
1124          */
1125         reset_node_present_pages(pgdat);
1126
1127         return pgdat;
1128 }
1129
1130 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1131 {
1132         arch_refresh_nodedata(nid, NULL);
1133         arch_free_nodedata(pgdat);
1134         return;
1135 }
1136
1137
1138 /**
1139  * try_online_node - online a node if offlined
1140  *
1141  * called by cpu_up() to online a node without onlined memory.
1142  */
1143 int try_online_node(int nid)
1144 {
1145         pg_data_t       *pgdat;
1146         int     ret;
1147
1148         if (node_online(nid))
1149                 return 0;
1150
1151         mem_hotplug_begin();
1152         pgdat = hotadd_new_pgdat(nid, 0);
1153         if (!pgdat) {
1154                 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1155                 ret = -ENOMEM;
1156                 goto out;
1157         }
1158         node_set_online(nid);
1159         ret = register_one_node(nid);
1160         BUG_ON(ret);
1161
1162         if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1163                 mutex_lock(&zonelists_mutex);
1164                 build_all_zonelists(NULL, NULL);
1165                 mutex_unlock(&zonelists_mutex);
1166         }
1167
1168 out:
1169         mem_hotplug_done();
1170         return ret;
1171 }
1172
1173 static int check_hotplug_memory_range(u64 start, u64 size)
1174 {
1175         u64 start_pfn = PFN_DOWN(start);
1176         u64 nr_pages = size >> PAGE_SHIFT;
1177
1178         /* Memory range must be aligned with section */
1179         if ((start_pfn & ~PAGE_SECTION_MASK) ||
1180             (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1181                 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1182                                 (unsigned long long)start,
1183                                 (unsigned long long)size);
1184                 return -EINVAL;
1185         }
1186
1187         return 0;
1188 }
1189
1190 /*
1191  * If movable zone has already been setup, newly added memory should be check.
1192  * If its address is higher than movable zone, it should be added as movable.
1193  * Without this check, movable zone may overlap with other zone.
1194  */
1195 static int should_add_memory_movable(int nid, u64 start, u64 size)
1196 {
1197         unsigned long start_pfn = start >> PAGE_SHIFT;
1198         pg_data_t *pgdat = NODE_DATA(nid);
1199         struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
1200
1201         if (zone_is_empty(movable_zone))
1202                 return 0;
1203
1204         if (movable_zone->zone_start_pfn <= start_pfn)
1205                 return 1;
1206
1207         return 0;
1208 }
1209
1210 int zone_for_memory(int nid, u64 start, u64 size, int zone_default)
1211 {
1212         if (should_add_memory_movable(nid, start, size))
1213                 return ZONE_MOVABLE;
1214
1215         return zone_default;
1216 }
1217
1218 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1219 int __ref add_memory(int nid, u64 start, u64 size)
1220 {
1221         pg_data_t *pgdat = NULL;
1222         bool new_pgdat;
1223         bool new_node;
1224         struct resource *res;
1225         int ret;
1226
1227         ret = check_hotplug_memory_range(start, size);
1228         if (ret)
1229                 return ret;
1230
1231         res = register_memory_resource(start, size);
1232         ret = -EEXIST;
1233         if (!res)
1234                 return ret;
1235
1236         {       /* Stupid hack to suppress address-never-null warning */
1237                 void *p = NODE_DATA(nid);
1238                 new_pgdat = !p;
1239         }
1240
1241         mem_hotplug_begin();
1242
1243         new_node = !node_online(nid);
1244         if (new_node) {
1245                 pgdat = hotadd_new_pgdat(nid, start);
1246                 ret = -ENOMEM;
1247                 if (!pgdat)
1248                         goto error;
1249         }
1250
1251         /* call arch's memory hotadd */
1252         ret = arch_add_memory(nid, start, size);
1253
1254         if (ret < 0)
1255                 goto error;
1256
1257         /* we online node here. we can't roll back from here. */
1258         node_set_online(nid);
1259
1260         if (new_node) {
1261                 ret = register_one_node(nid);
1262                 /*
1263                  * If sysfs file of new node can't create, cpu on the node
1264                  * can't be hot-added. There is no rollback way now.
