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 #include <linux/firmware-map.h>
  32
  33 #include <asm/tlbflush.h>
  34
  35 #include "internal.h"
  36
  37 /* add this memory to iomem resource */
  38 static struct resource *register_memory_resource(u64 start, u64 size)
  39 {
  40         struct resource *res;
  41         res = kzalloc(sizeof(struct resource), GFP_KERNEL);
  42         BUG_ON(!res);
  43
  44         res->name = "System RAM";
  45         res->start = start;
  46         res->end = start + size - 1;
  47         res->flags = IORESOURCE_MEM | IORESOURCE_BUSY;
  48         if (request_resource(&iomem_resource, res) < 0) {
  49                 printk("System RAM resource %llx - %llx cannot be added\n",
  50                 (unsigned long long)res->start, (unsigned long long)res->end);
  51                 kfree(res);
  52                 res = NULL;
  53         }
  54         return res;
  55 }
  56
  57 static void release_memory_resource(struct resource *res)
  58 {
  59         if (!res)
  60                 return;
  61         release_resource(res);
  62         kfree(res);
  63         return;
  64 }
  65
  66 #ifdef CONFIG_MEMORY_HOTPLUG_SPARSE
  67 #ifndef CONFIG_SPARSEMEM_VMEMMAP
  68 static void get_page_bootmem(unsigned long info,  struct page *page, int type)
  69 {
  70         atomic_set(&page->_mapcount, type);
  71         SetPagePrivate(page);
  72         set_page_private(page, info);
  73         atomic_inc(&page->_count);
  74 }
  75
  76 /* reference to __meminit __free_pages_bootmem is valid
  77  * so use __ref to tell modpost not to generate a warning */
  78 void __ref put_page_bootmem(struct page *page)
  79 {
  80         int type;
  81
  82         type = atomic_read(&page->_mapcount);
  83         BUG_ON(type >= -1);
  84
  85         if (atomic_dec_return(&page->_count) == 1) {
  86                 ClearPagePrivate(page);
  87                 set_page_private(page, 0);
  88                 reset_page_mapcount(page);
  89                 __free_pages_bootmem(page, 0);
  90         }
  91
  92 }
  93
  94 static void register_page_bootmem_info_section(unsigned long start_pfn)
  95 {
  96         unsigned long *usemap, mapsize, section_nr, i;
  97         struct mem_section *ms;
  98         struct page *page, *memmap;
  99
 100         if (!pfn_valid(start_pfn))
 101                 return;
 102
 103         section_nr = pfn_to_section_nr(start_pfn);
 104         ms = __nr_to_section(section_nr);
 105
 106         /* Get section's memmap address */
 107         memmap = sparse_decode_mem_map(ms->section_mem_map, section_nr);
 108
 109         /*
 110          * Get page for the memmap's phys address
 111          * XXX: need more consideration for sparse_vmemmap...
