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, pfn;
 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
 465         /* online_page_range is called later and expects pages reserved */
 466         for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) {
 467                 if (!pfn_valid(pfn))
 468                         continue;
 469
 470                 SetPageReserved(pfn_to_page(pfn));
 471         }
 472         return 0;
 473 }
 474
 475 static int __meminit __add_section(int nid, struct zone *zone,
 476                                         unsigned long phys_start_pfn)
 477 {
 478         int ret;
 479
 480         if (pfn_valid(phys_start_pfn))
 481                 return -EEXIST;
 482
 483         ret = sparse_add_one_section(zone, phys_start_pfn);
 484
 485         if (ret < 0)
 486                 return ret;
 487
 488         ret = __add_zone(zone, phys_start_pfn);
 489
 490         if (ret < 0)
 491                 return ret;
 492
 493         return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
 494 }
 495
 496 /*
 497  * Reasonably generic function for adding memory.  It is
 498  * expected that archs that support memory hotplug will
 499  * call this function after deciding the zone to which to
 500  * add the new pages.
 501  */
 502 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
 503                         unsigned long nr_pages)
 504 {
 505         unsigned long i;
 506         int err = 0;
 507         int start_sec, end_sec;
 508         /* during initialize mem_map, align hot-added range to section */
 509         start_sec = pfn_to_section_nr(phys_start_pfn);
 510         end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
 511
 512         for (i = start_sec; i <= end_sec; i++) {
 513                 err = __add_section(nid, zone, section_nr_to_pfn(i));
 514
 515                 /*
 516                  * EEXIST is finally dealt with by ioresource collision
 517                  * check. see add_memory() => register_memory_resource()
 518                  * Warning will be printed if there is collision.
 519                  */
 520                 if (err && (err != -EEXIST))
 521                         break;
 522                 err = 0;
 523         }
 524         vmemmap_populate_print_last();
 525
 526         return err;
 527 }
 528 EXPORT_SYMBOL_GPL(__add_pages);
 529
 530 #ifdef CONFIG_MEMORY_HOTREMOVE
 531 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
 532 static int find_smallest_section_pfn(int nid, struct zone *zone,
 533                                      unsigned long start_pfn,
 534                                      unsigned long end_pfn)
 535 {
 536         struct mem_section *ms;
 537
 538         for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
 539                 ms = __pfn_to_section(start_pfn);
 540
 541                 if (unlikely(!valid_section(ms)))
 542                         continue;
 543
 544                 if (unlikely(pfn_to_nid(start_pfn) != nid))
 545                         continue;
 546
 547                 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
 548                         continue;
 549
 550                 return start_pfn;
 551         }
 552
 553         return 0;
 554 }
 555
 556 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
 557 static int find_biggest_section_pfn(int nid, struct zone *zone,
 558                                     unsigned long start_pfn,
 559                                     unsigned long end_pfn)
 560 {
 561         struct mem_section *ms;
 562         unsigned long pfn;
 563
 564         /* pfn is the end pfn of a memory section. */
 565         pfn = end_pfn - 1;
 566         for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
 567                 ms = __pfn_to_section(pfn);
 568
 569                 if (unlikely(!valid_section(ms)))
 570                         continue;
 571
 572                 if (unlikely(pfn_to_nid(pfn) != nid))
 573                         continue;
 574
 575                 if (zone && zone != page_zone(pfn_to_page(pfn)))
 576                         continue;
 577
 578                 return pfn;
 579         }
 580
 581         return 0;
 582 }
 583
 584 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
 585                              unsigned long end_pfn)
 586 {
 587         unsigned long zone_start_pfn = zone->zone_start_pfn;
 588         unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
 589         unsigned long zone_end_pfn = z;
 590         unsigned long pfn;
 591         struct mem_section *ms;
 592         int nid = zone_to_nid(zone);
 593
 594         zone_span_writelock(zone);
 595         if (zone_start_pfn == start_pfn) {
 596                 /*
 597                  * If the section is smallest section in the zone, it need
 598                  * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
 599                  * In this case, we find second smallest valid mem_section
 600                  * for shrinking zone.
 601                  */
 602                 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
 603                                                 zone_end_pfn);
 604                 if (pfn) {
 605                         zone->zone_start_pfn = pfn;
 606                         zone->spanned_pages = zone_end_pfn - pfn;
 607                 }
 608         } else if (zone_end_pfn == end_pfn) {
 609                 /*
 610                  * If the section is biggest section in the zone, it need
 611                  * shrink zone->spanned_pages.
 612                  * In this case, we find second biggest valid mem_section for
 613                  * shrinking zone.
 614                  */
 615                 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
 616                                                start_pfn);
 617                 if (pfn)
 618                         zone->spanned_pages = pfn - zone_start_pfn + 1;
 619         }
 620
 621         /*
 622          * The section is not biggest or smallest mem_section in the zone, it
 623          * only creates a hole in the zone. So in this case, we need not
 624          * change the zone. But perhaps, the zone has only hole data. Thus
 625          * it check the zone has only hole or not.
 626          */
 627         pfn = zone_start_pfn;
 628         for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
 629                 ms = __pfn_to_section(pfn);
 630
 631                 if (unlikely(!valid_section(ms)))
 632                         continue;
 633
 634                 if (page_zone(pfn_to_page(pfn)) != zone)
 635                         continue;
 636
 637                  /* If the section is current section, it continues the loop */
 638                 if (start_pfn == pfn)
 639                         continue;
 640
 641                 /* If we find valid section, we have nothing to do */
 642                 zone_span_writeunlock(zone);
 643                 return;
 644         }
 645
 646         /* The zone has no valid section */
 647         zone->zone_start_pfn = 0;
 648         zone->spanned_pages = 0;
 649         zone_span_writeunlock(zone);
 650 }
 651
 652 static void shrink_pgdat_span(struct pglist_data *pgdat,
 653                               unsigned long start_pfn, unsigned long end_pfn)
 654 {
 655         unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
 656         unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
 657         unsigned long pgdat_end_pfn = p;
 658         unsigned long pfn;
 659         struct mem_section *ms;
 660         int nid = pgdat->node_id;
 661
 662         if (pgdat_start_pfn == start_pfn) {
 663                 /*
 664                  * If the section is smallest section in the pgdat, it need
 665                  * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
 666                  * In this case, we find second smallest valid mem_section
 667                  * for shrinking zone.
