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/dyn_pageflags.h>
  15 #include <linux/module.h>
  16 #include <linux/pagevec.h>
  17 #include <linux/writeback.h>
  18 #include <linux/slab.h>
  19 #include <linux/sysctl.h>
  20 #include <linux/cpu.h>
  21 #include <linux/memory.h>
  22 #include <linux/memory_hotplug.h>
  23 #include <linux/highmem.h>
  24 #include <linux/vmalloc.h>
  25 #include <linux/ioport.h>
  26 #include <linux/delay.h>
  27 #include <linux/migrate.h>
  28 #include <linux/page-isolation.h>
  29 #include <linux/pfn.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
 218         dyn_pageflags_hotplug(zone);
 219         return 0;
 220 }
 221
 222 static int __meminit __add_section(struct zone *zone, unsigned long phys_start_pfn)
 223 {
 224         int nr_pages = PAGES_PER_SECTION;
 225         int ret;
 226
 227         if (pfn_valid(phys_start_pfn))
 228                 return -EEXIST;
 229
 230         ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
 231
 232         if (ret < 0)
 233                 return ret;
 234
 235         ret = __add_zone(zone, phys_start_pfn);
 236
 237         if (ret < 0)
 238                 return ret;
 239
 240         return register_new_memory(__pfn_to_section(phys_start_pfn));
 241 }
 242
 243 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 244 static int __remove_section(struct zone *zone, struct mem_section *ms)
 245 {
 246         /*
 247          * XXX: Freeing memmap with vmemmap is not implement yet.
 248          *      This should be removed later.
 249          */
 250         return -EBUSY;
 251 }
 252 #else
 253 static int __remove_section(struct zone *zone, struct mem_section *ms)
 254 {
 255         unsigned long flags;
 256         struct pglist_data *pgdat = zone->zone_pgdat;
 257         int ret = -EINVAL;
 258
 259         if (!valid_section(ms))
 260                 return ret;
 261
 262         ret = unregister_memory_section(ms);
 263         if (ret)
 264                 return ret;
 265
 266         pgdat_resize_lock(pgdat, &flags);
 267         sparse_remove_one_section(zone, ms);
 268         pgdat_resize_unlock(pgdat, &flags);
 269         return 0;
 270 }
 271 #endif
 272
 273 /*
 274  * Reasonably generic function for adding memory.  It is
 275  * expected that archs that support memory hotplug will
 276  * call this function after deciding the zone to which to
 277  * add the new pages.
 278  */
 279 int __ref __add_pages(struct zone *zone, unsigned long phys_start_pfn,
 280                  unsigned long nr_pages)
 281 {
 282         unsigned long i;
 283         int err = 0;
 284         int start_sec, end_sec;
 285         /* during initialize mem_map, align hot-added range to section */
 286         start_sec = pfn_to_section_nr(phys_start_pfn);
 287         end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
 288
 289         for (i = start_sec; i <= end_sec; i++) {
 290                 err = __add_section(zone, i << PFN_SECTION_SHIFT);
 291
 292                 /*
 293                  * EEXIST is finally dealt with by ioresource collision
 294                  * check. see add_memory() => register_memory_resource()
 295                  * Warning will be printed if there is collision.
 296                  */
 297                 if (err && (err != -EEXIST))
 298                         break;
 299                 err = 0;
 300         }
 301
 302         return err;
 303 }
 304 EXPORT_SYMBOL_GPL(__add_pages);
 305
 306 /**
 307  * __remove_pages() - remove sections of pages from a zone
 308  * @zone: zone from which pages need to be removed
 309  * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
 310  * @nr_pages: number of pages to remove (must be multiple of section size)
 311  *
 312  * Generic helper function to remove section mappings and sysfs entries
 313  * for the section of the memory we are removing. Caller needs to make
 314  * sure that pages are marked reserved and zones are adjust properly by
 315  * calling offline_pages().
