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 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 __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 __add_section(struct zone *zone, unsigned long phys_start_pfn)
 220 {
 221         int nr_pages = PAGES_PER_SECTION;
 222         int ret;
 223
 224         if (pfn_valid(phys_start_pfn))
 225                 return -EEXIST;
 226
 227         ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
 228
 229         if (ret < 0)
 230                 return ret;
 231
 232         ret = __add_zone(zone, phys_start_pfn);
 233
 234         if (ret < 0)
 235                 return ret;
 236
 237         return register_new_memory(__pfn_to_section(phys_start_pfn));
 238 }
 239
 240 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 241 static int __remove_section(struct zone *zone, struct mem_section *ms)
 242 {
 243         /*
 244          * XXX: Freeing memmap with vmemmap is not implement yet.
 245          *      This should be removed later.
 246          */
 247         return -EBUSY;
 248 }
 249 #else
 250 static int __remove_section(struct zone *zone, struct mem_section *ms)
 251 {
 252         unsigned long flags;
 253         struct pglist_data *pgdat = zone->zone_pgdat;
 254         int ret = -EINVAL;
 255
 256         if (!valid_section(ms))
 257                 return ret;
 258
 259         ret = unregister_memory_section(ms);
 260         if (ret)
 261                 return ret;
 262
 263         pgdat_resize_lock(pgdat, &flags);
 264         sparse_remove_one_section(zone, ms);
 265         pgdat_resize_unlock(pgdat, &flags);
 266         return 0;
 267 }
 268 #endif
 269
 270 /*
 271  * Reasonably generic function for adding memory.  It is
 272  * expected that archs that support memory hotplug will
 273  * call this function after deciding the zone to which to
 274  * add the new pages.
 275  */
 276 int __add_pages(struct zone *zone, unsigned long phys_start_pfn,
 277                  unsigned long nr_pages)
 278 {
 279         unsigned long i;
 280         int err = 0;
 281         int start_sec, end_sec;
 282         /* during initialize mem_map, align hot-added range to section */
 283         start_sec = pfn_to_section_nr(phys_start_pfn);
 284         end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
 285
 286         for (i = start_sec; i <= end_sec; i++) {
 287                 err = __add_section(zone, i << PFN_SECTION_SHIFT);
 288
 289                 /*
 290                  * EEXIST is finally dealt with by ioresource collision
 291                  * check. see add_memory() => register_memory_resource()
 292                  * Warning will be printed if there is collision.
 293                  */
 294                 if (err && (err != -EEXIST))
 295                         break;
 296                 err = 0;
 297         }
 298
 299         return err;
 300 }
 301 EXPORT_SYMBOL_GPL(__add_pages);
 302
 303 /**
 304  * __remove_pages() - remove sections of pages from a zone
 305  * @zone: zone from which pages need to be removed
 306  * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
 307  * @nr_pages: number of pages to remove (must be multiple of section size)
 308  *
 309  * Generic helper function to remove section mappings and sysfs entries
 310  * for the section of the memory we are removing. Caller needs to make
 311  * sure that pages are marked reserved and zones are adjust properly by
 312  * calling offline_pages().
