linux/mm/page_alloc.c
<<
>>
Prefs
   1/*
   2 *  linux/mm/page_alloc.c
   3 *
   4 *  Manages the free list, the system allocates free pages here.
   5 *  Note that kmalloc() lives in slab.c
   6 *
   7 *  Copyright (C) 1991, 1992, 1993, 1994  Linus Torvalds
   8 *  Swap reorganised 29.12.95, Stephen Tweedie
   9 *  Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
  10 *  Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
  11 *  Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
  12 *  Zone balancing, Kanoj Sarcar, SGI, Jan 2000
  13 *  Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
  14 *          (lots of bits borrowed from Ingo Molnar & Andrew Morton)
  15 */
  16
  17#include <linux/stddef.h>
  18#include <linux/mm.h>
  19#include <linux/swap.h>
  20#include <linux/interrupt.h>
  21#include <linux/pagemap.h>
  22#include <linux/jiffies.h>
  23#include <linux/bootmem.h>
  24#include <linux/compiler.h>
  25#include <linux/kernel.h>
  26#include <linux/module.h>
  27#include <linux/suspend.h>
  28#include <linux/pagevec.h>
  29#include <linux/blkdev.h>
  30#include <linux/slab.h>
  31#include <linux/oom.h>
  32#include <linux/notifier.h>
  33#include <linux/topology.h>
  34#include <linux/sysctl.h>
  35#include <linux/cpu.h>
  36#include <linux/cpuset.h>
  37#include <linux/memory_hotplug.h>
  38#include <linux/nodemask.h>
  39#include <linux/vmalloc.h>
  40#include <linux/mempolicy.h>
  41#include <linux/stop_machine.h>
  42#include <linux/sort.h>
  43#include <linux/pfn.h>
  44#include <linux/backing-dev.h>
  45#include <linux/fault-inject.h>
  46#include <linux/page-isolation.h>
  47#include <linux/page_cgroup.h>
  48#include <linux/debugobjects.h>
  49
  50#include <asm/tlbflush.h>
  51#include <asm/div64.h>
  52#include "internal.h"
  53
  54/*
  55 * Array of node states.
  56 */
  57nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
  58        [N_POSSIBLE] = NODE_MASK_ALL,
  59        [N_ONLINE] = { { [0] = 1UL } },
  60#ifndef CONFIG_NUMA
  61        [N_NORMAL_MEMORY] = { { [0] = 1UL } },
  62#ifdef CONFIG_HIGHMEM
  63        [N_HIGH_MEMORY] = { { [0] = 1UL } },
  64#endif
  65        [N_CPU] = { { [0] = 1UL } },
  66#endif  /* NUMA */
  67};
  68EXPORT_SYMBOL(node_states);
  69
  70unsigned long totalram_pages __read_mostly;
  71unsigned long totalreserve_pages __read_mostly;
  72unsigned long highest_memmap_pfn __read_mostly;
  73int percpu_pagelist_fraction;
  74
  75#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
  76int pageblock_order __read_mostly;
  77#endif
  78
  79static void __free_pages_ok(struct page *page, unsigned int order);
  80
  81/*
  82 * results with 256, 32 in the lowmem_reserve sysctl:
  83 *      1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
  84 *      1G machine -> (16M dma, 784M normal, 224M high)
  85 *      NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
  86 *      HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
  87 *      HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
  88 *
  89 * TBD: should special case ZONE_DMA32 machines here - in those we normally
  90 * don't need any ZONE_NORMAL reservation
  91 */
  92int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
  93#ifdef CONFIG_ZONE_DMA
  94         256,
  95#endif
  96#ifdef CONFIG_ZONE_DMA32
  97         256,
  98#endif
  99#ifdef CONFIG_HIGHMEM
 100         32,
 101#endif
 102         32,
 103};
 104
 105EXPORT_SYMBOL(totalram_pages);
 106
 107static char * const zone_names[MAX_NR_ZONES] = {
 108#ifdef CONFIG_ZONE_DMA
 109         "DMA",
 110#endif
 111#ifdef CONFIG_ZONE_DMA32
 112         "DMA32",
 113#endif
 114         "Normal",
 115#ifdef CONFIG_HIGHMEM
 116         "HighMem",
 117#endif
 118         "Movable",
 119};
 120
 121int min_free_kbytes = 1024;
 122
 123unsigned long __meminitdata nr_kernel_pages;
 124unsigned long __meminitdata nr_all_pages;
 125static unsigned long __meminitdata dma_reserve;
 126
 127#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
 128  /*
 129   * MAX_ACTIVE_REGIONS determines the maximum number of distinct
 130   * ranges of memory (RAM) that may be registered with add_active_range().