 668                  */
 669                 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
 670                                                 pgdat_end_pfn);
 671                 if (pfn) {
 672                         pgdat->node_start_pfn = pfn;
 673                         pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
 674                 }
 675         } else if (pgdat_end_pfn == end_pfn) {
 676                 /*
 677                  * If the section is biggest section in the pgdat, it need
 678                  * shrink pgdat->node_spanned_pages.
 679                  * In this case, we find second biggest valid mem_section for
 680                  * shrinking zone.
 681                  */
 682                 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
 683                                                start_pfn);
 684                 if (pfn)
 685                         pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
 686         }
 687
 688         /*
 689          * If the section is not biggest or smallest mem_section in the pgdat,
 690          * it only creates a hole in the pgdat. So in this case, we need not
 691          * change the pgdat.
 692          * But perhaps, the pgdat has only hole data. Thus it check the pgdat
 693          * has only hole or not.
 694          */
 695         pfn = pgdat_start_pfn;
 696         for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
 697                 ms = __pfn_to_section(pfn);
 698
 699                 if (unlikely(!valid_section(ms)))
 700                         continue;
 701
 702                 if (pfn_to_nid(pfn) != nid)
 703                         continue;
 704
 705                  /* If the section is current section, it continues the loop */
 706                 if (start_pfn == pfn)
 707                         continue;
 708
 709                 /* If we find valid section, we have nothing to do */
 710                 return;
 711         }
 712
 713         /* The pgdat has no valid section */
 714         pgdat->node_start_pfn = 0;
 715         pgdat->node_spanned_pages = 0;
 716 }
 717
 718 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
 719 {
 720         struct pglist_data *pgdat = zone->zone_pgdat;
 721         int nr_pages = PAGES_PER_SECTION;
 722         int zone_type;
 723         unsigned long flags;
 724
 725         zone_type = zone - pgdat->node_zones;
 726
 727         pgdat_resize_lock(zone->zone_pgdat, &flags);
 728         shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
 729         shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
 730         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 731 }
 732
 733 static int __remove_section(struct zone *zone, struct mem_section *ms)
 734 {
 735         unsigned long start_pfn;
 736         int scn_nr;
 737         int ret = -EINVAL;
 738
 739         if (!valid_section(ms))
 740                 return ret;
 741
 742         ret = unregister_memory_section(ms);
 743         if (ret)
 744                 return ret;
 745
 746         scn_nr = __section_nr(ms);
 747         start_pfn = section_nr_to_pfn(scn_nr);
 748         __remove_zone(zone, start_pfn);
 749
 750         sparse_remove_one_section(zone, ms);
 751         return 0;
 752 }
 753
 754 /**
 755  * __remove_pages() - remove sections of pages from a zone
 756  * @zone: zone from which pages need to be removed
 757  * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
 758  * @nr_pages: number of pages to remove (must be multiple of section size)
 759  *
 760  * Generic helper function to remove section mappings and sysfs entries
 761  * for the section of the memory we are removing. Caller needs to make
 762  * sure that pages are marked reserved and zones are adjust properly by
 763  * calling offline_pages().
 764  */
 765 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
 766                  unsigned long nr_pages)
 767 {
 768         unsigned long i;
 769         int sections_to_remove;
 770         resource_size_t start, size;
 771         int ret = 0;
 772
 773         /*
 774          * We can only remove entire sections
 775          */
 776         BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
 777         BUG_ON(nr_pages % PAGES_PER_SECTION);
 778
 779         start = phys_start_pfn << PAGE_SHIFT;
 780         size = nr_pages * PAGE_SIZE;
 781         ret = release_mem_region_adjustable(&iomem_resource, start, size);
 782         if (ret) {
 783                 resource_size_t endres = start + size - 1;
 784
 785                 pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
 786                                 &start, &endres, ret);
 787         }
 788
 789         sections_to_remove = nr_pages / PAGES_PER_SECTION;
 790         for (i = 0; i < sections_to_remove; i++) {
 791                 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
 792                 ret = __remove_section(zone, __pfn_to_section(pfn));
 793                 if (ret)
 794                         break;
 795         }
 796         return ret;
 797 }
 798 EXPORT_SYMBOL_GPL(__remove_pages);
 799 #endif /* CONFIG_MEMORY_HOTREMOVE */
 800
 801 int set_online_page_callback(online_page_callback_t callback)
 802 {
 803         int rc = -EINVAL;
 804
 805         get_online_mems();
 806         mutex_lock(&online_page_callback_lock);
 807
 808         if (online_page_callback == generic_online_page) {
 809                 online_page_callback = callback;
 810                 rc = 0;
 811         }
 812
 813         mutex_unlock(&online_page_callback_lock);
 814         put_online_mems();
 815
 816         return rc;
 817 }
 818 EXPORT_SYMBOL_GPL(set_online_page_callback);
 819
 820 int restore_online_page_callback(online_page_callback_t callback)
 821 {
 822         int rc = -EINVAL;
 823
 824         get_online_mems();
 825         mutex_lock(&online_page_callback_lock);
 826
 827         if (online_page_callback == callback) {
 828                 online_page_callback = generic_online_page;
 829                 rc = 0;
 830         }
 831
 832         mutex_unlock(&online_page_callback_lock);
 833         put_online_mems();
 834
 835         return rc;
 836 }
 837 EXPORT_SYMBOL_GPL(restore_online_page_callback);
 838
 839 void __online_page_set_limits(struct page *page)
 840 {
 841 }
 842 EXPORT_SYMBOL_GPL(__online_page_set_limits);
 843
 844 void __online_page_increment_counters(struct page *page)
 845 {
 846         adjust_managed_page_count(page, 1);
 847 }
 848 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
 849
 850 void __online_page_free(struct page *page)
 851 {
 852         __free_reserved_page(page);
 853 }
 854 EXPORT_SYMBOL_GPL(__online_page_free);
 855
 856 static void generic_online_page(struct page *page)
 857 {
 858         __online_page_set_limits(page);
 859         __online_page_increment_counters(page);
 860         __online_page_free(page);
 861 }
 862
 863 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
 864                         void *arg)
 865 {
 866         unsigned long i;
 867         unsigned long onlined_pages = *(unsigned long *)arg;
 868         struct page *page;
 869         if (PageReserved(pfn_to_page(start_pfn)))
 870                 for (i = 0; i < nr_pages; i++) {
 871                         page = pfn_to_page(start_pfn + i);
 872                         (*online_page_callback)(page);
 873                         onlined_pages++;
 874                 }
 875         *(unsigned long *)arg = onlined_pages;
 876         return 0;
 877 }
 878
 879 #ifdef CONFIG_MOVABLE_NODE
 880 /*
 881  * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
 882  * normal memory.