 112          */
 113         page = virt_to_page(memmap);
 114         mapsize = sizeof(struct page) * PAGES_PER_SECTION;
 115         mapsize = PAGE_ALIGN(mapsize) >> PAGE_SHIFT;
 116
 117         /* remember memmap's page */
 118         for (i = 0; i < mapsize; i++, page++)
 119                 get_page_bootmem(section_nr, page, SECTION_INFO);
 120
 121         usemap = __nr_to_section(section_nr)->pageblock_flags;
 122         page = virt_to_page(usemap);
 123
 124         mapsize = PAGE_ALIGN(usemap_size()) >> PAGE_SHIFT;
 125
 126         for (i = 0; i < mapsize; i++, page++)
 127                 get_page_bootmem(section_nr, page, MIX_SECTION_INFO);
 128
 129 }
 130
 131 void register_page_bootmem_info_node(struct pglist_data *pgdat)
 132 {
 133         unsigned long i, pfn, end_pfn, nr_pages;
 134         int node = pgdat->node_id;
 135         struct page *page;
 136         struct zone *zone;
 137
 138         nr_pages = PAGE_ALIGN(sizeof(struct pglist_data)) >> PAGE_SHIFT;
 139         page = virt_to_page(pgdat);
 140
 141         for (i = 0; i < nr_pages; i++, page++)
 142                 get_page_bootmem(node, page, NODE_INFO);
 143
 144         zone = &pgdat->node_zones[0];
 145         for (; zone < pgdat->node_zones + MAX_NR_ZONES - 1; zone++) {
 146                 if (zone->wait_table) {
 147                         nr_pages = zone->wait_table_hash_nr_entries
 148                                 * sizeof(wait_queue_head_t);
 149                         nr_pages = PAGE_ALIGN(nr_pages) >> PAGE_SHIFT;
 150                         page = virt_to_page(zone->wait_table);
 151
 152                         for (i = 0; i < nr_pages; i++, page++)
 153                                 get_page_bootmem(node, page, NODE_INFO);
 154                 }
 155         }
 156
 157         pfn = pgdat->node_start_pfn;
 158         end_pfn = pfn + pgdat->node_spanned_pages;
 159
 160         /* register_section info */
 161         for (; pfn < end_pfn; pfn += PAGES_PER_SECTION)
 162                 register_page_bootmem_info_section(pfn);
 163
 164 }
 165 #endif /* !CONFIG_SPARSEMEM_VMEMMAP */
 166
 167 static void grow_zone_span(struct zone *zone, unsigned long start_pfn,
 168                            unsigned long end_pfn)
 169 {
 170         unsigned long old_zone_end_pfn;
 171
 172         zone_span_writelock(zone);
 173
 174         old_zone_end_pfn = zone->zone_start_pfn + zone->spanned_pages;
 175         if (start_pfn < zone->zone_start_pfn)
 176                 zone->zone_start_pfn = start_pfn;
 177
 178         zone->spanned_pages = max(old_zone_end_pfn, end_pfn) -
 179                                 zone->zone_start_pfn;
 180
 181         zone_span_writeunlock(zone);
 182 }
 183
 184 static void grow_pgdat_span(struct pglist_data *pgdat, unsigned long start_pfn,
 185                             unsigned long end_pfn)
 186 {
 187         unsigned long old_pgdat_end_pfn =
 188                 pgdat->node_start_pfn + pgdat->node_spanned_pages;
 189
 190         if (start_pfn < pgdat->node_start_pfn)
 191                 pgdat->node_start_pfn = start_pfn;
 192
 193         pgdat->node_spanned_pages = max(old_pgdat_end_pfn, end_pfn) -
 194                                         pgdat->node_start_pfn;
 195 }
 196
 197 static int __meminit __add_zone(struct zone *zone, unsigned long phys_start_pfn)
 198 {
 199         struct pglist_data *pgdat = zone->zone_pgdat;
 200         int nr_pages = PAGES_PER_SECTION;
 201         int nid = pgdat->node_id;
 202         int zone_type;
 203         unsigned long flags;
 204
 205         zone_type = zone - pgdat->node_zones;
 206         if (!zone->wait_table) {
 207                 int ret;
 208
 209                 ret = init_currently_empty_zone(zone, phys_start_pfn,
 210                                                 nr_pages, MEMMAP_HOTPLUG);
 211                 if (ret)
 212                         return ret;
 213         }
 214         pgdat_resize_lock(zone->zone_pgdat, &flags);
 215         grow_zone_span(zone, phys_start_pfn, phys_start_pfn + nr_pages);
 216         grow_pgdat_span(zone->zone_pgdat, phys_start_pfn,
 217                         phys_start_pfn + nr_pages);
 218         pgdat_resize_unlock(zone->zone_pgdat, &flags);
 219         memmap_init_zone(nr_pages, nid, zone_type,
 220                          phys_start_pfn, MEMMAP_HOTPLUG);
 221         return 0;
 222 }
 223
 224 static int __meminit __add_section(int nid, struct zone *zone,
 225                                         unsigned long phys_start_pfn)
 226 {
 227         int nr_pages = PAGES_PER_SECTION;
 228         int ret;
 229
 230         if (pfn_valid(phys_start_pfn))
 231                 return -EEXIST;
 232
 233         ret = sparse_add_one_section(zone, phys_start_pfn, nr_pages);
 234
 235         if (ret < 0)
 236                 return ret;
 237
 238         ret = __add_zone(zone, phys_start_pfn);
 239
 240         if (ret < 0)
 241                 return ret;
 242
 243         return register_new_memory(nid, __pfn_to_section(phys_start_pfn));
 244 }
 245
 246 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 247 static int __remove_section(struct zone *zone, struct mem_section *ms)
 248 {
 249         /*
 250          * XXX: Freeing memmap with vmemmap is not implement yet.