 313  */
 314 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
 315                  unsigned long nr_pages)
 316 {
 317         unsigned long i, ret = 0;
 318         int sections_to_remove;
 319
 320         /*
 321          * We can only remove entire sections
 322          */
 323         BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
 324         BUG_ON(nr_pages % PAGES_PER_SECTION);
 325
 326         release_mem_region(phys_start_pfn << PAGE_SHIFT, nr_pages * PAGE_SIZE);
 327
 328         sections_to_remove = nr_pages / PAGES_PER_SECTION;
 329         for (i = 0; i < sections_to_remove; i++) {
 330                 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
 331                 ret = __remove_section(zone, __pfn_to_section(pfn));
 332                 if (ret)
 333                         break;
 334         }
 335         return ret;
 336 }
 337 EXPORT_SYMBOL_GPL(__remove_pages);
 338
 339 void online_page(struct page *page)
 340 {
 341         totalram_pages++;
 342         num_physpages++;
 343
 344 #ifdef CONFIG_HIGHMEM
 345         if (PageHighMem(page))
 346                 totalhigh_pages++;
 347 #endif
 348
 349 #ifdef CONFIG_FLATMEM
 350         max_mapnr = max(page_to_pfn(page), max_mapnr);
 351 #endif
 352
 353         ClearPageReserved(page);
 354         init_page_count(page);
 355         __free_page(page);
 356 }
 357
 358 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
 359                         void *arg)
 360 {
 361         unsigned long i;
 362         unsigned long onlined_pages = *(unsigned long *)arg;
 363         struct page *page;
 364         if (PageReserved(pfn_to_page(start_pfn)))
 365                 for (i = 0; i < nr_pages; i++) {
 366                         page = pfn_to_page(start_pfn + i);
 367                         online_page(page);
 368                         onlined_pages++;
 369                 }
 370         *(unsigned long *)arg = onlined_pages;
 371         return 0;
 372 }
 373
 374
 375 int online_pages(unsigned long pfn, unsigned long nr_pages)
 376 {
 377         unsigned long onlined_pages = 0;
 378         struct zone *zone;
 379         int need_zonelists_rebuild = 0;
 380         int nid;
 381         int ret;
 382         struct memory_notify arg;
 383
 384         arg.start_pfn = pfn;
 385         arg.nr_pages = nr_pages;
 386         arg.status_change_nid = -1;
 387
 388         nid = page_to_nid(pfn_to_page(pfn));
 389         if (node_present_pages(nid) == 0)
 390                 arg.status_change_nid = nid;
 391
 392         ret = memory_notify(MEM_GOING_ONLINE, &arg);
 393         ret = notifier_to_errno(ret);
 394         if (ret) {
 395                 memory_notify(MEM_CANCEL_ONLINE, &arg);
 396                 return ret;
 397         }
 398         /*
 399          * This doesn't need a lock to do pfn_to_page().
 400          * The section can't be removed here because of the
 401          * memory_block->state_mutex.
 402          */
 403         zone = page_zone(pfn_to_page(pfn));
 404         /*
 405          * If this zone is not populated, then it is not in zonelist.
 406          * This means the page allocator ignores this zone.
 407          * So, zonelist must be updated after online.
 408          */
 409         if (!populated_zone(zone))
 410                 need_zonelists_rebuild = 1;
 411
 412         ret = walk_memory_resource(pfn, nr_pages, &onlined_pages,
 413                 online_pages_range);
 414         if (ret) {
 415                 printk(KERN_DEBUG "online_pages %lx at %lx failed\n",
 416                         nr_pages, pfn);
 417                 memory_notify(MEM_CANCEL_ONLINE, &arg);
 418                 return ret;
 419         }
 420
 421         zone->present_pages += onlined_pages;
 422         zone->zone_pgdat->node_present_pages += onlined_pages;
 423
 424         setup_per_zone_pages_min();
 425         if (onlined_pages) {
 426                 kswapd_run(zone_to_nid(zone));
 427                 node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
 428         }
 429
 430         if (need_zonelists_rebuild)
 431                 build_all_zonelists();
 432         else
 433                 vm_total_pages = nr_free_pagecache_pages();
 434
 435         writeback_set_ratelimit();
 436
 437         if (onlined_pages)
 438                 memory_notify(MEM_ONLINE, &arg);
 439
 440         return 0;
 441 }
 442 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
 443
 444 static pg_data_t *hotadd_new_pgdat(int nid, u64 start)
 445 {
 446         struct pglist_data *pgdat;
 447         unsigned long zones_size[MAX_NR_ZONES] = {0};
 448         unsigned long zholes_size[MAX_NR_ZONES] = {0};
 449         unsigned long start_pfn = start >> PAGE_SHIFT;
 450
 451         pgdat = arch_alloc_nodedata(nid);
 452         if (!pgdat)
 453                 return NULL;
 454
 455         arch_refresh_nodedata(nid, pgdat);
 456
 457         /* we can use NODE_DATA(nid) from here */
 458
 459         /* init node's zones as empty zones, we don't have any present pages.*/
 460         free_area_init_node(nid, zones_size, start_pfn, zholes_size);
 461
 462         return pgdat;
 463 }
 464
 465 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
 466 {
 467         arch_refresh_nodedata(nid, NULL);
 468         arch_free_nodedata(pgdat);
 469         return;
 470 }
 471
 472
 473 int add_memory(int nid, u64 start, u64 size)
 474 {
 475         pg_data_t *pgdat = NULL;
 476         int new_pgdat = 0;
 477         struct resource *res;
 478         int ret;
 479
 480         res = register_memory_resource(start, size);
 481         if (!res)
 482                 return -EEXIST;
 483
 484         if (!node_online(nid)) {
 485                 pgdat = hotadd_new_pgdat(nid, start);
 486                 if (!pgdat)
 487                         return -ENOMEM;
 488                 new_pgdat = 1;
 489         }
 490
 491         /* call arch's memory hotadd */
 492         ret = arch_add_memory(nid, start, size);
 493
 494         if (ret < 0)
 495                 goto error;
 496
 497         /* we online node here. we can't roll back from here. */
 498         node_set_online(nid);
 499
 500         cpuset_track_online_nodes();
 501
 502         if (new_pgdat) {
 503                 ret = register_one_node(nid);
 504                 /*
 505                  * If sysfs file of new node can't create, cpu on the node
 506                  * can't be hot-added. There is no rollback way now.
