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