 316  */
 317 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
 318                  unsigned long nr_pages)
 319 {
 320         unsigned long i, ret = 0;
 321         int sections_to_remove;
 322
 323         /*
 324          * We can only remove entire sections
 325          */
 326         BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
 327         BUG_ON(nr_pages % PAGES_PER_SECTION);
 328
 329         sections_to_remove = nr_pages / PAGES_PER_SECTION;
 330         for (i = 0; i < sections_to_remove; i++) {
 331                 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
 332                 release_mem_region(pfn << PAGE_SHIFT,
 333                                    PAGES_PER_SECTION << PAGE_SHIFT);
 334                 ret = __remove_section(zone, __pfn_to_section(pfn));
 335                 if (ret)
 336                         break;
 337         }
 338         return ret;
 339 }
 340 EXPORT_SYMBOL_GPL(__remove_pages);
 341
 342 void online_page(struct page *page)
 343 {
 344         totalram_pages++;
 345         num_physpages++;
 346
 347 #ifdef CONFIG_HIGHMEM
 348         if (PageHighMem(page))
 349                 totalhigh_pages++;
 350 #endif
 351
 352 #ifdef CONFIG_FLATMEM
 353         max_mapnr = max(page_to_pfn(page), max_mapnr);
 354 #endif
 355
 356         ClearPageReserved(page);
 357         init_page_count(page);
 358         __free_page(page);
 359 }
 360
 361 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
 362                         void *arg)
 363 {
 364         unsigned long i;
 365         unsigned long onlined_pages = *(unsigned long *)arg;
 366         struct page *page;
 367         if (PageReserved(pfn_to_page(start_pfn)))
 368                 for (i = 0; i < nr_pages; i++) {
 369                         page = pfn_to_page(start_pfn + i);
 370                         online_page(page);
 371                         onlined_pages++;
 372                 }
 373         *(unsigned long *)arg = onlined_pages;
 374         return 0;
 375 }
 376
 377
 378 int online_pages(unsigned long pfn, unsigned long nr_pages)
 379 {
 380         unsigned long onlined_pages = 0;
 381         struct zone *zone;
 382         int need_zonelists_rebuild = 0;
 383         int nid;
 384         int ret;
 385         struct memory_notify arg;
 386
 387         arg.start_pfn = pfn;
 388         arg.nr_pages = nr_pages;
 389         arg.status_change_nid = -1;
 390
 391         nid = page_to_nid(pfn_to_page(pfn));
 392         if (node_present_pages(nid) == 0)
 393                 arg.status_change_nid = nid;
 394
 395         ret = memory_notify(MEM_GOING_ONLINE, &arg);
 396         ret = notifier_to_errno(ret);
 397         if (ret) {
 398                 memory_notify(MEM_CANCEL_ONLINE, &arg);
 399                 return ret;
 400         }
 401         /*
 402          * This doesn't need a lock to do pfn_to_page().
 403          * The section can't be removed here because of the
 404          * memory_block->state_mutex.
 405          */
 406         zone = page_zone(pfn_to_page(pfn));
 407         /*
 408          * If this zone is not populated, then it is not in zonelist.
 409          * This means the page allocator ignores this zone.
 410          * So, zonelist must be updated after online.