 507                  * So, check by BUG_ON() to catch it reluctantly..
 508                  */
 509                 BUG_ON(ret);
 510         }
 511
 512         return ret;
 513 error:
 514         /* rollback pgdat allocation and others */
 515         if (new_pgdat)
 516                 rollback_node_hotadd(nid, pgdat);
 517         if (res)
 518                 release_memory_resource(res);
 519
 520         return ret;
 521 }
 522 EXPORT_SYMBOL_GPL(add_memory);
 523
 524 #ifdef CONFIG_MEMORY_HOTREMOVE
 525 /*
 526  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
 527  * set and the size of the free page is given by page_order(). Using this,
 528  * the function determines if the pageblock contains only free pages.
 529  * Due to buddy contraints, a free page at least the size of a pageblock will
 530  * be located at the start of the pageblock
 531  */
 532 static inline int pageblock_free(struct page *page)
 533 {
 534         return PageBuddy(page) && page_order(page) >= pageblock_order;
 535 }
 536
 537 /* Return the start of the next active pageblock after a given page */
 538 static struct page *next_active_pageblock(struct page *page)
 539 {
 540         int pageblocks_stride;
 541
 542         /* Ensure the starting page is pageblock-aligned */
 543         BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
 544
 545         /* Move forward by at least 1 * pageblock_nr_pages */
 546         pageblocks_stride = 1;
 547
 548         /* If the entire pageblock is free, move to the end of free page */
 549         if (pageblock_free(page))
 550                 pageblocks_stride += page_order(page) - pageblock_order;
 551
 552         return page + (pageblocks_stride * pageblock_nr_pages);
 553 }
 554
 555 /* Checks if this range of memory is likely to be hot-removable. */
 556 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
 557 {
 558         int type;
 559         struct page *page = pfn_to_page(start_pfn);
 560         struct page *end_page = page + nr_pages;
 561
 562         /* Check the starting page of each pageblock within the range */
 563         for (; page < end_page; page = next_active_pageblock(page)) {
 564                 type = get_pageblock_migratetype(page);
 565
 566                 /*
 567                  * A pageblock containing MOVABLE or free pages is considered
 568                  * removable
 569                  */
 570                 if (type != MIGRATE_MOVABLE && !pageblock_free(page))
 571                         return 0;
 572
 573                 /*
 574                  * A pageblock starting with a PageReserved page is not
 575                  * considered removable.
 576                  */
 577                 if (PageReserved(page))
 578                         return 0;
 579         }
 580
 581         /* All pageblocks in the memory block are likely to be hot-removable */
 582         return 1;
 583 }
 584
 585 /*
 586  * Confirm all pages in a range [start, end) is belongs to the same zone.
 587  */
 588 static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
 589 {
 590         unsigned long pfn;
 591         struct zone *zone = NULL;
 592         struct page *page;
 593         int i;
 594         for (pfn = start_pfn;
 595              pfn < end_pfn;
 596              pfn += MAX_ORDER_NR_PAGES) {
 597                 i = 0;
 598                 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
 599                 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
 600                         i++;
 601                 if (i == MAX_ORDER_NR_PAGES)
 602                         continue;
 603                 page = pfn_to_page(pfn + i);
 604                 if (zone && page_zone(page) != zone)
 605                         return 0;
 606                 zone = page_zone(page);
 607         }
 608         return 1;
 609 }
 610
 611 /*
 612  * Scanning pfn is much easier than scanning lru list.
