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/module.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
  33 #include <asm/tlbflush.h>
  34
  35 #include "internal.h"
  36
  37 /* add this memory to iomem resource */
  38 static struct resource *register_memory_resource(u64 start, u64 size)
  39 {
  40         struct resource *res;
  41         res = kzalloc(sizeof(struct resource), GFP_KERNEL);
  42         BUG_ON(!res);
  43
  44         res->name = "System RAM";
  45         res->start = start;
  46         res->end = start + size - 1;
  47         res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
  48         if (request_resource(&iomem_resource, res) < 0) {
  49                 printk("System RAM resource %llx - %llx cannot be added\n",
  50                 (unsigned long long)res->start, (unsigned long long)res->end);
  51                 kfree(res);
  52                 res = NULL;
  53         }
  54         return res;
  55 }
  56
  57 static void release_memory_resource(struct resource *res)
  58 {
  59         if (!res)
  60                 return;
  61         release_resource(res);
  62         kfree(res);
  63         return;
  64 }
  65
  66 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
  67 #ifndef CONFIG_SPARSEMEM_VMEMMAP
  68 static void get_page_bootmem(unsigned long info,  struct page *page, int type)
  69 {
  70         atomic_set(&page->_mapcount, type);
  71         SetPagePrivate(page);
  72         set_page_private(page, info);
  73         atomic_inc(&page->_count);
  74 }
  75
  76 /* reference to __meminit __free_pages_bootmem is valid
  77  * so use __ref to tell modpost not to generate a warning */
  78 void __ref put_page_bootmem(struct page *page)
  79 {
  80         int type;
  81
  82         type = atomic_read(&page->_mapcount);
  83         BUG_ON(type >= -1);
  84
  85         if (atomic_dec_return(&page->_count) == 1) {
  86                 ClearPagePrivate(page);
  87                 set_page_private(page, 0);
  88                 reset_page_mapcount(page);
  89                 __free_pages_bootmem(page, 0);
  90         }
  91
  92 }
  93
  94 static void register_page_bootmem_info_section(unsigned long start_pfn)
  95 {
  96         unsigned long *usemap, mapsize, section_nr, i;
  97         struct mem_section *ms;
  98         struct page *page, *memmap;
  99
 100         if (!pfn_valid(start_pfn))
 101                 return;
 102
 103         section_nr = pfn_to_section_nr(start_pfn);
 104         ms = __nr_to_section(section_nr);
 105
 106         /* Get section's memmap address */
 107         memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
 108
 109         /*
 110          * Get page for the memmap's phys address
 111          * XXX: need more consideration for sparse_vmemmap...
 112          */
 113         page = virt_to_page(memmap);
 114         mapsize = sizeof(struct page) * PAGES_PER_SECTION;
 115         mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
 116
 117         /* remember memmap's page */
 118         for (i = 0; i < mapsize; i++, page++)
 119                 get_page_bootmem(section_nr, page, SECTION_INFO);
 120
 121         usemap = __nr_to_section(section_nr)->pageblock_flags;
 122         page = virt_to_page(usemap);
 123
 124         mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
 125
 126         for (i = 0; i < mapsize; i++, page++)
 127                 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
 128
 129 }
 130
 131 void register_page_bootmem_info_node(struct pglist_data *pgdat)
 132 {
 133         unsigned long i, pfn, end_pfn, nr_pages;
 134         int node = pgdat->node_id;
 135         struct page *page;
 136         struct zone *zone;
 137
 138         nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
 139         page = virt_to_page(pgdat);
 140
 141         for (i = 0; i < nr_pages; i++, page++)
 142                 get_page_bootmem(node, page, NODE_INFO);
 143
 144         zone = &pgdat->node_zones[0];
 145         for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
 146                 if (zone->wait_table) {
 147                         nr_pages = zone->wait_table_hash_nr_entries
 148                                 * sizeof(wait_queue_head_t);
 149                         nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
 150                         page = virt_to_page(zone->wait_table);
 151
 152                         for (i = 0; i < nr_pages; i++, page++)