1265                  * So, check by BUG_ON() to catch it reluctantly..
1266                  */
1267                 BUG_ON(ret);
1268         }
1269
1270         /* create new memmap entry */
1271         firmware_map_add_hotplug(start, start + size, "System RAM");
1272
1273         goto out;
1274
1275 error:
1276         /* rollback pgdat allocation and others */
1277         if (new_pgdat)
1278                 rollback_node_hotadd(nid, pgdat);
1279         release_memory_resource(res);
1280
1281 out:
1282         mem_hotplug_done();
1283         return ret;
1284 }
1285 EXPORT_SYMBOL_GPL(add_memory);
1286
1287 #ifdef CONFIG_MEMORY_HOTREMOVE
1288 /*
1289  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1290  * set and the size of the free page is given by page_order(). Using this,
1291  * the function determines if the pageblock contains only free pages.
1292  * Due to buddy contraints, a free page at least the size of a pageblock will
1293  * be located at the start of the pageblock
1294  */
1295 static inline int pageblock_free(struct page *page)
1296 {
1297         return PageBuddy(page) && page_order(page) >= pageblock_order;
1298 }
1299
1300 /* Return the start of the next active pageblock after a given page */
1301 static struct page *next_active_pageblock(struct page *page)
1302 {
1303         /* Ensure the starting page is pageblock-aligned */
1304         BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1305
1306         /* If the entire pageblock is free, move to the end of free page */
1307         if (pageblock_free(page)) {
1308                 int order;
1309                 /* be careful. we don't have locks, page_order can be changed.*/
1310                 order = page_order(page);
1311                 if ((order < MAX_ORDER) && (order >= pageblock_order))
1312                         return page + (1 << order);
1313         }
1314
1315         return page + pageblock_nr_pages;
1316 }
1317
1318 /* Checks if this range of memory is likely to be hot-removable. */
1319 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1320 {
1321         struct page *page = pfn_to_page(start_pfn);
1322         struct page *end_page = page + nr_pages;
1323
1324         /* Check the starting page of each pageblock within the range */
1325         for (; page < end_page; page = next_active_pageblock(page)) {
1326                 if (!is_pageblock_removable_nolock(page))
1327                         return 0;
1328                 cond_resched();
1329         }
1330
1331         /* All pageblocks in the memory block are likely to be hot-removable */
1332         return 1;
1333 }
1334
1335 /*
1336  * Confirm all pages in a range [start, end) is belongs to the same zone.
1337  */
1338 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
1339 {
1340         unsigned long pfn;
1341         struct zone *zone = NULL;
1342         struct page *page;
1343         int i;
1344         for (pfn = start_pfn;
1345              pfn < end_pfn;
1346              pfn += MAX_ORDER_NR_PAGES) {
1347                 i = 0;
1348                 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1349                 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
1350                         i++;
1351                 if (i == MAX_ORDER_NR_PAGES)
1352                         continue;
1353                 page = pfn_to_page(pfn + i);
1354                 if (zone && page_zone(page) != zone)
1355                         return 0;
1356                 zone = page_zone(page);
1357         }
1358         return 1;
1359 }
1360
1361 /*
1362  * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1363  * and hugepages). We scan pfn because it's much easier than scanning over
1364  * linked list. This function returns the pfn of the first found movable
1365  * page if it's found, otherwise 0.
