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/cpuset.h>
  26 #include <linux/delay.h>
  27 #include <linux/migrate.h>
  28 #include <linux/page-isolation.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 magic)
  66 {
  67         atomic_set(&page->_mapcount, magic);
  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 magic;
  76
  77         magic = atomic_read(&page->_mapcount);
  78         BUG_ON(magic >= -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 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_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 int __add_zone(struct zone *zone, unsigned long phys_start_pfn)
 163 {
 164         struct pglist_data *pgdat = zone->zone_pgdat;
 165         int nr_pages = PAGES_PER_SECTION;
 166         int nid = pgdat->node_id;
 167         int zone_type;
 168
 169         zone_type = zone - pgdat->node_zones;
 170         if (!zone->wait_table) {
 171                 int ret = 0;
 172                 ret = init_currently_empty_zone(zone, phys_start_pfn,
 173                                                 nr_pages, MEMMAP_HOTPLUG);
 174                 if (ret < 0)
 175                         return ret;
 176         }
 177         memmap_init_zone(nr_pages, nid, zone_type,
 178                          phys_start_pfn, MEMMAP_HOTPLUG);
 179         return 0;
 180 }
 181
 182 static int __add_section(struct zone *zone, unsigned long phys_start_pfn)
 183 {
 184         int nr_pages = PAGES_PER_SECTION;
 185         int ret;
 186
 187         if (pfn_valid(phys_start_pfn))
 188                 return -EEXIST;
 189
 190         ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
 191
 192         if (ret < 0)
 193                 return ret;
 194
 195         ret = __add_zone(zone, phys_start_pfn);
 196
 197         if (ret < 0)
 198                 return ret;
 199
 200         return register_new_memory(__pfn_to_section(phys_start_pfn));
 201 }
 202
 203 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 204 static int __remove_section(struct zone *zone, struct mem_section *ms)
 205 {
 206         /*
 207          * XXX: Freeing memmap with vmemmap is not implement yet.
 208          *      This should be removed later.
 209          */
 210         return -EBUSY;
 211 }
 212 #else
 213 static int __remove_section(struct zone *zone, struct mem_section *ms)
 214 {
 215         unsigned long flags;
 216         struct pglist_data *pgdat = zone->zone_pgdat;
 217         int ret = -EINVAL;
 218
 219         if (!valid_section(ms))
 220                 return ret;
 221
 222         ret = unregister_memory_section(ms);
 223         if (ret)
 224                 return ret;
 225
 226         pgdat_resize_lock(pgdat, &flags);
 227         sparse_remove_one_section(zone, ms);
 228         pgdat_resize_unlock(pgdat, &flags);
 229         return 0;
 230 }
 231 #endif
 232
 233 /*
 234  * Reasonably generic function for adding memory.  It is
 235  * expected that archs that support memory hotplug will
 236  * call this function after deciding the zone to which to
 237  * add the new pages.
 238  */
 239 int __add_pages(struct zone *zone, unsigned long phys_start_pfn,
 240                  unsigned long nr_pages)
 241 {
 242         unsigned long i;
 243         int err = 0;
 244         int start_sec, end_sec;
 245         /* during initialize mem_map, align hot-added range to section */
 246         start_sec = pfn_to_section_nr(phys_start_pfn);
 247         end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
 248
 249         for (i = start_sec; i <= end_sec; i++) {
 250                 err = __add_section(zone, i << PFN_SECTION_SHIFT);
 251
 252                 /*
 253                  * EEXIST is finally dealt with by ioresource collision
 254                  * check. see add_memory() => register_memory_resource()
 255                  * Warning will be printed if there is collision.
 256                  */
 257                 if (err && (err != -EEXIST))
 258                         break;
 259                 err = 0;
 260         }
 261
 262         return err;
 263 }
 264 EXPORT_SYMBOL_GPL(__add_pages);
 265
 266 /**
 267  * __remove_pages() - remove sections of pages from a zone
 268  * @zone: zone from which pages need to be removed
 269  * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
 270  * @nr_pages: number of pages to remove (must be multiple of section size)
 271  *
 272  * Generic helper function to remove section mappings and sysfs entries
 273  * for the section of the memory we are removing. Caller needs to make
 274  * sure that pages are marked reserved and zones are adjust properly by
 275  * calling offline_pages().