 153                                 get_page_bootmem(node, page, NODE_INFO);
 154                 }
 155         }
 156
 157         pfn = pgdat->node_start_pfn;
 158         end_pfn = pfn + pgdat->node_spanned_pages;
 159
 160         /* register_section info */
 161         for (; pfn < end_pfn; pfn += PAGES_PER_SECTION)
 162                 register_page_bootmem_info_section(pfn);
 163
 164 }
 165 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
 166
 167 static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
 168                            unsigned long end_pfn)
 169 {
 170         unsigned long old_zone_end_pfn;
 171
 172         zone_span_writelock(zone);
 173
 174         old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
 175         if (start_pfn < zone->zone_start_pfn)
 176                 zone->zone_start_pfn = start_pfn;
 177
 178         zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
 179                                 zone->zone_start_pfn;
 180
 181         zone_span_writeunlock(zone);
 182 }
 183
 184 static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
 185                             unsigned long end_pfn)
 186 {
 187         unsigned long old_pgdat_end_pfn =
 188                 pgdat->node_start_pfn + pgdat->node_spanned_pages;
 189
 190         if (start_pfn < pgdat->node_start_pfn)
 191                 pgdat->node_start_pfn = start_pfn;
 192
 193         pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
 194                                         pgdat->node_start_pfn;
 195 }
 196
 197 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
 198 {
 199         struct pglist_data *pgdat = zone->zone_pgdat;
 200         int nr_pages = PAGES_PER_SECTION;
 201         int nid = pgdat->node_id;
 202         int zone_type;
 203         unsigned long flags;
 204
 205         zone_type = zone - pgdat->node_zones;
 206         if (!zone->wait_table) {
 207                 int ret;
 208
 209                 ret = init_currently_empty_zone(zone, phys_start_pfn,
 210                                                 nr_pages, MEMMAP_HOTPLUG);
 211                 if (ret)
 212                         return ret;
 213         }
 214         pgdat_resize_lock(zone->zone_pgdat, &flags);
 215         grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
 216         grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
 217                         phys_start_pfn + nr_pages);
 218         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 219         memmap_init_zone(nr_pages, nid, zone_type,
 220                          phys_start_pfn, MEMMAP_HOTPLUG);
 221         return 0;
 222 }
 223
 224 static int __meminit __add_section(int nid, struct zone *zone,
 225                                         unsigned long phys_start_pfn)
 226 {
 227         int nr_pages = PAGES_PER_SECTION;
 228         int ret;
 229
 230         if (pfn_valid(phys_start_pfn))
 231                 return -EEXIST;
 232
 233         ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
 234
 235         if (ret < 0)
 236                 return ret;
 237
 238         ret = __add_zone(zone, phys_start_pfn);
 239
 240         if (ret < 0)
 241                 return ret;
 242
 243         return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
 244 }
 245
 246 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 247 static int __remove_section(struct zone *zone, struct mem_section *ms)
 248 {
 249         /*
 250          * XXX: Freeing memmap with vmemmap is not implement yet.
 251          *      This should be removed later.
 252          */
 253         return -EBUSY;
 254 }
 255 #else
 256 static int __remove_section(struct zone *zone, struct mem_section *ms)
 257 {
 258         unsigned long flags;
 259         struct pglist_data *pgdat = zone->zone_pgdat;
 260         int ret = -EINVAL;
 261
 262         if (!valid_section(ms))
 263                 return ret;
 264
 265         ret = unregister_memory_section(ms);
 266         if (ret)
 267                 return ret;
 268
 269         pgdat_resize_lock(pgdat, &flags);
 270         sparse_remove_one_section(zone, ms);
 271         pgdat_resize_unlock(pgdat, &flags);
 272         return 0;
 273 }
 274 #endif
 275
 276 /*
 277  * Reasonably generic function for adding memory.  It is
 278  * expected that archs that support memory hotplug will
 279  * call this function after deciding the zone to which to
 280  * add the new pages.
