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