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/memremap.h>
  21 #include <linux/memory_hotplug.h>
  22 #include <linux/highmem.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/ioport.h>
  25 #include <linux/delay.h>
  26 #include <linux/migrate.h>
  27 #include <linux/page-isolation.h>
  28 #include <linux/pfn.h>
  29 #include <linux/suspend.h>
  30 #include <linux/mm_inline.h>
  31 #include <linux/firmware-map.h>
  32 #include <linux/stop_machine.h>
  33 #include <linux/hugetlb.h>
  34 #include <linux/memblock.h>
  35 #include <linux/bootmem.h>
  36
  37 #include <asm/tlbflush.h>
  38
  39 #include "internal.h"
  40
  41 /*
  42  * online_page_callback contains pointer to current page onlining function.
  43  * Initially it is generic_online_page(). If it is required it could be
  44  * changed by calling set_online_page_callback() for callback registration
  45  * and restore_online_page_callback() for generic callback restore.
  46  */
  47
  48 static void generic_online_page(struct page *page);
  49
  50 static online_page_callback_t online_page_callback = generic_online_page;
  51 static DEFINE_MUTEX(online_page_callback_lock);
  52
  53 /* The same as the cpu_hotplug lock, but for memory hotplug. */
  54 static struct {
  55         struct task_struct *active_writer;
  56         struct mutex lock; /* Synchronizes accesses to refcount, */
  57         /*
  58          * Also blocks the new readers during
  59          * an ongoing mem hotplug operation.
  60          */
  61         int refcount;
  62
  63 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  64         struct lockdep_map dep_map;
  65 #endif
  66 } mem_hotplug = {
  67         .active_writer = NULL,
  68         .lock = __MUTEX_INITIALIZER(mem_hotplug.lock),
  69         .refcount = 0,
  70 #ifdef CONFIG_DEBUG_LOCK_ALLOC
  71         .dep_map = {.name = "mem_hotplug.lock" },
  72 #endif
  73 };
  74
  75 /* Lockdep annotations for get/put_online_mems() and mem_hotplug_begin/end() */
  76 #define memhp_lock_acquire_read() lock_map_acquire_read(&mem_hotplug.dep_map)
  77 #define memhp_lock_acquire()      lock_map_acquire(&mem_hotplug.dep_map)
  78 #define memhp_lock_release()      lock_map_release(&mem_hotplug.dep_map)
  79
  80 void get_online_mems(void)
  81 {
  82         might_sleep();
  83         if (mem_hotplug.active_writer == current)
  84                 return;
  85         memhp_lock_acquire_read();
  86         mutex_lock(&mem_hotplug.lock);
  87         mem_hotplug.refcount++;
  88         mutex_unlock(&mem_hotplug.lock);
  89
  90 }
  91
  92 void put_online_mems(void)
  93 {
  94         if (mem_hotplug.active_writer == current)
  95                 return;
  96         mutex_lock(&mem_hotplug.lock);
  97
  98         if (WARN_ON(!mem_hotplug.refcount))
  99                 mem_hotplug.refcount++; /* try to fix things up */
 100
 101         if (!--mem_hotplug.refcount && unlikely(mem_hotplug.active_writer))
 102                 wake_up_process(mem_hotplug.active_writer);
 103         mutex_unlock(&mem_hotplug.lock);
 104         memhp_lock_release();
 105
 106 }
 107
 108 void mem_hotplug_begin(void)
 109 {
 110         mem_hotplug.active_writer = current;
 111
 112         memhp_lock_acquire();
 113         for (;;) {
 114                 mutex_lock(&mem_hotplug.lock);
 115                 if (likely(!mem_hotplug.refcount))
 116                         break;
 117                 __set_current_state(TASK_UNINTERRUPTIBLE);
 118                 mutex_unlock(&mem_hotplug.lock);
 119                 schedule();
 120         }
 121 }
 122
 123 void mem_hotplug_done(void)
 124 {
 125         mem_hotplug.active_writer = NULL;
 126         mutex_unlock(&mem_hotplug.lock);
 127         memhp_lock_release();
 128 }
 129
 130 /* add this memory to iomem resource */
 131 static struct resource *register_memory_resource(u64 start, u64 size)
 132 {
 133         struct resource *res;
 134         res = kzalloc(sizeof(struct resource), GFP_KERNEL);
 135         if (!res)
 136                 return ERR_PTR(-ENOMEM);
 137
 138         res->name = "System RAM";
 139         res->start = start;
 140         res->end = start + size - 1;
 141         res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
 142         if (request_resource(&iomem_resource, res) < 0) {
 143                 pr_debug("System RAM resource %pR cannot be added\n", res);
 144                 kfree(res);
 145                 return ERR_PTR(-EEXIST);
 146         }
 147         return res;
 148 }
 149
 150 static void release_memory_resource(struct resource *res)
 151 {
 152         if (!res)
 153                 return;
 154         release_resource(res);
 155         kfree(res);
 156         return;
 157 }
 158
 159 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
 160 void get_page_bootmem(unsigned long info,  struct page *page,
 161                       unsigned long type)
 162 {
 163         page->lru.next = (struct list_head *) type;
 164         SetPagePrivate(page);
 165         set_page_private(page, info);
 166         atomic_inc(&page->_count);
 167 }
 168
 169 void put_page_bootmem(struct page *page)
 170 {
 171         unsigned long type;
 172
 173         type = (unsigned long) page->lru.next;
 174         BUG_ON(type < MEMORY_HOTPLUG_MIN_BOOTMEM_TYPE ||
 175                type > MEMORY_HOTPLUG_MAX_BOOTMEM_TYPE);
 176
 177         if (atomic_dec_return(&page->_count) == 1) {
 178                 ClearPagePrivate(page);
 179                 set_page_private(page, 0);
 180                 INIT_LIST_HEAD(&page->lru);
 181                 free_reserved_page(page);
 182         }
 183 }
 184
 185 #ifdef CONFIG_HAVE_BOOTMEM_INFO_NODE
 186 #ifndef CONFIG_SPARSEMEM_VMEMMAP
 187 static void register_page_bootmem_info_section(unsigned long start_pfn)
 188 {
 189         unsigned long *usemap, mapsize, section_nr, i;
 190         struct mem_section *ms;
 191         struct page *page, *memmap;
 192
 193         section_nr = pfn_to_section_nr(start_pfn);
 194         ms = __nr_to_section(section_nr);
 195
 196         /* Get section's memmap address */
 197         memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
 198
 199         /*
 200          * Get page for the memmap's phys address
 201          * XXX: need more consideration for sparse_vmemmap...
 202          */
 203         page = virt_to_page(memmap);
 204         mapsize = sizeof(struct page) * PAGES_PER_SECTION;
 205         mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
 206
 207         /* remember memmap's page */
 208         for (i = 0; i < mapsize; i++, page++)
 209                 get_page_bootmem(section_nr, page, SECTION_INFO);
 210
 211         usemap = __nr_to_section(section_nr)->pageblock_flags;
 212         page = virt_to_page(usemap);
 213
 214         mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
 215
 216         for (i = 0; i < mapsize; i++, page++)
 217                 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
 218
 219 }
 220 #else /* CONFIG_SPARSEMEM_VMEMMAP */
 221 static void register_page_bootmem_info_section(unsigned long start_pfn)
 222 {
 223         unsigned long *usemap, mapsize, section_nr, i;
 224         struct mem_section *ms;
 225         struct page *page, *memmap;
 226
 227         if (!pfn_valid(start_pfn))
 228                 return;
 229
 230         section_nr = pfn_to_section_nr(start_pfn);
 231         ms = __nr_to_section(section_nr);
 232
 233         memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
 234
 235         register_page_bootmem_memmap(section_nr, memmap, PAGES_PER_SECTION);
 236
 237         usemap = __nr_to_section(section_nr)->pageblock_flags;
 238         page = virt_to_page(usemap);
 239
 240         mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
 241
 242         for (i = 0; i < mapsize; i++, page++)
 243                 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
 244 }
 245 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
 246
 247 void register_page_bootmem_info_node(struct pglist_data *pgdat)
 248 {
 249         unsigned long i, pfn, end_pfn, nr_pages;
 250         int node = pgdat->node_id;
 251         struct page *page;
 252         struct zone *zone;
 253
 254         nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
 255         page = virt_to_page(pgdat);
 256
 257         for (i = 0; i < nr_pages; i++, page++)
 258                 get_page_bootmem(node, page, NODE_INFO);
 259
 260         zone = &pgdat->node_zones[0];
 261         for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
 262                 if (zone_is_initialized(zone)) {
 263                         nr_pages = zone->wait_table_hash_nr_entries
 264                                 * sizeof(wait_queue_head_t);
 265                         nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
 266                         page = virt_to_page(zone->wait_table);
 267
 268                         for (i = 0; i < nr_pages; i++, page++)
 269                                 get_page_bootmem(node, page, NODE_INFO);
 270                 }
 271         }
 272
 273         pfn = pgdat->node_start_pfn;
 274         end_pfn = pgdat_end_pfn(pgdat);
 275
 276         /* register section info */
 277         for (; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
 278                 /*
 279                  * Some platforms can assign the same pfn to multiple nodes - on
 280                  * node0 as well as nodeN.  To avoid registering a pfn against
 281                  * multiple nodes we check that this pfn does not already
 282                  * reside in some other nodes.