 251          *      This should be removed later.
 252          */
 253         return -EBUSY;
 254 }
 255 #else
 256 static int __remove_section(struct zone *zone, struct mem_section *ms)
 257 {
 258         unsigned long flags;
 259         struct pglist_data *pgdat = zone->zone_pgdat;
 260         int ret = -EINVAL;
 261
 262         if (!valid_section(ms))
 263                 return ret;
 264
 265         ret = unregister_memory_section(ms);
 266         if (ret)
 267                 return ret;
 268
 269         pgdat_resize_lock(pgdat, &flags);
 270         sparse_remove_one_section(zone, ms);
 271         pgdat_resize_unlock(pgdat, &flags);
 272         return 0;
 273 }
 274 #endif
 275
 276 /*
 277  * Reasonably generic function for adding memory.  It is
 278  * expected that archs that support memory hotplug will
 279  * call this function after deciding the zone to which to
 280  * add the new pages.
 281  */
 282 int __ref __add_pages(int nid, struct zone *zone, unsigned long phys_start_pfn,
 283                         unsigned long nr_pages)
 284 {
 285         unsigned long i;
 286         int err = 0;
 287         int start_sec, end_sec;
 288         /* during initialize mem_map, align hot-added range to section */
 289         start_sec = pfn_to_section_nr(phys_start_pfn);
 290         end_sec = pfn_to_section_nr(phys_start_pfn + nr_pages - 1);
 291
 292         for (i = start_sec; i <= end_sec; i++) {
 293                 err = __add_section(nid, zone, i << PFN_SECTION_SHIFT);
 294
 295                 /*
 296                  * EEXIST is finally dealt with by ioresource collision
 297                  * check. see add_memory() => register_memory_resource()
 298                  * Warning will be printed if there is collision.
 299                  */
 300                 if (err && (err != -EEXIST))
 301                         break;
 302                 err = 0;
 303         }
 304
 305         return err;
 306 }
 307 EXPORT_SYMBOL_GPL(__add_pages);
 308
 309 /**
 310  * __remove_pages() - remove sections of pages from a zone
 311  * @zone: zone from which pages need to be removed
 312  * @phys_start_pfn: starting pageframe (must be aligned to start of a section)
 313  * @nr_pages: number of pages to remove (must be multiple of section size)
 314  *
 315  * Generic helper function to remove section mappings and sysfs entries
 316  * for the section of the memory we are removing. Caller needs to make
 317  * sure that pages are marked reserved and zones are adjust properly by
 318  * calling offline_pages().
