mm/memory_hotplug.c

   1 /*
   2  *  linux/mm/memory_hotplug.c
   3  *
   4  *  Copyright (C)
   5  */
   6
   7 #include <linux/stddef.h>
   8 #include <linux/mm.h>
   9 #include <linux/swap.h>
  10 #include <linux/interrupt.h>
  11 #include <linux/pagemap.h>
  12 #include <linux/bootmem.h>
  13 #include <linux/compiler.h>
  14 #include <linux/module.h>
  15 #include <linux/pagevec.h>
  16 #include <linux/writeback.h>
  17 #include <linux/slab.h>
  18 #include <linux/sysctl.h>
  19 #include <linux/cpu.h>
  20 #include <linux/memory.h>
  21 #include <linux/memory_hotplug.h>
  22 #include <linux/highmem.h>
  23 #include <linux/vmalloc.h>
  24 #include <linux/ioport.h>
  25 #include <linux/delay.h>
  26 #include <linux/migrate.h>
  27 #include <linux/page-isolation.h>
  28 #include <linux/pfn.h>
  29 #include <linux/suspend.h>
  30 #include <linux/mm_inline.h>
  31
  32 #include <asm/tlbflush.h>
  33
  34 #include "internal.h"
  35
  36 /* add this memory to iomem resource */
  37 static struct resource *register_memory_resource(u64 start, u64 size)
  38 {
  39         struct resource *res;
  40         res = kzalloc(sizeof(struct resource), GFP_KERNEL);
  41         BUG_ON(!res);
  42
  43         res->name = "System RAM";
  44         res->start = start;
  45         res->end = start + size - 1;
  46         res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
  47         if (request_resource(&iomem_resource, res) < 0) {
  48                 printk("System RAM resource %llx - %llx cannot be added\n",
  49                 (unsigned long long)res->start, (unsigned long long)res->end);
  50                 kfree(res);
  51                 res = NULL;
  52         }
  53         return res;
  54 }
  55
  56 static void release_memory_resource(struct resource *res)
  57 {
  58         if (!res)
  59                 return;
  60         release_resource(res);
  61         kfree(res);
  62         return;
  63 }
  64
  65 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
  66 #ifndef CONFIG_SPARSEMEM_VMEMMAP
  67 static void get_page_bootmem(unsigned long info,  struct page *page, int type)
  68 {
  69         atomic_set(&page->_mapcount, type);
  70         SetPagePrivate(page);
  71         set_page_private(page, info);
  72         atomic_inc(&page->_count);
  73 }
  74
  75 /* reference to __meminit __free_pages_bootmem is valid
  76  * so use __ref to tell modpost not to generate a warning */
  77 void __ref put_page_bootmem(struct page *page)
  78 {
  79         int type;
  80
  81         type = atomic_read(&page->_mapcount);
  82         BUG_ON(type >= -1);
  83
  84         if (atomic_dec_return(&page->_count) == 1) {
  85                 ClearPagePrivate(page);
  86                 set_page_private(page, 0);
  87                 reset_page_mapcount(page);
  88                 __free_pages_bootmem(page, 0);
  89         }
  90
  91 }
  92
  93 static void register_page_bootmem_info_section(unsigned long start_pfn)
  94 {
  95         unsigned long *usemap, mapsize, section_nr, i;
  96         struct mem_section *ms;
  97         struct page *page, *memmap;
  98
  99         if (!pfn_valid(start_pfn))
 100                 return;
 101
 102         section_nr = pfn_to_section_nr(start_pfn);
 103         ms = __nr_to_section(section_nr);
 104
 105         /* Get section's memmap address */
 106         memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
 107
 108         /*
 109          * Get page for the memmap's phys address
 110          * XXX: need more consideration for sparse_vmemmap...