 283                  */
 284                 if (pfn_valid(pfn) && (pfn_to_nid(pfn) == node))
 285                         register_page_bootmem_info_section(pfn);
 286         }
 287 }
 288 #endif /* CONFIG_HAVE_BOOTMEM_INFO_NODE */
 289
 290 static void __meminit grow_zone_span(struct zone *zone, unsigned long start_pfn,
 291                                      unsigned long end_pfn)
 292 {
 293         unsigned long old_zone_end_pfn;
 294
 295         zone_span_writelock(zone);
 296
 297         old_zone_end_pfn = zone_end_pfn(zone);
 298         if (zone_is_empty(zone) || start_pfn < zone->zone_start_pfn)
 299                 zone->zone_start_pfn = start_pfn;
 300
 301         zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
 302                                 zone->zone_start_pfn;
 303
 304         zone_span_writeunlock(zone);
 305 }
 306
 307 static void resize_zone(struct zone *zone, unsigned long start_pfn,
 308                 unsigned long end_pfn)
 309 {
 310         zone_span_writelock(zone);
 311
 312         if (end_pfn - start_pfn) {
 313                 zone->zone_start_pfn = start_pfn;
 314                 zone->spanned_pages = end_pfn - start_pfn;
 315         } else {
 316                 /*
 317                  * make it consist as free_area_init_core(),
 318                  * if spanned_pages = 0, then keep start_pfn = 0
 319                  */
 320                 zone->zone_start_pfn = 0;
 321                 zone->spanned_pages = 0;
 322         }
 323
 324         zone_span_writeunlock(zone);
 325 }
 326
 327 static void fix_zone_id(struct zone *zone, unsigned long start_pfn,
 328                 unsigned long end_pfn)
 329 {
 330         enum zone_type zid = zone_idx(zone);
 331         int nid = zone->zone_pgdat->node_id;
 332         unsigned long pfn;
 333
 334         for (pfn = start_pfn; pfn < end_pfn; pfn++)
 335                 set_page_links(pfn_to_page(pfn), zid, nid, pfn);
 336 }
 337
 338 /* Can fail with -ENOMEM from allocating a wait table with vmalloc() or
 339  * alloc_bootmem_node_nopanic()/memblock_virt_alloc_node_nopanic() */
 340 static int __ref ensure_zone_is_initialized(struct zone *zone,
 341                         unsigned long start_pfn, unsigned long num_pages)
 342 {
 343         if (!zone_is_initialized(zone))
 344                 return init_currently_empty_zone(zone, start_pfn, num_pages);
 345
 346         return 0;
 347 }
 348
 349 static int __meminit move_pfn_range_left(struct zone *z1, struct zone *z2,
 350                 unsigned long start_pfn, unsigned long end_pfn)
 351 {
 352         int ret;
 353         unsigned long flags;
 354         unsigned long z1_start_pfn;
 355
 356         ret = ensure_zone_is_initialized(z1, start_pfn, end_pfn - start_pfn);
 357         if (ret)
 358                 return ret;
 359
 360         pgdat_resize_lock(z1->zone_pgdat, &flags);
 361
 362         /* can't move pfns which are higher than @z2 */
 363         if (end_pfn > zone_end_pfn(z2))
 364                 goto out_fail;
 365         /* the move out part must be at the left most of @z2 */
 366         if (start_pfn > z2->zone_start_pfn)
 367                 goto out_fail;
 368         /* must included/overlap */
 369         if (end_pfn <= z2->zone_start_pfn)
 370                 goto out_fail;
 371
 372         /* use start_pfn for z1's start_pfn if z1 is empty */
 373         if (!zone_is_empty(z1))
 374                 z1_start_pfn = z1->zone_start_pfn;
 375         else
 376                 z1_start_pfn = start_pfn;
 377
 378         resize_zone(z1, z1_start_pfn, end_pfn);
 379         resize_zone(z2, end_pfn, zone_end_pfn(z2));
 380
 381         pgdat_resize_unlock(z1->zone_pgdat, &flags);
 382
 383         fix_zone_id(z1, start_pfn, end_pfn);
 384
 385         return 0;
 386 out_fail:
 387         pgdat_resize_unlock(z1->zone_pgdat, &flags);
 388         return -1;
 389 }
 390
 391 static int __meminit move_pfn_range_right(struct zone *z1, struct zone *z2,
 392                 unsigned long start_pfn, unsigned long end_pfn)
 393 {
 394         int ret;
 395         unsigned long flags;
 396         unsigned long z2_end_pfn;
 397
 398         ret = ensure_zone_is_initialized(z2, start_pfn, end_pfn - start_pfn);
 399         if (ret)
 400                 return ret;
 401
 402         pgdat_resize_lock(z1->zone_pgdat, &flags);
 403
 404         /* can't move pfns which are lower than @z1 */
 405         if (z1->zone_start_pfn > start_pfn)
 406                 goto out_fail;
 407         /* the move out part mast at the right most of @z1 */
 408         if (zone_end_pfn(z1) >  end_pfn)
 409                 goto out_fail;
 410         /* must included/overlap */
 411         if (start_pfn >= zone_end_pfn(z1))
 412                 goto out_fail;
 413
 414         /* use end_pfn for z2's end_pfn if z2 is empty */
 415         if (!zone_is_empty(z2))
 416                 z2_end_pfn = zone_end_pfn(z2);
 417         else
 418                 z2_end_pfn = end_pfn;
 419
 420         resize_zone(z1, z1->zone_start_pfn, start_pfn);
 421         resize_zone(z2, start_pfn, z2_end_pfn);
 422
 423         pgdat_resize_unlock(z1->zone_pgdat, &flags);
 424
 425         fix_zone_id(z2, start_pfn, end_pfn);
 426
 427         return 0;
 428 out_fail:
 429         pgdat_resize_unlock(z1->zone_pgdat, &flags);
 430         return -1;
 431 }
 432
 433 static void __meminit grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
 434                                       unsigned long end_pfn)
 435 {
 436         unsigned long old_pgdat_end_pfn = pgdat_end_pfn(pgdat);
 437
 438         if (!pgdat->node_spanned_pages || start_pfn < pgdat->node_start_pfn)
 439                 pgdat->node_start_pfn = start_pfn;
 440
 441         pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
 442                                         pgdat->node_start_pfn;
 443 }
 444
 445 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
 446 {
 447         struct pglist_data *pgdat = zone->zone_pgdat;
 448         int nr_pages = PAGES_PER_SECTION;
 449         int nid = pgdat->node_id;
 450         int zone_type;
 451         unsigned long flags, pfn;
 452         int ret;
 453
 454         zone_type = zone - pgdat->node_zones;
 455         ret = ensure_zone_is_initialized(zone, phys_start_pfn, nr_pages);
 456         if (ret)
 457                 return ret;
 458
 459         pgdat_resize_lock(zone->zone_pgdat, &flags);
 460         grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
 461         grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
 462                         phys_start_pfn + nr_pages);
 463         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 464         memmap_init_zone(nr_pages, nid, zone_type,
 465                          phys_start_pfn, MEMMAP_HOTPLUG);
 466
 467         /* online_page_range is called later and expects pages reserved */
 468         for (pfn = phys_start_pfn; pfn < phys_start_pfn + nr_pages; pfn++) {
 469                 if (!pfn_valid(pfn))
 470                         continue;
 471
 472                 SetPageReserved(pfn_to_page(pfn));
 473         }
 474         return 0;
 475 }
 476
 477 static int __meminit __add_section(int nid, struct zone *zone,
 478                                         unsigned long phys_start_pfn)
 479 {
 480         int ret;
 481
 482         if (pfn_valid(phys_start_pfn))
 483                 return -EEXIST;
 484
 485         ret = sparse_add_one_section(zone, phys_start_pfn);
 486
 487         if (ret < 0)
 488                 return ret;
 489
 490         ret = __add_zone(zone, phys_start_pfn);
 491
 492         if (ret < 0)
 493                 return ret;
 494
 495         return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
 496 }
 497
 498 /*
 499  * Reasonably generic function for adding memory.  It is
 500  * expected that archs that support memory hotplug will
 501  * call this function after deciding the zone to which to
 502  * add the new pages.
 503  */
 504 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
 505                         unsigned long nr_pages)
 506 {
 507         unsigned long i;
 508         int err = 0;
 509         int start_sec, end_sec;
 510         struct vmem_altmap *altmap;
 511
 512         /* during initialize mem_map, align hot-added range to section */
 513         start_sec = pfn_to_section_nr(phys_start_pfn);
 514         end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
 515
 516         altmap = to_vmem_altmap((unsigned long) pfn_to_page(phys_start_pfn));
 517         if (altmap) {
 518                 /*
 519                  * Validate altmap is within bounds of the total request
 520                  */
 521                 if (altmap->base_pfn != phys_start_pfn
 522                                 || vmem_altmap_offset(altmap) > nr_pages) {
 523                         pr_warn_once("memory add fail, invalid altmap\n");
 524                         return -EINVAL;
 525                 }
 526                 altmap->alloc = 0;
 527         }
 528
 529         for (i = start_sec; i <= end_sec; i++) {
 530                 err = __add_section(nid, zone, section_nr_to_pfn(i));
 531
 532                 /*
 533                  * EEXIST is finally dealt with by ioresource collision
 534                  * check. see add_memory() => register_memory_resource()
 535                  * Warning will be printed if there is collision.