 131   * Ranges passed to add_active_range() will be merged if possible
 132   * so the number of times add_active_range() can be called is
 133   * related to the number of nodes and the number of holes
 134   */
 135  #ifdef CONFIG_MAX_ACTIVE_REGIONS
 136    /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
 137    #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
 138  #else
 139    #if MAX_NUMNODES >= 32
 140      /* If there can be many nodes, allow up to 50 holes per node */
 141      #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
 142    #else
 143      /* By default, allow up to 256 distinct regions */
 144      #define MAX_ACTIVE_REGIONS 256
 145    #endif
 146  #endif
 147
 148  static struct node_active_region __meminitdata early_node_map[MAX_ACTIVE_REGIONS];
 149  static int __meminitdata nr_nodemap_entries;
 150  static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
 151  static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
 152#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
 153  static unsigned long __meminitdata node_boundary_start_pfn[MAX_NUMNODES];
 154  static unsigned long __meminitdata node_boundary_end_pfn[MAX_NUMNODES];
 155#endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
 156  static unsigned long __initdata required_kernelcore;
 157  static unsigned long __initdata required_movablecore;
 158  static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
 159
 160  /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
 161  int movable_zone;
 162  EXPORT_SYMBOL(movable_zone);
 163#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
 164
 165#if MAX_NUMNODES > 1
 166int nr_node_ids __read_mostly = MAX_NUMNODES;
 167EXPORT_SYMBOL(nr_node_ids);
 168#endif
 169
 170int page_group_by_mobility_disabled __read_mostly;
 171
 172static void set_pageblock_migratetype(struct page *page, int migratetype)
 173{
 174        set_pageblock_flags_group(page, (unsigned long)migratetype,
 175                                        PB_migrate, PB_migrate_end);
 176}
 177
 178#ifdef CONFIG_DEBUG_VM
 179static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
 180{
 181        int ret = 0;
 182        unsigned seq;
 183        unsigned long pfn = page_to_pfn(page);
 184
 185        do {
 186                seq = zone_span_seqbegin(zone);
 187                if (pfn >= zone->zone_start_pfn + zone->spanned_pages)
 188                        ret = 1;
 189                else if (pfn < zone->zone_start_pfn)
 190                        ret = 1;
 191        } while (zone_span_seqretry(zone, seq));
 192
 193        return ret;
 194}
 195
 196static int page_is_consistent(struct zone *zone, struct page *page)
 197{
 198        if (!pfn_valid_within(page_to_pfn(page)))
 199                return 0;
 200        if (zone != page_zone(page))
 201                return 0;
 202
 203        return 1;
 204}
 205/*
 206 * Temporary debugging check for pages not lying within a given zone.
 207 */
 208static int bad_range(struct zone *zone, struct page *page)
 209{
 210        if (page_outside_zone_boundaries(zone, page))
 211                return 1;
 212        if (!page_is_consistent(zone, page))
 213                return 1;
 214
 215        return 0;
 216}
 217#else
 218static inline int bad_range(struct zone *zone, struct page *page)
 219{
 220        return 0;
 221}
 222#endif
 223
 224static void bad_page(struct page *page)
 225{
 226        static unsigned long resume;
 227        static unsigned long nr_shown;
 228        static unsigned long nr_unshown;
 229
 230        /*
 231         * Allow a burst of 60 reports, then keep quiet for that minute;
 232         * or allow a steady drip of one report per second.