 111          */
 112         page = virt_to_page(memmap);
 113         mapsize = sizeof(struct page) * PAGES_PER_SECTION;
 114         mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
 115
 116         /* remember memmap's page */
 117         for (i = 0; i < mapsize; i++, page++)
 118                 get_page_bootmem(section_nr, page, SECTION_INFO);
 119
 120         usemap = __nr_to_section(section_nr)->pageblock_flags;
 121         page = virt_to_page(usemap);
 122
 123         mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
 124
 125         for (i = 0; i < mapsize; i++, page++)
 126                 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
 127
 128 }
 129
 130 void register_page_bootmem_info_node(struct pglist_data *pgdat)
 131 {
 132         unsigned long i, pfn, end_pfn, nr_pages;
 133         int node = pgdat->node_id;
 134         struct page *page;
 135         struct zone *zone;
 136
 137         nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
 138         page = virt_to_page(pgdat);
 139
 140         for (i = 0; i < nr_pages; i++, page++)
 141                 get_page_bootmem(node, page, NODE_INFO);
 142
 143         zone = &pgdat->node_zones[0];
 144         for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
 145                 if (zone->wait_table) {
 146                         nr_pages = zone->wait_table_hash_nr_entries
 147                                 * sizeof(wait_queue_head_t);
 148                         nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
 149                         page = virt_to_page(zone->wait_table);
 150
 151                         for (i = 0; i < nr_pages; i++, page++)
 152                                 get_page_bootmem(node, page, NODE_INFO);
 153                 }
 154         }
 155
 156         pfn = pgdat->node_start_pfn;
 157         end_pfn = pfn + pgdat->node_spanned_pages;
 158
 159         /* register_section info */
 160         for (; pfn < end_pfn; pfn += PAGES_PER_SECTION)
 161                 register_page_bootmem_info_section(pfn);
 162
 163 }
 164 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
 165
 166 static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
 167                            unsigned long end_pfn)
 168 {
 169         unsigned long old_zone_end_pfn;
 170
 171         zone_span_writelock(zone);
 172
 173         old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
 174         if (start_pfn < zone->zone_start_pfn)
 175                 zone->zone_start_pfn = start_pfn;
 176
 177         zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
 178                                 zone->zone_start_pfn;
 179
 180         zone_span_writeunlock(zone);
 181 }
 182
 183 static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
 184                             unsigned long end_pfn)
 185 {
 186         unsigned long old_pgdat_end_pfn =
 187                 pgdat->node_start_pfn + pgdat->node_spanned_pages;
 188
 189         if (start_pfn < pgdat->node_start_pfn)
 190                 pgdat->node_start_pfn = start_pfn;
 191
 192         pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
 193                                         pgdat->node_start_pfn;
 194 }
 195
 196 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
 197 {
 198         struct pglist_data *pgdat = zone->zone_pgdat;
 199         int nr_pages = PAGES_PER_SECTION;
 200         int nid = pgdat->node_id;
 201         int zone_type;
 202         unsigned long flags;
 203
 204         zone_type = zone - pgdat->node_zones;
 205         if (!zone->wait_table) {
 206                 int ret;
 207
 208                 ret = init_currently_empty_zone(zone, phys_start_pfn,
 209                                                 nr_pages, MEMMAP_HOTPLUG);
 210                 if (ret)
 211                         return ret;
 212         }
 213         pgdat_resize_lock(zone->zone_pgdat, &flags);
 214         grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
 215         grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
 216                         phys_start_pfn + nr_pages);
 217         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 218         memmap_init_zone(nr_pages, nid, zone_type,
 219                          phys_start_pfn, MEMMAP_HOTPLUG);
 220         return 0;
 221 }
 222
 223 static int __meminit __add_section(int nid, struct zone *zone,
 224                                         unsigned long phys_start_pfn)
 225 {
 226         int nr_pages = PAGES_PER_SECTION;
 227         int ret;
 228
 229         if (pfn_valid(phys_start_pfn))
 230                 return -EEXIST;
 231
 232         ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
 233
 234         if (ret < 0)
 235                 return ret;
 236
 237         ret = __add_zone(zone, phys_start_pfn);
 238
 239         if (ret < 0)
 240                 return ret;
 241
 242         return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
 243 }
 244
 245 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 246 static int __remove_section(struct zone *zone, struct mem_section *ms)
 247 {
 248         /*
 249          * XXX: Freeing memmap with vmemmap is not implement yet.
 250          *      This should be removed later.