 536                  */
 537                 if (err && (err != -EEXIST))
 538                         break;
 539                 err = 0;
 540         }
 541         vmemmap_populate_print_last();
 542
 543         return err;
 544 }
 545 EXPORT_SYMBOL_GPL(__add_pages);
 546
 547 #ifdef CONFIG_MEMORY_HOTREMOVE
 548 /* find the smallest valid pfn in the range [start_pfn, end_pfn) */
 549 static int find_smallest_section_pfn(int nid, struct zone *zone,
 550                                      unsigned long start_pfn,
 551                                      unsigned long end_pfn)
 552 {
 553         struct mem_section *ms;
 554
 555         for (; start_pfn < end_pfn; start_pfn += PAGES_PER_SECTION) {
 556                 ms = __pfn_to_section(start_pfn);
 557
 558                 if (unlikely(!valid_section(ms)))
 559                         continue;
 560
 561                 if (unlikely(pfn_to_nid(start_pfn) != nid))
 562                         continue;
 563
 564                 if (zone && zone != page_zone(pfn_to_page(start_pfn)))
 565                         continue;
 566
 567                 return start_pfn;
 568         }
 569
 570         return 0;
 571 }
 572
 573 /* find the biggest valid pfn in the range [start_pfn, end_pfn). */
 574 static int find_biggest_section_pfn(int nid, struct zone *zone,
 575                                     unsigned long start_pfn,
 576                                     unsigned long end_pfn)
 577 {
 578         struct mem_section *ms;
 579         unsigned long pfn;
 580
 581         /* pfn is the end pfn of a memory section. */
 582         pfn = end_pfn - 1;
 583         for (; pfn >= start_pfn; pfn -= PAGES_PER_SECTION) {
 584                 ms = __pfn_to_section(pfn);
 585
 586                 if (unlikely(!valid_section(ms)))
 587                         continue;
 588
 589                 if (unlikely(pfn_to_nid(pfn) != nid))
 590                         continue;
 591
 592                 if (zone && zone != page_zone(pfn_to_page(pfn)))
 593                         continue;
 594
 595                 return pfn;
 596         }
 597
 598         return 0;
 599 }
 600
 601 static void shrink_zone_span(struct zone *zone, unsigned long start_pfn,
 602                              unsigned long end_pfn)
 603 {
 604         unsigned long zone_start_pfn = zone->zone_start_pfn;
 605         unsigned long z = zone_end_pfn(zone); /* zone_end_pfn namespace clash */
 606         unsigned long zone_end_pfn = z;
 607         unsigned long pfn;
 608         struct mem_section *ms;
 609         int nid = zone_to_nid(zone);
 610
 611         zone_span_writelock(zone);
 612         if (zone_start_pfn == start_pfn) {
 613                 /*
 614                  * If the section is smallest section in the zone, it need
 615                  * shrink zone->zone_start_pfn and zone->zone_spanned_pages.
 616                  * In this case, we find second smallest valid mem_section
 617                  * for shrinking zone.
 618                  */
 619                 pfn = find_smallest_section_pfn(nid, zone, end_pfn,
 620                                                 zone_end_pfn);
 621                 if (pfn) {
 622                         zone->zone_start_pfn = pfn;
 623                         zone->spanned_pages = zone_end_pfn - pfn;
 624                 }
 625         } else if (zone_end_pfn == end_pfn) {
 626                 /*
 627                  * If the section is biggest section in the zone, it need
 628                  * shrink zone->spanned_pages.
 629                  * In this case, we find second biggest valid mem_section for
 630                  * shrinking zone.
 631                  */
 632                 pfn = find_biggest_section_pfn(nid, zone, zone_start_pfn,
 633                                                start_pfn);
 634                 if (pfn)
 635                         zone->spanned_pages = pfn - zone_start_pfn + 1;
 636         }
 637
 638         /*
 639          * The section is not biggest or smallest mem_section in the zone, it
 640          * only creates a hole in the zone. So in this case, we need not
 641          * change the zone. But perhaps, the zone has only hole data. Thus
 642          * it check the zone has only hole or not.
 643          */
 644         pfn = zone_start_pfn;
 645         for (; pfn < zone_end_pfn; pfn += PAGES_PER_SECTION) {
 646                 ms = __pfn_to_section(pfn);
 647
 648                 if (unlikely(!valid_section(ms)))
 649                         continue;
 650
 651                 if (page_zone(pfn_to_page(pfn)) != zone)
 652                         continue;
 653
 654                  /* If the section is current section, it continues the loop */
 655                 if (start_pfn == pfn)
 656                         continue;
 657
 658                 /* If we find valid section, we have nothing to do */
 659                 zone_span_writeunlock(zone);
 660                 return;
 661         }
 662
 663         /* The zone has no valid section */
 664         zone->zone_start_pfn = 0;
 665         zone->spanned_pages = 0;
 666         zone_span_writeunlock(zone);
 667 }
 668
 669 static void shrink_pgdat_span(struct pglist_data *pgdat,
 670                               unsigned long start_pfn, unsigned long end_pfn)
 671 {
 672         unsigned long pgdat_start_pfn = pgdat->node_start_pfn;
 673         unsigned long p = pgdat_end_pfn(pgdat); /* pgdat_end_pfn namespace clash */
 674         unsigned long pgdat_end_pfn = p;
 675         unsigned long pfn;
 676         struct mem_section *ms;
 677         int nid = pgdat->node_id;
 678
 679         if (pgdat_start_pfn == start_pfn) {
 680                 /*
 681                  * If the section is smallest section in the pgdat, it need
 682                  * shrink pgdat->node_start_pfn and pgdat->node_spanned_pages.
 683                  * In this case, we find second smallest valid mem_section
 684                  * for shrinking zone.
 685                  */
 686                 pfn = find_smallest_section_pfn(nid, NULL, end_pfn,
 687                                                 pgdat_end_pfn);
 688                 if (pfn) {
 689                         pgdat->node_start_pfn = pfn;
 690                         pgdat->node_spanned_pages = pgdat_end_pfn - pfn;
 691                 }
 692         } else if (pgdat_end_pfn == end_pfn) {
 693                 /*
 694                  * If the section is biggest section in the pgdat, it need
 695                  * shrink pgdat->node_spanned_pages.
 696                  * In this case, we find second biggest valid mem_section for
 697                  * shrinking zone.
 698                  */
 699                 pfn = find_biggest_section_pfn(nid, NULL, pgdat_start_pfn,
 700                                                start_pfn);
 701                 if (pfn)
 702                         pgdat->node_spanned_pages = pfn - pgdat_start_pfn + 1;
 703         }
 704
 705         /*
 706          * If the section is not biggest or smallest mem_section in the pgdat,
 707          * it only creates a hole in the pgdat. So in this case, we need not
 708          * change the pgdat.
 709          * But perhaps, the pgdat has only hole data. Thus it check the pgdat
 710          * has only hole or not.
 711          */
 712         pfn = pgdat_start_pfn;
 713         for (; pfn < pgdat_end_pfn; pfn += PAGES_PER_SECTION) {
 714                 ms = __pfn_to_section(pfn);
 715
 716                 if (unlikely(!valid_section(ms)))
 717                         continue;
 718
 719                 if (pfn_to_nid(pfn) != nid)
 720                         continue;
 721
 722                  /* If the section is current section, it continues the loop */
 723                 if (start_pfn == pfn)
 724                         continue;
 725
 726                 /* If we find valid section, we have nothing to do */
 727                 return;
 728         }
 729
 730         /* The pgdat has no valid section */
 731         pgdat->node_start_pfn = 0;
 732         pgdat->node_spanned_pages = 0;
 733 }
 734
 735 static void __remove_zone(struct zone *zone, unsigned long start_pfn)
 736 {
 737         struct pglist_data *pgdat = zone->zone_pgdat;
 738         int nr_pages = PAGES_PER_SECTION;
 739         int zone_type;
 740         unsigned long flags;
 741
 742         zone_type = zone - pgdat->node_zones;
 743
 744         pgdat_resize_lock(zone->zone_pgdat, &flags);
 745         shrink_zone_span(zone, start_pfn, start_pfn + nr_pages);
 746         shrink_pgdat_span(pgdat, start_pfn, start_pfn + nr_pages);
 747         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 748 }
 749
 750 static int __remove_section(struct zone *zone, struct mem_section *ms,
 751                 unsigned long map_offset)
 752 {
 753         unsigned long start_pfn;
 754         int scn_nr;
 755         int ret = -EINVAL;
 756
 757         if (!valid_section(ms))
 758                 return ret;
 759
 760         ret = unregister_memory_section(ms);
 761         if (ret)
 762                 return ret;
 763
 764         scn_nr = __section_nr(ms);
 765         start_pfn = section_nr_to_pfn(scn_nr);
 766         __remove_zone(zone, start_pfn);
 767
 768         sparse_remove_one_section(zone, ms, map_offset);
 769         return 0;
 770 }
 771
 772 /**
 773  * __remove_pages() - remove sections of pages from a zone
 774  * @zone: zone from which pages need to be removed
 775  * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
 776  * @nr_pages: number of pages to remove (must be multiple of section size)
 777  *
 778  * Generic helper function to remove section mappings and sysfs entries
 779  * for the section of the memory we are removing. Caller needs to make
 780  * sure that pages are marked reserved and zones are adjust properly by
 781  * calling offline_pages().