 276  */
 277 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
 278                  unsigned long nr_pages)
 279 {
 280         unsigned long i, ret = 0;
 281         int sections_to_remove;
 282
 283         /*
 284          * We can only remove entire sections
 285          */
 286         BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
 287         BUG_ON(nr_pages % PAGES_PER_SECTION);
 288
 289         release_mem_region(phys_start_pfn << PAGE_SHIFT, nr_pages * PAGE_SIZE);
 290
 291         sections_to_remove = nr_pages / PAGES_PER_SECTION;
 292         for (i = 0; i < sections_to_remove; i++) {
 293                 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
 294                 ret = __remove_section(zone, __pfn_to_section(pfn));
 295                 if (ret)
 296                         break;
 297         }
 298         return ret;
 299 }
 300 EXPORT_SYMBOL_GPL(__remove_pages);
 301
 302 static void grow_zone_span(struct zone *zone,
 303                 unsigned long start_pfn, unsigned long end_pfn)
 304 {
 305         unsigned long old_zone_end_pfn;
 306
 307         zone_span_writelock(zone);
 308
 309         old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
 310         if (start_pfn < zone->zone_start_pfn)
 311                 zone->zone_start_pfn = start_pfn;
 312
 313         zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
 314                                 zone->zone_start_pfn;
 315
 316         zone_span_writeunlock(zone);
 317 }
 318
 319 static void grow_pgdat_span(struct pglist_data *pgdat,
 320                 unsigned long start_pfn, unsigned long end_pfn)
 321 {
 322         unsigned long old_pgdat_end_pfn =
 323                 pgdat->node_start_pfn + pgdat->node_spanned_pages;
 324
 325         if (start_pfn < pgdat->node_start_pfn)
 326                 pgdat->node_start_pfn = start_pfn;
 327
 328         pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
 329                                         pgdat->node_start_pfn;
 330 }
 331
 332 void online_page(struct page *page)
 333 {
 334         totalram_pages++;
 335         num_physpages++;
 336
 337 #ifdef CONFIG_HIGHMEM
 338         if (PageHighMem(page))
 339                 totalhigh_pages++;
 340 #endif
 341
 342 #ifdef CONFIG_FLATMEM
 343         max_mapnr = max(page_to_pfn(page), max_mapnr);
 344 #endif
 345
 346         ClearPageReserved(page);
 347         init_page_count(page);
 348         __free_page(page);
 349 }
 350
 351 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
 352                         void *arg)
 353 {
 354         unsigned long i;
 355         unsigned long onlined_pages = *(unsigned long *)arg;
 356         struct page *page;
 357         if (PageReserved(pfn_to_page(start_pfn)))
 358                 for (i = 0; i < nr_pages; i++) {
 359                         page = pfn_to_page(start_pfn + i);
 360                         online_page(page);
 361                         onlined_pages++;
 362                 }
 363         *(unsigned long *)arg = onlined_pages;
 364         return 0;
 365 }
 366
 367
 368 int online_pages(unsigned long pfn, unsigned long nr_pages)
 369 {
 370         unsigned long flags;
 371         unsigned long onlined_pages = 0;
 372         struct zone *zone;
 373         int need_zonelists_rebuild = 0;
 374         int nid;
 375         int ret;
 376         struct memory_notify arg;
 377
 378         arg.start_pfn = pfn;
 379         arg.nr_pages = nr_pages;
 380         arg.status_change_nid = -1;
 381
 382         nid = page_to_nid(pfn_to_page(pfn));
 383         if (node_present_pages(nid) == 0)
 384                 arg.status_change_nid = nid;
 385
 386         ret = memory_notify(MEM_GOING_ONLINE, &arg);
 387         ret = notifier_to_errno(ret);
 388         if (ret) {
 389                 memory_notify(MEM_CANCEL_ONLINE, &arg);
 390                 return ret;
 391         }
 392         /*
 393          * This doesn't need a lock to do pfn_to_page().
 394          * The section can't be removed here because of the
 395          * memory_block->state_mutex.
 396          */
 397         zone = page_zone(pfn_to_page(pfn));
 398         pgdat_resize_lock(zone->zone_pgdat, &flags);
 399         grow_zone_span(zone, pfn, pfn + nr_pages);
 400         grow_pgdat_span(zone->zone_pgdat, pfn, pfn + nr_pages);
 401         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 402
 403         /*
 404          * If this zone is not populated, then it is not in zonelist.