 319  */
 320 int __remove_pages(struct zone *zone, unsigned long phys_start_pfn,
 321                  unsigned long nr_pages)
 322 {
 323         unsigned long i, ret = 0;
 324         int sections_to_remove;
 325
 326         /*
 327          * We can only remove entire sections
 328          */
 329         BUG_ON(phys_start_pfn & ~PAGE_SECTION_MASK);
 330         BUG_ON(nr_pages % PAGES_PER_SECTION);
 331
 332         sections_to_remove = nr_pages / PAGES_PER_SECTION;
 333         for (i = 0; i < sections_to_remove; i++) {
 334                 unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION;
 335                 release_mem_region(pfn << PAGE_SHIFT,
 336                                    PAGES_PER_SECTION << PAGE_SHIFT);
 337                 ret = __remove_section(zone, __pfn_to_section(pfn));
 338                 if (ret)
 339                         break;
 340         }
 341         return ret;
 342 }
 343 EXPORT_SYMBOL_GPL(__remove_pages);
 344
 345 void online_page(struct page *page)
 346 {
 347         unsigned long pfn = page_to_pfn(page);
 348
 349         totalram_pages++;
 350         if (pfn >= num_physpages)
 351                 num_physpages = pfn + 1;
 352
 353 #ifdef CONFIG_HIGHMEM
 354         if (PageHighMem(page))
 355                 totalhigh_pages++;
 356 #endif
 357
 358 #ifdef CONFIG_FLATMEM
 359         max_mapnr = max(page_to_pfn(page), max_mapnr);
 360 #endif
 361
 362         ClearPageReserved(page);
 363         init_page_count(page);
 364         __free_page(page);
 365 }
 366
 367 static int online_pages_range(unsigned long start_pfn, unsigned long nr_pages,
 368                         void *arg)
 369 {
 370         unsigned long i;
 371         unsigned long onlined_pages = *(unsigned long *)arg;
 372         struct page *page;
 373         if (PageReserved(pfn_to_page(start_pfn)))
 374                 for (i = 0; i < nr_pages; i++) {
 375                         page = pfn_to_page(start_pfn + i);
 376                         online_page(page);
 377                         onlined_pages++;
 378                 }
 379         *(unsigned long *)arg = onlined_pages;
 380         return 0;
 381 }
 382
 383
 384 int online_pages(unsigned long pfn, unsigned long nr_pages)
 385 {
 386         unsigned long onlined_pages = 0;
 387         struct zone *zone;
 388         int need_zonelists_rebuild = 0;
 389         int nid;
 390         int ret;
 391         struct memory_notify arg;
 392
 393         arg.start_pfn = pfn;
 394         arg.nr_pages = nr_pages;
 395         arg.status_change_nid = -1;
 396
 397         nid = page_to_nid(pfn_to_page(pfn));
 398         if (node_present_pages(nid) == 0)
 399                 arg.status_change_nid = nid;
 400
 401         ret = memory_notify(MEM_GOING_ONLINE, &arg);
 402         ret = notifier_to_errno(ret);
 403         if (ret) {
 404                 memory_notify(MEM_CANCEL_ONLINE, &arg);
 405                 return ret;
 406         }
 407         /*
 408          * This doesn't need a lock to do pfn_to_page().
 409          * The section can't be removed here because of the
 410          * memory_block->state_mutex.
 411          */
 412         zone = page_zone(pfn_to_page(pfn));
 413         /*
 414          * If this zone is not populated, then it is not in zonelist.
 415          * This means the page allocator ignores this zone.
 416          * So, zonelist must be updated after online.
 417          */
 418         mutex_lock(&zonelists_mutex);
 419         if (!populated_zone(zone))
 420                 need_zonelists_rebuild = 1;
 421
 422         ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
 423                 online_pages_range);
 424         if (ret) {
 425                 mutex_unlock(&zonelists_mutex);
 426                 printk(KERN_DEBUG "online_pages %lx at %lx failed\n",
 427                         nr_pages, pfn);
 428                 memory_notify(MEM_CANCEL_ONLINE, &arg);
 429                 return ret;
 430         }
 431
 432         zone->present_pages += onlined_pages;
 433         zone->zone_pgdat->node_present_pages += onlined_pages;
 434         if (need_zonelists_rebuild)
 435                 build_all_zonelists(zone);
 436         else
 437                 zone_pcp_update(zone);
 438
 439         mutex_unlock(&zonelists_mutex);
 440         setup_per_zone_wmarks();
 441         calculate_zone_inactive_ratio(zone);
 442         if (onlined_pages) {
 443                 kswapd_run(zone_to_nid(zone));
 444                 node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
 445         }
 446
 447         vm_total_pages = nr_free_pagecache_pages();
 448
 449         writeback_set_ratelimit();
 450
 451         if (onlined_pages)
 452                 memory_notify(MEM_ONLINE, &arg);
 453
 454         return 0;
 455 }
 456 #endif /* CONFIG_MEMORY_HOTPLUG_SPARSE */
 457
 458 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
 459 static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)
 460 {
 461         struct pglist_data *pgdat;
 462         unsigned long zones_size[MAX_NR_ZONES] = {0};
 463         unsigned long zholes_size[MAX_NR_ZONES] = {0};
 464         unsigned long start_pfn = start >> PAGE_SHIFT;
 465
 466         pgdat = arch_alloc_nodedata(nid);
 467         if (!pgdat)
 468                 return NULL;
 469
 470         arch_refresh_nodedata(nid, pgdat);
 471
 472         /* we can use NODE_DATA(nid) from here */
 473
 474         /* init node's zones as empty zones, we don't have any present pages.*/
 475         free_area_init_node(nid, zones_size, start_pfn, zholes_size);
 476
 477         return pgdat;
 478 }
 479
 480 static void rollback_node_hotadd(int nid, pg_data_t *pgdat)
 481 {
 482         arch_refresh_nodedata(nid, NULL);
 483         arch_free_nodedata(pgdat);
 484         return;
 485 }
 486
 487
 488 /*
 489  * called by cpu_up() to online a node without onlined memory.