 782  */
 783 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
 784                  unsigned long nr_pages)
 785 {
 786         unsigned long i;
 787         unsigned long map_offset = 0;
 788         int sections_to_remove, ret = 0;
 789
 790         /* In the ZONE_DEVICE case device driver owns the memory region */
 791         if (is_dev_zone(zone)) {
 792                 struct page *page = pfn_to_page(phys_start_pfn);
 793                 struct vmem_altmap *altmap;
 794
 795                 altmap = to_vmem_altmap((unsigned long) page);
 796                 if (altmap)
 797                         map_offset = vmem_altmap_offset(altmap);
 798         } else {
 799                 resource_size_t start, size;
 800
 801                 start = phys_start_pfn << PAGE_SHIFT;
 802                 size = nr_pages * PAGE_SIZE;
 803
 804                 ret = release_mem_region_adjustable(&iomem_resource, start,
 805                                         size);
 806                 if (ret) {
 807                         resource_size_t endres = start + size - 1;
 808
 809                         pr_warn("Unable to release resource <%pa-%pa> (%d)\n",
 810                                         &start, &endres, ret);
 811                 }
 812         }
 813
 814         /*
 815          * We can only remove entire sections
 816          */
 817         BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
 818         BUG_ON(nr_pages % PAGES_PER_SECTION);
 819
 820         sections_to_remove = nr_pages / PAGES_PER_SECTION;
 821         for (i = 0; i < sections_to_remove; i++) {
 822                 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
 823
 824                 ret = __remove_section(zone, __pfn_to_section(pfn), map_offset);
 825                 map_offset = 0;
 826                 if (ret)
 827                         break;
 828         }
 829         return ret;
 830 }
 831 EXPORT_SYMBOL_GPL(__remove_pages);
 832 #endif /* CONFIG_MEMORY_HOTREMOVE */
 833
 834 int set_online_page_callback(online_page_callback_t callback)
 835 {
 836         int rc = -EINVAL;
 837
 838         get_online_mems();
 839         mutex_lock(&online_page_callback_lock);
 840
 841         if (online_page_callback == generic_online_page) {
 842                 online_page_callback = callback;
 843                 rc = 0;
 844         }
 845
 846         mutex_unlock(&online_page_callback_lock);
 847         put_online_mems();
 848
 849         return rc;
 850 }
 851 EXPORT_SYMBOL_GPL(set_online_page_callback);
 852
 853 int restore_online_page_callback(online_page_callback_t callback)
 854 {
 855         int rc = -EINVAL;
 856
 857         get_online_mems();
 858         mutex_lock(&online_page_callback_lock);
 859
 860         if (online_page_callback == callback) {
 861                 online_page_callback = generic_online_page;
 862                 rc = 0;
 863         }
 864
 865         mutex_unlock(&online_page_callback_lock);
 866         put_online_mems();
 867
 868         return rc;
 869 }
 870 EXPORT_SYMBOL_GPL(restore_online_page_callback);
 871
 872 void __online_page_set_limits(struct page *page)
 873 {
 874 }
 875 EXPORT_SYMBOL_GPL(__online_page_set_limits);
 876
 877 void __online_page_increment_counters(struct page *page)
 878 {
 879         adjust_managed_page_count(page, 1);
 880 }
 881 EXPORT_SYMBOL_GPL(__online_page_increment_counters);
 882
 883 void __online_page_free(struct page *page)
 884 {
 885         __free_reserved_page(page);
 886 }
 887 EXPORT_SYMBOL_GPL(__online_page_free);
 888
 889 static void generic_online_page(struct page *page)
 890 {
 891         __online_page_set_limits(page);
 892         __online_page_increment_counters(page);
 893         __online_page_free(page);
 894 }
 895
 896 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
 897                         void *arg)
 898 {
 899         unsigned long i;
 900         unsigned long onlined_pages = *(unsigned long *)arg;
 901         struct page *page;
 902         if (PageReserved(pfn_to_page(start_pfn)))
 903                 for (i = 0; i < nr_pages; i++) {
 904                         page = pfn_to_page(start_pfn + i);
 905                         (*online_page_callback)(page);
 906                         onlined_pages++;
 907                 }
 908         *(unsigned long *)arg = onlined_pages;
 909         return 0;
 910 }
 911
 912 #ifdef CONFIG_MOVABLE_NODE
 913 /*
 914  * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't have
 915  * normal memory.
 916  */
 917 static bool can_online_high_movable(struct zone *zone)
 918 {
 919         return true;
 920 }
 921 #else /* CONFIG_MOVABLE_NODE */
 922 /* ensure every online node has NORMAL memory */
 923 static bool can_online_high_movable(struct zone *zone)
 924 {
 925         return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
 926 }
 927 #endif /* CONFIG_MOVABLE_NODE */
 928
 929 /* check which state of node_states will be changed when online memory */
 930 static void node_states_check_changes_online(unsigned long nr_pages,
 931         struct zone *zone, struct memory_notify *arg)
 932 {
 933         int nid = zone_to_nid(zone);
 934         enum zone_type zone_last = ZONE_NORMAL;
 935
 936         /*
 937          * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
 938          * contains nodes which have zones of 0...ZONE_NORMAL,
 939          * set zone_last to ZONE_NORMAL.
 940          *
 941          * If we don't have HIGHMEM nor movable node,
 942          * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
 943          * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
 944          */
 945         if (N_MEMORY == N_NORMAL_MEMORY)
 946                 zone_last = ZONE_MOVABLE;
 947
 948         /*
 949          * if the memory to be online is in a zone of 0...zone_last, and
 950          * the zones of 0...zone_last don't have memory before online, we will
 951          * need to set the node to node_states[N_NORMAL_MEMORY] after
 952          * the memory is online.
 953          */
 954         if (zone_idx(zone) <= zone_last && !node_state(nid, N_NORMAL_MEMORY))
 955                 arg->status_change_nid_normal = nid;
 956         else
 957                 arg->status_change_nid_normal = -1;
 958
 959 #ifdef CONFIG_HIGHMEM
 960         /*
 961          * If we have movable node, node_states[N_HIGH_MEMORY]
 962          * contains nodes which have zones of 0...ZONE_HIGHMEM,
 963          * set zone_last to ZONE_HIGHMEM.
 964          *
 965          * If we don't have movable node, node_states[N_NORMAL_MEMORY]
 966          * contains nodes which have zones of 0...ZONE_MOVABLE,
 967          * set zone_last to ZONE_MOVABLE.
 968          */
 969         zone_last = ZONE_HIGHMEM;
 970         if (N_MEMORY == N_HIGH_MEMORY)
 971                 zone_last = ZONE_MOVABLE;
 972
 973         if (zone_idx(zone) <= zone_last && !node_state(nid, N_HIGH_MEMORY))
 974                 arg->status_change_nid_high = nid;
 975         else
 976                 arg->status_change_nid_high = -1;
 977 #else
 978         arg->status_change_nid_high = arg->status_change_nid_normal;
 979 #endif
 980
 981         /*
 982          * if the node don't have memory befor online, we will need to
 983          * set the node to node_states[N_MEMORY] after the memory
 984          * is online.
 985          */
 986         if (!node_state(nid, N_MEMORY))
 987                 arg->status_change_nid = nid;
 988         else
 989                 arg->status_change_nid = -1;
 990 }
 991
 992 static void node_states_set_node(int node, struct memory_notify *arg)
 993 {
 994         if (arg->status_change_nid_normal >= 0)
 995                 node_set_state(node, N_NORMAL_MEMORY);
 996
 997         if (arg->status_change_nid_high >= 0)
 998                 node_set_state(node, N_HIGH_MEMORY);
 999
1000         node_set_state(node, N_MEMORY);
1001 }
1002
1003
1004 /* Must be protected by mem_hotplug_begin() */
1005 int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_type)
1006 {
1007         unsigned long flags;
1008         unsigned long onlined_pages = 0;
1009         struct zone *zone;
1010         int need_zonelists_rebuild = 0;
1011         int nid;
1012         int ret;
1013         struct memory_notify arg;
1014
1015         /*
1016          * This doesn't need a lock to do pfn_to_page().
1017          * The section can't be removed here because of the
1018          * memory_block->state_mutex.
1019          */
1020         zone = page_zone(pfn_to_page(pfn));
1021
1022         if ((zone_idx(zone) > ZONE_NORMAL ||
1023             online_type == MMOP_ONLINE_MOVABLE) &&
1024             !can_online_high_movable(zone))
1025                 return -EINVAL;
1026
1027         if (online_type == MMOP_ONLINE_KERNEL &&
1028             zone_idx(zone) == ZONE_MOVABLE) {
1029                 if (move_pfn_range_left(zone - 1, zone, pfn, pfn + nr_pages))
1030                         return -EINVAL;
1031         }
1032         if (online_type == MMOP_ONLINE_MOVABLE &&
1033             zone_idx(zone) == ZONE_MOVABLE - 1) {
1034                 if (move_pfn_range_right(zone, zone + 1, pfn, pfn + nr_pages))
1035                         return -EINVAL;
1036         }
1037
1038         /* Previous code may changed the zone of the pfn range */
1039         zone = page_zone(pfn_to_page(pfn));
1040
1041         arg.start_pfn = pfn;
1042         arg.nr_pages = nr_pages;
1043         node_states_check_changes_online(nr_pages, zone, &arg);
1044
1045         nid = pfn_to_nid(pfn);
1046
1047         ret = memory_notify(MEM_GOING_ONLINE, &arg);
1048         ret = notifier_to_errno(ret);
1049         if (ret) {
1050                 memory_notify(MEM_CANCEL_ONLINE, &arg);
1051                 return ret;
1052         }
1053         /*
1054          * If this zone is not populated, then it is not in zonelist.
