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
  31 #include <asm/tlbflush.h>
  32
  33 #include "internal.h"
  34
  35 /* add this memory to iomem resource */
  36 static struct resource *register_memory_resource(u64 start, u64 size)
  37 {
  38         struct resource *res;
  39         res = kzalloc(sizeof(struct resource), GFP_KERNEL);
  40         BUG_ON(!res);
  41
  42         res->name = "System RAM";
  43         res->start = start;
  44         res->end = start + size - 1;
  45         res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
  46         if (request_resource(&iomem_resource, res) < 0) {
  47                 printk("System RAM resource %llx - %llx cannot be added\n",
  48                 (unsigned long long)res->start, (unsigned long long)res->end);
  49                 kfree(res);
  50                 res = NULL;
  51         }
  52         return res;
  53 }
  54
  55 static void release_memory_resource(struct resource *res)
  56 {
  57         if (!res)
  58                 return;
  59         release_resource(res);
  60         kfree(res);
  61         return;
  62 }
  63
  64 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
  65 #ifndef CONFIG_SPARSEMEM_VMEMMAP
  66 static void get_page_bootmem(unsigned long info,  struct page *page, int type)
  67 {
  68         atomic_set(&page->_mapcount, type);
  69         SetPagePrivate(page);
  70         set_page_private(page, info);
  71         atomic_inc(&page->_count);
  72 }
  73
  74 void put_page_bootmem(struct page *page)
  75 {
  76         int type;
  77
  78         type = atomic_read(&page->_mapcount);
  79         BUG_ON(type >= -1);
  80
  81         if (atomic_dec_return(&page->_count) == 1) {
  82                 ClearPagePrivate(page);
  83                 set_page_private(page, 0);
  84                 reset_page_mapcount(page);
  85                 __free_pages_bootmem(page, 0);
  86         }
  87
  88 }
  89
  90 static void register_page_bootmem_info_section(unsigned long start_pfn)
  91 {
  92         unsigned long *usemap, mapsize, section_nr, i;
  93         struct mem_section *ms;
  94         struct page *page, *memmap;
  95
  96         if (!pfn_valid(start_pfn))
  97                 return;
  98
  99         section_nr = pfn_to_section_nr(start_pfn);
 100         ms = __nr_to_section(section_nr);
 101
 102         /* Get section's memmap address */
 103         memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
 104
 105         /*
 106          * Get page for the memmap's phys address
 107          * XXX: need more consideration for sparse_vmemmap...
 108          */
 109         page = virt_to_page(memmap);
 110         mapsize = sizeof(struct page) * PAGES_PER_SECTION;
 111         mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
 112
 113         /* remember memmap's page */
 114         for (i = 0; i < mapsize; i++, page++)
 115                 get_page_bootmem(section_nr, page, SECTION_INFO);
 116
 117         usemap = __nr_to_section(section_nr)->pageblock_flags;
 118         page = virt_to_page(usemap);
 119
 120         mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
 121
 122         for (i = 0; i < mapsize; i++, page++)
 123                 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
 124
 125 }
 126
 127 void register_page_bootmem_info_node(struct pglist_data *pgdat)
 128 {
 129         unsigned long i, pfn, end_pfn, nr_pages;
 130         int node = pgdat->node_id;
 131         struct page *page;
 132         struct zone *zone;
 133
 134         nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
 135         page = virt_to_page(pgdat);
 136
 137         for (i = 0; i < nr_pages; i++, page++)
 138                 get_page_bootmem(node, page, NODE_INFO);
 139
 140         zone = &pgdat->node_zones[0];
 141         for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
 142                 if (zone->wait_table) {
 143                         nr_pages = zone->wait_table_hash_nr_entries
 144                                 * sizeof(wait_queue_head_t);
 145                         nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
 146                         page = virt_to_page(zone->wait_table);
 147
 148                         for (i = 0; i < nr_pages; i++, page++)
 149                                 get_page_bootmem(node, page, NODE_INFO);
 150                 }