1366  */
1367 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1368 {
1369         unsigned long pfn;
1370         struct page *page;
1371         for (pfn = start; pfn < end; pfn++) {
1372                 if (pfn_valid(pfn)) {
1373                         page = pfn_to_page(pfn);
1374                         if (PageLRU(page))
1375                                 return pfn;
1376                         if (PageHuge(page)) {
1377                                 if (page_huge_active(page))
1378                                         return pfn;
1379                                 else
1380                                         pfn = round_up(pfn + 1,
1381                                                 1 << compound_order(page)) - 1;
1382                         }
1383                 }
1384         }
1385         return 0;
1386 }
1387
1388 #define NR_OFFLINE_AT_ONCE_PAGES        (256)
1389 static int
1390 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1391 {
1392         unsigned long pfn;
1393         struct page *page;
1394         int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1395         int not_managed = 0;
1396         int ret = 0;
1397         LIST_HEAD(source);
1398
1399         for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1400                 if (!pfn_valid(pfn))
1401                         continue;
1402                 page = pfn_to_page(pfn);
1403
1404                 if (PageHuge(page)) {
1405                         struct page *head = compound_head(page);
1406                         pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1407                         if (compound_order(head) > PFN_SECTION_SHIFT) {
1408                                 ret = -EBUSY;
1409                                 break;
1410                         }
1411                         if (isolate_huge_page(page, &source))
1412                                 move_pages -= 1 << compound_order(head);
1413                         continue;
1414                 }
1415
1416                 if (!get_page_unless_zero(page))
1417                         continue;
1418                 /*
1419                  * We can skip free pages. And we can only deal with pages on
1420                  * LRU.
1421                  */
1422                 ret = isolate_lru_page(page);
1423                 if (!ret) { /* Success */
1424                         put_page(page);
1425                         list_add_tail(&page->lru, &source);
1426                         move_pages--;
1427                         inc_zone_page_state(page, NR_ISOLATED_ANON +
1428                                             page_is_file_cache(page));
1429
1430                 } else {
1431 #ifdef CONFIG_DEBUG_VM
1432                         printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
1433                                pfn);
1434                         dump_page(page, "failed to remove from LRU");
1435 #endif
1436                         put_page(page);
1437                         /* Because we don't have big zone->lock. we should
1438                            check this again here. */
1439                         if (page_count(page)) {
1440                                 not_managed++;
1441                                 ret = -EBUSY;
1442                                 break;
1443                         }
1444                 }
1445         }
1446         if (!list_empty(&source)) {
1447                 if (not_managed) {
1448                         putback_movable_pages(&source);
1449                         goto out;
1450                 }
1451
1452                 /*
1453                  * alloc_migrate_target should be improooooved!!
1454                  * migrate_pages returns # of failed pages.
1455                  */
1456                 ret = migrate_pages(&source, alloc_migrate_target, NULL, 0,
1457                                         MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1458                 if (ret)
1459                         putback_movable_pages(&source);
1460         }
1461 out:
1462         return ret;
1463 }
1464
1465 /*
1466  * remove from free_area[] and mark all as Reserved.
1467  */
1468 static int
1469 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1470                         void *data)
1471 {
1472         __offline_isolated_pages(start, start + nr_pages);
1473         return 0;
1474 }
1475
1476 static void
1477 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1478 {
1479         walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1480                                 offline_isolated_pages_cb);
1481 }
1482
1483 /*
1484  * Check all pages in range, recoreded as memory resource, are isolated.
1485  */
1486 static int
1487 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1488                         void *data)
1489 {
1490         int ret;
1491         long offlined = *(long *)data;
1492         ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1493         offlined = nr_pages;
1494         if (!ret)
1495                 *(long *)data += offlined;
1496         return ret;
1497 }
1498
1499 static long
1500 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1501 {
1502         long offlined = 0;
1503         int ret;
1504
1505         ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1506                         check_pages_isolated_cb);
1507         if (ret < 0)
1508                 offlined = (long)ret;
1509         return offlined;
1510 }
1511
1512 #ifdef CONFIG_MOVABLE_NODE
1513 /*
1514  * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1515  * normal memory.
1516  */
1517 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1518 {
1519         return true;
1520 }
1521 #else /* CONFIG_MOVABLE_NODE */
1522 /* ensure the node has NORMAL memory if it is still online */
1523 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1524 {
1525         struct pglist_data *pgdat = zone->zone_pgdat;
1526         unsigned long present_pages = 0;
1527         enum zone_type zt;
1528
1529         for (zt = 0; zt <= ZONE_NORMAL; zt++)
1530                 present_pages += pgdat->node_zones[zt].present_pages;
1531
1532         if (present_pages > nr_pages)
1533                 return true;
1534
1535         present_pages = 0;
1536         for (; zt <= ZONE_MOVABLE; zt++)
1537                 present_pages += pgdat->node_zones[zt].present_pages;
1538
1539         /*
1540          * we can't offline the last normal memory until all
1541          * higher memory is offlined.