 251          */
 252         return -EBUSY;
 253 }
 254 #else
 255 static int __remove_section(struct zone *zone, struct mem_section *ms)
 256 {
 257         unsigned long flags;
 258         struct pglist_data *pgdat = zone->zone_pgdat;
 259         int ret = -EINVAL;
 260
 261         if (!valid_section(ms))
 262                 return ret;
 263
 264         ret = unregister_memory_section(ms);
 265         if (ret)
 266                 return ret;
 267
 268         pgdat_resize_lock(pgdat, &flags);
 269         sparse_remove_one_section(zone, ms);
 270         pgdat_resize_unlock(pgdat, &flags);
 271         return 0;
 272 }
 273 #endif
 274
 275 /*
 276  * Reasonably generic function for adding memory.  It is
 277  * expected that archs that support memory hotplug will
 278  * call this function after deciding the zone to which to
 279  * add the new pages.
 280  */
 281 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
 282                         unsigned long nr_pages)
 283 {
 284         unsigned long i;
 285         int err = 0;
 286         int start_sec, end_sec;
 287         /* during initialize mem_map, align hot-added range to section */
 288         start_sec = pfn_to_section_nr(phys_start_pfn);
 289         end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
 290
 291         for (i = start_sec; i <= end_sec; i++) {
 292                 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
 293
 294                 /*
 295                  * EEXIST is finally dealt with by ioresource collision
 296                  * check. see add_memory() => register_memory_resource()
 297                  * Warning will be printed if there is collision.
 298                  */
 299                 if (err && (err != -EEXIST))
 300                         break;
 301                 err = 0;
 302         }
 303
 304         return err;
 305 }
 306 EXPORT_SYMBOL_GPL(__add_pages);
 307
 308 /**
 309  * __remove_pages() - remove sections of pages from a zone
 310  * @zone: zone from which pages need to be removed
 311  * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
 312  * @nr_pages: number of pages to remove (must be multiple of section size)
 313  *
 314  * Generic helper function to remove section mappings and sysfs entries
 315  * for the section of the memory we are removing. Caller needs to make
 316  * sure that pages are marked reserved and zones are adjust properly by
 317  * calling offline_pages().
 318  */
 319 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
 320                  unsigned long nr_pages)
 321 {
 322         unsigned long i, ret = 0;
 323         int sections_to_remove;
 324
 325         /*
 326          * We can only remove entire sections
 327          */
 328         BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
 329         BUG_ON(nr_pages % PAGES_PER_SECTION);
 330
 331         sections_to_remove = nr_pages / PAGES_PER_SECTION;
 332         for (i = 0; i < sections_to_remove; i++) {
 333                 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
 334                 release_mem_region(pfn << PAGE_SHIFT,
 335                                    PAGES_PER_SECTION << PAGE_SHIFT);
 336                 ret = __remove_section(zone, __pfn_to_section(pfn));
 337                 if (ret)
 338                         break;
 339         }
 340         return ret;
 341 }
 342 EXPORT_SYMBOL_GPL(__remove_pages);
 343
 344 void online_page(struct page *page)
 345 {
 346         unsigned long pfn = page_to_pfn(page);
 347
 348         totalram_pages++;
 349         if (pfn >= num_physpages)
 350                 num_physpages = pfn + 1;
 351
 352 #ifdef CONFIG_HIGHMEM
 353         if (PageHighMem(page))
 354                 totalhigh_pages++;
 355 #endif
 356
 357 #ifdef CONFIG_FLATMEM
 358         max_mapnr = max(page_to_pfn(page), max_mapnr);
 359 #endif
 360
 361         ClearPageReserved(page);
 362         init_page_count(page);
 363         __free_page(page);
 364 }
 365
 366 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
 367                         void *arg)
 368 {
 369         unsigned long i;
 370         unsigned long onlined_pages = *(unsigned long *)arg;
 371         struct page *page;
 372         if (PageReserved(pfn_to_page(start_pfn)))
 373                 for (i = 0; i < nr_pages; i++) {
 374                         page = pfn_to_page(start_pfn + i);
 375                         online_page(page);
 376                         onlined_pages++;
 377                 }
 378         *(unsigned long *)arg = onlined_pages;
 379         return 0;
 380 }
 381
 382
 383 int online_pages(unsigned long pfn, unsigned long nr_pages)
 384 {
 385         unsigned long onlined_pages = 0;
 386         struct zone *zone;
 387         int need_zonelists_rebuild = 0;
 388         int nid;
 389         int ret;
 390         struct memory_notify arg;
 391
 392         arg.start_pfn = pfn;
 393         arg.nr_pages = nr_pages;
 394         arg.status_change_nid = -1;
 395
 396         nid = page_to_nid(pfn_to_page(pfn));
 397         if (node_present_pages(nid) == 0)
 398                 arg.status_change_nid = nid;
 399
 400         ret = memory_notify(MEM_GOING_ONLINE, &arg);
 401         ret = notifier_to_errno(ret);
 402         if (ret) {
 403                 memory_notify(MEM_CANCEL_ONLINE, &arg);
 404                 return ret;
 405         }
 406         /*
 407          * This doesn't need a lock to do pfn_to_page().