 883  */
 884 static bool can_online_high_movable(struct zone *zone)
 885 {
 886         return true;
 887 }
 888 #else /* CONFIG_MOVABLE_NODE */
 889 /* ensure every online node has NORMAL memory */
 890 static bool can_online_high_movable(struct zone *zone)
 891 {
 892         return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
 893 }
 894 #endif /* CONFIG_MOVABLE_NODE */
 895
 896 /* check which state of node_states will be changed when online memory */
 897 static void node_states_check_changes_online(unsigned long nr_pages,
 898         struct zone *zone, struct memory_notify *arg)
 899 {
 900         int nid = zone_to_nid(zone);
 901         enum zone_type zone_last = ZONE_NORMAL;
 902
 903         /*
 904          * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
 905          * contains nodes which have zones of 0...ZONE_NORMAL,
 906          * set zone_last to ZONE_NORMAL.
 907          *
 908          * If we don't have HIGHMEM nor movable node,
 909          * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
 910          * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
 911          */
 912         if (N_MEMORY == N_NORMAL_MEMORY)
 913                 zone_last = ZONE_MOVABLE;
 914
 915         /*
 916          * if the memory to be online is in a zone of 0...zone_last, and
 917          * the zones of 0...zone_last don't have memory before online, we will
 918          * need to set the node to node_states[N_NORMAL_MEMORY] after
 919          * the memory is online.
 920          */
 921         if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
 922                 arg->status_change_nid_normal = nid;
 923         else
 924                 arg->status_change_nid_normal = -1;
 925
 926 #ifdef CONFIG_HIGHMEM
 927         /*
 928          * If we have movable node, node_states[N_HIGH_MEMORY]
 929          * contains nodes which have zones of 0...ZONE_HIGHMEM,
 930          * set zone_last to ZONE_HIGHMEM.
 931          *
 932          * If we don't have movable node, node_states[N_NORMAL_MEMORY]
 933          * contains nodes which have zones of 0...ZONE_MOVABLE,
 934          * set zone_last to ZONE_MOVABLE.
 935          */
 936         zone_last = ZONE_HIGHMEM;
 937         if (N_MEMORY == N_HIGH_MEMORY)
 938                 zone_last = ZONE_MOVABLE;
 939
 940         if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
 941                 arg->status_change_nid_high = nid;
 942         else
 943                 arg->status_change_nid_high = -1;
 944 #else
 945         arg->status_change_nid_high = arg->status_change_nid_normal;
 946 #endif
 947
 948         /*
 949          * if the node don't have memory befor online, we will need to
 950          * set the node to node_states[N_MEMORY] after the memory
 951          * is online.
 952          */
 953         if (!node_state(nid, N_MEMORY))
 954                 arg->status_change_nid = nid;
 955         else
 956                 arg->status_change_nid = -1;
 957 }
 958
 959 static void node_states_set_node(int node, struct memory_notify *arg)
 960 {
 961         if (arg->status_change_nid_normal >= 0)
 962                 node_set_state(node, N_NORMAL_MEMORY);
 963
 964         if (arg->status_change_nid_high >= 0)
 965                 node_set_state(node, N_HIGH_MEMORY);
 966
 967         node_set_state(node, N_MEMORY);
 968 }
 969
 970
 971 /* Must be protected by mem_hotplug_begin() */
 972 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
 973 {
 974         unsigned long flags;
 975         unsigned long onlined_pages = 0;
 976         struct zone *zone;
 977         int need_zonelists_rebuild = 0;
 978         int nid;
 979         int ret;
 980         struct memory_notify arg;
 981
 982         /*
 983          * This doesn't need a lock to do pfn_to_page().
 984          * The section can't be removed here because of the
 985          * memory_block->state_mutex.
 986          */
 987         zone = page_zone(pfn_to_page(pfn));
 988
 989         if ((zone_idx(zone) > ZONE_NORMAL ||
 990             online_type == MMOP_ONLINE_MOVABLE) &&
 991             !can_online_high_movable(zone))
 992                 return -EINVAL;
 993
 994         if (online_type == MMOP_ONLINE_KERNEL &&
 995             zone_idx(zone) == ZONE_MOVABLE) {
 996                 if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages))
 997                         return -EINVAL;
 998         }
 999         if (online_type == MMOP_ONLINE_MOVABLE &&
1000             zone_idx(zone) == ZONE_MOVABLE - 1) {
1001                 if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages))
1002                         return -EINVAL;
1003         }
1004
1005         /* Previous code may changed the zone of the pfn range */
1006         zone = page_zone(pfn_to_page(pfn));
1007
1008         arg.start_pfn = pfn;
1009         arg.nr_pages = nr_pages;
1010         node_states_check_changes_online(nr_pages, zone, &arg);
1011
1012         nid = pfn_to_nid(pfn);
1013
1014         ret = memory_notify(MEM_GOING_ONLINE, &arg);
1015         ret = notifier_to_errno(ret);
1016         if (ret) {
1017                 memory_notify(MEM_CANCEL_ONLINE, &arg);
1018                 return ret;
1019         }
1020         /*
1021          * If this zone is not populated, then it is not in zonelist.
1022          * This means the page allocator ignores this zone.
1023          * So, zonelist must be updated after online.