 281  */
 282 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
 283                         unsigned long nr_pages)
 284 {
 285         unsigned long i;
 286         int err = 0;
 287         int start_sec, end_sec;
 288         /* during initialize mem_map, align hot-added range to section */
 289         start_sec = pfn_to_section_nr(phys_start_pfn);
 290         end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
 291
 292         for (i = start_sec; i <= end_sec; i++) {
 293                 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
 294
 295                 /*
 296                  * EEXIST is finally dealt with by ioresource collision
 297                  * check. see add_memory() => register_memory_resource()
 298                  * Warning will be printed if there is collision.
 299                  */
 300                 if (err && (err != -EEXIST))
 301                         break;
 302                 err = 0;
 303         }
 304
 305         return err;
 306 }
 307 EXPORT_SYMBOL_GPL(__add_pages);
 308
 309 /**
 310  * __remove_pages() - remove sections of pages from a zone
 311  * @zone: zone from which pages need to be removed
 312  * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
 313  * @nr_pages: number of pages to remove (must be multiple of section size)
 314  *
 315  * Generic helper function to remove section mappings and sysfs entries
 316  * for the section of the memory we are removing. Caller needs to make
 317  * sure that pages are marked reserved and zones are adjust properly by
 318  * calling offline_pages().
 319  */
 320 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
 321                  unsigned long nr_pages)
 322 {
 323         unsigned long i, ret = 0;
 324         int sections_to_remove;
 325
 326         /*
 327          * We can only remove entire sections
 328          */
 329         BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
 330         BUG_ON(nr_pages % PAGES_PER_SECTION);
 331
 332         sections_to_remove = nr_pages / PAGES_PER_SECTION;
 333         for (i = 0; i < sections_to_remove; i++) {
 334                 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
 335                 release_mem_region(pfn << PAGE_SHIFT,
 336                                    PAGES_PER_SECTION << PAGE_SHIFT);
 337                 ret = __remove_section(zone, __pfn_to_section(pfn));
 338                 if (ret)
 339                         break;
 340         }
 341         return ret;
 342 }
 343 EXPORT_SYMBOL_GPL(__remove_pages);
 344
 345 void online_page(struct page *page)
 346 {
 347         unsigned long pfn = page_to_pfn(page);
 348
 349         totalram_pages++;
 350         if (pfn >= num_physpages)
 351                 num_physpages = pfn + 1;
 352
 353 #ifdef CONFIG_HIGHMEM
 354         if (PageHighMem(page))
 355                 totalhigh_pages++;
 356 #endif
 357
 358 #ifdef CONFIG_FLATMEM
 359         max_mapnr = max(page_to_pfn(page), max_mapnr);
 360 #endif
 361
 362         ClearPageReserved(page);
 363         init_page_count(page);
 364         __free_page(page);
 365 }
 366
 367 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
 368                         void *arg)
 369 {
 370         unsigned long i;
 371         unsigned long onlined_pages = *(unsigned long *)arg;
 372         struct page *page;
 373         if (PageReserved(pfn_to_page(start_pfn)))
 374                 for (i = 0; i < nr_pages; i++) {
 375                         page = pfn_to_page(start_pfn + i);
 376                         online_page(page);
 377                         onlined_pages++;
 378                 }
 379         *(unsigned long *)arg = onlined_pages;
 380         return 0;
 381 }
 382
 383
 384 int online_pages(unsigned long pfn, unsigned long nr_pages)
 385 {
 386         unsigned long onlined_pages = 0;
 387         struct zone *zone;
 388         int need_zonelists_rebuild = 0;
 389         int nid;
 390         int ret;
 391         struct memory_notify arg;
 392
 393         arg.start_pfn = pfn;
 394         arg.nr_pages = nr_pages;
 395         arg.status_change_nid = -1;
 396
 397         nid = page_to_nid(pfn_to_page(pfn));
 398         if (node_present_pages(nid) == 0)
 399                 arg.status_change_nid = nid;
 400
 401         ret = memory_notify(MEM_GOING_ONLINE, &arg);
 402         ret = notifier_to_errno(ret);
 403         if (ret) {
 404                 memory_notify(MEM_CANCEL_ONLINE, &arg);
 405                 return ret;
 406         }
 407         /*
 408          * This doesn't need a lock to do pfn_to_page().
 409          * The section can't be removed here because of the
 410          * memory_block->state_mutex.