 405          * This means the page allocator ignores this zone.
 406          * So, zonelist must be updated after online.
 407          */
 408         if (!populated_zone(zone))
 409                 need_zonelists_rebuild = 1;
 410
 411         walk_memory_resource(pfn, nr_pages, &onlined_pages,
 412                 online_pages_range);
 413         zone->present_pages += onlined_pages;
 414         zone->zone_pgdat->node_present_pages += onlined_pages;
 415
 416         setup_per_zone_pages_min();
 417         if (onlined_pages) {
 418                 kswapd_run(zone_to_nid(zone));
 419                 node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
 420         }
 421
 422         if (need_zonelists_rebuild)
 423                 build_all_zonelists();
 424         vm_total_pages = nr_free_pagecache_pages();
 425         writeback_set_ratelimit();
 426
 427         if (onlined_pages)
 428                 memory_notify(MEM_ONLINE, &arg);
 429
 430         return 0;
 431 }
 432 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
 433
 434 static pg_data_t *hotadd_new_pgdat(int nid, u64 start)
 435 {
 436         struct pglist_data *pgdat;
 437         unsigned long zones_size[MAX_NR_ZONES] = {0};
 438         unsigned long zholes_size[MAX_NR_ZONES] = {0};
 439         unsigned long start_pfn = start >> PAGE_SHIFT;
 440
 441         pgdat = arch_alloc_nodedata(nid);
 442         if (!pgdat)
 443                 return NULL;
 444
 445         arch_refresh_nodedata(nid, pgdat);
 446
 447         /* we can use NODE_DATA(nid) from here */
 448
 449         /* init node's zones as empty zones, we don't have any present pages.*/
 450         free_area_init_node(nid, pgdat, zones_size, start_pfn, zholes_size);
 451
 452         return pgdat;
 453 }
 454
 455 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
 456 {
 457         arch_refresh_nodedata(nid, NULL);
 458         arch_free_nodedata(pgdat);
 459         return;
 460 }
 461
 462
 463 int add_memory(int nid, u64 start, u64 size)
 464 {
 465         pg_data_t *pgdat = NULL;
 466         int new_pgdat = 0;
 467         struct resource *res;
 468         int ret;
 469
 470         res = register_memory_resource(start, size);
 471         if (!res)
 472                 return -EEXIST;
 473
 474         if (!node_online(nid)) {
 475                 pgdat = hotadd_new_pgdat(nid, start);
 476                 if (!pgdat)
 477                         return -ENOMEM;
 478                 new_pgdat = 1;
 479         }
 480
 481         /* call arch's memory hotadd */
 482         ret = arch_add_memory(nid, start, size);
 483
 484         if (ret < 0)
 485                 goto error;
 486
 487         /* we online node here. we can't roll back from here. */
 488         node_set_online(nid);
 489
 490         cpuset_track_online_nodes();
 491
 492         if (new_pgdat) {
 493                 ret = register_one_node(nid);
 494                 /*
 495                  * If sysfs file of new node can't create, cpu on the node
 496                  * can't be hot-added. There is no rollback way now.
 497                  * So, check by BUG_ON() to catch it reluctantly..
 498                  */
 499                 BUG_ON(ret);
 500         }
 501
 502         return ret;
 503 error:
 504         /* rollback pgdat allocation and others */
 505         if (new_pgdat)
 506                 rollback_node_hotadd(nid, pgdat);
 507         if (res)
 508                 release_memory_resource(res);
 509
 510         return ret;
 511 }
 512 EXPORT_SYMBOL_GPL(add_memory);
 513
 514 #ifdef CONFIG_MEMORY_HOTREMOVE
 515 /*
 516  * Confirm all pages in a range [start, end) is belongs to the same zone.
 517  */
 518 static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
 519 {
 520         unsigned long pfn;
 521         struct zone *zone = NULL;
 522         struct page *page;
 523         int i;
 524         for (pfn = start_pfn;
 525              pfn < end_pfn;
 526              pfn += MAX_ORDER_NR_PAGES) {
 527                 i = 0;
 528                 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
 529                 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
 530                         i++;
 531                 if (i == MAX_ORDER_NR_PAGES)
 532                         continue;
 533                 page = pfn_to_page(pfn + i);
 534                 if (zone && page_zone(page) != zone)
 535                         return 0;
 536                 zone = page_zone(page);
 537         }
 538         return 1;
 539 }
 540
 541 /*
 542  * Scanning pfn is much easier than scanning lru list.