1024          */
1025         mutex_lock(&zonelists_mutex);
1026         if (!populated_zone(zone)) {
1027                 need_zonelists_rebuild = 1;
1028                 build_all_zonelists(NULL, zone);
1029         }
1030
1031         ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
1032                 online_pages_range);
1033         if (ret) {
1034                 if (need_zonelists_rebuild)
1035                         zone_pcp_reset(zone);
1036                 mutex_unlock(&zonelists_mutex);
1037                 printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n",
1038                        (unsigned long long) pfn << PAGE_SHIFT,
1039                        (((unsigned long long) pfn + nr_pages)
1040                             << PAGE_SHIFT) - 1);
1041                 memory_notify(MEM_CANCEL_ONLINE, &arg);
1042                 return ret;
1043         }
1044
1045         zone->present_pages += onlined_pages;
1046
1047         pgdat_resize_lock(zone->zone_pgdat, &flags);
1048         zone->zone_pgdat->node_present_pages += onlined_pages;
1049         pgdat_resize_unlock(zone->zone_pgdat, &flags);
1050
1051         if (onlined_pages) {
1052                 node_states_set_node(zone_to_nid(zone), &arg);
1053                 if (need_zonelists_rebuild)
1054                         build_all_zonelists(NULL, NULL);
1055                 else
1056                         zone_pcp_update(zone);
1057         }
1058
1059         mutex_unlock(&zonelists_mutex);
1060
1061         init_per_zone_wmark_min();
1062
1063         if (onlined_pages)
1064                 kswapd_run(zone_to_nid(zone));
1065
1066         vm_total_pages = nr_free_pagecache_pages();
1067
1068         writeback_set_ratelimit();
1069
1070         if (onlined_pages)
1071                 memory_notify(MEM_ONLINE, &arg);
1072         return 0;
1073 }
1074 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1075
1076 static void reset_node_present_pages(pg_data_t *pgdat)
1077 {
1078         struct zone *z;
1079
1080         for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1081                 z->present_pages = 0;
1082
1083         pgdat->node_present_pages = 0;
1084 }
1085
1086 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1087 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1088 {
1089         struct pglist_data *pgdat;
1090         unsigned long zones_size[MAX_NR_ZONES] = {0};
1091         unsigned long zholes_size[MAX_NR_ZONES] = {0};
1092         unsigned long start_pfn = PFN_DOWN(start);
1093
1094         pgdat = NODE_DATA(nid);
1095         if (!pgdat) {
1096                 pgdat = arch_alloc_nodedata(nid);
1097                 if (!pgdat)
1098                         return NULL;
1099
1100                 arch_refresh_nodedata(nid, pgdat);
1101         } else {
1102                 /* Reset the nr_zones and classzone_idx to 0 before reuse */
1103                 pgdat->nr_zones = 0;
1104                 pgdat->classzone_idx = 0;
1105         }
1106
1107         /* we can use NODE_DATA(nid) from here */
1108
1109         /* init node's zones as empty zones, we don't have any present pages.*/
1110         free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1111
1112         /*
1113          * The node we allocated has no zone fallback lists. For avoiding
1114          * to access not-initialized zonelist, build here.
1115          */
1116         mutex_lock(&zonelists_mutex);
1117         build_all_zonelists(pgdat, NULL);
1118         mutex_unlock(&zonelists_mutex);
1119
1120         /*
1121          * zone->managed_pages is set to an approximate value in
1122          * free_area_init_core(), which will cause
1123          * /sys/device/system/node/nodeX/meminfo has wrong data.
1124          * So reset it to 0 before any memory is onlined.
1125          */
1126         reset_node_managed_pages(pgdat);
1127
1128         /*
1129          * When memory is hot-added, all the memory is in offline state. So
1130          * clear all zones' present_pages because they will be updated in
1131          * online_pages() and offline_pages().
1132          */
1133         reset_node_present_pages(pgdat);
1134
1135         return pgdat;
1136 }
1137
1138 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1139 {
1140         arch_refresh_nodedata(nid, NULL);
1141         arch_free_nodedata(pgdat);
1142         return;
1143 }
1144
1145
1146 /**
1147  * try_online_node - online a node if offlined
1148  *
1149  * called by cpu_up() to online a node without onlined memory.
1150  */
1151 int try_online_node(int nid)
1152 {
1153         pg_data_t       *pgdat;
1154         int     ret;
1155
1156         if (node_online(nid))
1157                 return 0;
1158
1159         mem_hotplug_begin();
1160         pgdat = hotadd_new_pgdat(nid, 0);
1161         if (!pgdat) {
1162                 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1163                 ret = -ENOMEM;
1164                 goto out;
1165         }
1166         node_set_online(nid);
1167         ret = register_one_node(nid);
1168         BUG_ON(ret);
1169
1170         if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1171                 mutex_lock(&zonelists_mutex);
1172                 build_all_zonelists(NULL, NULL);
1173                 mutex_unlock(&zonelists_mutex);
1174         }
1175
1176 out:
1177         mem_hotplug_done();
1178         return ret;
1179 }
1180
1181 static int check_hotplug_memory_range(u64 start, u64 size)
1182 {
1183         u64 start_pfn = PFN_DOWN(start);
1184         u64 nr_pages = size >> PAGE_SHIFT;
1185
1186         /* Memory range must be aligned with section */
1187         if ((start_pfn & ~PAGE_SECTION_MASK) ||
1188             (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1189                 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1190                                 (unsigned long long)start,
1191                                 (unsigned long long)size);
1192                 return -EINVAL;
1193         }
1194
1195         return 0;
1196 }
1197
1198 /*
1199  * If movable zone has already been setup, newly added memory should be check.
1200  * If its address is higher than movable zone, it should be added as movable.
1201  * Without this check, movable zone may overlap with other zone.
1202  */
1203 static int should_add_memory_movable(int nid, u64 start, u64 size)
1204 {
1205         unsigned long start_pfn = start >> PAGE_SHIFT;
1206         pg_data_t *pgdat = NODE_DATA(nid);
1207         struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
1208
1209         if (zone_is_empty(movable_zone))
1210                 return 0;
1211
1212         if (movable_zone->zone_start_pfn <= start_pfn)
1213                 return 1;
1214
1215         return 0;
1216 }
1217
1218 int zone_for_memory(int nid, u64 start, u64 size, int zone_default)
1219 {
1220         if (should_add_memory_movable(nid, start, size))
1221                 return ZONE_MOVABLE;
1222
1223         return zone_default;
1224 }
1225
1226 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1227 int __ref add_memory(int nid, u64 start, u64 size)
1228 {
1229         pg_data_t *pgdat = NULL;
1230         bool new_pgdat;
1231         bool new_node;
1232         struct resource *res;
1233         int ret;
1234
1235         ret = check_hotplug_memory_range(start, size);
1236         if (ret)
1237                 return ret;
1238
1239         res = register_memory_resource(start, size);
1240         ret = -EEXIST;
1241         if (!res)
1242                 return ret;
1243
1244         {       /* Stupid hack to suppress address-never-null warning */
1245                 void *p = NODE_DATA(nid);
1246                 new_pgdat = !p;
1247         }
1248
1249         mem_hotplug_begin();
1250
1251         new_node = !node_online(nid);
1252         if (new_node) {
1253                 pgdat = hotadd_new_pgdat(nid, start);
1254                 ret = -ENOMEM;
1255                 if (!pgdat)
1256                         goto error;
1257         }
1258
1259         /* call arch's memory hotadd */
1260         ret = arch_add_memory(nid, start, size);
1261
1262         if (ret < 0)
1263                 goto error;
1264
1265         /* we online node here. we can't roll back from here. */
1266         node_set_online(nid);
1267
1268         if (new_node) {
1269                 ret = register_one_node(nid);
1270                 /*
1271                  * If sysfs file of new node can't create, cpu on the node
1272                  * can't be hot-added. There is no rollback way now.