 411          */
 412         zone = page_zone(pfn_to_page(pfn));
 413         /*
 414          * If this zone is not populated, then it is not in zonelist.
 415          * This means the page allocator ignores this zone.
 416          * So, zonelist must be updated after online.
 417          */
 418         if (!populated_zone(zone))
 419                 need_zonelists_rebuild = 1;
 420
 421         ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
 422                 online_pages_range);
 423         if (ret) {
 424                 printk(KERN_DEBUG "online_pages %lx at %lx failed\n",
 425                         nr_pages, pfn);
 426                 memory_notify(MEM_CANCEL_ONLINE, &arg);
 427                 return ret;
 428         }
 429
 430         zone->present_pages += onlined_pages;
 431         zone->zone_pgdat->node_present_pages += onlined_pages;
 432
 433         zone_pcp_update(zone);
 434         setup_per_zone_wmarks();
 435         calculate_zone_inactive_ratio(zone);
 436         if (onlined_pages) {
 437                 kswapd_run(zone_to_nid(zone));
 438                 node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
 439         }
 440
 441         if (need_zonelists_rebuild)
 442                 build_all_zonelists();
 443         else
 444                 vm_total_pages = nr_free_pagecache_pages();
 445
 446         writeback_set_ratelimit();
 447
 448         if (onlined_pages)
 449                 memory_notify(MEM_ONLINE, &arg);
 450
 451         return 0;
 452 }
 453 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
 454
 455 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
 456 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
 457 {
 458         struct pglist_data *pgdat;
 459         unsigned long zones_size[MAX_NR_ZONES] = {0};
 460         unsigned long zholes_size[MAX_NR_ZONES] = {0};
 461         unsigned long start_pfn = start >> PAGE_SHIFT;
 462
 463         pgdat = arch_alloc_nodedata(nid);
 464         if (!pgdat)
 465                 return NULL;
 466
 467         arch_refresh_nodedata(nid, pgdat);
 468
 469         /* we can use NODE_DATA(nid) from here */
 470
 471         /* init node's zones as empty zones, we don't have any present pages.*/
 472         free_area_init_node(nid, zones_size, start_pfn, zholes_size);
 473
 474         return pgdat;
 475 }
 476
 477 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
 478 {
 479         arch_refresh_nodedata(nid, NULL);
 480         arch_free_nodedata(pgdat);
 481         return;
 482 }
 483
 484
 485 /*
 486  * called by cpu_up() to online a node without onlined memory.
 487  */
 488 int mem_online_node(int nid)
 489 {
 490         pg_data_t       *pgdat;
 491         int     ret;
 492
 493         lock_system_sleep();
 494         pgdat = hotadd_new_pgdat(nid, 0);
 495         if (pgdat) {
 496                 ret = -ENOMEM;
 497                 goto out;
 498         }
 499         node_set_online(nid);
 500         ret = register_one_node(nid);
 501         BUG_ON(ret);
 502
 503 out:
 504         unlock_system_sleep();
 505         return ret;
 506 }
 507
 508 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
 509 int __ref add_memory(int nid, u64 start, u64 size)
 510 {
 511         pg_data_t *pgdat = NULL;
 512         int new_pgdat = 0;
 513         struct resource *res;
 514         int ret;
 515
 516         lock_system_sleep();
 517
 518         res = register_memory_resource(start, size);
 519         ret = -EEXIST;
 520         if (!res)
 521                 goto out;
 522
 523         if (!node_online(nid)) {
 524                 pgdat = hotadd_new_pgdat(nid, start);
 525                 ret = -ENOMEM;
 526                 if (!pgdat)
 527                         goto out;
 528                 new_pgdat = 1;
 529         }
 530
 531         /* call arch's memory hotadd */
 532         ret = arch_add_memory(nid, start, size);
 533
 534         if (ret < 0)
 535                 goto error;
 536
 537         /* we online node here. we can't roll back from here. */
 538         node_set_online(nid);
 539
 540         if (new_pgdat) {
 541                 ret = register_one_node(nid);
 542                 /*
 543                  * If sysfs file of new node can't create, cpu on the node
 544                  * can't be hot-added. There is no rollback way now.