1542          */
1543         return present_pages == 0;
1544 }
1545 #endif /* CONFIG_MOVABLE_NODE */
1546
1547 static int __init cmdline_parse_movable_node(char *p)
1548 {
1549 #ifdef CONFIG_MOVABLE_NODE
1550         /*
1551          * Memory used by the kernel cannot be hot-removed because Linux
1552          * cannot migrate the kernel pages. When memory hotplug is
1553          * enabled, we should prevent memblock from allocating memory
1554          * for the kernel.
1555          *
1556          * ACPI SRAT records all hotpluggable memory ranges. But before
1557          * SRAT is parsed, we don't know about it.
1558          *
1559          * The kernel image is loaded into memory at very early time. We
1560          * cannot prevent this anyway. So on NUMA system, we set any
1561          * node the kernel resides in as un-hotpluggable.
1562          *
1563          * Since on modern servers, one node could have double-digit
1564          * gigabytes memory, we can assume the memory around the kernel
1565          * image is also un-hotpluggable. So before SRAT is parsed, just
1566          * allocate memory near the kernel image to try the best to keep
1567          * the kernel away from hotpluggable memory.
1568          */
1569         memblock_set_bottom_up(true);
1570         movable_node_enabled = true;
1571 #else
1572         pr_warn("movable_node option not supported\n");
1573 #endif
1574         return 0;
1575 }
1576 early_param("movable_node", cmdline_parse_movable_node);
1577
1578 /* check which state of node_states will be changed when offline memory */
1579 static void node_states_check_changes_offline(unsigned long nr_pages,
1580                 struct zone *zone, struct memory_notify *arg)
1581 {
1582         struct pglist_data *pgdat = zone->zone_pgdat;
1583         unsigned long present_pages = 0;
1584         enum zone_type zt, zone_last = ZONE_NORMAL;
1585
1586         /*
1587          * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1588          * contains nodes which have zones of 0...ZONE_NORMAL,
1589          * set zone_last to ZONE_NORMAL.
1590          *
1591          * If we don't have HIGHMEM nor movable node,
1592          * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1593          * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1594          */
1595         if (N_MEMORY == N_NORMAL_MEMORY)
1596                 zone_last = ZONE_MOVABLE;
1597
1598         /*
1599          * check whether node_states[N_NORMAL_MEMORY] will be changed.
1600          * If the memory to be offline is in a zone of 0...zone_last,
1601          * and it is the last present memory, 0...zone_last will
1602          * become empty after offline , thus we can determind we will
1603          * need to clear the node from node_states[N_NORMAL_MEMORY].
1604          */
1605         for (zt = 0; zt <= zone_last; zt++)
1606                 present_pages += pgdat->node_zones[zt].present_pages;
1607         if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1608                 arg->status_change_nid_normal = zone_to_nid(zone);
1609         else
1610                 arg->status_change_nid_normal = -1;
1611
1612 #ifdef CONFIG_HIGHMEM
1613         /*
1614          * If we have movable node, node_states[N_HIGH_MEMORY]
1615          * contains nodes which have zones of 0...ZONE_HIGHMEM,
1616          * set zone_last to ZONE_HIGHMEM.
1617          *
1618          * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1619          * contains nodes which have zones of 0...ZONE_MOVABLE,
1620          * set zone_last to ZONE_MOVABLE.
1621          */
1622         zone_last = ZONE_HIGHMEM;
1623         if (N_MEMORY == N_HIGH_MEMORY)
1624                 zone_last = ZONE_MOVABLE;
1625
1626         for (; zt <= zone_last; zt++)
1627                 present_pages += pgdat->node_zones[zt].present_pages;
1628         if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1629                 arg->status_change_nid_high = zone_to_nid(zone);
1630         else
1631                 arg->status_change_nid_high = -1;
1632 #else
1633         arg->status_change_nid_high = arg->status_change_nid_normal;
1634 #endif
1635
1636         /*
1637          * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1638          */
1639         zone_last = ZONE_MOVABLE;
1640
1641         /*
1642          * check whether node_states[N_HIGH_MEMORY] will be changed
1643          * If we try to offline the last present @nr_pages from the node,
1644          * we can determind we will need to clear the node from
1645          * node_states[N_HIGH_MEMORY].