 233         */
 234        if (nr_shown == 60) {
 235                if (time_before(jiffies, resume)) {
 236                        nr_unshown++;
 237                        goto out;
 238                }
 239                if (nr_unshown) {
 240                        printk(KERN_ALERT
 241                              "BUG: Bad page state: %lu messages suppressed\n",
 242                                nr_unshown);
 243                        nr_unshown = 0;
 244                }
 245                nr_shown = 0;
 246        }
 247        if (nr_shown++ == 0)
 248                resume = jiffies + 60 * HZ;
 249
 250        printk(KERN_ALERT "BUG: Bad page state in process %s  pfn:%05lx\n",
 251                current->comm, page_to_pfn(page));
 252        printk(KERN_ALERT
 253                "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
 254                page, (void *)page->flags, page_count(page),
 255                page_mapcount(page), page->mapping, page->index);
 256
 257        dump_stack();
 258out:
 259        /* Leave bad fields for debug, except PageBuddy could make trouble */
 260        __ClearPageBuddy(page);
 261        add_taint(TAINT_BAD_PAGE);
 262}
 263
 264/*
 265 * Higher-order pages are called "compound pages".  They are structured thusly:
 266 *
 267 * The first PAGE_SIZE page is called the "head page".
 268 *
 269 * The remaining PAGE_SIZE pages are called "tail pages".
 270 *
 271 * All pages have PG_compound set.  All pages have their ->private pointing at
 272 * the head page (even the head page has this).
 273 *
 274 * The first tail page's ->lru.next holds the address of the compound page's
 275 * put_page() function.  Its ->lru.prev holds the order of allocation.
 276 * This usage means that zero-order pages may not be compound.
 277 */
 278
 279static void free_compound_page(struct page *page)
 280{
 281        __free_pages_ok(page, compound_order(page));
 282}
 283
 284void prep_compound_page(struct page *page, unsigned long order)
 285{
 286        int i;
 287        int nr_pages = 1 << order;
 288
 289        set_compound_page_dtor(page, free_compound_page);
 290        set_compound_order(page, order);
 291        __SetPageHead(page);
 292        for (i = 1; i < nr_pages; i++) {
 293                struct page *p = page + i;
 294
 295                __SetPageTail(p);
 296                p->first_page = page;
 297        }
 298}
 299
 300#ifdef CONFIG_HUGETLBFS
 301void prep_compound_gigantic_page(struct page *page, unsigned long order)
 302{
 303        int i;
 304        int nr_pages = 1 << order;
 305        struct page *p = page + 1;
 306
 307        set_compound_page_dtor(page, free_compound_page);
 308        set_compound_order(page, order);
 309        __SetPageHead(page);
 310        for (i = 1; i < nr_pages; i++, p = mem_map_next(p, page, i)) {
 311                __SetPageTail(p);
 312                p->first_page = page;
 313        }
 314}
 315#endif
 316
 317static int destroy_compound_page(struct page *page, unsigned long order)
 318{
 319        int i;
 320        int nr_pages = 1 << order;
 321        int bad = 0;
 322
 323        if (unlikely(compound_order(page) != order) ||
 324            unlikely(!PageHead(page))) {
 325                bad_page(page);
 326                bad++;
 327        }
 328
 329        __ClearPageHead(page);
 330
 331        for (i = 1; i < nr_pages; i++) {
 332                struct page *p = page + i;
 333
 334                if (unlikely(!PageTail(p) || (p->first_page != page))) {
 335                        bad_page(page);
 336                        bad++;
 337                }
 338                __ClearPageTail(p);
 339        }
 340
 341        return bad;
 342}
 343
 344static inline void prep_zero_page(struct page *page, int order, gfp_t gfp_flags)
 345{
 346        int i;
 347
 348        /*
 349         * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
 350         * and __GFP_HIGHMEM from hard or soft interrupt context.
 351         */
 352        VM_BUG_ON((gfp_flags & __GFP_HIGHMEM) && in_interrupt());
 353        for (i = 0; i < (1 << order); i++)
 354                clear_highpage(page + i);
 355}
 356
 357static inline void set_page_order(struct page *page, int order)
 358{
 359        set_page_private(page, order);
 360        __SetPageBuddy(page);
 361}
 362
 363static inline void rmv_page_order(struct page *page)
 364{
 365        __ClearPageBuddy(page);
 366        set_page_private(page, 0);
 367}
 368
 369/*
 370 * Locate the struct page for both the matching buddy in our
 371 * pair (buddy1) and the combined O(n+1) page they form (page).