1273                  * So, check by BUG_ON() to catch it reluctantly..
1274                  */
1275                 BUG_ON(ret);
1276         }
1277
1278         /* create new memmap entry */
1279         firmware_map_add_hotplug(start, start + size, "System RAM");
1280
1281         goto out;
1282
1283 error:
1284         /* rollback pgdat allocation and others */
1285         if (new_pgdat)
1286                 rollback_node_hotadd(nid, pgdat);
1287         release_memory_resource(res);
1288
1289 out:
1290         mem_hotplug_done();
1291         return ret;
1292 }
1293 EXPORT_SYMBOL_GPL(add_memory);
1294
1295 #ifdef CONFIG_MEMORY_HOTREMOVE
1296 /*
1297  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1298  * set and the size of the free page is given by page_order(). Using this,
1299  * the function determines if the pageblock contains only free pages.
1300  * Due to buddy contraints, a free page at least the size of a pageblock will
1301  * be located at the start of the pageblock
1302  */
1303 static inline int pageblock_free(struct page *page)
1304 {
1305         return PageBuddy(page) && page_order(page) >= pageblock_order;
1306 }
1307
1308 /* Return the start of the next active pageblock after a given page */
1309 static struct page *next_active_pageblock(struct page *page)
1310 {
1311         /* Ensure the starting page is pageblock-aligned */
1312         BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1313
1314         /* If the entire pageblock is free, move to the end of free page */
1315         if (pageblock_free(page)) {
1316                 int order;
1317                 /* be careful. we don't have locks, page_order can be changed.*/
1318                 order = page_order(page);
1319                 if ((order < MAX_ORDER) && (order >= pageblock_order))
1320                         return page + (1 << order);
1321         }
1322
1323         return page + pageblock_nr_pages;
1324 }
1325
1326 /* Checks if this range of memory is likely to be hot-removable. */
1327 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1328 {
1329         struct page *page = pfn_to_page(start_pfn);
1330         struct page *end_page = page + nr_pages;
1331
1332         /* Check the starting page of each pageblock within the range */
1333         for (; page < end_page; page = next_active_pageblock(page)) {
1334                 if (!is_pageblock_removable_nolock(page))
1335                         return 0;
1336                 cond_resched();
1337         }
1338
1339         /* All pageblocks in the memory block are likely to be hot-removable */
1340         return 1;
1341 }
1342
1343 /*
1344  * Confirm all pages in a range [start, end) is belongs to the same zone.
1345  */
1346 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
1347 {
1348         unsigned long pfn;
1349         struct zone *zone = NULL;
1350         struct page *page;
1351         int i;
1352         for (pfn = start_pfn;
1353              pfn < end_pfn;
1354              pfn += MAX_ORDER_NR_PAGES) {
1355                 i = 0;
1356                 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1357                 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
1358                         i++;
1359                 if (i == MAX_ORDER_NR_PAGES)
1360                         continue;
1361                 page = pfn_to_page(pfn + i);
1362                 if (zone && page_zone(page) != zone)
1363                         return 0;
1364                 zone = page_zone(page);
1365         }
1366         return 1;
1367 }
1368
1369 /*
1370  * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1371  * and hugepages). We scan pfn because it's much easier than scanning over
1372  * linked list. This function returns the pfn of the first found movable
1373  * page if it's found, otherwise 0.
1374  */
1375 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1376 {
1377         unsigned long pfn;
1378         struct page *page;
1379         for (pfn = start; pfn < end; pfn++) {
1380                 if (pfn_valid(pfn)) {
1381                         page = pfn_to_page(pfn);
1382                         if (PageLRU(page))
1383                                 return pfn;
1384                         if (PageHuge(page)) {
1385                                 if (page_huge_active(page))
1386                                         return pfn;
1387                                 else
1388                                         pfn = round_up(pfn + 1,
1389                                                 1 << compound_order(page)) - 1;
1390                         }
1391                 }
1392         }
1393         return 0;
1394 }
1395
1396 #define NR_OFFLINE_AT_ONCE_PAGES        (256)
1397 static int
1398 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1399 {
1400         unsigned long pfn;
1401         struct page *page;
1402         int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1403         int not_managed = 0;
1404         int ret = 0;
1405         LIST_HEAD(source);
1406
1407         for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1408                 if (!pfn_valid(pfn))
1409                         continue;
1410                 page = pfn_to_page(pfn);
1411
1412                 if (PageHuge(page)) {
1413                         struct page *head = compound_head(page);
1414                         pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1415                         if (compound_order(head) > PFN_SECTION_SHIFT) {
1416                                 ret = -EBUSY;
1417                                 break;
1418                         }
1419                         if (isolate_huge_page(page, &source))
1420                                 move_pages -= 1 << compound_order(head);
1421                         continue;
1422                 }
1423
1424                 if (!get_page_unless_zero(page))
1425                         continue;
1426                 /*
1427                  * We can skip free pages. And we can only deal with pages on
1428                  * LRU.