 151         }
 152
 153         pfn = pgdat->node_start_pfn;
 154         end_pfn = pfn + pgdat->node_spanned_pages;
 155
 156         /* register_section info */
 157         for (; pfn < end_pfn; pfn += PAGES_PER_SECTION)
 158                 register_page_bootmem_info_section(pfn);
 159
 160 }
 161 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
 162
 163 static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
 164                            unsigned long end_pfn)
 165 {
 166         unsigned long old_zone_end_pfn;
 167
 168         zone_span_writelock(zone);
 169
 170         old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
 171         if (start_pfn < zone->zone_start_pfn)
 172                 zone->zone_start_pfn = start_pfn;
 173
 174         zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
 175                                 zone->zone_start_pfn;
 176
 177         zone_span_writeunlock(zone);
 178 }
 179
 180 static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
 181                             unsigned long end_pfn)
 182 {
 183         unsigned long old_pgdat_end_pfn =
 184                 pgdat->node_start_pfn + pgdat->node_spanned_pages;
 185
 186         if (start_pfn < pgdat->node_start_pfn)
 187                 pgdat->node_start_pfn = start_pfn;
 188
 189         pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
 190                                         pgdat->node_start_pfn;
 191 }
 192
 193 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
 194 {
 195         struct pglist_data *pgdat = zone->zone_pgdat;
 196         int nr_pages = PAGES_PER_SECTION;
 197         int nid = pgdat->node_id;
 198         int zone_type;
 199         unsigned long flags;
 200
 201         zone_type = zone - pgdat->node_zones;
 202         if (!zone->wait_table) {
 203                 int ret;
 204
 205                 ret = init_currently_empty_zone(zone, phys_start_pfn,
 206                                                 nr_pages, MEMMAP_HOTPLUG);
 207                 if (ret)
 208                         return ret;
 209         }
 210         pgdat_resize_lock(zone->zone_pgdat, &flags);
 211         grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
 212         grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
 213                         phys_start_pfn + nr_pages);
 214         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 215         memmap_init_zone(nr_pages, nid, zone_type,
 216                          phys_start_pfn, MEMMAP_HOTPLUG);
 217         return 0;
 218 }
 219
 220 static int __meminit __add_section(int nid, struct zone *zone,
 221                                         unsigned long phys_start_pfn)
 222 {
 223         int nr_pages = PAGES_PER_SECTION;
 224         int ret;
 225
 226         if (pfn_valid(phys_start_pfn))
 227                 return -EEXIST;
 228
 229         ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
 230
 231         if (ret < 0)
 232                 return ret;
 233
 234         ret = __add_zone(zone, phys_start_pfn);
 235
 236         if (ret < 0)
 237                 return ret;
 238
 239         return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
 240 }
 241
 242 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 243 static int __remove_section(struct zone *zone, struct mem_section *ms)
 244 {
 245         /*
 246          * XXX: Freeing memmap with vmemmap is not implement yet.
 247          *      This should be removed later.
 248          */
 249         return -EBUSY;
 250 }
 251 #else
 252 static int __remove_section(struct zone *zone, struct mem_section *ms)
 253 {
 254         unsigned long flags;
 255         struct pglist_data *pgdat = zone->zone_pgdat;
 256         int ret = -EINVAL;
 257
 258         if (!valid_section(ms))
 259                 return ret;
 260
 261         ret = unregister_memory_section(ms);
 262         if (ret)
 263                 return ret;
 264
 265         pgdat_resize_lock(pgdat, &flags);
 266         sparse_remove_one_section(zone, ms);
 267         pgdat_resize_unlock(pgdat, &flags);
 268         return 0;
 269 }
 270 #endif
 271
 272 /*
 273  * Reasonably generic function for adding memory.  It is
 274  * expected that archs that support memory hotplug will
 275  * call this function after deciding the zone to which to
 276  * add the new pages.