 372 *
 373 * 1) Any buddy B1 will have an order O twin B2 which satisfies
 374 * the following equation:
 375 *     B2 = B1 ^ (1 << O)
 376 * For example, if the starting buddy (buddy2) is #8 its order
 377 * 1 buddy is #10:
 378 *     B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
 379 *
 380 * 2) Any buddy B will have an order O+1 parent P which
 381 * satisfies the following equation:
 382 *     P = B & ~(1 << O)
 383 *
 384 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
 385 */
 386static inline struct page *
 387__page_find_buddy(struct page *page, unsigned long page_idx, unsigned int order)
 388{
 389        unsigned long buddy_idx = page_idx ^ (1 << order);
 390
 391        return page + (buddy_idx - page_idx);
 392}
 393
 394static inline unsigned long
 395__find_combined_index(unsigned long page_idx, unsigned int order)
 396{
 397        return (page_idx & ~(1 << order));
 398}
 399
 400/*
 401 * This function checks whether a page is free && is the buddy
 402 * we can do coalesce a page and its buddy if
 403 * (a) the buddy is not in a hole &&
 404 * (b) the buddy is in the buddy system &&
 405 * (c) a page and its buddy have the same order &&
 406 * (d) a page and its buddy are in the same zone.
 407 *
 408 * For recording whether a page is in the buddy system, we use PG_buddy.
 409 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
 410 *
 411 * For recording page's order, we use page_private(page).
 412 */
 413static inline int page_is_buddy(struct page *page, struct page *buddy,
 414                                                                int order)
 415{
 416        if (!pfn_valid_within(page_to_pfn(buddy)))
 417                return 0;
 418
 419        if (page_zone_id(page) != page_zone_id(buddy))
 420                return 0;
 421
 422        if (PageBuddy(buddy) && page_order(buddy) == order) {
 423                BUG_ON(page_count(buddy) != 0);
 424                return 1;
 425        }
 426        return 0;
 427}
 428
 429/*
 430 * Freeing function for a buddy system allocator.
 431 *
 432 * The concept of a buddy system is to maintain direct-mapped table
 433 * (containing bit values) for memory blocks of various "orders".
 434 * The bottom level table contains the map for the smallest allocatable
 435 * units of memory (here, pages), and each level above it describes
 436 * pairs of units from the levels below, hence, "buddies".
 437 * At a high level, all that happens here is marking the table entry
 438 * at the bottom level available, and propagating the changes upward
 439 * as necessary, plus some accounting needed to play nicely with other
 440 * parts of the VM system.
 441 * At each level, we keep a list of pages, which are heads of continuous
 442 * free pages of length of (1 << order) and marked with PG_buddy. Page's
 443 * order is recorded in page_private(page) field.
 444 * So when we are allocating or freeing one, we can derive the state of the
 445 * other.  That is, if we allocate a small block, and both were   
 446 * free, the remainder of the region must be split into blocks.   
 447 * If a block is freed, and its buddy is also free, then this
 448 * triggers coalescing into a block of larger size.            
 449 *
 450 * -- wli
 451 */
 452
 453static inline void __free_one_page(struct page *page,
 454                struct zone *zone, unsigned int order)
 455{
 456        unsigned long page_idx;
 457        int order_size = 1 << order;
 458        int migratetype = get_pageblock_migratetype(page);
 459
 460        if (unlikely(PageCompound(page)))
 461                if (unlikely(destroy_compound_page(page, order)))
 462                        return;
 463
 464        page_idx = page_to_pfn(page) & ((1 << MAX_ORDER) - 1);
 465
 466        VM_BUG_ON(page_idx & (order_size - 1));
 467        VM_BUG_ON(bad_range(zone, page));
 468
 469        __mod_zone_page_state(zone, NR_FREE_PAGES, order_size);
 470        while (order < MAX_ORDER-1) {
 471                unsigned long combined_idx;
 472                struct page *buddy;
 473
 474                buddy = __page_find_buddy(page, page_idx, order);
 475                if (!page_is_buddy(page, buddy, order))
 476                        break;
 477
 478                /* Our buddy is free, merge with it and move up one order. */
 479                list_del(&buddy->lru);
 480                zone->free_area[order].nr_free--;
 481                rmv_page_order(buddy);
 482                combined_idx = __find_combined_index(page_idx, order);
 483                page = page + (combined_idx - page_idx);
 484                page_idx = combined_idx;
 485                order++;
 486        }
 487        set_page_order(page, order);
 488        list_add(&page->lru,
 489                &zone->free_area[order].free_list[migratetype]);
 490        zone->free_area[order].nr_free++;
 491}
 492
 493static inline int free_pages_check(struct page *page)
 494{
 495        free_page_mlock(page);
 496        if (unlikely(page_mapcount(page) |
 497                (page->mapping != NULL)  |
 498                (page_count(page) != 0)  |
 499                (page->flags & PAGE_FLAGS_CHECK_AT_FREE))) {
 500                bad_page(page);
 501                return 1;
 502        }
 503        if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
 504                page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
 505        return 0;
 506}
 507
 508/*
 509 * Frees a list of pages. 