 408          * The section can't be removed here because of the
 409          * memory_block->state_mutex.
 410          */
 411         zone = page_zone(pfn_to_page(pfn));
 412         /*
 413          * If this zone is not populated, then it is not in zonelist.
 414          * This means the page allocator ignores this zone.
 415          * So, zonelist must be updated after online.
 416          */
 417         if (!populated_zone(zone))
 418                 need_zonelists_rebuild = 1;
 419
 420         ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
 421                 online_pages_range);
 422         if (ret) {
 423                 printk(KERN_DEBUG "online_pages %lx at %lx failed\n",
 424                         nr_pages, pfn);
 425                 memory_notify(MEM_CANCEL_ONLINE, &arg);
 426                 return ret;
 427         }
 428
 429         zone->present_pages += onlined_pages;
 430         zone->zone_pgdat->node_present_pages += onlined_pages;
 431
 432         zone_pcp_update(zone);
 433         setup_per_zone_wmarks();
 434         calculate_zone_inactive_ratio(zone);
 435         if (onlined_pages) {
 436                 kswapd_run(zone_to_nid(zone));
 437                 node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
 438         }
 439
 440         if (need_zonelists_rebuild)
 441                 build_all_zonelists();
 442         else
 443                 vm_total_pages = nr_free_pagecache_pages();
 444
 445         writeback_set_ratelimit();
 446
 447         if (onlined_pages)
 448                 memory_notify(MEM_ONLINE, &arg);
 449
 450         return 0;
 451 }
 452 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
 453
 454 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
 455 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
 456 {
 457         struct pglist_data *pgdat;
 458         unsigned long zones_size[MAX_NR_ZONES] = {0};
 459         unsigned long zholes_size[MAX_NR_ZONES] = {0};
 460         unsigned long start_pfn = start >> PAGE_SHIFT;
 461
 462         pgdat = arch_alloc_nodedata(nid);
 463         if (!pgdat)
 464                 return NULL;
 465
 466         arch_refresh_nodedata(nid, pgdat);
 467
 468         /* we can use NODE_DATA(nid) from here */
 469
 470         /* init node's zones as empty zones, we don't have any present pages.*/
 471         free_area_init_node(nid, zones_size, start_pfn, zholes_size);
 472
 473         return pgdat;
 474 }
 475
 476 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
 477 {
 478         arch_refresh_nodedata(nid, NULL);
 479         arch_free_nodedata(pgdat);
 480         return;
 481 }
 482
 483
 484 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
 485 int __ref add_memory(int nid, u64 start, u64 size)
 486 {
 487         pg_data_t *pgdat = NULL;
 488         int new_pgdat = 0;
 489         struct resource *res;
 490         int ret;
 491
 492         lock_system_sleep();
 493
 494         res = register_memory_resource(start, size);
 495         ret = -EEXIST;
 496         if (!res)
 497                 goto out;
 498
 499         if (!node_online(nid)) {
 500                 pgdat = hotadd_new_pgdat(nid, start);
 501                 ret = -ENOMEM;
 502                 if (!pgdat)
 503                         goto out;
 504                 new_pgdat = 1;
 505         }
 506
 507         /* call arch's memory hotadd */
 508         ret = arch_add_memory(nid, start, size);
 509
 510         if (ret < 0)
 511                 goto error;
 512
 513         /* we online node here. we can't roll back from here. */
 514         node_set_online(nid);
 515
 516         if (new_pgdat) {
 517                 ret = register_one_node(nid);
 518                 /*
 519                  * If sysfs file of new node can't create, cpu on the node
 520                  * can't be hot-added. There is no rollback way now.