 277  */
 278 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
 279                         unsigned long nr_pages)
 280 {
 281         unsigned long i;
 282         int err = 0;
 283         int start_sec, end_sec;
 284         /* during initialize mem_map, align hot-added range to section */
 285         start_sec = pfn_to_section_nr(phys_start_pfn);
 286         end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
 287
 288         for (i = start_sec; i <= end_sec; i++) {
 289                 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
 290
 291                 /*
 292                  * EEXIST is finally dealt with by ioresource collision
 293                  * check. see add_memory() => register_memory_resource()
 294                  * Warning will be printed if there is collision.
 295                  */
 296                 if (err && (err != -EEXIST))
 297                         break;
 298                 err = 0;
 299         }
 300
 301         return err;
 302 }
 303 EXPORT_SYMBOL_GPL(__add_pages);
 304
 305 /**
 306  * __remove_pages() - remove sections of pages from a zone
 307  * @zone: zone from which pages need to be removed
 308  * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
 309  * @nr_pages: number of pages to remove (must be multiple of section size)
 310  *
 311  * Generic helper function to remove section mappings and sysfs entries
 312  * for the section of the memory we are removing. Caller needs to make
 313  * sure that pages are marked reserved and zones are adjust properly by
 314  * calling offline_pages().
 315  */
 316 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
 317                  unsigned long nr_pages)
 318 {
 319         unsigned long i, ret = 0;
 320         int sections_to_remove;
 321
 322         /*
 323          * We can only remove entire sections
 324          */
 325         BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
 326         BUG_ON(nr_pages % PAGES_PER_SECTION);
 327
 328         sections_to_remove = nr_pages / PAGES_PER_SECTION;
 329         for (i = 0; i < sections_to_remove; i++) {
 330                 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
 331                 release_mem_region(pfn << PAGE_SHIFT,
 332                                    PAGES_PER_SECTION << PAGE_SHIFT);
 333                 ret = __remove_section(zone, __pfn_to_section(pfn));
 334                 if (ret)
 335                         break;
 336         }
 337         return ret;
 338 }
 339 EXPORT_SYMBOL_GPL(__remove_pages);
 340
 341 void online_page(struct page *page)
 342 {
 343         unsigned long pfn = page_to_pfn(page);
 344
 345         totalram_pages++;
 346         if (pfn >= num_physpages)
 347                 num_physpages = pfn + 1;
 348
 349 #ifdef CONFIG_HIGHMEM
 350         if (PageHighMem(page))
 351                 totalhigh_pages++;
 352 #endif
 353
 354 #ifdef CONFIG_FLATMEM
 355         max_mapnr = max(page_to_pfn(page), max_mapnr);
 356 #endif
 357
 358         ClearPageReserved(page);
 359         init_page_count(page);
 360         __free_page(page);
 361 }
 362
 363 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
 364                         void *arg)
 365 {
 366         unsigned long i;
 367         unsigned long onlined_pages = *(unsigned long *)arg;
 368         struct page *page;
 369         if (PageReserved(pfn_to_page(start_pfn)))
 370                 for (i = 0; i < nr_pages; i++) {
 371                         page = pfn_to_page(start_pfn + i);
 372                         online_page(page);
 373                         onlined_pages++;
 374                 }
 375         *(unsigned long *)arg = onlined_pages;
 376         return 0;
 377 }
 378
 379
 380 int online_pages(unsigned long pfn, unsigned long nr_pages)
 381 {
 382         unsigned long onlined_pages = 0;
 383         struct zone *zone;
 384         int need_zonelists_rebuild = 0;
 385         int nid;
 386         int ret;
 387         struct memory_notify arg;
 388
 389         arg.start_pfn = pfn;
 390         arg.nr_pages = nr_pages;
 391         arg.status_change_nid = -1;
 392
 393         nid = page_to_nid(pfn_to_page(pfn));
 394         if (node_present_pages(nid) == 0)
 395                 arg.status_change_nid = nid;
 396
 397         ret = memory_notify(MEM_GOING_ONLINE, &arg);
 398         ret = notifier_to_errno(ret);
 399         if (ret) {
 400                 memory_notify(MEM_CANCEL_ONLINE, &arg);
 401                 return ret;
 402         }
 403         /*
 404          * This doesn't need a lock to do pfn_to_page().