 510 * Assumes all pages on list are in same zone, and of same order.
 511 * count is the number of pages to free.
 512 *
 513 * If the zone was previously in an "all pages pinned" state then look to
 514 * see if this freeing clears that state.
 515 *
 516 * And clear the zone's pages_scanned counter, to hold off the "all pages are
 517 * pinned" detection logic.
 518 */
 519static void free_pages_bulk(struct zone *zone, int count,
 520                                        struct list_head *list, int order)
 521{
 522        spin_lock(&zone->lock);
 523        zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
 524        zone->pages_scanned = 0;
 525        while (count--) {
 526                struct page *page;
 527
 528                VM_BUG_ON(list_empty(list));
 529                page = list_entry(list->prev, struct page, lru);
 530                /* have to delete it as __free_one_page list manipulates */
 531                list_del(&page->lru);
 532                __free_one_page(page, zone, order);
 533        }
 534        spin_unlock(&zone->lock);
 535}
 536
 537static void free_one_page(struct zone *zone, struct page *page, int order)
 538{
 539        spin_lock(&zone->lock);
 540        zone_clear_flag(zone, ZONE_ALL_UNRECLAIMABLE);
 541        zone->pages_scanned = 0;
 542        __free_one_page(page, zone, order);
 543        spin_unlock(&zone->lock);
 544}
 545
 546static void __free_pages_ok(struct page *page, unsigned int order)
 547{
 548        unsigned long flags;
 549        int i;
 550        int bad = 0;
 551
 552        for (i = 0 ; i < (1 << order) ; ++i)
 553                bad += free_pages_check(page + i);
 554        if (bad)
 555                return;
 556
 557        if (!PageHighMem(page)) {
 558                debug_check_no_locks_freed(page_address(page),PAGE_SIZE<<order);
 559                debug_check_no_obj_freed(page_address(page),
 560                                           PAGE_SIZE << order);
 561        }
 562        arch_free_page(page, order);
 563        kernel_map_pages(page, 1 << order, 0);
 564
 565        local_irq_save(flags);
 566        __count_vm_events(PGFREE, 1 << order);
 567        free_one_page(page_zone(page), page, order);
 568        local_irq_restore(flags);
 569}
 570
 571/*
 572 * permit the bootmem allocator to evade page validation on high-order frees
 573 */
 574void __meminit __free_pages_bootmem(struct page *page, unsigned int order)
 575{
 576        if (order == 0) {
 577                __ClearPageReserved(page);
 578                set_page_count(page, 0);
 579                set_page_refcounted(page);
 580                __free_page(page);
 581        } else {
 582                int loop;
 583
 584                prefetchw(page);
 585                for (loop = 0; loop < BITS_PER_LONG; loop++) {
 586                        struct page *p = &page[loop];
 587
 588                        if (loop + 1 < BITS_PER_LONG)
 589                                prefetchw(p + 1);
 590                        __ClearPageReserved(p);
 591                        set_page_count(p, 0);
 592                }
 593
 594                set_page_refcounted(page);
 595                __free_pages(page, order);
 596        }
 597}
 598
 599
 600/*
 601 * The order of subdivision here is critical for the IO subsystem.
 602 * Please do not alter this order without good reasons and regression
 603 * testing. Specifically, as large blocks of memory are subdivided,
 604 * the order in which smaller blocks are delivered depends on the order
 605 * they're subdivided in this function. This is the primary factor
 606 * influencing the order in which pages are delivered to the IO
 607 * subsystem according to empirical testing, and this is also justified
 608 * by considering the behavior of a buddy system containing a single
 609 * large block of memory acted on by a series of small allocations.
 610 * This behavior is a critical factor in sglist merging's success.