1429                  */
1430                 ret = isolate_lru_page(page);
1431                 if (!ret) { /* Success */
1432                         put_page(page);
1433                         list_add_tail(&page->lru, &source);
1434                         move_pages--;
1435                         inc_zone_page_state(page, NR_ISOLATED_ANON +
1436                                             page_is_file_cache(page));
1437
1438                 } else {
1439 #ifdef CONFIG_DEBUG_VM
1440                         printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
1441                                pfn);
1442                         dump_page(page, "failed to remove from LRU");
1443 #endif
1444                         put_page(page);
1445                         /* Because we don't have big zone->lock. we should
1446                            check this again here. */
1447                         if (page_count(page)) {
1448                                 not_managed++;
1449                                 ret = -EBUSY;
1450                                 break;
1451                         }
1452                 }
1453         }
1454         if (!list_empty(&source)) {
1455                 if (not_managed) {
1456                         putback_movable_pages(&source);
1457                         goto out;
1458                 }
1459
1460                 /*
1461                  * alloc_migrate_target should be improooooved!!
1462                  * migrate_pages returns # of failed pages.
1463                  */
1464                 ret = migrate_pages(&source, alloc_migrate_target, NULL, 0,
1465                                         MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1466                 if (ret)
1467                         putback_movable_pages(&source);
1468         }
1469 out:
1470         return ret;
1471 }
1472
1473 /*
1474  * remove from free_area[] and mark all as Reserved.
1475  */
1476 static int
1477 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1478                         void *data)
1479 {
1480         __offline_isolated_pages(start, start + nr_pages);
1481         return 0;
1482 }
1483
1484 static void
1485 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1486 {
1487         walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1488                                 offline_isolated_pages_cb);
1489 }
1490
1491 /*
1492  * Check all pages in range, recoreded as memory resource, are isolated.
1493  */
1494 static int
1495 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1496                         void *data)
1497 {
1498         int ret;
1499         long offlined = *(long *)data;
1500         ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1501         offlined = nr_pages;
1502         if (!ret)
1503                 *(long *)data += offlined;
1504         return ret;
1505 }
1506
1507 static long
1508 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1509 {
1510         long offlined = 0;
1511         int ret;
1512
1513         ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1514                         check_pages_isolated_cb);
1515         if (ret < 0)
1516                 offlined = (long)ret;
1517         return offlined;
1518 }
1519
1520 #ifdef CONFIG_MOVABLE_NODE
1521 /*
1522  * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1523  * normal memory.
1524  */
1525 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1526 {
1527         return true;
1528 }
1529 #else /* CONFIG_MOVABLE_NODE */
1530 /* ensure the node has NORMAL memory if it is still online */
1531 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1532 {
1533         struct pglist_data *pgdat = zone->zone_pgdat;
1534         unsigned long present_pages = 0;
1535         enum zone_type zt;
1536
1537         for (zt = 0; zt <= ZONE_NORMAL; zt++)
1538                 present_pages += pgdat->node_zones[zt].present_pages;
1539
1540         if (present_pages > nr_pages)
1541                 return true;
1542
1543         present_pages = 0;
1544         for (; zt <= ZONE_MOVABLE; zt++)
1545                 present_pages += pgdat->node_zones[zt].present_pages;
1546
1547         /*
1548          * we can't offline the last normal memory until all
1549          * higher memory is offlined.
1550          */
1551         return present_pages == 0;
1552 }
1553 #endif /* CONFIG_MOVABLE_NODE */
1554
1555 static int __init cmdline_parse_movable_node(char *p)
1556 {
1557 #ifdef CONFIG_MOVABLE_NODE
1558         /*
1559          * Memory used by the kernel cannot be hot-removed because Linux
1560          * cannot migrate the kernel pages. When memory hotplug is
1561          * enabled, we should prevent memblock from allocating memory
1562          * for the kernel.
1563          *
1564          * ACPI SRAT records all hotpluggable memory ranges. But before
1565          * SRAT is parsed, we don't know about it.
1566          *
1567          * The kernel image is loaded into memory at very early time. We
1568          * cannot prevent this anyway. So on NUMA system, we set any
1569          * node the kernel resides in as un-hotpluggable.
1570          *
1571          * Since on modern servers, one node could have double-digit
1572          * gigabytes memory, we can assume the memory around the kernel
1573          * image is also un-hotpluggable. So before SRAT is parsed, just
1574          * allocate memory near the kernel image to try the best to keep
1575          * the kernel away from hotpluggable memory.
1576          */
1577         memblock_set_bottom_up(true);
1578         movable_node_enabled = true;
1579 #else
1580         pr_warn("movable_node option not supported\n");
1581 #endif
1582         return 0;
1583 }
1584 early_param("movable_node", cmdline_parse_movable_node);
1585
1586 /* check which state of node_states will be changed when offline memory */
1587 static void node_states_check_changes_offline(unsigned long nr_pages,
1588                 struct zone *zone, struct memory_notify *arg)
1589 {
1590         struct pglist_data *pgdat = zone->zone_pgdat;
1591         unsigned long present_pages = 0;
1592         enum zone_type zt, zone_last = ZONE_NORMAL;
1593
1594         /*
1595          * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1596          * contains nodes which have zones of 0...ZONE_NORMAL,
1597          * set zone_last to ZONE_NORMAL.
1598          *
1599          * If we don't have HIGHMEM nor movable node,
1600          * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1601          * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1602          */
1603         if (N_MEMORY == N_NORMAL_MEMORY)
1604                 zone_last = ZONE_MOVABLE;
1605
1606         /*
1607          * check whether node_states[N_NORMAL_MEMORY] will be changed.
1608          * If the memory to be offline is in a zone of 0...zone_last,
1609          * and it is the last present memory, 0...zone_last will
1610          * become empty after offline , thus we can determind we will
1611          * need to clear the node from node_states[N_NORMAL_MEMORY].
1612          */
1613         for (zt = 0; zt <= zone_last; zt++)
1614                 present_pages += pgdat->node_zones[zt].present_pages;
1615         if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1616                 arg->status_change_nid_normal = zone_to_nid(zone);
1617         else
1618                 arg->status_change_nid_normal = -1;
1619
1620 #ifdef CONFIG_HIGHMEM
1621         /*
1622          * If we have movable node, node_states[N_HIGH_MEMORY]
1623          * contains nodes which have zones of 0...ZONE_HIGHMEM,
1624          * set zone_last to ZONE_HIGHMEM.
1625          *
1626          * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1627          * contains nodes which have zones of 0...ZONE_MOVABLE,
1628          * set zone_last to ZONE_MOVABLE.