 405          * The section can't be removed here because of the
 406          * memory_block->state_mutex.
 407          */
 408         zone = page_zone(pfn_to_page(pfn));
 409         /*
 410          * If this zone is not populated, then it is not in zonelist.
 411          * This means the page allocator ignores this zone.
 412          * So, zonelist must be updated after online.
 413          */
 414         if (!populated_zone(zone))
 415                 need_zonelists_rebuild = 1;
 416
 417         ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
 418                 online_pages_range);
 419         if (ret) {
 420                 printk(KERN_DEBUG "online_pages %lx at %lx failed\n",
 421                         nr_pages, pfn);
 422                 memory_notify(MEM_CANCEL_ONLINE, &arg);
 423                 return ret;
 424         }
 425
 426         zone->present_pages += onlined_pages;
 427         zone->zone_pgdat->node_present_pages += onlined_pages;
 428
 429         zone_pcp_update(zone);
 430         setup_per_zone_wmarks();
 431         calculate_zone_inactive_ratio(zone);
 432         if (onlined_pages) {
 433                 kswapd_run(zone_to_nid(zone));
 434                 node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
 435         }
 436
 437         if (need_zonelists_rebuild)
 438                 build_all_zonelists();
 439         else
 440                 vm_total_pages = nr_free_pagecache_pages();
 441
 442         writeback_set_ratelimit();
 443
 444         if (onlined_pages)
 445                 memory_notify(MEM_ONLINE, &arg);
 446
 447         return 0;
 448 }
 449 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
 450
 451 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
 452 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
 453 {
 454         struct pglist_data *pgdat;
 455         unsigned long zones_size[MAX_NR_ZONES] = {0};
 456         unsigned long zholes_size[MAX_NR_ZONES] = {0};
 457         unsigned long start_pfn = start >> PAGE_SHIFT;
 458
 459         pgdat = arch_alloc_nodedata(nid);
 460         if (!pgdat)
 461                 return NULL;
 462
 463         arch_refresh_nodedata(nid, pgdat);
 464
 465         /* we can use NODE_DATA(nid) from here */
 466
 467         /* init node's zones as empty zones, we don't have any present pages.*/
 468         free_area_init_node(nid, zones_size, start_pfn, zholes_size);
 469
 470         return pgdat;
 471 }
 472
 473 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
 474 {
 475         arch_refresh_nodedata(nid, NULL);
 476         arch_free_nodedata(pgdat);
 477         return;
 478 }
 479
 480
 481 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
 482 int __ref add_memory(int nid, u64 start, u64 size)
 483 {
 484         pg_data_t *pgdat = NULL;
 485         int new_pgdat = 0;
 486         struct resource *res;
 487         int ret;
 488
 489         lock_system_sleep();
 490
 491         res = register_memory_resource(start, size);
 492         ret = -EEXIST;
 493         if (!res)
 494                 goto out;
 495
 496         if (!node_online(nid)) {
 497                 pgdat = hotadd_new_pgdat(nid, start);
 498                 ret = -ENOMEM;
 499                 if (!pgdat)
 500                         goto out;
 501                 new_pgdat = 1;
 502         }
 503
 504         /* call arch's memory hotadd */
 505         ret = arch_add_memory(nid, start, size);
 506
 507         if (ret < 0)
 508                 goto error;
 509
 510         /* we online node here. we can't roll back from here. */
 511         node_set_online(nid);
 512
 513         if (new_pgdat) {
 514                 ret = register_one_node(nid);
 515                 /*
 516                  * If sysfs file of new node can't create, cpu on the node
 517                  * can't be hot-added. There is no rollback way now.
 518                  * So, check by BUG_ON() to catch it reluctantly..