 521                  * So, check by BUG_ON() to catch it reluctantly..
 522                  */
 523                 BUG_ON(ret);
 524         }
 525
 526         goto out;
 527
 528 error:
 529         /* rollback pgdat allocation and others */
 530         if (new_pgdat)
 531                 rollback_node_hotadd(nid, pgdat);
 532         if (res)
 533                 release_memory_resource(res);
 534
 535 out:
 536         unlock_system_sleep();
 537         return ret;
 538 }
 539 EXPORT_SYMBOL_GPL(add_memory);
 540
 541 #ifdef CONFIG_MEMORY_HOTREMOVE
 542 /*
 543  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
 544  * set and the size of the free page is given by page_order(). Using this,
 545  * the function determines if the pageblock contains only free pages.
 546  * Due to buddy contraints, a free page at least the size of a pageblock will
 547  * be located at the start of the pageblock
 548  */
 549 static inline int pageblock_free(struct page *page)
 550 {
 551         return PageBuddy(page) && page_order(page) >= pageblock_order;
 552 }
 553
 554 /* Return the start of the next active pageblock after a given page */
 555 static struct page *next_active_pageblock(struct page *page)
 556 {
 557         int pageblocks_stride;
 558
 559         /* Ensure the starting page is pageblock-aligned */
 560         BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
 561
 562         /* Move forward by at least 1 * pageblock_nr_pages */
 563         pageblocks_stride = 1;
 564
 565         /* If the entire pageblock is free, move to the end of free page */
 566         if (pageblock_free(page))
 567                 pageblocks_stride += page_order(page) - pageblock_order;
 568
 569         return page + (pageblocks_stride * pageblock_nr_pages);
 570 }
 571
 572 /* Checks if this range of memory is likely to be hot-removable. */
 573 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
 574 {
 575         int type;
 576         struct page *page = pfn_to_page(start_pfn);
 577         struct page *end_page = page + nr_pages;
 578
 579         /* Check the starting page of each pageblock within the range */
 580         for (; page < end_page; page = next_active_pageblock(page)) {
 581                 type = get_pageblock_migratetype(page);
 582
 583                 /*
 584                  * A pageblock containing MOVABLE or free pages is considered
 585                  * removable
 586                  */
 587                 if (type != MIGRATE_MOVABLE && !pageblock_free(page))
 588                         return 0;
 589
 590                 /*
 591                  * A pageblock starting with a PageReserved page is not
 592                  * considered removable.
 593                  */
 594                 if (PageReserved(page))
 595                         return 0;
 596         }
 597
 598         /* All pageblocks in the memory block are likely to be hot-removable */
 599         return 1;
 600 }
 601
 602 /*
 603  * Confirm all pages in a range [start, end) is belongs to the same zone.
 604  */
 605 static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
 606 {
 607         unsigned long pfn;
 608         struct zone *zone = NULL;
 609         struct page *page;
 610         int i;
 611         for (pfn = start_pfn;
 612              pfn < end_pfn;
 613              pfn += MAX_ORDER_NR_PAGES) {
 614                 i = 0;
 615                 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
 616                 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
 617                         i++;
 618                 if (i == MAX_ORDER_NR_PAGES)
 619                         continue;
 620                 page = pfn_to_page(pfn + i);
 621                 if (zone && page_zone(page) != zone)
 622                         return 0;
 623                 zone = page_zone(page);
 624         }
 625         return 1;
 626 }
 627
 628 /*
 629  * Scanning pfn is much easier than scanning lru list.
 630  * Scan pfn from start to end and Find LRU page.