 613  * Scan pfn from start to end and Find LRU page.
 614  */
 615 int scan_lru_pages(unsigned long start, unsigned long end)
 616 {
 617         unsigned long pfn;
 618         struct page *page;
 619         for (pfn = start; pfn < end; pfn++) {
 620                 if (pfn_valid(pfn)) {
 621                         page = pfn_to_page(pfn);
 622                         if (PageLRU(page))
 623                                 return pfn;
 624                 }
 625         }
 626         return 0;
 627 }
 628
 629 static struct page *
 630 hotremove_migrate_alloc(struct page *page,
 631                         unsigned long private,
 632                         int **x)
 633 {
 634         /* This should be improoooooved!! */
 635         return alloc_page(GFP_HIGHUSER_PAGECACHE);
 636 }
 637
 638
 639 #define NR_OFFLINE_AT_ONCE_PAGES        (256)
 640 static int
 641 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
 642 {
 643         unsigned long pfn;
 644         struct page *page;
 645         int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
 646         int not_managed = 0;
 647         int ret = 0;
 648         LIST_HEAD(source);
 649
 650         for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
 651                 if (!pfn_valid(pfn))
 652                         continue;
 653                 page = pfn_to_page(pfn);
 654                 if (!page_count(page))
 655                         continue;
 656                 /*
 657                  * We can skip free pages. And we can only deal with pages on
 658                  * LRU.
 659                  */
 660                 ret = isolate_lru_page(page, &source);
 661                 if (!ret) { /* Success */
 662                         move_pages--;
 663                 } else {
 664                         /* Becasue we don't have big zone->lock. we should
 665                            check this again here. */
 666                         if (page_count(page))
 667                                 not_managed++;
 668 #ifdef CONFIG_DEBUG_VM
 669                         printk(KERN_INFO "removing from LRU failed"
 670                                          " %lx/%d/%lx\n",
 671                                 pfn, page_count(page), page->flags);
 672 #endif
 673                 }
 674         }
 675         ret = -EBUSY;
 676         if (not_managed) {
 677                 if (!list_empty(&source))
 678                         putback_lru_pages(&source);
 679                 goto out;
 680         }
 681         ret = 0;
 682         if (list_empty(&source))
 683                 goto out;
 684         /* this function returns # of failed pages */
 685         ret = migrate_pages(&source, hotremove_migrate_alloc, 0);
 686
 687 out:
 688         return ret;
 689 }
 690
 691 /*
 692  * remove from free_area[] and mark all as Reserved.
 693  */
 694 static int
 695 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
 696                         void *data)
 697 {
 698         __offline_isolated_pages(start, start + nr_pages);
 699         return 0;
 700 }
 701
 702 static void
 703 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
 704 {
 705         walk_memory_resource(start_pfn, end_pfn - start_pfn, NULL,
 706                                 offline_isolated_pages_cb);
 707 }
 708
 709 /*
 710  * Check all pages in range, recoreded as memory resource, are isolated.
 711  */
 712 static int
 713 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
 714                         void *data)
 715 {
 716         int ret;
 717         long offlined = *(long *)data;
 718         ret = test_pages_isolated(start_pfn, start_pfn + nr_pages);
 719         offlined = nr_pages;
 720         if (!ret)
 721                 *(long *)data += offlined;
 722         return ret;
 723 }
 724
 725 static long
 726 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
 727 {
 728         long offlined = 0;
 729         int ret;
 730
 731         ret = walk_memory_resource(start_pfn, end_pfn - start_pfn, &offlined,
 732                         check_pages_isolated_cb);
 733         if (ret < 0)
 734                 offlined = (long)ret;
 735         return offlined;
 736 }
 737
 738 int offline_pages(unsigned long start_pfn,
 739                   unsigned long end_pfn, unsigned long timeout)
 740 {
 741         unsigned long pfn, nr_pages, expire;
 742         long offlined_pages;
 743         int ret, drain, retry_max, node;
 744         struct zone *zone;
 745         struct memory_notify arg;
 746
 747         BUG_ON(start_pfn >= end_pfn);
 748         /* at least, alignment against pageblock is necessary */
 749         if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
 750                 return -EINVAL;
 751         if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
 752                 return -EINVAL;
 753         /* This makes hotplug much easier...and readable.