 543  * Scan pfn from start to end and Find LRU page.
 544  */
 545 int scan_lru_pages(unsigned long start, unsigned long end)
 546 {
 547         unsigned long pfn;
 548         struct page *page;
 549         for (pfn = start; pfn < end; pfn++) {
 550                 if (pfn_valid(pfn)) {
 551                         page = pfn_to_page(pfn);
 552                         if (PageLRU(page))
 553                                 return pfn;
 554                 }
 555         }
 556         return 0;
 557 }
 558
 559 static struct page *
 560 hotremove_migrate_alloc(struct page *page,
 561                         unsigned long private,
 562                         int **x)
 563 {
 564         /* This should be improoooooved!! */
 565         return alloc_page(GFP_HIGHUSER_PAGECACHE);
 566 }
 567
 568
 569 #define NR_OFFLINE_AT_ONCE_PAGES        (256)
 570 static int
 571 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
 572 {
 573         unsigned long pfn;
 574         struct page *page;
 575         int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
 576         int not_managed = 0;
 577         int ret = 0;
 578         LIST_HEAD(source);
 579
 580         for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
 581                 if (!pfn_valid(pfn))
 582                         continue;
 583                 page = pfn_to_page(pfn);
 584                 if (!page_count(page))
 585                         continue;
 586                 /*
 587                  * We can skip free pages. And we can only deal with pages on
 588                  * LRU.
 589                  */
 590                 ret = isolate_lru_page(page, &source);
 591                 if (!ret) { /* Success */
 592                         move_pages--;
 593                 } else {
 594                         /* Becasue we don't have big zone->lock. we should
 595                            check this again here. */
 596                         if (page_count(page))
 597                                 not_managed++;
 598 #ifdef CONFIG_DEBUG_VM
 599                         printk(KERN_INFO "removing from LRU failed"
 600                                          " %lx/%d/%lx\n",
 601                                 pfn, page_count(page), page->flags);
 602 #endif
 603                 }
 604         }
 605         ret = -EBUSY;
 606         if (not_managed) {
 607                 if (!list_empty(&source))
 608                         putback_lru_pages(&source);
 609                 goto out;
 610         }
 611         ret = 0;
 612         if (list_empty(&source))
 613                 goto out;
 614         /* this function returns # of failed pages */
 615         ret = migrate_pages(&source, hotremove_migrate_alloc, 0);
 616
 617 out:
 618         return ret;
 619 }
 620
 621 /*
 622  * remove from free_area[] and mark all as Reserved.
 623  */
 624 static int
 625 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
 626                         void *data)
 627 {
 628         __offline_isolated_pages(start, start + nr_pages);
 629         return 0;
 630 }
 631
 632 static void
 633 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
 634 {
 635         walk_memory_resource(start_pfn, end_pfn - start_pfn, NULL,
 636                                 offline_isolated_pages_cb);
 637 }
 638
 639 /*
 640  * Check all pages in range, recoreded as memory resource, are isolated.
 641  */
 642 static int
 643 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
 644                         void *data)
 645 {
 646         int ret;
 647         long offlined = *(long *)data;
 648         ret = test_pages_isolated(start_pfn, start_pfn + nr_pages);
 649         offlined = nr_pages;
 650         if (!ret)
 651                 *(long *)data += offlined;
 652         return ret;
 653 }
 654
 655 static long
 656 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
 657 {
 658         long offlined = 0;
 659         int ret;
 660
 661         ret = walk_memory_resource(start_pfn, end_pfn - start_pfn, &offlined,
 662                         check_pages_isolated_cb);
 663         if (ret < 0)
 664                 offlined = (long)ret;
 665         return offlined;
 666 }
 667
 668 int offline_pages(unsigned long start_pfn,
 669                   unsigned long end_pfn, unsigned long timeout)
 670 {
 671         unsigned long pfn, nr_pages, expire;
 672         long offlined_pages;
 673         int ret, drain, retry_max, node;
 674         struct zone *zone;
 675         struct memory_notify arg;
 676
 677         BUG_ON(start_pfn >= end_pfn);
 678         /* at least, alignment against pageblock is necessary */
 679         if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
 680                 return -EINVAL;
 681         if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
 682                 return -EINVAL;
 683         /* This makes hotplug much easier...and readable.