 490  */
 491 int mem_online_node(int nid)
 492 {
 493         pg_data_t       *pgdat;
 494         int     ret;
 495
 496         lock_system_sleep();
 497         pgdat = hotadd_new_pgdat(nid, 0);
 498         if (pgdat) {
 499                 ret = -ENOMEM;
 500                 goto out;
 501         }
 502         node_set_online(nid);
 503         ret = register_one_node(nid);
 504         BUG_ON(ret);
 505
 506 out:
 507         unlock_system_sleep();
 508         return ret;
 509 }
 510
 511 /* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */
 512 int __ref add_memory(int nid, u64 start, u64 size)
 513 {
 514         pg_data_t *pgdat = NULL;
 515         int new_pgdat = 0;
 516         struct resource *res;
 517         int ret;
 518
 519         lock_system_sleep();
 520
 521         res = register_memory_resource(start, size);
 522         ret = -EEXIST;
 523         if (!res)
 524                 goto out;
 525
 526         if (!node_online(nid)) {
 527                 pgdat = hotadd_new_pgdat(nid, start);
 528                 ret = -ENOMEM;
 529                 if (!pgdat)
 530                         goto out;
 531                 new_pgdat = 1;
 532         }
 533
 534         /* call arch's memory hotadd */
 535         ret = arch_add_memory(nid, start, size);
 536
 537         if (ret < 0)
 538                 goto error;
 539
 540         /* we online node here. we can't roll back from here. */
 541         node_set_online(nid);
 542
 543         if (new_pgdat) {
 544                 ret = register_one_node(nid);
 545                 /*
 546                  * If sysfs file of new node can't create, cpu on the node
 547                  * can't be hot-added. There is no rollback way now.
 548                  * So, check by BUG_ON() to catch it reluctantly..
 549                  */
 550                 BUG_ON(ret);
 551         }
 552
 553         /* create new memmap entry */
 554         firmware_map_add_hotplug(start, start + size, "System RAM");
 555
 556         goto out;
 557
 558 error:
 559         /* rollback pgdat allocation and others */
 560         if (new_pgdat)
 561                 rollback_node_hotadd(nid, pgdat);
 562         if (res)
 563                 release_memory_resource(res);
 564
 565 out:
 566         unlock_system_sleep();
 567         return ret;
 568 }
 569 EXPORT_SYMBOL_GPL(add_memory);
 570
 571 #ifdef CONFIG_MEMORY_HOTREMOVE
 572 /*
 573  * A free page on the buddy free lists (not the per-cpu lists) has PageBuddy
 574  * set and the size of the free page is given by page_order(). Using this,
 575  * the function determines if the pageblock contains only free pages.