 631  */
 632 int scan_lru_pages(unsigned long start, unsigned long end)
 633 {
 634         unsigned long pfn;
 635         struct page *page;
 636         for (pfn = start; pfn < end; pfn++) {
 637                 if (pfn_valid(pfn)) {
 638                         page = pfn_to_page(pfn);
 639                         if (PageLRU(page))
 640                                 return pfn;
 641                 }
 642         }
 643         return 0;
 644 }
 645
 646 static struct page *
 647 hotremove_migrate_alloc(struct page *page, unsigned long private, int **x)
 648 {
 649         /* This should be improooooved!! */
 650         return alloc_page(GFP_HIGHUSER_MOVABLE);
 651 }
 652
 653 #define NR_OFFLINE_AT_ONCE_PAGES        (256)
 654 static int
 655 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
 656 {
 657         unsigned long pfn;
 658         struct page *page;
 659         int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
 660         int not_managed = 0;
 661         int ret = 0;
 662         LIST_HEAD(source);
 663
 664         for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
 665                 if (!pfn_valid(pfn))
 666                         continue;
 667                 page = pfn_to_page(pfn);
 668                 if (!page_count(page))
 669                         continue;
 670                 /*
 671                  * We can skip free pages. And we can only deal with pages on
 672                  * LRU.
 673                  */
 674                 ret = isolate_lru_page(page);
 675                 if (!ret) { /* Success */
 676                         list_add_tail(&page->lru, &source);
 677                         move_pages--;
 678                         inc_zone_page_state(page, NR_ISOLATED_ANON +
 679                                             page_is_file_cache(page));
 680
 681                 } else {
 682                         /* Becasue we don't have big zone->lock. we should
 683                            check this again here. */
 684                         if (page_count(page))
 685                                 not_managed++;
 686 #ifdef CONFIG_DEBUG_VM
 687                         printk(KERN_INFO "removing from LRU failed"
 688                                          " %lx/%d/%lx\n",
 689                                 pfn, page_count(page), page->flags);
 690 #endif
 691                 }
 692         }
 693         ret = -EBUSY;
 694         if (not_managed) {
 695                 if (!list_empty(&source))
 696                         putback_lru_pages(&source);
 697                 goto out;
 698         }
 699         ret = 0;
 700         if (list_empty(&source))
 701                 goto out;
 702         /* this function returns # of failed pages */
 703         ret = migrate_pages(&source, hotremove_migrate_alloc, 0, 1);
 704
 705 out:
 706         return ret;
 707 }
 708
 709 /*
 710  * remove from free_area[] and mark all as Reserved.
 711  */
 712 static int
 713 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
 714                         void *data)
 715 {
 716         __offline_isolated_pages(start, start + nr_pages);
 717         return 0;
 718 }
 719
 720 static void
 721 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
 722 {
 723         walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
 724                                 offline_isolated_pages_cb);
 725 }
 726
 727 /*
 728  * Check all pages in range, recoreded as memory resource, are isolated.
 729  */
 730 static int
 731 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
 732                         void *data)
 733 {
 734         int ret;
 735         long offlined = *(long *)data;
 736         ret = test_pages_isolated(start_pfn, start_pfn + nr_pages);
 737         offlined = nr_pages;
 738         if (!ret)
 739                 *(long *)data += offlined;
 740         return ret;
 741 }
 742
 743 static long
 744 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
 745 {
 746         long offlined = 0;
 747         int ret;
 748
 749         ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
 750                         check_pages_isolated_cb);
 751         if (ret < 0)
 752                 offlined = (long)ret;
 753         return offlined;
 754 }
 755
 756 static int offline_pages(unsigned long start_pfn,
 757                   unsigned long end_pfn, unsigned long timeout)
 758 {
 759         unsigned long pfn, nr_pages, expire;
 760         long offlined_pages;
 761         int ret, drain, retry_max, node;
 762         struct zone *zone;
 763         struct memory_notify arg;
 764
 765         BUG_ON(start_pfn >= end_pfn);
 766         /* at least, alignment against pageblock is necessary */
 767         if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
 768                 return -EINVAL;
 769         if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
 770                 return -EINVAL;
 771         /* This makes hotplug much easier...and readable.