 611 *
 612 * -- wli
 613 */
 614static inline void expand(struct zone *zone, struct page *page,
 615        int low, int high, struct free_area *area,
 616        int migratetype)
 617{
 618        unsigned long size = 1 << high;
 619
 620        while (high > low) {
 621                area--;
 622                high--;
 623                size >>= 1;
 624                VM_BUG_ON(bad_range(zone, &page[size]));
 625                list_add(&page[size].lru, &area->free_list[migratetype]);
 626                area->nr_free++;
 627                set_page_order(&page[size], high);
 628        }
 629}
 630
 631/*
 632 * This page is about to be returned from the page allocator
 633 */
 634static int prep_new_page(struct page *page, int order, gfp_t gfp_flags)
 635{
 636        if (unlikely(page_mapcount(page) |
 637                (page->mapping != NULL)  |
 638                (page_count(page) != 0)  |
 639                (page->flags & PAGE_FLAGS_CHECK_AT_PREP))) {
 640                bad_page(page);
 641                return 1;
 642        }
 643
 644        set_page_private(page, 0);
 645        set_page_refcounted(page);
 646
 647        arch_alloc_page(page, order);
 648        kernel_map_pages(page, 1 << order, 1);
 649
 650        if (gfp_flags & __GFP_ZERO)
 651                prep_zero_page(page, order, gfp_flags);
 652
 653        if (order && (gfp_flags & __GFP_COMP))
 654                prep_compound_page(page, order);
 655
 656        return 0;
 657}
 658
 659/*
 660 * Go through the free lists for the given migratetype and remove
 661 * the smallest available page from the freelists
 662 */
 663static struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
 664                                                int migratetype)
 665{
 666        unsigned int current_order;
 667        struct free_area * area;
 668        struct page *page;
 669
 670        /* Find a page of the appropriate size in the preferred list */
 671        for (current_order = order; current_order < MAX_ORDER; ++current_order) {
 672                area = &(zone->free_area[current_order]);
 673                if (list_empty(&area->free_list[migratetype]))
 674                        continue;
 675
 676                page = list_entry(area->free_list[migratetype].next,
 677                                                        struct page, lru);
 678                list_del(&page->lru);
 679                rmv_page_order(page);
 680                area->nr_free--;
 681                __mod_zone_page_state(zone, NR_FREE_PAGES, - (1UL << order));
 682                expand(zone, page, order, current_order, area, migratetype);
 683                return page;
 684        }
 685
 686        return NULL;
 687}
 688
 689
 690/*
 691 * This array describes the order lists are fallen back to when
 692 * the free lists for the desirable migrate type are depleted
 693 */
 694static int fallbacks[MIGRATE_TYPES][MIGRATE_TYPES-1] = {
 695        [MIGRATE_UNMOVABLE]   = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE,   MIGRATE_RESERVE },
 696        [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE,   MIGRATE_MOVABLE,   MIGRATE_RESERVE },
 697        [MIGRATE_MOVABLE]     = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_RESERVE },
 698        [MIGRATE_RESERVE]     = { MIGRATE_RESERVE,     MIGRATE_RESERVE,   MIGRATE_RESERVE }, /* Never used */
 699};
 700
 701/*
 702 * Move the free pages in a range to the free lists of the requested type.
 703 * Note that start_page and end_pages are not aligned on a pageblock
 704 * boundary. If alignment is required, use move_freepages_block()
 705 */
 706static int move_freepages(struct zone *zone,
 707                          struct page *start_page, struct page *end_page,
 708                          int migratetype)
 709{
 710        struct page *page;
 711        unsigned long order;
 712        int pages_moved = 0;
 713
 714#ifndef CONFIG_HOLES_IN_ZONE
 715        /*
 716         * page_zone is not safe to call in this context when
 717         * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
 718         * anyway as we check zone boundaries in move_freepages_block().