 576  * Due to buddy contraints, a free page at least the size of a pageblock will
 577  * be located at the start of the pageblock
 578  */
 579 static inline int pageblock_free(struct page *page)
 580 {
 581         return PageBuddy(page) && page_order(page) >= pageblock_order;
 582 }
 583
 584 /* Return the start of the next active pageblock after a given page */
 585 static struct page *next_active_pageblock(struct page *page)
 586 {
 587         /* Ensure the starting page is pageblock-aligned */
 588         BUG_ON(page_to_pfn(page) & (pageblock_nr_pages - 1));
 589
 590         /* If the entire pageblock is free, move to the end of free page */
 591         if (pageblock_free(page)) {
 592                 int order;
 593                 /* be careful. we don't have locks, page_order can be changed.*/
 594                 order = page_order(page);
 595                 if ((order < MAX_ORDER) && (order >= pageblock_order))
 596                         return page + (1 << order);
 597         }
 598
 599         return page + pageblock_nr_pages;
 600 }
 601
 602 /* Checks if this range of memory is likely to be hot-removable. */
 603 int is_mem_section_removable(unsigned long start_pfn, unsigned long nr_pages)
 604 {
 605         struct page *page = pfn_to_page(start_pfn);
 606         struct page *end_page = page + nr_pages;
 607
 608         /* Check the starting page of each pageblock within the range */
 609         for (; page < end_page; page = next_active_pageblock(page)) {
 610                 if (!is_pageblock_removable_nolock(page))
 611                         return 0;
 612                 cond_resched();
 613         }
 614
 615         /* All pageblocks in the memory block are likely to be hot-removable */
 616         return 1;
 617 }
 618
 619 /*
 620  * Confirm all pages in a range [start, end) is belongs to the same zone.
 621  */
 622 static int test_pages_in_a_zone(unsigned long start_pfn, unsigned long end_pfn)
 623 {
 624         unsigned long pfn;
 625         struct zone *zone = NULL;
 626         struct page *page;
 627         int i;
 628         for (pfn = start_pfn;
 629              pfn < end_pfn;
 630              pfn += MAX_ORDER_NR_PAGES) {
 631                 i = 0;
 632                 /* This is just a CONFIG_HOLES_IN_ZONE check.*/
 633                 while ((i < MAX_ORDER_NR_PAGES) && !pfn_valid_within(pfn + i))
 634                         i++;
 635                 if (i == MAX_ORDER_NR_PAGES)
 636                         continue;
 637                 page = pfn_to_page(pfn + i);
 638                 if (zone && page_zone(page) != zone)
 639                         return 0;
 640                 zone = page_zone(page);
 641         }
 642         return 1;
 643 }
 644
 645 /*
 646  * Scanning pfn is much easier than scanning lru list.
 647  * Scan pfn from start to end and Find LRU page.
 648  */
 649 static unsigned long scan_lru_pages(unsigned long start, unsigned long end)
 650 {
 651         unsigned long pfn;
 652         struct page *page;
 653         for (pfn = start; pfn < end; pfn++) {
 654                 if (pfn_valid(pfn)) {
 655                         page = pfn_to_page(pfn);
 656                         if (PageLRU(page))
 657                                 return pfn;
 658                 }
 659         }
 660         return 0;
 661 }
 662
 663 static struct page *
 664 hotremove_migrate_alloc(struct page *page, unsigned long private, int **x)
 665 {
 666         /* This should be improooooved!! */
 667         return alloc_page(GFP_HIGHUSER_MOVABLE);
 668 }
 669
 670 #define NR_OFFLINE_AT_ONCE_PAGES        (256)
 671 static int
 672 do_migrate_range(unsigned long start_pfn, unsigned long end_pfn)
 673 {
 674         unsigned long pfn;
 675         struct page *page;
 676         int move_pages = NR_OFFLINE_AT_ONCE_PAGES;
 677         int not_managed = 0;
 678         int ret = 0;
 679         LIST_HEAD(source);
 680
 681         for (pfn = start_pfn; pfn < end_pfn && move_pages > 0; pfn++) {
 682                 if (!pfn_valid(pfn))
 683                         continue;
 684                 page = pfn_to_page(pfn);
 685                 if (!page_count(page))
 686                         continue;
 687                 /*
 688                  * We can skip free pages. And we can only deal with pages on
 689                  * LRU.