1646          */
1647         for (; zt <= zone_last; zt++)
1648                 present_pages += pgdat->node_zones[zt].present_pages;
1649         if (nr_pages >= present_pages)
1650                 arg->status_change_nid = zone_to_nid(zone);
1651         else
1652                 arg->status_change_nid = -1;
1653 }
1654
1655 static void node_states_clear_node(int node, struct memory_notify *arg)
1656 {
1657         if (arg->status_change_nid_normal >= 0)
1658                 node_clear_state(node, N_NORMAL_MEMORY);
1659
1660         if ((N_MEMORY != N_NORMAL_MEMORY) &&
1661             (arg->status_change_nid_high >= 0))
1662                 node_clear_state(node, N_HIGH_MEMORY);
1663
1664         if ((N_MEMORY != N_HIGH_MEMORY) &&
1665             (arg->status_change_nid >= 0))
1666                 node_clear_state(node, N_MEMORY);
1667 }
1668
1669 static int __ref __offline_pages(unsigned long start_pfn,
1670                   unsigned long end_pfn, unsigned long timeout)
1671 {
1672         unsigned long pfn, nr_pages, expire;
1673         long offlined_pages;
1674         int ret, drain, retry_max, node;
1675         unsigned long flags;
1676         struct zone *zone;
1677         struct memory_notify arg;
1678
1679         /* at least, alignment against pageblock is necessary */
1680         if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1681                 return -EINVAL;
1682         if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1683                 return -EINVAL;
1684         /* This makes hotplug much easier...and readable.
1685            we assume this for now. .*/
1686         if (!test_pages_in_a_zone(start_pfn, end_pfn))
1687                 return -EINVAL;
1688
1689         zone = page_zone(pfn_to_page(start_pfn));
1690         node = zone_to_nid(zone);
1691         nr_pages = end_pfn - start_pfn;
1692
1693         if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1694                 return -EINVAL;
1695
1696         /* set above range as isolated */
1697         ret = start_isolate_page_range(start_pfn, end_pfn,
1698                                        MIGRATE_MOVABLE, true);
1699         if (ret)
1700                 return ret;
1701
1702         arg.start_pfn = start_pfn;
1703         arg.nr_pages = nr_pages;
1704         node_states_check_changes_offline(nr_pages, zone, &arg);
1705
1706         ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1707         ret = notifier_to_errno(ret);
1708         if (ret)
1709                 goto failed_removal;
1710
1711         pfn = start_pfn;
1712         expire = jiffies + timeout;
1713         drain = 0;
1714         retry_max = 5;
1715 repeat:
1716         /* start memory hot removal */
1717         ret = -EAGAIN;
1718         if (time_after(jiffies, expire))
1719                 goto failed_removal;
1720         ret = -EINTR;
1721         if (signal_pending(current))
1722                 goto failed_removal;
1723         ret = 0;
1724         if (drain) {
1725                 lru_add_drain_all();
1726                 cond_resched();
1727                 drain_all_pages(zone);
1728         }
1729
1730         pfn = scan_movable_pages(start_pfn, end_pfn);
1731         if (pfn) { /* We have movable pages */
1732                 ret = do_migrate_range(pfn, end_pfn);
1733                 if (!ret) {
1734                         drain = 1;
1735                         goto repeat;
1736                 } else {
1737                         if (ret < 0)
1738                                 if (--retry_max == 0)
1739                                         goto failed_removal;
1740                         yield();
1741                         drain = 1;
1742                         goto repeat;
1743                 }
1744         }
1745         /* drain all zone's lru pagevec, this is asynchronous... */
1746         lru_add_drain_all();
1747         yield();
1748         /* drain pcp pages, this is synchronous. */
1749         drain_all_pages(zone);
1750         /*
1751          * dissolve free hugepages in the memory block before doing offlining
1752          * actually in order to make hugetlbfs's object counting consistent.