 772            we assume this for now. .*/
 773         if (!test_pages_in_a_zone(start_pfn, end_pfn))
 774                 return -EINVAL;
 775
 776         lock_system_sleep();
 777
 778         zone = page_zone(pfn_to_page(start_pfn));
 779         node = zone_to_nid(zone);
 780         nr_pages = end_pfn - start_pfn;
 781
 782         /* set above range as isolated */
 783         ret = start_isolate_page_range(start_pfn, end_pfn);
 784         if (ret)
 785                 goto out;
 786
 787         arg.start_pfn = start_pfn;
 788         arg.nr_pages = nr_pages;
 789         arg.status_change_nid = -1;
 790         if (nr_pages >= node_present_pages(node))
 791                 arg.status_change_nid = node;
 792
 793         ret = memory_notify(MEM_GOING_OFFLINE, &arg);
 794         ret = notifier_to_errno(ret);
 795         if (ret)
 796                 goto failed_removal;
 797
 798         pfn = start_pfn;
 799         expire = jiffies + timeout;
 800         drain = 0;
 801         retry_max = 5;
 802 repeat:
 803         /* start memory hot removal */
 804         ret = -EAGAIN;
 805         if (time_after(jiffies, expire))
 806                 goto failed_removal;
 807         ret = -EINTR;
 808         if (signal_pending(current))
 809                 goto failed_removal;
 810         ret = 0;
 811         if (drain) {
 812                 lru_add_drain_all();
 813                 flush_scheduled_work();
 814                 cond_resched();
 815                 drain_all_pages();
 816         }
 817
 818         pfn = scan_lru_pages(start_pfn, end_pfn);
 819         if (pfn) { /* We have page on LRU */
 820                 ret = do_migrate_range(pfn, end_pfn);
 821                 if (!ret) {
 822                         drain = 1;
 823                         goto repeat;
 824                 } else {
 825                         if (ret < 0)
 826                                 if (--retry_max == 0)
 827                                         goto failed_removal;
 828                         yield();
 829                         drain = 1;
 830                         goto repeat;
 831                 }
 832         }
 833         /* drain all zone's lru pagevec, this is asyncronous... */
 834         lru_add_drain_all();
 835         flush_scheduled_work();
 836         yield();
 837         /* drain pcp pages , this is synchrouns. */
 838         drain_all_pages();
 839         /* check again */
 840         offlined_pages = check_pages_isolated(start_pfn, end_pfn);
 841         if (offlined_pages < 0) {
 842                 ret = -EBUSY;
 843                 goto failed_removal;
 844         }
 845         printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
 846         /* Ok, all of our target is islaoted.
 847            We cannot do rollback at this point. */
 848         offline_isolated_pages(start_pfn, end_pfn);
 849         /* reset pagetype flags and makes migrate type to be MOVABLE */
 850         undo_isolate_page_range(start_pfn, end_pfn);
 851         /* removal success */
 852         zone->present_pages -= offlined_pages;
 853         zone->zone_pgdat->node_present_pages -= offlined_pages;
 854         totalram_pages -= offlined_pages;
 855
 856         setup_per_zone_wmarks();
 857         calculate_zone_inactive_ratio(zone);
 858         if (!node_present_pages(node)) {
 859                 node_clear_state(node, N_HIGH_MEMORY);
 860                 kswapd_stop(node);
 861         }
 862
 863         vm_total_pages = nr_free_pagecache_pages();
 864         writeback_set_ratelimit();
 865
 866         memory_notify(MEM_OFFLINE, &arg);
 867         unlock_system_sleep();
 868         return 0;
 869
 870 failed_removal:
 871         printk(KERN_INFO "memory offlining %lx to %lx failed\n",
 872                 start_pfn, end_pfn);
 873         memory_notify(MEM_CANCEL_OFFLINE, &arg);
 874         /* pushback to free area */
 875         undo_isolate_page_range(start_pfn, end_pfn);
 876
 877 out:
 878         unlock_system_sleep();
 879         return ret;
 880 }
 881
 882 int remove_memory(u64 start, u64 size)
 883 {
 884         unsigned long start_pfn, end_pfn;
 885
 886         start_pfn = PFN_DOWN(start);
 887         end_pfn = start_pfn + PFN_DOWN(size);
 888         return offline_pages(start_pfn, end_pfn, 120 * HZ);
 889 }
 890 #else
 891 int remove_memory(u64 start, u64 size)
 892 {
 893         return -EINVAL;
 894 }
 895 #endif /* CONFIG_MEMORY_HOTREMOVE */
 896 EXPORT_SYMBOL_GPL(remove_memory);