 545                  * So, check by BUG_ON() to catch it reluctantly..
 546                  */
 547                 BUG_ON(ret);
 548         }
 549
 550         /* create new memmap entry */
 551         firmware_map_add_hotplug(start, start + size, "System RAM");
 552
 553         goto out;
 554
 555 error:
 556         /* rollback pgdat allocation and others */
 557         if (new_pgdat)
 558                 rollback_node_hotadd(nid, pgdat);
 559         if (res)
 560                 release_memory_resource(res);
 561
 562 out:
 563         unlock_system_sleep();
 564         return ret;
 565 }
 566 EXPORT_SYMBOL_GPL(add_memory);
 567
 568 #ifdef CONFIG_MEMORY_HOTREMOVE
 569 /*
 570  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
 571  * set and the size of the free page is given by page_order(). Using this,
 572  * the function determines if the pageblock contains only free pages.
 573  * Due to buddy contraints, a free page at least the size of a pageblock will
 574  * be located at the start of the pageblock
 575  */
 576 static inline int pageblock_free(struct page *page)
 577 {
 578         return PageBuddy(page) && page_order(page) >= pageblock_order;
 579 }
 580
 581 /* Return the start of the next active pageblock after a given page */
 582 static struct page *next_active_pageblock(struct page *page)
 583 {
 584         int pageblocks_stride;
 585
 586         /* Ensure the starting page is pageblock-aligned */
 587         BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
 588
 589         /* Move forward by at least 1 * pageblock_nr_pages */
 590         pageblocks_stride = 1;
 591
 592         /* If the entire pageblock is free, move to the end of free page */
 593         if (pageblock_free(page))
 594                 pageblocks_stride += page_order(page) - pageblock_order;
 595
 596         return page + (pageblocks_stride * pageblock_nr_pages);
 597 }
 598
 599 /* Checks if this range of memory is likely to be hot-removable. */
 600 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
 601 {
 602         int type;
 603         struct page *page = pfn_to_page(start_pfn);
 604         struct page *end_page = page + nr_pages;
 605
 606         /* Check the starting page of each pageblock within the range */
 607         for (; page < end_page; page = next_active_pageblock(page)) {
 608                 type = get_pageblock_migratetype(page);
 609
 610                 /*
 611                  * A pageblock containing MOVABLE or free pages is considered
 612                  * removable
 613                  */
 614                 if (type != MIGRATE_MOVABLE && !pageblock_free(page))
 615                         return 0;
 616
 617                 /*
 618                  * A pageblock starting with a PageReserved page is not
 619                  * considered removable.
 620                  */
 621                 if (PageReserved(page))
 622                         return 0;
 623         }
 624
 625         /* All pageblocks in the memory block are likely to be hot-removable */
 626         return 1;
 627 }
 628
 629 /*
 630  * Confirm all pages in a range [start, end) is belongs to the same zone.
 631  */
 632 static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
 633 {
 634         unsigned long pfn;
 635         struct zone *zone = NULL;
 636         struct page *page;
 637         int i;
 638         for (pfn = start_pfn;
 639              pfn < end_pfn;
 640              pfn += MAX_ORDER_NR_PAGES) {
 641                 i = 0;
 642                 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
 643                 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
 644                         i++;
 645                 if (i == MAX_ORDER_NR_PAGES)
 646                         continue;
 647                 page = pfn_to_page(pfn + i);
 648                 if (zone && page_zone(page) != zone)
 649                         return 0;
 650                 zone = page_zone(page);
 651         }
 652         return 1;
 653 }
 654
 655 /*
 656  * Scanning pfn is much easier than scanning lru list.
 657  * Scan pfn from start to end and Find LRU page.