 684            we assume this for now. .*/
 685         if (!test_pages_in_a_zone(start_pfn, end_pfn))
 686                 return -EINVAL;
 687
 688         zone = page_zone(pfn_to_page(start_pfn));
 689         node = zone_to_nid(zone);
 690         nr_pages = end_pfn - start_pfn;
 691
 692         /* set above range as isolated */
 693         ret = start_isolate_page_range(start_pfn, end_pfn);
 694         if (ret)
 695                 return ret;
 696
 697         arg.start_pfn = start_pfn;
 698         arg.nr_pages = nr_pages;
 699         arg.status_change_nid = -1;
 700         if (nr_pages >= node_present_pages(node))
 701                 arg.status_change_nid = node;
 702
 703         ret = memory_notify(MEM_GOING_OFFLINE, &arg);
 704         ret = notifier_to_errno(ret);
 705         if (ret)
 706                 goto failed_removal;
 707
 708         pfn = start_pfn;
 709         expire = jiffies + timeout;
 710         drain = 0;
 711         retry_max = 5;
 712 repeat:
 713         /* start memory hot removal */
 714         ret = -EAGAIN;
 715         if (time_after(jiffies, expire))
 716                 goto failed_removal;
 717         ret = -EINTR;
 718         if (signal_pending(current))
 719                 goto failed_removal;
 720         ret = 0;
 721         if (drain) {
 722                 lru_add_drain_all();
 723                 flush_scheduled_work();
 724                 cond_resched();
 725                 drain_all_pages();
 726         }
 727
 728         pfn = scan_lru_pages(start_pfn, end_pfn);
 729         if (pfn) { /* We have page on LRU */
 730                 ret = do_migrate_range(pfn, end_pfn);
 731                 if (!ret) {
 732                         drain = 1;
 733                         goto repeat;
 734                 } else {
 735                         if (ret < 0)
 736                                 if (--retry_max == 0)
 737                                         goto failed_removal;
 738                         yield();
 739                         drain = 1;
 740                         goto repeat;
 741                 }
 742         }
 743         /* drain all zone's lru pagevec, this is asyncronous... */
 744         lru_add_drain_all();
 745         flush_scheduled_work();
 746         yield();
 747         /* drain pcp pages , this is synchrouns. */
 748         drain_all_pages();
 749         /* check again */
 750         offlined_pages = check_pages_isolated(start_pfn, end_pfn);
 751         if (offlined_pages < 0) {
 752                 ret = -EBUSY;
 753                 goto failed_removal;
 754         }
 755         printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
 756         /* Ok, all of our target is islaoted.
 757            We cannot do rollback at this point. */
 758         offline_isolated_pages(start_pfn, end_pfn);
 759         /* reset pagetype flags and makes migrate type to be MOVABLE */
 760         undo_isolate_page_range(start_pfn, end_pfn);
 761         /* removal success */
 762         zone->present_pages -= offlined_pages;
 763         zone->zone_pgdat->node_present_pages -= offlined_pages;
 764         totalram_pages -= offlined_pages;
 765         num_physpages -= offlined_pages;
 766
 767         vm_total_pages = nr_free_pagecache_pages();
 768         writeback_set_ratelimit();
 769
 770         memory_notify(MEM_OFFLINE, &arg);
 771         return 0;
 772
 773 failed_removal:
 774         printk(KERN_INFO "memory offlining %lx to %lx failed\n",
 775                 start_pfn, end_pfn);
 776         memory_notify(MEM_CANCEL_OFFLINE, &arg);
 777         /* pushback to free area */
 778         undo_isolate_page_range(start_pfn, end_pfn);
 779
 780         return ret;
 781 }
 782 #else
 783 int remove_memory(u64 start, u64 size)
 784 {
 785         return -EINVAL;
 786 }
 787 EXPORT_SYMBOL_GPL(remove_memory);
 788 #endif /* CONFIG_MEMORY_HOTREMOVE */