 719         * Remove at a later date when no bug reports exist related to
 720         * grouping pages by mobility
 721         */
 722        BUG_ON(page_zone(start_page) != page_zone(end_page));
 723#endif
 724
 725        for (page = start_page; page <= end_page;) {
 726                /* Make sure we are not inadvertently changing nodes */
 727                VM_BUG_ON(page_to_nid(page) != zone_to_nid(zone));
 728
 729                if (!pfn_valid_within(page_to_pfn(page))) {
 730                        page++;
 731                        continue;
 732                }
 733
 734                if (!PageBuddy(page)) {
 735                        page++;
 736                        continue;
 737                }
 738
 739                order = page_order(page);
 740                list_del(&page->lru);
 741                list_add(&page->lru,
 742                        &zone->free_area[order].free_list[migratetype]);
 743                page += 1 << order;
 744                pages_moved += 1 << order;
 745        }
 746
 747        return pages_moved;
 748}
 749
 750static int move_freepages_block(struct zone *zone, struct page *page,
 751                                int migratetype)
 752{
 753        unsigned long start_pfn, end_pfn;
 754        struct page *start_page, *end_page;
 755
 756        start_pfn = page_to_pfn(page);
 757        start_pfn = start_pfn & ~(pageblock_nr_pages-1);
 758        start_page = pfn_to_page(start_pfn);
 759        end_page = start_page + pageblock_nr_pages - 1;
 760        end_pfn = start_pfn + pageblock_nr_pages - 1;
 761
 762        /* Do not cross zone boundaries */
 763        if (start_pfn < zone->zone_start_pfn)
 764                start_page = page;
 765        if (end_pfn >= zone->zone_start_pfn + zone->spanned_pages)
 766                return 0;
 767
 768        return move_freepages(zone, start_page, end_page, migratetype);
 769}
 770
 771/* Remove an element from the buddy allocator from the fallback list */
 772static struct page *__rmqueue_fallback(struct zone *zone, int order,
 773                                                int start_migratetype)
 774{
 775        struct free_area * area;
 776        int current_order;
 777        struct page *page;
 778        int migratetype, i;
 779
 780        /* Find the largest possible block of pages in the other list */
 781        for (current_order = MAX_ORDER-1; current_order >= order;
 782                                                --current_order) {
 783                for (i = 0; i < MIGRATE_TYPES - 1; i++) {
 784                        migratetype = fallbacks[start_migratetype][i];
 785
 786                        /* MIGRATE_RESERVE handled later if necessary */
 787                        if (migratetype == MIGRATE_RESERVE)
 788                                continue;
 789
 790                        area = &(zone->free_area[current_order]);
 791                        if (list_empty(&area->free_list[migratetype]))
 792                                continue;
 793
 794                        page = list_entry(area->free_list[migratetype].next,
 795                                        struct page, lru);
 796                        area->nr_free--;
 797
 798                        /*
 799                         * If breaking a large block of pages, move all free
 800                         * pages to the preferred allocation list. If falling
 801                         * back for a reclaimable kernel allocation, be more
 802                         * agressive about taking ownership of free pages
 803                         */
 804                        if (unlikely(current_order >= (pageblock_order >> 1)) ||
 805                                        start_migratetype == MIGRATE_RECLAIMABLE) {
 806                                unsigned long pages;
 807                                pages = move_freepages_block(zone, page,
 808                                                                start_migratetype);
 809
 810                                /* Claim the whole block if over half of it is free */
 811                                if (pages >= (1 << (pageblock_order-1)))
 812                                        set_pageblock_migratetype(page,
 813                                                                start_migratetype);
 814
 815                                migratetype = start_migratetype;
 816                        }
 817
 818                        /* Remove the page from the freelists */
 819                        list_del(&page->lru);
 820                        rmv_page_order(page);
 821                        __mod_zone_page_state(zone, NR_FREE_PAGES,
 822                                                        -(1UL << order));
 823
 824                        if (current_order == pageblock_order)
 825                                set_pageblock_migratetype(page,
 826                                                        start_migratetype);
 827
 828                        expand(zone, page, order, current_order, area, migratetype);
 829                        return page;
 830                }
 831        }
 832
 833        /* Use MIGRATE_RESERVE rather than fail an allocation */
 834        return __rmqueue_smallest(zone, order, MIGRATE_RESERVE);
 835}
 836
 837/*
 838 * Do the hard work of removing an element from the buddy allocator.
 839 * Call me with the zone->lock already held.
 840 */
 841static struct page *__rmqueue(struct zone *zone, unsigned int order,
 842                                                int migratetype)
 843{
 844        struct page *page;
 845
 846        page = __rmqueue_smallest(zone, order, migratetype);
 847
 848        if (unlikely(!page))
 849                page = __rmqueue_fallback(zone, order, migratetype);
 850
 851        return page;
 852}
 853
 854/* 
 855 * Obtain a specified number of elements from the buddy allocator, all under
 856 * a single hold of the lock, for efficiency.  Add them to the supplied list.