 411          */
 412         if (!populated_zone(zone))
 413                 need_zonelists_rebuild = 1;
 414
 415         ret = walk_memory_resource(pfn, nr_pages, &onlined_pages,
 416                 online_pages_range);
 417         if (ret) {
 418                 printk(KERN_DEBUG "online_pages %lx at %lx failed\n",
 419                         nr_pages, pfn);
 420                 memory_notify(MEM_CANCEL_ONLINE, &arg);
 421                 return ret;
 422         }
 423
 424         zone->present_pages += onlined_pages;
 425         zone->zone_pgdat->node_present_pages += onlined_pages;
 426
 427         setup_per_zone_pages_min();
 428         if (onlined_pages) {
 429                 kswapd_run(zone_to_nid(zone));
 430                 node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
 431         }
 432
 433         if (need_zonelists_rebuild)
 434                 build_all_zonelists();
 435         else
 436                 vm_total_pages = nr_free_pagecache_pages();
 437
 438         writeback_set_ratelimit();
 439
 440         if (onlined_pages)
 441                 memory_notify(MEM_ONLINE, &arg);
 442
 443         return 0;
 444 }
 445 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
 446
 447 static pg_data_t *hotadd_new_pgdat(int nid, u64 start)
 448 {
 449         struct pglist_data *pgdat;
 450         unsigned long zones_size[MAX_NR_ZONES] = {0};
 451         unsigned long zholes_size[MAX_NR_ZONES] = {0};
 452         unsigned long start_pfn = start >> PAGE_SHIFT;
 453
 454         pgdat = arch_alloc_nodedata(nid);
 455         if (!pgdat)
 456                 return NULL;
 457
 458         arch_refresh_nodedata(nid, pgdat);
 459
 460         /* we can use NODE_DATA(nid) from here */
 461
 462         /* init node's zones as empty zones, we don't have any present pages.*/
 463         free_area_init_node(nid, zones_size, start_pfn, zholes_size);
 464
 465         return pgdat;
 466 }
 467
 468 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
 469 {
 470         arch_refresh_nodedata(nid, NULL);
 471         arch_free_nodedata(pgdat);
 472         return;
 473 }
 474
 475
 476 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
 477 int __ref add_memory(int nid, u64 start, u64 size)
 478 {
 479         pg_data_t *pgdat = NULL;
 480         int new_pgdat = 0;
 481         struct resource *res;
 482         int ret;
 483
 484         res = register_memory_resource(start, size);
 485         if (!res)
 486                 return -EEXIST;
 487
 488         if (!node_online(nid)) {
 489                 pgdat = hotadd_new_pgdat(nid, start);
 490                 if (!pgdat)
 491                         return -ENOMEM;
 492                 new_pgdat = 1;
 493         }
 494
 495         /* call arch's memory hotadd */
 496         ret = arch_add_memory(nid, start, size);
 497
 498         if (ret < 0)
 499                 goto error;
 500
 501         /* we online node here. we can't roll back from here. */
 502         node_set_online(nid);
 503
 504         if (new_pgdat) {
 505                 ret = register_one_node(nid);
 506                 /*
 507                  * If sysfs file of new node can't create, cpu on the node
 508                  * can't be hot-added. There is no rollback way now.
 509                  * So, check by BUG_ON() to catch it reluctantly..
 510                  */
 511                 BUG_ON(ret);
 512         }
 513
 514         return ret;
 515 error:
 516         /* rollback pgdat allocation and others */
 517         if (new_pgdat)
 518                 rollback_node_hotadd(nid, pgdat);
 519         if (res)
 520                 release_memory_resource(res);
 521
 522         return ret;
 523 }
 524 EXPORT_SYMBOL_GPL(add_memory);
 525
 526 #ifdef CONFIG_MEMORY_HOTREMOVE
 527 /*
 528  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
 529  * set and the size of the free page is given by page_order(). Using this,
 530  * the function determines if the pageblock contains only free pages.