 690                  */
 691                 ret = isolate_lru_page(page);
 692                 if (!ret) { /* Success */
 693                         list_add_tail(&page->lru, &source);
 694                         move_pages--;
 695                         inc_zone_page_state(page, NR_ISOLATED_ANON +
 696                                             page_is_file_cache(page));
 697
 698                 } else {
 699 #ifdef CONFIG_DEBUG_VM
 700                         printk(KERN_ALERT "removing pfn %lx from LRU failed\n",
 701                                pfn);
 702                         dump_page(page);
 703 #endif
 704                         /* Becasue we don't have big zone->lock. we should
 705                            check this again here. */
 706                         if (page_count(page)) {
 707                                 not_managed++;
 708                                 ret = -EBUSY;
 709                                 break;
 710                         }
 711                 }
 712         }
 713         if (!list_empty(&source)) {
 714                 if (not_managed) {
 715                         putback_lru_pages(&source);
 716                         goto out;
 717                 }
 718                 /* this function returns # of failed pages */
 719                 ret = migrate_pages(&source, hotremove_migrate_alloc, 0, 1);
 720                 if (ret)
 721                         putback_lru_pages(&source);
 722         }
 723 out:
 724         return ret;
 725 }
 726
 727 /*
 728  * remove from free_area[] and mark all as Reserved.
 729  */
 730 static int
 731 offline_isolated_pages_cb(unsigned long start, unsigned long nr_pages,
 732                         void *data)
 733 {
 734         __offline_isolated_pages(start, start + nr_pages);
 735         return 0;
 736 }
 737
 738 static void
 739 offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
 740 {
 741         walk_system_ram_range(start_pfn, end_pfn - start_pfn, NULL,
 742                                 offline_isolated_pages_cb);
 743 }
 744
 745 /*
 746  * Check all pages in range, recoreded as memory resource, are isolated.
 747  */
 748 static int
 749 check_pages_isolated_cb(unsigned long start_pfn, unsigned long nr_pages,
 750                         void *data)
 751 {
 752         int ret;
 753         long offlined = *(long *)data;
 754         ret = test_pages_isolated(start_pfn, start_pfn + nr_pages);
 755         offlined = nr_pages;
 756         if (!ret)
 757                 *(long *)data += offlined;
 758         return ret;
 759 }
 760
 761 static long
 762 check_pages_isolated(unsigned long start_pfn, unsigned long end_pfn)
 763 {
 764         long offlined = 0;
 765         int ret;
 766
 767         ret = walk_system_ram_range(start_pfn, end_pfn - start_pfn, &offlined,
 768                         check_pages_isolated_cb);
 769         if (ret < 0)
 770                 offlined = (long)ret;
 771         return offlined;
 772 }
 773
 774 static int offline_pages(unsigned long start_pfn,
 775                   unsigned long end_pfn, unsigned long timeout)
 776 {
 777         unsigned long pfn, nr_pages, expire;
 778         long offlined_pages;
 779         int ret, drain, retry_max, node;
 780         struct zone *zone;
 781         struct memory_notify arg;
 782
 783         BUG_ON(start_pfn >= end_pfn);
 784         /* at least, alignment against pageblock is necessary */
 785         if (!IS_ALIGNED(start_pfn, pageblock_nr_pages))
 786                 return -EINVAL;
 787         if (!IS_ALIGNED(end_pfn, pageblock_nr_pages))
 788                 return -EINVAL;
 789         /* This makes hotplug much easier...and readable.