 519                  */
 520                 BUG_ON(ret);
 521         }
 522
 523         goto out;
 524
 525 error:
 526         /* rollback pgdat allocation and others */
 527         if (new_pgdat)
 528                 rollback_node_hotadd(nid, pgdat);
 529         if (res)
 530                 release_memory_resource(res);
 531
 532 out:
 533         unlock_system_sleep();
 534         return ret;
 535 }
 536 EXPORT_SYMBOL_GPL(add_memory);
 537
 538 #ifdef CONFIG_MEMORY_HOTREMOVE
 539 /*
 540  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
 541  * set and the size of the free page is given by page_order(). Using this,
 542  * the function determines if the pageblock contains only free pages.
 543  * Due to buddy contraints, a free page at least the size of a pageblock will
 544  * be located at the start of the pageblock
 545  */
 546 static inline int pageblock_free(struct page *page)
 547 {
 548         return PageBuddy(page) && page_order(page) >= pageblock_order;
 549 }
 550
 551 /* Return the start of the next active pageblock after a given page */
 552 static struct page *next_active_pageblock(struct page *page)
 553 {
 554         /* Ensure the starting page is pageblock-aligned */
 555         BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
 556
 557         /* If the entire pageblock is free, move to the end of free page */
 558         if (pageblock_free(page)) {
 559                 int order;
 560                 /* be careful. we don't have locks, page_order can be changed.*/
 561                 order = page_order(page);
 562                 if ((order < MAX_ORDER) && (order >= pageblock_order))
 563                         return page + (1 << order);
 564         }
 565
 566         return page + pageblock_nr_pages;
 567 }
 568
 569 /* Checks if this range of memory is likely to be hot-removable. */
 570 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
 571 {
 572         int type;
 573         struct page *page = pfn_to_page(start_pfn);
 574         struct page *end_page = page + nr_pages;
 575
 576         /* Check the starting page of each pageblock within the range */
 577         for (; page < end_page; page = next_active_pageblock(page)) {
 578                 type = get_pageblock_migratetype(page);
 579
 580                 /*
 581                  * A pageblock containing MOVABLE or free pages is considered
 582                  * removable
 583                  */
 584                 if (type != MIGRATE_MOVABLE && !pageblock_free(page))
 585                         return 0;
 586
 587                 /*
 588                  * A pageblock starting with a PageReserved page is not
 589                  * considered removable.
 590                  */
 591                 if (PageReserved(page))
 592                         return 0;
 593         }
 594
 595         /* All pageblocks in the memory block are likely to be hot-removable */
 596         return 1;
 597 }
 598
 599 /*
 600  * Confirm all pages in a range [start, end) is belongs to the same zone.
 601  */
 602 static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
 603 {
 604         unsigned long pfn;
 605         struct zone *zone = NULL;
 606         struct page *page;
 607         int i;
 608         for (pfn = start_pfn;
 609              pfn < end_pfn;
 610              pfn += MAX_ORDER_NR_PAGES) {
 611                 i = 0;
 612                 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
 613                 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
 614                         i++;
 615                 if (i == MAX_ORDER_NR_PAGES)
 616                         continue;
 617                 page = pfn_to_page(pfn + i);
 618                 if (zone && page_zone(page) != zone)
 619                         return 0;
 620                 zone = page_zone(page);
 621         }
 622         return 1;
 623 }
 624
 625 /*
 626  * Scanning pfn is much easier than scanning lru list.
 627  * Scan pfn from start to end and Find LRU page.
