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