 790            we assume this for now. .*/
 791         if (!test_pages_in_a_zone(start_pfn, end_pfn))
 792                 return -EINVAL;
 793
 794         lock_system_sleep();
 795
 796         zone = page_zone(pfn_to_page(start_pfn));
 797         node = zone_to_nid(zone);
 798         nr_pages = end_pfn - start_pfn;
 799
 800         /* set above range as isolated */
 801         ret = start_isolate_page_range(start_pfn, end_pfn);
 802         if (ret)
 803                 goto out;
 804
 805         arg.start_pfn = start_pfn;
 806         arg.nr_pages = nr_pages;
 807         arg.status_change_nid = -1;
 808         if (nr_pages >= node_present_pages(node))
 809                 arg.status_change_nid = node;
 810
 811         ret = memory_notify(MEM_GOING_OFFLINE, &arg);
 812         ret = notifier_to_errno(ret);
 813         if (ret)
 814                 goto failed_removal;
 815
 816         pfn = start_pfn;
 817         expire = jiffies + timeout;
 818         drain = 0;
 819         retry_max = 5;
 820 repeat:
 821         /* start memory hot removal */
 822         ret = -EAGAIN;
 823         if (time_after(jiffies, expire))
 824                 goto failed_removal;
 825         ret = -EINTR;
 826         if (signal_pending(current))
 827                 goto failed_removal;
 828         ret = 0;
 829         if (drain) {
 830                 lru_add_drain_all();
 831                 cond_resched();
 832                 drain_all_pages();
 833         }
 834
 835         pfn = scan_lru_pages(start_pfn, end_pfn);
 836         if (pfn) { /* We have page on LRU */
 837                 ret = do_migrate_range(pfn, end_pfn);
 838                 if (!ret) {
 839                         drain = 1;
 840                         goto repeat;
 841                 } else {
 842                         if (ret < 0)
 843                                 if (--retry_max == 0)
 844                                         goto failed_removal;
 845                         yield();
 846                         drain = 1;
 847                         goto repeat;
 848                 }
 849         }
 850         /* drain all zone's lru pagevec, this is asyncronous... */
 851         lru_add_drain_all();
 852         yield();
 853         /* drain pcp pages , this is synchrouns. */
 854         drain_all_pages();
 855         /* check again */
 856         offlined_pages = check_pages_isolated(start_pfn, end_pfn);
 857         if (offlined_pages < 0) {
 858                 ret = -EBUSY;
 859                 goto failed_removal;
 860         }
 861         printk(KERN_INFO "Offlined Pages %ld\n", offlined_pages);
 862         /* Ok, all of our target is islaoted.
 863            We cannot do rollback at this point. */
 864         offline_isolated_pages(start_pfn, end_pfn);
 865         /* reset pagetype flags and makes migrate type to be MOVABLE */
 866         undo_isolate_page_range(start_pfn, end_pfn);
 867         /* removal success */
 868         zone->present_pages -= offlined_pages;
 869         zone->zone_pgdat->node_present_pages -= offlined_pages;
 870         totalram_pages -= offlined_pages;
 871
 872         setup_per_zone_wmarks();
 873         calculate_zone_inactive_ratio(zone);
 874         if (!node_present_pages(node)) {
 875                 node_clear_state(node, N_HIGH_MEMORY);
 876                 kswapd_stop(node);
 877         }
 878
 879         vm_total_pages = nr_free_pagecache_pages();
 880         writeback_set_ratelimit();
 881
 882         memory_notify(MEM_OFFLINE, &arg);
 883         unlock_system_sleep();
 884         return 0;
 885
 886 failed_removal:
 887         printk(KERN_INFO "memory offlining %lx to %lx failed\n",
 888                 start_pfn, end_pfn);
 889         memory_notify(MEM_CANCEL_OFFLINE, &arg);
 890         /* pushback to free area */
 891         undo_isolate_page_range(start_pfn, end_pfn);
 892
 893 out:
 894         unlock_system_sleep();
 895         return ret;
 896 }
 897
 898 int remove_memory(u64 start, u64 size)
 899 {
 900         unsigned long start_pfn, end_pfn;
 901
 902         start_pfn = PFN_DOWN(start);
 903         end_pfn = start_pfn + PFN_DOWN(size);
 904         return offline_pages(start_pfn, end_pfn, 120 * HZ);
 905 }
 906 #else
 907 int remove_memory(u64 start, u64 size)
 908 {
 909         return -EINVAL;
 910 }
 911 #endif /* CONFIG_MEMORY_HOTREMOVE */
 912 EXPORT_SYMBOL_GPL(remove_memory);