1055          * This means the page allocator ignores this zone.
1056          * So, zonelist must be updated after online.
1057          */
1058         mutex_lock(&zonelists_mutex);
1059         if (!populated_zone(zone)) {
1060                 need_zonelists_rebuild = 1;
1061                 build_all_zonelists(NULL, zone);
1062         }
1063
1064         ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
1065                 online_pages_range);
1066         if (ret) {
1067                 if (need_zonelists_rebuild)
1068                         zone_pcp_reset(zone);
1069                 mutex_unlock(&zonelists_mutex);
1070                 printk(KERN_DEBUG "online_pages [mem %#010llx-%#010llx] failed\n",
1071                        (unsigned long long) pfn << PAGE_SHIFT,
1072                        (((unsigned long long) pfn + nr_pages)
1073                             << PAGE_SHIFT) - 1);
1074                 memory_notify(MEM_CANCEL_ONLINE, &arg);
1075                 return ret;
1076         }
1077
1078         zone->present_pages += onlined_pages;
1079
1080         pgdat_resize_lock(zone->zone_pgdat, &flags);
1081         zone->zone_pgdat->node_present_pages += onlined_pages;
1082         pgdat_resize_unlock(zone->zone_pgdat, &flags);
1083
1084         if (onlined_pages) {
1085                 node_states_set_node(zone_to_nid(zone), &arg);
1086                 if (need_zonelists_rebuild)
1087                         build_all_zonelists(NULL, NULL);
1088                 else
1089                         zone_pcp_update(zone);
1090         }
1091
1092         mutex_unlock(&zonelists_mutex);
1093
1094         init_per_zone_wmark_min();
1095
1096         if (onlined_pages)
1097                 kswapd_run(zone_to_nid(zone));
1098
1099         vm_total_pages = nr_free_pagecache_pages();
1100
1101         writeback_set_ratelimit();
1102
1103         if (onlined_pages)
1104                 memory_notify(MEM_ONLINE, &arg);
1105         return 0;
1106 }
1107 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
1108
1109 static void reset_node_present_pages(pg_data_t *pgdat)
1110 {
1111         struct zone *z;
1112
1113         for (z = pgdat->node_zones; z < pgdat->node_zones + MAX_NR_ZONES; z++)
1114                 z->present_pages = 0;
1115
1116         pgdat->node_present_pages = 0;
1117 }
1118
1119 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1120 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
1121 {
1122         struct pglist_data *pgdat;
1123         unsigned long zones_size[MAX_NR_ZONES] = {0};
1124         unsigned long zholes_size[MAX_NR_ZONES] = {0};
1125         unsigned long start_pfn = PFN_DOWN(start);
1126
1127         pgdat = NODE_DATA(nid);
1128         if (!pgdat) {
1129                 pgdat = arch_alloc_nodedata(nid);
1130                 if (!pgdat)
1131                         return NULL;
1132
1133                 arch_refresh_nodedata(nid, pgdat);
1134         } else {
1135                 /* Reset the nr_zones and classzone_idx to 0 before reuse */
1136                 pgdat->nr_zones = 0;
1137                 pgdat->classzone_idx = 0;
1138         }
1139
1140         /* we can use NODE_DATA(nid) from here */
1141
1142         /* init node's zones as empty zones, we don't have any present pages.*/
1143         free_area_init_node(nid, zones_size, start_pfn, zholes_size);
1144
1145         /*
1146          * The node we allocated has no zone fallback lists. For avoiding
1147          * to access not-initialized zonelist, build here.
1148          */
1149         mutex_lock(&zonelists_mutex);
1150         build_all_zonelists(pgdat, NULL);
1151         mutex_unlock(&zonelists_mutex);
1152
1153         /*
1154          * zone->managed_pages is set to an approximate value in
1155          * free_area_init_core(), which will cause
1156          * /sys/device/system/node/nodeX/meminfo has wrong data.
1157          * So reset it to 0 before any memory is onlined.
1158          */
1159         reset_node_managed_pages(pgdat);
1160
1161         /*
1162          * When memory is hot-added, all the memory is in offline state. So
1163          * clear all zones' present_pages because they will be updated in
1164          * online_pages() and offline_pages().
1165          */
1166         reset_node_present_pages(pgdat);
1167
1168         return pgdat;
1169 }
1170
1171 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
1172 {
1173         arch_refresh_nodedata(nid, NULL);
1174         arch_free_nodedata(pgdat);
1175         return;
1176 }
1177
1178
1179 /**
1180  * try_online_node - online a node if offlined
1181  *
1182  * called by cpu_up() to online a node without onlined memory.
1183  */
1184 int try_online_node(int nid)
1185 {
1186         pg_data_t       *pgdat;
1187         int     ret;
1188
1189         if (node_online(nid))
1190                 return 0;
1191
1192         mem_hotplug_begin();
1193         pgdat = hotadd_new_pgdat(nid, 0);
1194         if (!pgdat) {
1195                 pr_err("Cannot online node %d due to NULL pgdat\n", nid);
1196                 ret = -ENOMEM;
1197                 goto out;
1198         }
1199         node_set_online(nid);
1200         ret = register_one_node(nid);
1201         BUG_ON(ret);
1202
1203         if (pgdat->node_zonelists->_zonerefs->zone == NULL) {
1204                 mutex_lock(&zonelists_mutex);
1205                 build_all_zonelists(NULL, NULL);
1206                 mutex_unlock(&zonelists_mutex);
1207         }
1208
1209 out:
1210         mem_hotplug_done();
1211         return ret;
1212 }
1213
1214 static int check_hotplug_memory_range(u64 start, u64 size)
1215 {
1216         u64 start_pfn = PFN_DOWN(start);
1217         u64 nr_pages = size >> PAGE_SHIFT;
1218
1219         /* Memory range must be aligned with section */
1220         if ((start_pfn & ~PAGE_SECTION_MASK) ||
1221             (nr_pages % PAGES_PER_SECTION) || (!nr_pages)) {
1222                 pr_err("Section-unaligned hotplug range: start 0x%llx, size 0x%llx\n",
1223                                 (unsigned long long)start,
1224                                 (unsigned long long)size);
1225                 return -EINVAL;
1226         }
1227
1228         return 0;
1229 }
1230
1231 /*
1232  * If movable zone has already been setup, newly added memory should be check.
1233  * If its address is higher than movable zone, it should be added as movable.
1234  * Without this check, movable zone may overlap with other zone.
1235  */
1236 static int should_add_memory_movable(int nid, u64 start, u64 size)
1237 {
1238         unsigned long start_pfn = start >> PAGE_SHIFT;
1239         pg_data_t *pgdat = NODE_DATA(nid);
1240         struct zone *movable_zone = pgdat->node_zones + ZONE_MOVABLE;
1241
1242         if (zone_is_empty(movable_zone))
1243                 return 0;
1244
1245         if (movable_zone->zone_start_pfn <= start_pfn)
1246                 return 1;
1247
1248         return 0;
1249 }
1250
1251 int zone_for_memory(int nid, u64 start, u64 size, int zone_default,
1252                 bool for_device)
1253 {
1254 #ifdef CONFIG_ZONE_DEVICE
1255         if (for_device)
1256                 return ZONE_DEVICE;
1257 #endif
1258         if (should_add_memory_movable(nid, start, size))
1259                 return ZONE_MOVABLE;
1260
1261         return zone_default;
1262 }
1263
1264 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
1265 int __ref add_memory_resource(int nid, struct resource *res)
1266 {
1267         u64 start, size;
1268         pg_data_t *pgdat = NULL;
1269         bool new_pgdat;
1270         bool new_node;
1271         int ret;
1272
1273         start = res->start;
1274         size = resource_size(res);
1275
1276         ret = check_hotplug_memory_range(start, size);
1277         if (ret)
1278                 return ret;
1279
1280         {       /* Stupid hack to suppress address-never-null warning */
1281                 void *p = NODE_DATA(nid);
1282                 new_pgdat = !p;
1283         }
1284
1285         mem_hotplug_begin();
1286
1287         /*
1288          * Add new range to memblock so that when hotadd_new_pgdat() is called
1289          * to allocate new pgdat, get_pfn_range_for_nid() will be able to find
1290          * this new range and calculate total pages correctly.  The range will
1291          * be removed at hot-remove time.
1292          */
1293         memblock_add_node(start, size, nid);
1294
1295         new_node = !node_online(nid);
1296         if (new_node) {
1297                 pgdat = hotadd_new_pgdat(nid, start);
1298                 ret = -ENOMEM;
1299                 if (!pgdat)
1300                         goto error;
1301         }
1302
1303         /* call arch's memory hotadd */
1304         ret = arch_add_memory(nid, start, size, false);
1305
1306         if (ret < 0)
1307                 goto error;
1308
1309         /* we online node here. we can't roll back from here. */
1310         node_set_online(nid);
1311
1312         if (new_node) {
1313                 ret = register_one_node(nid);
1314                 /*
1315                  * If sysfs file of new node can't create, cpu on the node
1316                  * can't be hot-added. There is no rollback way now.