 658  */
 659 int scan_lru_pages(unsigned long start, unsigned long end)
 660 {
 661         unsigned long pfn;
 662         struct page *page;
 663         for (pfn = start; pfn < end; pfn++) {
 664                 if (pfn_valid(pfn)) {
 665                         page = pfn_to_page(pfn);
 666                         if (PageLRU(page))
 667                                 return pfn;
 668                 }
 669         }
 670         return 0;
 671 }
 672
 673 static struct page *
 674 hotremove_migrate_alloc(struct page *page, unsigned long private, int **x)
 675 {
 676         /* This should be improooooved!! */
 677         return alloc_page(GFP_HIGHUSER_MOVABLE);
 678 }
 679
 680 #define NR_OFFLINE_AT_ONCE_PAGES        (256)
 681 static int
 682 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
 683 {
 684         unsigned long pfn;
 685         struct page *page;
 686         int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
 687         int not_managed = 0;
 688         int ret = 0;
 689         LIST_HEAD(source);
 690
 691         for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
 692                 if (!pfn_valid(pfn))
 693                         continue;
 694                 page = pfn_to_page(pfn);
 695                 if (!page_count(page))
 696                         continue;
 697                 /*
 698                  * We can skip free pages. And we can only deal with pages on
 699                  * LRU.
 700                  */
 701                 ret = isolate_lru_page(page);
 702                 if (!ret) { /* Success */
 703                         list_add_tail(&page->lru, &source);
 704                         move_pages--;
 705                         inc_zone_page_state(page, NR_ISOLATED_ANON +
 706                                             page_is_file_cache(page));
 707
 708                 } else {
 709                         /* Becasue we don't have big zone->lock. we should
 710                            check this again here. */
 711                         if (page_count(page))
 712                                 not_managed++;
 713 #ifdef CONFIG_DEBUG_VM
 714                         printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
 715                                pfn);
 716                         dump_page(page);
 717 #endif
 718                 }
 719         }
 720         ret = -EBUSY;
 721         if (not_managed) {
 722                 if (!list_empty(&source))
 723                         putback_lru_pages(&source);
 724                 goto out;
 725         }
 726         ret = 0;
 727         if (list_empty(&source))
 728                 goto out;
 729         /* this function returns # of failed pages */
 730         ret = migrate_pages(&source, hotremove_migrate_alloc, 0, 1);
 731
 732 out:
 733         return ret;
 734 }
 735
 736 /*
 737  * remove from free_area[] and mark all as Reserved.
 738  */
 739 static int
 740 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
 741                         void *data)
 742 {
 743         __offline_isolated_pages(start, start + nr_pages);
 744         return 0;
 745 }
 746
 747 static void
 748 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
 749 {
 750         walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
 751                                 offline_isolated_pages_cb);
 752 }
 753
 754 /*
 755  * Check all pages in range, recoreded as memory resource, are isolated.
 756  */
 757 static int
 758 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
 759                         void *data)
 760 {
 761         int ret;
 762         long offlined = *(long *)data;
 763         ret = test_pages_isolated(start_pfn, start_pfn + nr_pages);
 764         offlined = nr_pages;
 765         if (!ret)
 766                 *(long *)data += offlined;
 767         return ret;
 768 }
 769
 770 static long
 771 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
 772 {
 773         long offlined = 0;
 774         int ret;
 775
 776         ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
 777                         check_pages_isolated_cb);
 778         if (ret < 0)
 779                 offlined = (long)ret;
 780         return offlined;
 781 }
 782
 783 static int offline_pages(unsigned long start_pfn,
 784                   unsigned long end_pfn, unsigned long timeout)
 785 {
 786         unsigned long pfn, nr_pages, expire;
 787         long offlined_pages;
 788         int ret, drain, retry_max, node;
 789         struct zone *zone;
 790         struct memory_notify arg;
 791
 792         BUG_ON(start_pfn >= end_pfn);
 793         /* at least, alignment against pageblock is necessary */
 794         if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
 795                 return -EINVAL;
 796         if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
 797                 return -EINVAL;
 798         /* This makes hotplug much easier...and readable.