1629          */
1630         zone_last = ZONE_HIGHMEM;
1631         if (N_MEMORY == N_HIGH_MEMORY)
1632                 zone_last = ZONE_MOVABLE;
1633
1634         for (; zt <= zone_last; zt++)
1635                 present_pages += pgdat->node_zones[zt].present_pages;
1636         if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1637                 arg->status_change_nid_high = zone_to_nid(zone);
1638         else
1639                 arg->status_change_nid_high = -1;
1640 #else
1641         arg->status_change_nid_high = arg->status_change_nid_normal;
1642 #endif
1643
1644         /*
1645          * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1646          */
1647         zone_last = ZONE_MOVABLE;
1648
1649         /*
1650          * check whether node_states[N_HIGH_MEMORY] will be changed
1651          * If we try to offline the last present @nr_pages from the node,
1652          * we can determind we will need to clear the node from
1653          * node_states[N_HIGH_MEMORY].
1654          */
1655         for (; zt <= zone_last; zt++)
1656                 present_pages += pgdat->node_zones[zt].present_pages;
1657         if (nr_pages >= present_pages)
1658                 arg->status_change_nid = zone_to_nid(zone);
1659         else
1660                 arg->status_change_nid = -1;
1661 }
1662
1663 static void node_states_clear_node(int node, struct memory_notify *arg)
1664 {
1665         if (arg->status_change_nid_normal >= 0)
1666                 node_clear_state(node, N_NORMAL_MEMORY);
1667
1668         if ((N_MEMORY != N_NORMAL_MEMORY) &&
1669             (arg->status_change_nid_high >= 0))
1670                 node_clear_state(node, N_HIGH_MEMORY);
1671
1672         if ((N_MEMORY != N_HIGH_MEMORY) &&
1673             (arg->status_change_nid >= 0))
1674                 node_clear_state(node, N_MEMORY);
1675 }
1676
1677 static int __ref __offline_pages(unsigned long start_pfn,
1678                   unsigned long end_pfn, unsigned long timeout)
1679 {
1680         unsigned long pfn, nr_pages, expire;
1681         long offlined_pages;
1682         int ret, drain, retry_max, node;
1683         unsigned long flags;
1684         struct zone *zone;
1685         struct memory_notify arg;
1686
1687         /* at least, alignment against pageblock is necessary */
1688         if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1689                 return -EINVAL;
1690         if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1691                 return -EINVAL;
1692         /* This makes hotplug much easier...and readable.
1693            we assume this for now. .*/
1694         if (!test_pages_in_a_zone(start_pfn, end_pfn))
1695                 return -EINVAL;
1696
1697         zone = page_zone(pfn_to_page(start_pfn));
1698         node = zone_to_nid(zone);
1699         nr_pages = end_pfn - start_pfn;
1700
1701         if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1702                 return -EINVAL;
1703
1704         /* set above range as isolated */
1705         ret = start_isolate_page_range(start_pfn, end_pfn,
1706                                        MIGRATE_MOVABLE, true);
1707         if (ret)
1708                 return ret;
1709
1710         arg.start_pfn = start_pfn;
1711         arg.nr_pages = nr_pages;
1712         node_states_check_changes_offline(nr_pages, zone, &arg);
1713
1714         ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1715         ret = notifier_to_errno(ret);
1716         if (ret)
1717                 goto failed_removal;
1718
1719         pfn = start_pfn;
1720         expire = jiffies + timeout;
1721         drain = 0;
1722         retry_max = 5;
1723 repeat:
1724         /* start memory hot removal */
1725         ret = -EAGAIN;
1726         if (time_after(jiffies, expire))
1727                 goto failed_removal;
1728         ret = -EINTR;
1729         if (signal_pending(current))
1730                 goto failed_removal;
1731         ret = 0;
1732         if (drain) {
1733                 lru_add_drain_all();
1734                 cond_resched();
1735                 drain_all_pages(zone);
1736         }
1737
1738         pfn = scan_movable_pages(start_pfn, end_pfn);
1739         if (pfn) { /* We have movable pages */
1740                 ret = do_migrate_range(pfn, end_pfn);
1741                 if (!ret) {
1742                         drain = 1;
1743                         goto repeat;
1744                 } else {
1745                         if (ret < 0)
1746                                 if (--retry_max == 0)
1747                                         goto failed_removal;
1748                         yield();
1749                         drain = 1;
1750                         goto repeat;
1751                 }
1752         }
1753         /* drain all zone's lru pagevec, this is asynchronous... */
1754         lru_add_drain_all();
1755         yield();
1756         /* drain pcp pages, this is synchronous. */
1757         drain_all_pages(zone);
1758         /*
1759          * dissolve free hugepages in the memory block before doing offlining
1760          * actually in order to make hugetlbfs's object counting consistent.
1761          */
1762         dissolve_free_huge_pages(start_pfn, end_pfn);
1763         /* check again */
1764         offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1765         if (offlined_pages < 0) {
1766                 ret = -EBUSY;
1767                 goto failed_removal;
1768         }
1769         printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
1770         /* Ok, all of our target is isolated.
1771            We cannot do rollback at this point. */
1772         offline_isolated_pages(start_pfn, end_pfn);
1773         /* reset pagetype flags and makes migrate type to be MOVABLE */
1774         undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1775         /* removal success */
1776         adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1777         zone->present_pages -= offlined_pages;
1778
1779         pgdat_resize_lock(zone->zone_pgdat, &flags);
1780         zone->zone_pgdat->node_present_pages -= offlined_pages;
1781         pgdat_resize_unlock(zone->zone_pgdat, &flags);
1782
1783         init_per_zone_wmark_min();
1784
1785         if (!populated_zone(zone)) {
1786                 zone_pcp_reset(zone);
1787                 mutex_lock(&zonelists_mutex);
1788                 build_all_zonelists(NULL, NULL);
1789                 mutex_unlock(&zonelists_mutex);
1790         } else
1791                 zone_pcp_update(zone);
1792
1793         node_states_clear_node(node, &arg);
1794         if (arg.status_change_nid >= 0)
1795                 kswapd_stop(node);
1796
1797         vm_total_pages = nr_free_pagecache_pages();
1798         writeback_set_ratelimit();
1799
1800         memory_notify(MEM_OFFLINE, &arg);
1801         return 0;
1802
1803 failed_removal:
1804         printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
1805                (unsigned long long) start_pfn << PAGE_SHIFT,
1806                ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1807         memory_notify(MEM_CANCEL_OFFLINE, &arg);
1808         /* pushback to free area */
1809         undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1810         return ret;
1811 }
1812
1813 /* Must be protected by mem_hotplug_begin() */
1814 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1815 {
1816         return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1817 }
1818 #endif /* CONFIG_MEMORY_HOTREMOVE */
1819
1820 /**
1821  * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1822  * @start_pfn: start pfn of the memory range
1823  * @end_pfn: end pfn of the memory range
1824  * @arg: argument passed to func
1825  * @func: callback for each memory section walked
1826  *
1827  * This function walks through all present mem sections in range
1828  * [start_pfn, end_pfn) and call func on each mem section.