1317                  * So, check by BUG_ON() to catch it reluctantly..
1318                  */
1319                 BUG_ON(ret);
1320         }
1321
1322         /* create new memmap entry */
1323         firmware_map_add_hotplug(start, start + size, "System RAM");
1324
1325         goto out;
1326
1327 error:
1328         /* rollback pgdat allocation and others */
1329         if (new_pgdat)
1330                 rollback_node_hotadd(nid, pgdat);
1331         memblock_remove(start, size);
1332
1333 out:
1334         mem_hotplug_done();
1335         return ret;
1336 }
1337 EXPORT_SYMBOL_GPL(add_memory_resource);
1338
1339 int __ref add_memory(int nid, u64 start, u64 size)
1340 {
1341         struct resource *res;
1342         int ret;
1343
1344         res = register_memory_resource(start, size);
1345         if (IS_ERR(res))
1346                 return PTR_ERR(res);
1347
1348         ret = add_memory_resource(nid, res);
1349         if (ret < 0)
1350                 release_memory_resource(res);
1351         return ret;
1352 }
1353 EXPORT_SYMBOL_GPL(add_memory);
1354
1355 #ifdef CONFIG_MEMORY_HOTREMOVE
1356 /*
1357  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
1358  * set and the size of the free page is given by page_order(). Using this,
1359  * the function determines if the pageblock contains only free pages.
1360  * Due to buddy contraints, a free page at least the size of a pageblock will
1361  * be located at the start of the pageblock
1362  */
1363 static inline int pageblock_free(struct page *page)
1364 {
1365         return PageBuddy(page) && page_order(page) >= pageblock_order;
1366 }
1367
1368 /* Return the start of the next active pageblock after a given page */
1369 static struct page *next_active_pageblock(struct page *page)
1370 {
1371         /* Ensure the starting page is pageblock-aligned */
1372         BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
1373
1374         /* If the entire pageblock is free, move to the end of free page */
1375         if (pageblock_free(page)) {
1376                 int order;
1377                 /* be careful. we don't have locks, page_order can be changed.*/
1378                 order = page_order(page);
1379                 if ((order < MAX_ORDER) && (order >= pageblock_order))
1380                         return page + (1 << order);
1381         }
1382
1383         return page + pageblock_nr_pages;
1384 }
1385
1386 /* Checks if this range of memory is likely to be hot-removable. */
1387 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
1388 {
1389         struct page *page = pfn_to_page(start_pfn);
1390         struct page *end_page = page + nr_pages;
1391
1392         /* Check the starting page of each pageblock within the range */
1393         for (; page < end_page; page = next_active_pageblock(page)) {
1394                 if (!is_pageblock_removable_nolock(page))
1395                         return 0;
1396                 cond_resched();
1397         }
1398
1399         /* All pageblocks in the memory block are likely to be hot-removable */
1400         return 1;
1401 }
1402
1403 /*
1404  * Confirm all pages in a range [start, end) is belongs to the same zone.
1405  */
1406 int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
1407 {
1408         unsigned long pfn, sec_end_pfn;
1409         struct zone *zone = NULL;
1410         struct page *page;
1411         int i;
1412         for (pfn = start_pfn, sec_end_pfn = SECTION_ALIGN_UP(start_pfn);
1413              pfn < end_pfn;
1414              pfn = sec_end_pfn + 1, sec_end_pfn += PAGES_PER_SECTION) {
1415                 /* Make sure the memory section is present first */
1416                 if (!present_section_nr(pfn_to_section_nr(pfn)))
1417                         continue;
1418                 for (; pfn < sec_end_pfn && pfn < end_pfn;
1419                      pfn += MAX_ORDER_NR_PAGES) {
1420                         i = 0;
1421                         /* This is just a CONFIG_HOLES_IN_ZONE check.*/
1422                         while ((i < MAX_ORDER_NR_PAGES) &&
1423                                 !pfn_valid_within(pfn + i))
1424                                 i++;
1425                         if (i == MAX_ORDER_NR_PAGES)
1426                                 continue;
1427                         page = pfn_to_page(pfn + i);
1428                         if (zone && page_zone(page) != zone)
1429                                 return 0;
1430                         zone = page_zone(page);
1431                 }
1432         }
1433         return 1;
1434 }
1435
1436 /*
1437  * Scan pfn range [start,end) to find movable/migratable pages (LRU pages
1438  * and hugepages). We scan pfn because it's much easier than scanning over
1439  * linked list. This function returns the pfn of the first found movable
1440  * page if it's found, otherwise 0.
1441  */
1442 static unsigned long scan_movable_pages(unsigned long start, unsigned long end)
1443 {
1444         unsigned long pfn;
1445         struct page *page;
1446         for (pfn = start; pfn < end; pfn++) {
1447                 if (pfn_valid(pfn)) {
1448                         page = pfn_to_page(pfn);
1449                         if (PageLRU(page))
1450                                 return pfn;
1451                         if (PageHuge(page)) {
1452                                 if (page_huge_active(page))
1453                                         return pfn;
1454                                 else
1455                                         pfn = round_up(pfn + 1,
1456                                                 1 << compound_order(page)) - 1;
1457                         }
1458                 }
1459         }
1460         return 0;
1461 }
1462
1463 #define NR_OFFLINE_AT_ONCE_PAGES        (256)
1464 static int
1465 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
1466 {
1467         unsigned long pfn;
1468         struct page *page;
1469         int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
1470         int not_managed = 0;
1471         int ret = 0;
1472         LIST_HEAD(source);
1473
1474         for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
1475                 if (!pfn_valid(pfn))
1476                         continue;
1477                 page = pfn_to_page(pfn);
1478
1479                 if (PageHuge(page)) {
1480                         struct page *head = compound_head(page);
1481                         pfn = page_to_pfn(head) + (1<<compound_order(head)) - 1;
1482                         if (compound_order(head) > PFN_SECTION_SHIFT) {
1483                                 ret = -EBUSY;
1484                                 break;
1485                         }
1486                         if (isolate_huge_page(page, &source))
1487                                 move_pages -= 1 << compound_order(head);
1488                         continue;
1489                 }
1490
1491                 if (!get_page_unless_zero(page))
1492                         continue;
1493                 /*
1494                  * We can skip free pages. And we can only deal with pages on
1495                  * LRU.
1496                  */
1497                 ret = isolate_lru_page(page);
1498                 if (!ret) { /* Success */
1499                         put_page(page);
1500                         list_add_tail(&page->lru, &source);
1501                         move_pages--;
1502                         inc_zone_page_state(page, NR_ISOLATED_ANON +
1503                                             page_is_file_cache(page));
1504
1505                 } else {
1506 #ifdef CONFIG_DEBUG_VM
1507                         printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
1508                                pfn);
1509                         dump_page(page, "failed to remove from LRU");
1510 #endif
1511                         put_page(page);
1512                         /* Because we don't have big zone->lock. we should
1513                            check this again here. */
1514                         if (page_count(page)) {
1515                                 not_managed++;
1516                                 ret = -EBUSY;
1517                                 break;
1518                         }
1519                 }
1520         }
1521         if (!list_empty(&source)) {
1522                 if (not_managed) {
1523                         putback_movable_pages(&source);
1524                         goto out;
1525                 }
1526
1527                 /*
1528                  * alloc_migrate_target should be improooooved!!
1529                  * migrate_pages returns # of failed pages.
1530                  */
1531                 ret = migrate_pages(&source, alloc_migrate_target, NULL, 0,
1532                                         MIGRATE_SYNC, MR_MEMORY_HOTPLUG);
1533                 if (ret)
1534                         putback_movable_pages(&source);
1535         }
1536 out:
1537         return ret;
1538 }
1539
1540 /*
1541  * remove from free_area[] and mark all as Reserved.
1542  */
1543 static int
1544 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
1545                         void *data)
1546 {
1547         __offline_isolated_pages(start, start + nr_pages);
1548         return 0;
1549 }
1550
1551 static void
1552 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
1553 {
1554         walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
1555                                 offline_isolated_pages_cb);
1556 }
1557
1558 /*
1559  * Check all pages in range, recoreded as memory resource, are isolated.
1560  */
1561 static int
1562 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
1563                         void *data)
1564 {
1565         int ret;
1566         long offlined = *(long *)data;
1567         ret = test_pages_isolated(start_pfn, start_pfn + nr_pages, true);
1568         offlined = nr_pages;
1569         if (!ret)
1570                 *(long *)data += offlined;
1571         return ret;
1572 }
1573
1574 static long
1575 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
1576 {
1577         long offlined = 0;
1578         int ret;
1579
1580         ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
1581                         check_pages_isolated_cb);
1582         if (ret < 0)
1583                 offlined = (long)ret;
1584         return offlined;
1585 }
1586
1587 #ifdef CONFIG_MOVABLE_NODE
1588 /*
1589  * When CONFIG_MOVABLE_NODE, we permit offlining of a node which doesn't have
1590  * normal memory.