1753          */
1754         dissolve_free_huge_pages(start_pfn, end_pfn);
1755         /* check again */
1756         offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1757         if (offlined_pages < 0) {
1758                 ret = -EBUSY;
1759                 goto failed_removal;
1760         }
1761         printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
1762         /* Ok, all of our target is isolated.
1763            We cannot do rollback at this point. */
1764         offline_isolated_pages(start_pfn, end_pfn);
1765         /* reset pagetype flags and makes migrate type to be MOVABLE */
1766         undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1767         /* removal success */
1768         adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1769         zone->present_pages -= offlined_pages;
1770
1771         pgdat_resize_lock(zone->zone_pgdat, &flags);
1772         zone->zone_pgdat->node_present_pages -= offlined_pages;
1773         pgdat_resize_unlock(zone->zone_pgdat, &flags);
1774
1775         init_per_zone_wmark_min();
1776
1777         if (!populated_zone(zone)) {
1778                 zone_pcp_reset(zone);
1779                 mutex_lock(&zonelists_mutex);
1780                 build_all_zonelists(NULL, NULL);
1781                 mutex_unlock(&zonelists_mutex);
1782         } else
1783                 zone_pcp_update(zone);
1784
1785         node_states_clear_node(node, &arg);
1786         if (arg.status_change_nid >= 0)
1787                 kswapd_stop(node);
1788
1789         vm_total_pages = nr_free_pagecache_pages();
1790         writeback_set_ratelimit();
1791
1792         memory_notify(MEM_OFFLINE, &arg);
1793         return 0;
1794
1795 failed_removal:
1796         printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
1797                (unsigned long long) start_pfn << PAGE_SHIFT,
1798                ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1799         memory_notify(MEM_CANCEL_OFFLINE, &arg);
1800         /* pushback to free area */
1801         undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1802         return ret;
1803 }
1804
1805 /* Must be protected by mem_hotplug_begin() */
1806 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1807 {
1808         return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1809 }
1810 #endif /* CONFIG_MEMORY_HOTREMOVE */
1811
1812 /**
1813  * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1814  * @start_pfn: start pfn of the memory range
1815  * @end_pfn: end pfn of the memory range
1816  * @arg: argument passed to func
1817  * @func: callback for each memory section walked
1818  *
1819  * This function walks through all present mem sections in range
1820  * [start_pfn, end_pfn) and call func on each mem section.
1821  *
1822  * Returns the return value of func.
1823  */
1824 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1825                 void *arg, int (*func)(struct memory_block *, void *))
1826 {
1827         struct memory_block *mem = NULL;
1828         struct mem_section *section;
1829         unsigned long pfn, section_nr;
1830         int ret;
1831
1832         for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1833                 section_nr = pfn_to_section_nr(pfn);
1834                 if (!present_section_nr(section_nr))
1835                         continue;
1836
1837                 section = __nr_to_section(section_nr);
1838                 /* same memblock? */
1839                 if (mem)
1840                         if ((section_nr >= mem->start_section_nr) &&
1841                             (section_nr <= mem->end_section_nr))
1842                                 continue;
1843
1844                 mem = find_memory_block_hinted(section, mem);
1845                 if (!mem)
1846                         continue;
1847
1848                 ret = func(mem, arg);
1849                 if (ret) {
1850                         kobject_put(&mem->dev.kobj);
1851                         return ret;
1852                 }
1853         }
1854
1855         if (mem)
1856                 kobject_put(&mem->dev.kobj);
1857
1858         return 0;
1859 }
1860
1861 #ifdef CONFIG_MEMORY_HOTREMOVE
1862 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1863 {
1864         int ret = !is_memblock_offlined(mem);
1865
1866         if (unlikely(ret)) {
1867                 phys_addr_t beginpa, endpa;
1868
1869                 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1870                 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1871                 pr_warn("removing memory fails, because memory "
1872                         "[%pa-%pa] is onlined\n",
1873                         &beginpa, &endpa);
1874         }
1875
1876         return ret;
1877 }
1878
1879 static int check_cpu_on_node(pg_data_t *pgdat)
1880 {
1881         int cpu;
1882
1883         for_each_present_cpu(cpu) {
1884                 if (cpu_to_node(cpu) == pgdat->node_id)
1885                         /*
1886                          * the cpu on this node isn't removed, and we can't
1887                          * offline this node.