 531  * Due to buddy contraints, a free page at least the size of a pageblock will
 532  * be located at the start of the pageblock
 533  */
 534 static inline int pageblock_free(struct page *page)
 535 {
 536         return PageBuddy(page) && page_order(page) >= pageblock_order;
 537 }
 538
 539 /* Return the start of the next active pageblock after a given page */
 540 static struct page *next_active_pageblock(struct page *page)
 541 {
 542         int pageblocks_stride;
 543
 544         /* Ensure the starting page is pageblock-aligned */
 545         BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
 546
 547         /* Move forward by at least 1 * pageblock_nr_pages */
 548         pageblocks_stride = 1;
 549
 550         /* If the entire pageblock is free, move to the end of free page */
 551         if (pageblock_free(page))
 552                 pageblocks_stride += page_order(page) - pageblock_order;
 553
 554         return page + (pageblocks_stride * pageblock_nr_pages);
 555 }
 556
 557 /* Checks if this range of memory is likely to be hot-removable. */
 558 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
 559 {
 560         int type;
 561         struct page *page = pfn_to_page(start_pfn);
 562         struct page *end_page = page + nr_pages;
 563
 564         /* Check the starting page of each pageblock within the range */
 565         for (; page < end_page; page = next_active_pageblock(page)) {
 566                 type = get_pageblock_migratetype(page);
 567
 568                 /*
 569                  * A pageblock containing MOVABLE or free pages is considered
 570                  * removable
 571                  */
 572                 if (type != MIGRATE_MOVABLE && !pageblock_free(page))
 573                         return 0;
 574
 575                 /*
 576                  * A pageblock starting with a PageReserved page is not
 577                  * considered removable.
 578                  */
 579                 if (PageReserved(page))
 580                         return 0;
 581         }
 582
 583         /* All pageblocks in the memory block are likely to be hot-removable */
 584         return 1;
 585 }
 586
 587 /*
 588  * Confirm all pages in a range [start, end) is belongs to the same zone.
 589  */
 590 static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
 591 {
 592         unsigned long pfn;
 593         struct zone *zone = NULL;
 594         struct page *page;
 595         int i;
 596         for (pfn = start_pfn;
 597              pfn < end_pfn;
 598              pfn += MAX_ORDER_NR_PAGES) {
 599                 i = 0;
 600                 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
 601                 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
 602                         i++;
 603                 if (i == MAX_ORDER_NR_PAGES)
 604                         continue;
 605                 page = pfn_to_page(pfn + i);
 606                 if (zone && page_zone(page) != zone)
 607                         return 0;
 608                 zone = page_zone(page);
 609         }
 610         return 1;
 611 }
 612
 613 /*
 614  * Scanning pfn is much easier than scanning lru list.
 615  * Scan pfn from start to end and Find LRU page.
 616  */
 617 int scan_lru_pages(unsigned long start, unsigned long end)
 618 {
 619         unsigned long pfn;
 620         struct page *page;
 621         for (pfn = start; pfn < end; pfn++) {
 622                 if (pfn_valid(pfn)) {
 623                         page = pfn_to_page(pfn);
 624                         if (PageLRU(page))
 625                                 return pfn;
 626                 }
 627         }
 628         return 0;
 629 }
 630
 631 static struct page *
 632 hotremove_migrate_alloc(struct page *page,
 633                         unsigned long private,
 634                         int **x)
 635 {
 636         /* This should be improoooooved!! */
 637         return alloc_page(GFP_HIGHUSER_PAGECACHE);
 638 }
 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_memory_resource(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_memory_resource(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         num_physpages -= offlined_pages;
 839
 840         vm_total_pages = nr_free_pagecache_pages();
 841         writeback_set_ratelimit();
 842
 843         memory_notify(MEM_OFFLINE, &arg);
 844         return 0;
 845
 846 failed_removal:
 847         printk(KERN_INFO "memory offlining %lx to %lx failed\n",
 848                 start_pfn, end_pfn);
 849         memory_notify(MEM_CANCEL_OFFLINE, &arg);
 850         /* pushback to free area */
 851         undo_isolate_page_range(start_pfn, end_pfn);
 852
 853         return ret;
 854 }
 855
 856 int remove_memory(u64 start, u64 size)
 857 {
 858         unsigned long start_pfn, end_pfn;
 859
 860         start_pfn = PFN_DOWN(start);
 861         end_pfn = start_pfn + PFN_DOWN(size);
 862         return offline_pages(start_pfn, end_pfn, 120 * HZ);
 863 }
 864 #else
 865 int remove_memory(u64 start, u64 size)
 866 {
 867         return -EINVAL;
 868 }
 869 #endif /* CONFIG_MEMORY_HOTREMOVE */
 870 EXPORT_SYMBOL_GPL(remove_memory);