 628  */
 629 unsigned long scan_lru_pages(unsigned long start, unsigned long end)
 630 {
 631         unsigned long pfn;
 632         struct page *page;
 633         for (pfn = start; pfn < end; pfn++) {
 634                 if (pfn_valid(pfn)) {
 635                         page = pfn_to_page(pfn);
 636                         if (PageLRU(page))
 637                                 return pfn;
 638                 }
 639         }
 640         return 0;
 641 }
 642
 643 static struct page *
 644 hotremove_migrate_alloc(struct page *page, unsigned long private, int **x)
 645 {
 646         /* This should be improooooved!! */
 647         return alloc_page(GFP_HIGHUSER_MOVABLE);
 648 }
 649
 650 #define NR_OFFLINE_AT_ONCE_PAGES        (256)
 651 static int
 652 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
 653 {
 654         unsigned long pfn;
 655         struct page *page;
 656         int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
 657         int not_managed = 0;
 658         int ret = 0;
 659         LIST_HEAD(source);
 660
 661         for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
 662                 if (!pfn_valid(pfn))
 663                         continue;
 664                 page = pfn_to_page(pfn);
 665                 if (!page_count(page))
 666                         continue;
 667                 /*
 668                  * We can skip free pages. And we can only deal with pages on
 669                  * LRU.
 670                  */
 671                 ret = isolate_lru_page(page);
 672                 if (!ret) { /* Success */
 673                         list_add_tail(&page->lru, &source);
 674                         move_pages--;
 675                 } else {
 676                         /* Becasue we don't have big zone->lock. we should
 677                            check this again here. */
 678                         if (page_count(page))
 679                                 not_managed++;
 680 #ifdef CONFIG_DEBUG_VM
 681                         printk(KERN_INFO "removing from LRU failed"
 682                                          " %lx/%d/%lx\n",
 683                                 pfn, page_count(page), page->flags);
 684 #endif
 685                 }
 686         }
 687         ret = -EBUSY;
 688         if (not_managed) {
 689                 if (!list_empty(&source))
 690                         putback_lru_pages(&source);
 691                 goto out;
 692         }
 693         ret = 0;
 694         if (list_empty(&source))
 695                 goto out;
 696         /* this function returns # of failed pages */
 697         ret = migrate_pages(&source, hotremove_migrate_alloc, 0);
 698
 699 out:
 700         return ret;
 701 }
 702
 703 /*
 704  * remove from free_area[] and mark all as Reserved.
 705  */
 706 static int
 707 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
 708                         void *data)
 709 {
 710         __offline_isolated_pages(start, start + nr_pages);
 711         return 0;
 712 }
 713
 714 static void
 715 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
 716 {
 717         walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
 718                                 offline_isolated_pages_cb);
 719 }
 720
 721 /*
 722  * Check all pages in range, recoreded as memory resource, are isolated.
 723  */
 724 static int
 725 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
 726                         void *data)
 727 {
 728         int ret;
 729         long offlined = *(long *)data;
 730         ret = test_pages_isolated(start_pfn, start_pfn + nr_pages);
 731         offlined = nr_pages;
 732         if (!ret)
 733                 *(long *)data += offlined;
 734         return ret;
 735 }
 736
 737 static long
 738 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
 739 {
 740         long offlined = 0;
 741         int ret;
 742
 743         ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
 744                         check_pages_isolated_cb);
 745         if (ret < 0)
 746                 offlined = (long)ret;
 747         return offlined;
 748 }
 749
 750 int offline_pages(unsigned long start_pfn,
 751                   unsigned long end_pfn, unsigned long timeout)
 752 {
 753         unsigned long pfn, nr_pages, expire;
 754         long offlined_pages;
 755         int ret, drain, retry_max, node;
 756         struct zone *zone;
 757         struct memory_notify arg;
 758
 759         BUG_ON(start_pfn >= end_pfn);
 760         /* at least, alignment against pageblock is necessary */
 761         if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
 762                 return -EINVAL;
 763         if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
 764                 return -EINVAL;
 765         /* This makes hotplug much easier...and readable.