1591  */
1592 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1593 {
1594         return true;
1595 }
1596 #else /* CONFIG_MOVABLE_NODE */
1597 /* ensure the node has NORMAL memory if it is still online */
1598 static bool can_offline_normal(struct zone *zone, unsigned long nr_pages)
1599 {
1600         struct pglist_data *pgdat = zone->zone_pgdat;
1601         unsigned long present_pages = 0;
1602         enum zone_type zt;
1603
1604         for (zt = 0; zt <= ZONE_NORMAL; zt++)
1605                 present_pages += pgdat->node_zones[zt].present_pages;
1606
1607         if (present_pages > nr_pages)
1608                 return true;
1609
1610         present_pages = 0;
1611         for (; zt <= ZONE_MOVABLE; zt++)
1612                 present_pages += pgdat->node_zones[zt].present_pages;
1613
1614         /*
1615          * we can't offline the last normal memory until all
1616          * higher memory is offlined.
1617          */
1618         return present_pages == 0;
1619 }
1620 #endif /* CONFIG_MOVABLE_NODE */
1621
1622 static int __init cmdline_parse_movable_node(char *p)
1623 {
1624 #ifdef CONFIG_MOVABLE_NODE
1625         /*
1626          * Memory used by the kernel cannot be hot-removed because Linux
1627          * cannot migrate the kernel pages. When memory hotplug is
1628          * enabled, we should prevent memblock from allocating memory
1629          * for the kernel.
1630          *
1631          * ACPI SRAT records all hotpluggable memory ranges. But before
1632          * SRAT is parsed, we don't know about it.
1633          *
1634          * The kernel image is loaded into memory at very early time. We
1635          * cannot prevent this anyway. So on NUMA system, we set any
1636          * node the kernel resides in as un-hotpluggable.
1637          *
1638          * Since on modern servers, one node could have double-digit
1639          * gigabytes memory, we can assume the memory around the kernel
1640          * image is also un-hotpluggable. So before SRAT is parsed, just
1641          * allocate memory near the kernel image to try the best to keep
1642          * the kernel away from hotpluggable memory.
1643          */
1644         memblock_set_bottom_up(true);
1645         movable_node_enabled = true;
1646 #else
1647         pr_warn("movable_node option not supported\n");
1648 #endif
1649         return 0;
1650 }
1651 early_param("movable_node", cmdline_parse_movable_node);
1652
1653 /* check which state of node_states will be changed when offline memory */
1654 static void node_states_check_changes_offline(unsigned long nr_pages,
1655                 struct zone *zone, struct memory_notify *arg)
1656 {
1657         struct pglist_data *pgdat = zone->zone_pgdat;
1658         unsigned long present_pages = 0;
1659         enum zone_type zt, zone_last = ZONE_NORMAL;
1660
1661         /*
1662          * If we have HIGHMEM or movable node, node_states[N_NORMAL_MEMORY]
1663          * contains nodes which have zones of 0...ZONE_NORMAL,
1664          * set zone_last to ZONE_NORMAL.
1665          *
1666          * If we don't have HIGHMEM nor movable node,
1667          * node_states[N_NORMAL_MEMORY] contains nodes which have zones of
1668          * 0...ZONE_MOVABLE, set zone_last to ZONE_MOVABLE.
1669          */
1670         if (N_MEMORY == N_NORMAL_MEMORY)
1671                 zone_last = ZONE_MOVABLE;
1672
1673         /*
1674          * check whether node_states[N_NORMAL_MEMORY] will be changed.
1675          * If the memory to be offline is in a zone of 0...zone_last,
1676          * and it is the last present memory, 0...zone_last will
1677          * become empty after offline , thus we can determind we will
1678          * need to clear the node from node_states[N_NORMAL_MEMORY].
1679          */
1680         for (zt = 0; zt <= zone_last; zt++)
1681                 present_pages += pgdat->node_zones[zt].present_pages;
1682         if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1683                 arg->status_change_nid_normal = zone_to_nid(zone);
1684         else
1685                 arg->status_change_nid_normal = -1;
1686
1687 #ifdef CONFIG_HIGHMEM
1688         /*
1689          * If we have movable node, node_states[N_HIGH_MEMORY]
1690          * contains nodes which have zones of 0...ZONE_HIGHMEM,
1691          * set zone_last to ZONE_HIGHMEM.
1692          *
1693          * If we don't have movable node, node_states[N_NORMAL_MEMORY]
1694          * contains nodes which have zones of 0...ZONE_MOVABLE,
1695          * set zone_last to ZONE_MOVABLE.
1696          */
1697         zone_last = ZONE_HIGHMEM;
1698         if (N_MEMORY == N_HIGH_MEMORY)
1699                 zone_last = ZONE_MOVABLE;
1700
1701         for (; zt <= zone_last; zt++)
1702                 present_pages += pgdat->node_zones[zt].present_pages;
1703         if (zone_idx(zone) <= zone_last && nr_pages >= present_pages)
1704                 arg->status_change_nid_high = zone_to_nid(zone);
1705         else
1706                 arg->status_change_nid_high = -1;
1707 #else
1708         arg->status_change_nid_high = arg->status_change_nid_normal;
1709 #endif
1710
1711         /*
1712          * node_states[N_HIGH_MEMORY] contains nodes which have 0...ZONE_MOVABLE
1713          */
1714         zone_last = ZONE_MOVABLE;
1715
1716         /*
1717          * check whether node_states[N_HIGH_MEMORY] will be changed
1718          * If we try to offline the last present @nr_pages from the node,
1719          * we can determind we will need to clear the node from
1720          * node_states[N_HIGH_MEMORY].
1721          */
1722         for (; zt <= zone_last; zt++)
1723                 present_pages += pgdat->node_zones[zt].present_pages;
1724         if (nr_pages >= present_pages)
1725                 arg->status_change_nid = zone_to_nid(zone);
1726         else
1727                 arg->status_change_nid = -1;
1728 }
1729
1730 static void node_states_clear_node(int node, struct memory_notify *arg)
1731 {
1732         if (arg->status_change_nid_normal >= 0)
1733                 node_clear_state(node, N_NORMAL_MEMORY);
1734
1735         if ((N_MEMORY != N_NORMAL_MEMORY) &&
1736             (arg->status_change_nid_high >= 0))
1737                 node_clear_state(node, N_HIGH_MEMORY);
1738
1739         if ((N_MEMORY != N_HIGH_MEMORY) &&
1740             (arg->status_change_nid >= 0))
1741                 node_clear_state(node, N_MEMORY);
1742 }
1743
1744 static int __ref __offline_pages(unsigned long start_pfn,
1745                   unsigned long end_pfn, unsigned long timeout)
1746 {
1747         unsigned long pfn, nr_pages, expire;
1748         long offlined_pages;
1749         int ret, drain, retry_max, node;
1750         unsigned long flags;
1751         struct zone *zone;
1752         struct memory_notify arg;
1753
1754         /* at least, alignment against pageblock is necessary */
1755         if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
1756                 return -EINVAL;
1757         if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
1758                 return -EINVAL;
1759         /* This makes hotplug much easier...and readable.
1760            we assume this for now. .*/
1761         if (!test_pages_in_a_zone(start_pfn, end_pfn))
1762                 return -EINVAL;
1763
1764         zone = page_zone(pfn_to_page(start_pfn));
1765         node = zone_to_nid(zone);
1766         nr_pages = end_pfn - start_pfn;
1767
1768         if (zone_idx(zone) <= ZONE_NORMAL && !can_offline_normal(zone, nr_pages))
1769                 return -EINVAL;
1770
1771         /* set above range as isolated */
1772         ret = start_isolate_page_range(start_pfn, end_pfn,
1773                                        MIGRATE_MOVABLE, true);
1774         if (ret)
1775                 return ret;
1776
1777         arg.start_pfn = start_pfn;
1778         arg.nr_pages = nr_pages;
1779         node_states_check_changes_offline(nr_pages, zone, &arg);
1780
1781         ret = memory_notify(MEM_GOING_OFFLINE, &arg);
1782         ret = notifier_to_errno(ret);
1783         if (ret)
1784                 goto failed_removal;
1785
1786         pfn = start_pfn;
1787         expire = jiffies + timeout;
1788         drain = 0;
1789         retry_max = 5;
1790 repeat:
1791         /* start memory hot removal */
1792         ret = -EAGAIN;
1793         if (time_after(jiffies, expire))
1794                 goto failed_removal;
1795         ret = -EINTR;
1796         if (signal_pending(current))
1797                 goto failed_removal;
1798         ret = 0;
1799         if (drain) {
1800                 lru_add_drain_all();
1801                 cond_resched();
1802                 drain_all_pages(zone);
1803         }
1804
1805         pfn = scan_movable_pages(start_pfn, end_pfn);
1806         if (pfn) { /* We have movable pages */
1807                 ret = do_migrate_range(pfn, end_pfn);
1808                 if (!ret) {
1809                         drain = 1;
1810                         goto repeat;
1811                 } else {
1812                         if (ret < 0)
1813                                 if (--retry_max == 0)
1814                                         goto failed_removal;
1815                         yield();
1816                         drain = 1;
1817                         goto repeat;
1818                 }
1819         }
1820         /* drain all zone's lru pagevec, this is asynchronous... */
1821         lru_add_drain_all();
1822         yield();
1823         /* drain pcp pages, this is synchronous. */
1824         drain_all_pages(zone);
1825         /*
1826          * dissolve free hugepages in the memory block before doing offlining
1827          * actually in order to make hugetlbfs's object counting consistent.