 754            we assume this for now. .*/
 755         if (!test_pages_in_a_zone(start_pfn, end_pfn))
 756                 return -EINVAL;
 757
 758         zone = page_zone(pfn_to_page(start_pfn));
 759         node = zone_to_nid(zone);
 760         nr_pages = end_pfn - start_pfn;
 761
 762         /* set above range as isolated */
 763         ret = start_isolate_page_range(start_pfn, end_pfn);
 764         if (ret)
 765                 return ret;
 766
 767         arg.start_pfn = start_pfn;
 768         arg.nr_pages = nr_pages;
 769         arg.status_change_nid = -1;
 770         if (nr_pages >= node_present_pages(node))
 771                 arg.status_change_nid = node;
 772
 773         ret = memory_notify(MEM_GOING_OFFLINE, &arg);
 774         ret = notifier_to_errno(ret);
 775         if (ret)
 776                 goto failed_removal;
 777
 778         pfn = start_pfn;
 779         expire = jiffies + timeout;
 780         drain = 0;
 781         retry_max = 5;
 782 repeat:
 783         /* start memory hot removal */
 784         ret = -EAGAIN;
 785         if (time_after(jiffies, expire))
 786                 goto failed_removal;
 787         ret = -EINTR;
 788         if (signal_pending(current))
 789                 goto failed_removal;
 790         ret = 0;
 791         if (drain) {
 792                 lru_add_drain_all();
 793                 flush_scheduled_work();
 794                 cond_resched();
 795                 drain_all_pages();
 796         }
 797
 798         pfn = scan_lru_pages(start_pfn, end_pfn);
 799         if (pfn) { /* We have page on LRU */
 800                 ret = do_migrate_range(pfn, end_pfn);
 801                 if (!ret) {
 802                         drain = 1;
 803                         goto repeat;
 804                 } else {
 805                         if (ret < 0)
 806                                 if (--retry_max == 0)
 807                                         goto failed_removal;
 808                         yield();
 809                         drain = 1;
 810                         goto repeat;
 811                 }
 812         }
 813         /* drain all zone's lru pagevec, this is asyncronous... */
 814         lru_add_drain_all();
 815         flush_scheduled_work();
 816         yield();
 817         /* drain pcp pages , this is synchrouns. */
 818         drain_all_pages();
 819         /* check again */
 820         offlined_pages = check_pages_isolated(start_pfn, end_pfn);
 821         if (offlined_pages < 0) {
 822                 ret = -EBUSY;
 823                 goto failed_removal;
 824         }
 825         printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
 826         /* Ok, all of our target is islaoted.
 827            We cannot do rollback at this point. */
 828         offline_isolated_pages(start_pfn, end_pfn);
 829         /* reset pagetype flags and makes migrate type to be MOVABLE */
 830         undo_isolate_page_range(start_pfn, end_pfn);
 831         /* removal success */
 832         zone->present_pages -= offlined_pages;
 833         zone->zone_pgdat->node_present_pages -= offlined_pages;
 834         totalram_pages -= offlined_pages;
 835         num_physpages -= offlined_pages;
 836
 837         vm_total_pages = nr_free_pagecache_pages();
 838         writeback_set_ratelimit();
 839
 840         memory_notify(MEM_OFFLINE, &arg);
 841         return 0;
 842
 843 failed_removal:
 844         printk(KERN_INFO "memory offlining %lx to %lx failed\n",
 845                 start_pfn, end_pfn);
 846         memory_notify(MEM_CANCEL_OFFLINE, &arg);
 847         /* pushback to free area */
 848         undo_isolate_page_range(start_pfn, end_pfn);
 849
 850         return ret;
 851 }
 852 #else
 853 int remove_memory(u64 start, u64 size)
 854 {
 855         return -EINVAL;
 856 }
 857 EXPORT_SYMBOL_GPL(remove_memory);
 858 #endif /* CONFIG_MEMORY_HOTREMOVE */