 799            we assume this for now. .*/
 800         if (!test_pages_in_a_zone(start_pfn, end_pfn))
 801                 return -EINVAL;
 802
 803         lock_system_sleep();
 804
 805         zone = page_zone(pfn_to_page(start_pfn));
 806         node = zone_to_nid(zone);
 807         nr_pages = end_pfn - start_pfn;
 808
 809         /* set above range as isolated */
 810         ret = start_isolate_page_range(start_pfn, end_pfn);
 811         if (ret)
 812                 goto out;
 813
 814         arg.start_pfn = start_pfn;
 815         arg.nr_pages = nr_pages;
 816         arg.status_change_nid = -1;
 817         if (nr_pages >= node_present_pages(node))
 818                 arg.status_change_nid = node;
 819
 820         ret = memory_notify(MEM_GOING_OFFLINE, &arg);
 821         ret = notifier_to_errno(ret);
 822         if (ret)
 823                 goto failed_removal;
 824
 825         pfn = start_pfn;
 826         expire = jiffies + timeout;
 827         drain = 0;
 828         retry_max = 5;
 829 repeat:
 830         /* start memory hot removal */
 831         ret = -EAGAIN;
 832         if (time_after(jiffies, expire))
 833                 goto failed_removal;
 834         ret = -EINTR;
 835         if (signal_pending(current))
 836                 goto failed_removal;
 837         ret = 0;
 838         if (drain) {
 839                 lru_add_drain_all();
 840                 flush_scheduled_work();
 841                 cond_resched();
 842                 drain_all_pages();
 843         }
 844
 845         pfn = scan_lru_pages(start_pfn, end_pfn);
 846         if (pfn) { /* We have page on LRU */
 847                 ret = do_migrate_range(pfn, end_pfn);
 848                 if (!ret) {
 849                         drain = 1;
 850                         goto repeat;
 851                 } else {
 852                         if (ret < 0)
 853                                 if (--retry_max == 0)
 854                                         goto failed_removal;
 855                         yield();
 856                         drain = 1;
 857                         goto repeat;
 858                 }
 859         }
 860         /* drain all zone's lru pagevec, this is asyncronous... */
 861         lru_add_drain_all();
 862         flush_scheduled_work();
 863         yield();
 864         /* drain pcp pages , this is synchrouns. */
 865         drain_all_pages();
 866         /* check again */
 867         offlined_pages = check_pages_isolated(start_pfn, end_pfn);
 868         if (offlined_pages < 0) {
 869                 ret = -EBUSY;
 870                 goto failed_removal;
 871         }
 872         printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
 873         /* Ok, all of our target is islaoted.
 874            We cannot do rollback at this point. */
 875         offline_isolated_pages(start_pfn, end_pfn);
 876         /* reset pagetype flags and makes migrate type to be MOVABLE */
 877         undo_isolate_page_range(start_pfn, end_pfn);
 878         /* removal success */
 879         zone->present_pages -= offlined_pages;
 880         zone->zone_pgdat->node_present_pages -= offlined_pages;
 881         totalram_pages -= offlined_pages;
 882
 883         setup_per_zone_wmarks();
 884         calculate_zone_inactive_ratio(zone);
 885         if (!node_present_pages(node)) {
 886                 node_clear_state(node, N_HIGH_MEMORY);
 887                 kswapd_stop(node);
 888         }
 889
 890         vm_total_pages = nr_free_pagecache_pages();
 891         writeback_set_ratelimit();
 892
 893         memory_notify(MEM_OFFLINE, &arg);
 894         unlock_system_sleep();
 895         return 0;
 896
 897 failed_removal:
 898         printk(KERN_INFO "memory offlining %lx to %lx failed\n",
 899                 start_pfn, end_pfn);
 900         memory_notify(MEM_CANCEL_OFFLINE, &arg);
 901         /* pushback to free area */
 902         undo_isolate_page_range(start_pfn, end_pfn);
 903
 904 out:
 905         unlock_system_sleep();
 906         return ret;
 907 }
 908
 909 int remove_memory(u64 start, u64 size)
 910 {
 911         unsigned long start_pfn, end_pfn;
 912
 913         start_pfn = PFN_DOWN(start);
 914         end_pfn = start_pfn + PFN_DOWN(size);
 915         return offline_pages(start_pfn, end_pfn, 120 * HZ);
 916 }
 917 #else
 918 int remove_memory(u64 start, u64 size)
 919 {
 920         return -EINVAL;
 921 }
 922 #endif /* CONFIG_MEMORY_HOTREMOVE */
 923 EXPORT_SYMBOL_GPL(remove_memory);