1888                          */
1889                         return -EBUSY;
1890         }
1891
1892         return 0;
1893 }
1894
1895 static void unmap_cpu_on_node(pg_data_t *pgdat)
1896 {
1897 #ifdef CONFIG_ACPI_NUMA
1898         int cpu;
1899
1900         for_each_possible_cpu(cpu)
1901                 if (cpu_to_node(cpu) == pgdat->node_id)
1902                         numa_clear_node(cpu);
1903 #endif
1904 }
1905
1906 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
1907 {
1908         int ret;
1909
1910         ret = check_cpu_on_node(pgdat);
1911         if (ret)
1912                 return ret;
1913
1914         /*
1915          * the node will be offlined when we come here, so we can clear
1916          * the cpu_to_node() now.
1917          */
1918
1919         unmap_cpu_on_node(pgdat);
1920         return 0;
1921 }
1922
1923 /**
1924  * try_offline_node
1925  *
1926  * Offline a node if all memory sections and cpus of the node are removed.
1927  *
1928  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1929  * and online/offline operations before this call.
1930  */
1931 void try_offline_node(int nid)
1932 {
1933         pg_data_t *pgdat = NODE_DATA(nid);
1934         unsigned long start_pfn = pgdat->node_start_pfn;
1935         unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1936         unsigned long pfn;
1937         int i;
1938
1939         for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1940                 unsigned long section_nr = pfn_to_section_nr(pfn);
1941
1942                 if (!present_section_nr(section_nr))
1943                         continue;
1944
1945                 if (pfn_to_nid(pfn) != nid)
1946                         continue;
1947
1948                 /*
1949                  * some memory sections of this node are not removed, and we
1950                  * can't offline node now.
1951                  */
1952                 return;
1953         }
1954
1955         if (check_and_unmap_cpu_on_node(pgdat))
1956                 return;
1957
1958         /*
1959          * all memory/cpu of this node are removed, we can offline this
1960          * node now.
1961          */
1962         node_set_offline(nid);
1963         unregister_one_node(nid);
1964
1965         /* free waittable in each zone */
1966         for (i = 0; i < MAX_NR_ZONES; i++) {
1967                 struct zone *zone = pgdat->node_zones + i;
1968
1969                 /*
1970                  * wait_table may be allocated from boot memory,
1971                  * here only free if it's allocated by vmalloc.
1972                  */
1973                 if (is_vmalloc_addr(zone->wait_table)) {
1974                         vfree(zone->wait_table);
1975                         zone->wait_table = NULL;
1976                 }
1977         }
1978 }
1979 EXPORT_SYMBOL(try_offline_node);
1980
1981 /**
1982  * remove_memory
1983  *
1984  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1985  * and online/offline operations before this call, as required by
1986  * try_offline_node().
1987  */
1988 void __ref remove_memory(int nid, u64 start, u64 size)
1989 {
1990         int ret;
1991
1992         BUG_ON(check_hotplug_memory_range(start, size));
1993
1994         mem_hotplug_begin();
1995
1996         /*
1997          * All memory blocks must be offlined before removing memory.  Check
1998          * whether all memory blocks in question are offline and trigger a BUG()
1999          * if this is not the case.
2000          */
2001         ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
2002                                 check_memblock_offlined_cb);
2003         if (ret)
2004                 BUG();
2005
2006         /* remove memmap entry */
2007         firmware_map_remove(start, start + size, "System RAM");
2008
2009         arch_remove_memory(start, size);
2010
2011         try_offline_node(nid);
2012
2013         mem_hotplug_done();
2014 }
2015 EXPORT_SYMBOL_GPL(remove_memory);
2016 #endif /* CONFIG_MEMORY_HOTREMOVE */