 766            we assume this for now. .*/
 767         if (!test_pages_in_a_zone(start_pfn, end_pfn))
 768                 return -EINVAL;
 769
 770         lock_system_sleep();
 771
 772         zone = page_zone(pfn_to_page(start_pfn));
 773         node = zone_to_nid(zone);
 774         nr_pages = end_pfn - start_pfn;
 775
 776         /* set above range as isolated */
 777         ret = start_isolate_page_range(start_pfn, end_pfn);
 778         if (ret)
 779                 goto out;
 780
 781         arg.start_pfn = start_pfn;
 782         arg.nr_pages = nr_pages;
 783         arg.status_change_nid = -1;
 784         if (nr_pages >= node_present_pages(node))
 785                 arg.status_change_nid = node;
 786
 787         ret = memory_notify(MEM_GOING_OFFLINE, &arg);
 788         ret = notifier_to_errno(ret);
 789         if (ret)
 790                 goto failed_removal;
 791
 792         pfn = start_pfn;
 793         expire = jiffies + timeout;
 794         drain = 0;
 795         retry_max = 5;
 796 repeat:
 797         /* start memory hot removal */
 798         ret = -EAGAIN;
 799         if (time_after(jiffies, expire))
 800                 goto failed_removal;
 801         ret = -EINTR;
 802         if (signal_pending(current))
 803                 goto failed_removal;
 804         ret = 0;
 805         if (drain) {
 806                 lru_add_drain_all();
 807                 flush_scheduled_work();
 808                 cond_resched();
 809                 drain_all_pages();
 810         }
 811
 812         pfn = scan_lru_pages(start_pfn, end_pfn);
 813         if (pfn) { /* We have page on LRU */
 814                 ret = do_migrate_range(pfn, end_pfn);
 815                 if (!ret) {
 816                         drain = 1;
 817                         goto repeat;
 818                 } else {
 819                         if (ret < 0)
 820                                 if (--retry_max == 0)
 821                                         goto failed_removal;
 822                         yield();
 823                         drain = 1;
 824                         goto repeat;
 825                 }
 826         }
 827         /* drain all zone's lru pagevec, this is asyncronous... */
 828         lru_add_drain_all();
 829         flush_scheduled_work();
 830         yield();
 831         /* drain pcp pages , this is synchrouns. */
 832         drain_all_pages();
 833         /* check again */
 834         offlined_pages = check_pages_isolated(start_pfn, end_pfn);
 835         if (offlined_pages < 0) {
 836                 ret = -EBUSY;
 837                 goto failed_removal;
 838         }
 839         printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
 840         /* Ok, all of our target is islaoted.
 841            We cannot do rollback at this point. */
 842         offline_isolated_pages(start_pfn, end_pfn);
 843         /* reset pagetype flags and makes migrate type to be MOVABLE */
 844         undo_isolate_page_range(start_pfn, end_pfn);
 845         /* removal success */
 846         zone->present_pages -= offlined_pages;
 847         zone->zone_pgdat->node_present_pages -= offlined_pages;
 848         totalram_pages -= offlined_pages;
 849
 850         setup_per_zone_wmarks();
 851         calculate_zone_inactive_ratio(zone);
 852
 853         vm_total_pages = nr_free_pagecache_pages();
 854         writeback_set_ratelimit();
 855
 856         memory_notify(MEM_OFFLINE, &arg);
 857         unlock_system_sleep();
 858         return 0;
 859
 860 failed_removal:
 861         printk(KERN_INFO "memory offlining %lx to %lx failed\n",
 862                 start_pfn, end_pfn);
 863         memory_notify(MEM_CANCEL_OFFLINE, &arg);
 864         /* pushback to free area */
 865         undo_isolate_page_range(start_pfn, end_pfn);
 866
 867 out:
 868         unlock_system_sleep();
 869         return ret;
 870 }
 871
 872 int remove_memory(u64 start, u64 size)
 873 {
 874         unsigned long start_pfn, end_pfn;
 875
 876         start_pfn = PFN_DOWN(start);
 877         end_pfn = start_pfn + PFN_DOWN(size);
 878         return offline_pages(start_pfn, end_pfn, 120 * HZ);
 879 }
 880 #else
 881 int remove_memory(u64 start, u64 size)
 882 {
 883         return -EINVAL;
 884 }
 885 #endif /* CONFIG_MEMORY_HOTREMOVE */
 886 EXPORT_SYMBOL_GPL(remove_memory);