1828          */
1829         dissolve_free_huge_pages(start_pfn, end_pfn);
1830         /* check again */
1831         offlined_pages = check_pages_isolated(start_pfn, end_pfn);
1832         if (offlined_pages < 0) {
1833                 ret = -EBUSY;
1834                 goto failed_removal;
1835         }
1836         printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
1837         /* Ok, all of our target is isolated.
1838            We cannot do rollback at this point. */
1839         offline_isolated_pages(start_pfn, end_pfn);
1840         /* reset pagetype flags and makes migrate type to be MOVABLE */
1841         undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1842         /* removal success */
1843         adjust_managed_page_count(pfn_to_page(start_pfn), -offlined_pages);
1844         zone->present_pages -= offlined_pages;
1845
1846         pgdat_resize_lock(zone->zone_pgdat, &flags);
1847         zone->zone_pgdat->node_present_pages -= offlined_pages;
1848         pgdat_resize_unlock(zone->zone_pgdat, &flags);
1849
1850         init_per_zone_wmark_min();
1851
1852         if (!populated_zone(zone)) {
1853                 zone_pcp_reset(zone);
1854                 mutex_lock(&zonelists_mutex);
1855                 build_all_zonelists(NULL, NULL);
1856                 mutex_unlock(&zonelists_mutex);
1857         } else
1858                 zone_pcp_update(zone);
1859
1860         node_states_clear_node(node, &arg);
1861         if (arg.status_change_nid >= 0)
1862                 kswapd_stop(node);
1863
1864         vm_total_pages = nr_free_pagecache_pages();
1865         writeback_set_ratelimit();
1866
1867         memory_notify(MEM_OFFLINE, &arg);
1868         return 0;
1869
1870 failed_removal:
1871         printk(KERN_INFO "memory offlining [mem %#010llx-%#010llx] failed\n",
1872                (unsigned long long) start_pfn << PAGE_SHIFT,
1873                ((unsigned long long) end_pfn << PAGE_SHIFT) - 1);
1874         memory_notify(MEM_CANCEL_OFFLINE, &arg);
1875         /* pushback to free area */
1876         undo_isolate_page_range(start_pfn, end_pfn, MIGRATE_MOVABLE);
1877         return ret;
1878 }
1879
1880 /* Must be protected by mem_hotplug_begin() */
1881 int offline_pages(unsigned long start_pfn, unsigned long nr_pages)
1882 {
1883         return __offline_pages(start_pfn, start_pfn + nr_pages, 120 * HZ);
1884 }
1885 #endif /* CONFIG_MEMORY_HOTREMOVE */
1886
1887 /**
1888  * walk_memory_range - walks through all mem sections in [start_pfn, end_pfn)
1889  * @start_pfn: start pfn of the memory range
1890  * @end_pfn: end pfn of the memory range
1891  * @arg: argument passed to func
1892  * @func: callback for each memory section walked
1893  *
1894  * This function walks through all present mem sections in range
1895  * [start_pfn, end_pfn) and call func on each mem section.
1896  *
1897  * Returns the return value of func.
1898  */
1899 int walk_memory_range(unsigned long start_pfn, unsigned long end_pfn,
1900                 void *arg, int (*func)(struct memory_block *, void *))
1901 {
1902         struct memory_block *mem = NULL;
1903         struct mem_section *section;
1904         unsigned long pfn, section_nr;
1905         int ret;
1906
1907         for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
1908                 section_nr = pfn_to_section_nr(pfn);
1909                 if (!present_section_nr(section_nr))
1910                         continue;
1911
1912                 section = __nr_to_section(section_nr);
1913                 /* same memblock? */
1914                 if (mem)
1915                         if ((section_nr >= mem->start_section_nr) &&
1916                             (section_nr <= mem->end_section_nr))
1917                                 continue;
1918
1919                 mem = find_memory_block_hinted(section, mem);
1920                 if (!mem)
1921                         continue;
1922
1923                 ret = func(mem, arg);
1924                 if (ret) {
1925                         kobject_put(&mem->dev.kobj);
1926                         return ret;
1927                 }
1928         }
1929
1930         if (mem)
1931                 kobject_put(&mem->dev.kobj);
1932
1933         return 0;
1934 }
1935
1936 #ifdef CONFIG_MEMORY_HOTREMOVE
1937 static int check_memblock_offlined_cb(struct memory_block *mem, void *arg)
1938 {
1939         int ret = !is_memblock_offlined(mem);
1940
1941         if (unlikely(ret)) {
1942                 phys_addr_t beginpa, endpa;
1943
1944                 beginpa = PFN_PHYS(section_nr_to_pfn(mem->start_section_nr));
1945                 endpa = PFN_PHYS(section_nr_to_pfn(mem->end_section_nr + 1))-1;
1946                 pr_warn("removing memory fails, because memory "
1947                         "[%pa-%pa] is onlined\n",
1948                         &beginpa, &endpa);
1949         }
1950
1951         return ret;
1952 }
1953
1954 static int check_cpu_on_node(pg_data_t *pgdat)
1955 {
1956         int cpu;
1957
1958         for_each_present_cpu(cpu) {
1959                 if (cpu_to_node(cpu) == pgdat->node_id)
1960                         /*
1961                          * the cpu on this node isn't removed, and we can't
1962                          * offline this node.
1963                          */
1964                         return -EBUSY;
1965         }
1966
1967         return 0;
1968 }
1969
1970 static void unmap_cpu_on_node(pg_data_t *pgdat)
1971 {
1972 #ifdef CONFIG_ACPI_NUMA
1973         int cpu;
1974
1975         for_each_possible_cpu(cpu)
1976                 if (cpu_to_node(cpu) == pgdat->node_id)
1977                         numa_clear_node(cpu);
1978 #endif
1979 }
1980
1981 static int check_and_unmap_cpu_on_node(pg_data_t *pgdat)
1982 {
1983         int ret;
1984
1985         ret = check_cpu_on_node(pgdat);
1986         if (ret)
1987                 return ret;
1988
1989         /*
1990          * the node will be offlined when we come here, so we can clear
1991          * the cpu_to_node() now.
1992          */
1993
1994         unmap_cpu_on_node(pgdat);
1995         return 0;
1996 }
1997
1998 /**
1999  * try_offline_node
2000  *
2001  * Offline a node if all memory sections and cpus of the node are removed.
2002  *
2003  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2004  * and online/offline operations before this call.
2005  */
2006 void try_offline_node(int nid)
2007 {
2008         pg_data_t *pgdat = NODE_DATA(nid);
2009         unsigned long start_pfn = pgdat->node_start_pfn;
2010         unsigned long end_pfn = start_pfn + pgdat->node_spanned_pages;
2011         unsigned long pfn;
2012         int i;
2013
2014         for (pfn = start_pfn; pfn < end_pfn; pfn += PAGES_PER_SECTION) {
2015                 unsigned long section_nr = pfn_to_section_nr(pfn);
2016
2017                 if (!present_section_nr(section_nr))
2018                         continue;
2019
2020                 if (pfn_to_nid(pfn) != nid)
2021                         continue;
2022
2023                 /*
2024                  * some memory sections of this node are not removed, and we
2025                  * can't offline node now.
2026                  */
2027                 return;
2028         }
2029
2030         if (check_and_unmap_cpu_on_node(pgdat))
2031                 return;
2032
2033         /*
2034          * all memory/cpu of this node are removed, we can offline this
2035          * node now.
2036          */
2037         node_set_offline(nid);
2038         unregister_one_node(nid);
2039
2040         /* free waittable in each zone */
2041         for (i = 0; i < MAX_NR_ZONES; i++) {
2042                 struct zone *zone = pgdat->node_zones + i;
2043
2044                 /*
2045                  * wait_table may be allocated from boot memory,
2046                  * here only free if it's allocated by vmalloc.
2047                  */
2048                 if (is_vmalloc_addr(zone->wait_table)) {
2049                         vfree(zone->wait_table);
2050                         zone->wait_table = NULL;
2051                 }
2052         }
2053 }
2054 EXPORT_SYMBOL(try_offline_node);
2055
2056 /**
2057  * remove_memory
2058  *
2059  * NOTE: The caller must call lock_device_hotplug() to serialize hotplug
2060  * and online/offline operations before this call, as required by
2061  * try_offline_node().
2062  */
2063 void __ref remove_memory(int nid, u64 start, u64 size)
2064 {
2065         int ret;
2066
2067         BUG_ON(check_hotplug_memory_range(start, size));
2068
2069         mem_hotplug_begin();
2070
2071         /*
2072          * All memory blocks must be offlined before removing memory.  Check
2073          * whether all memory blocks in question are offline and trigger a BUG()
2074          * if this is not the case.
2075          */
2076         ret = walk_memory_range(PFN_DOWN(start), PFN_UP(start + size - 1), NULL,
2077                                 check_memblock_offlined_cb);
2078         if (ret)
2079                 BUG();
2080
2081         /* remove memmap entry */
2082         firmware_map_remove(start, start + size, "System RAM");
2083         memblock_free(start, size);
2084         memblock_remove(start, size);
2085
2086         arch_remove_memory(start, size);
2087
2088         try_offline_node(nid);
2089
2090         mem_hotplug_done();
2091 }
2092 EXPORT_SYMBOL_GPL(remove_memory);
2093 #endif /* CONFIG_MEMORY_HOTREMOVE */