1829  *
1830  * Returns the return value of func.
1831  */
1832 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1833                 void *arg, int (*func)(struct memory_block *, void *))
1834 {
1835         struct memory_block *mem = NULL;
1836         struct mem_section *section;
1837         unsigned long pfn, section_nr;
1838         int ret;
1839
1840         for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1841                 section_nr = pfn_to_section_nr(pfn);
1842                 if (!present_section_nr(section_nr))
1843                         continue;
1844
1845                 section = __nr_to_section(section_nr);
1846                 /* same memblock? */
1847                 if (mem)
1848                         if ((section_nr >= mem->start_section_nr) &&
1849                             (section_nr <= mem->end_section_nr))
1850                                 continue;
1851
1852                 mem = find_memory_block_hinted(section, mem);
1853                 if (!mem)
1854                         continue;
1855
1856                 ret = func(mem, arg);
1857                 if (ret) {
1858                         kobject_put(&mem->dev.kobj);
1859                         return ret;
1860                 }
1861         }
1862
1863         if (mem)
1864                 kobject_put(&mem->dev.kobj);
1865
1866         return 0;
1867 }
1868
1869 #ifdef CONFIG_MEMORY_HOTREMOVE
1870 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1871 {
1872         int ret = !is_memblock_offlined(mem);
1873
1874         if (unlikely(ret)) {
1875                 phys_addr_t beginpa, endpa;
1876
1877                 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1878                 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1879                 pr_warn("removing memory fails, because memory "
1880                         "[%pa-%pa] is onlined\n",
1881                         &beginpa, &endpa);
1882         }
1883
1884         return ret;
1885 }
1886
1887 static int check_cpu_on_node(pg_data_t *pgdat)
1888 {
1889         int cpu;
1890
1891         for_each_present_cpu(cpu) {
1892                 if (cpu_to_node(cpu) == pgdat->node_id)
1893                         /*
1894                          * the cpu on this node isn't removed, and we can't
1895                          * offline this node.
1896                          */
1897                         return -EBUSY;
1898         }
1899
1900         return 0;
1901 }
1902
1903 static void unmap_cpu_on_node(pg_data_t *pgdat)
1904 {
1905 #ifdef CONFIG_ACPI_NUMA
1906         int cpu;
1907
1908         for_each_possible_cpu(cpu)
1909                 if (cpu_to_node(cpu) == pgdat->node_id)
1910                         numa_clear_node(cpu);
1911 #endif
1912 }
1913
1914 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
1915 {
1916         int ret;
1917
1918         ret = check_cpu_on_node(pgdat);
1919         if (ret)
1920                 return ret;
1921
1922         /*
1923          * the node will be offlined when we come here, so we can clear
1924          * the cpu_to_node() now.
1925          */
1926
1927         unmap_cpu_on_node(pgdat);
1928         return 0;
1929 }
1930
1931 /**
1932  * try_offline_node
1933  *
1934  * Offline a node if all memory sections and cpus of the node are removed.
1935  *
1936  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1937  * and online/offline operations before this call.
1938  */
1939 void try_offline_node(int nid)
1940 {
1941         pg_data_t *pgdat = NODE_DATA(nid);
1942         unsigned long start_pfn = pgdat->node_start_pfn;
1943         unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
1944         unsigned long pfn;
1945         int i;
1946
1947         for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1948                 unsigned long section_nr = pfn_to_section_nr(pfn);
1949
1950                 if (!present_section_nr(section_nr))
1951                         continue;
1952
1953                 if (pfn_to_nid(pfn) != nid)
1954                         continue;
1955
1956                 /*
1957                  * some memory sections of this node are not removed, and we
1958                  * can't offline node now.
1959                  */
1960                 return;
1961         }
1962
1963         if (check_and_unmap_cpu_on_node(pgdat))
1964                 return;
1965
1966         /*
1967          * all memory/cpu of this node are removed, we can offline this
1968          * node now.
1969          */
1970         node_set_offline(nid);
1971         unregister_one_node(nid);
1972
1973         /* free waittable in each zone */
1974         for (i = 0; i < MAX_NR_ZONES; i++) {
1975                 struct zone *zone = pgdat->node_zones + i;
1976
1977                 /*
1978                  * wait_table may be allocated from boot memory,
1979                  * here only free if it's allocated by vmalloc.
1980                  */
1981                 if (is_vmalloc_addr(zone->wait_table)) {
1982                         vfree(zone->wait_table);
1983                         zone->wait_table = NULL;
1984                 }
1985         }
1986 }
1987 EXPORT_SYMBOL(try_offline_node);
1988
1989 /**
1990  * remove_memory
1991  *
1992  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
1993  * and online/offline operations before this call, as required by
1994  * try_offline_node().
1995  */
1996 void __ref remove_memory(int nid, u64 start, u64 size)
1997 {
1998         int ret;
1999
2000         BUG_ON(check_hotplug_memory_range(start, size));
2001
2002         mem_hotplug_begin();
2003
2004         /*
2005          * All memory blocks must be offlined before removing memory.  Check
2006          * whether all memory blocks in question are offline and trigger a BUG()
2007          * if this is not the case.
2008          */
2009         ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
2010                                 check_memblock_offlined_cb);
2011         if (ret)
2012                 BUG();
2013
2014         /* remove memmap entry */
2015         firmware_map_remove(start, start + size, "System RAM");
2016
2017         arch_remove_memory(start, size);
2018
2019         try_offline_node(nid);
2020
2021         mem_hotplug_done();
2022 }
2023 EXPORT_SYMBOL_GPL(remove_memory);
2024 #endif /* CONFIG_MEMORY_HOTREMOVE */