 857 * Returns the number of new pages which were placed at *list.
 858 */
 859static int rmqueue_bulk(struct zone *zone, unsigned int order, 
 860                        unsigned long count, struct list_head *list,
 861                        int migratetype, int cold)
 862{
 863        int i;
 864        
 865        spin_lock(&zone->lock);
 866        for (i = 0; i < count; ++i) {
 867                struct page *page = __rmqueue(zone, order, migratetype);
 868                if (unlikely(page == NULL))
 869                        break;
 870
 871                /*
 872                 * Split buddy pages returned by expand() are received here
 873                 * in physical page order. The page is added to the callers and
 874                 * list and the list head then moves forward. From the callers
 875                 * perspective, the linked list is ordered by page number in
 876                 * some conditions. This is useful for IO devices that can
 877                 * merge IO requests if the physical pages are ordered
 878                 * properly.
 879                 */
 880                if (likely(cold == 0))
 881                        list_add(&page->lru, list);
 882                else
 883                        list_add_tail(&page->lru, list);
 884                set_page_private(page, migratetype);
 885                list = &page->lru;
 886        }
 887        spin_unlock(&zone->lock);
 888        return i;
 889}
 890
 891#ifdef CONFIG_NUMA
 892/*
 893 * Called from the vmstat counter updater to drain pagesets of this
 894 * currently executing processor on remote nodes after they have
 895 * expired.
 896 *
 897 * Note that this function must be called with the thread pinned to
 898 * a single processor.
 899 */
 900void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
 901{
 902        unsigned long flags;
 903        int to_drain;
 904
 905        local_irq_save(flags);
 906        if (pcp->count >= pcp->batch)
 907                to_drain = pcp->batch;
 908        else
 909                to_drain = pcp->count;
 910        free_pages_bulk(zone, to_drain, &pcp->list, 0);
 911        pcp->count -= to_drain;
 912        local_irq_restore(flags);
 913}
 914#endif
 915
 916/*
 917 * Drain pages of the indicated processor.
 918 *
 919 * The processor must either be the current processor and the
 920 * thread pinned to the current processor or a processor that
 921 * is not online.
 922 */
 923static void drain_pages(unsigned int cpu)
 924{
 925        unsigned long flags;
 926        struct zone *zone;
 927
 928        for_each_populated_zone(zone) {
 929                struct per_cpu_pageset *pset;
 930                struct per_cpu_pages *pcp;
 931
 932                pset = zone_pcp(zone, cpu);
 933
 934                pcp = &pset->pcp;
 935                local_irq_save(flags);
 936                free_pages_bulk(zone, pcp->count, &pcp->list, 0);
 937                pcp->count = 0;
 938                local_irq_restore(flags);
 939        }
 940}
 941
 942/*
 943 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
 944 */
 945void drain_local_pages(void *arg)
 946{
 947        drain_pages(smp_processor_id());
 948}
 949
 950/*
 951 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
 952 */
 953void drain_all_pages(void)
 954{
 955        on_each_cpu(drain_local_pages, NULL, 1);
 956}
 957
 958#ifdef CONFIG_HIBERNATION
 959
 960void mark_free_pages(struct zone *zone)
 961{
 962        unsigned long pfn, max_zone_pfn;
 963        unsigned long flags;
 964        int order, t;
 965        struct list_head *curr;
 966
 967        if (!zone->spanned_pages)
 968                return;
 969
 970        spin_lock_irqsave(&zone->lock, flags);
 971
 972        max_zone_pfn = zone->zone_start_pfn + zone->spanned_pages;
 973        for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
 974                if (pfn_valid(pfn)) {
 975                        struct page *page = pfn_to_page(pfn);
 976
 977                        if (!swsusp_page_is_forbidden(page))
 978                                swsusp_unset_page_free(page);
 979                }
 980
 981        for_each_migratetype_order(order, t) {
 982                list_for_each(curr, &zone->free_area[order].free_list[t]) {
 983                        unsigned long i;
 984
 985                        pfn = page_to_pfn(list_entry(curr, struct page, lru));
 986                        for (i = 0; i < (1UL << order); i++)
 987                                swsusp_set_page_free(pfn_to_page(pfn + i));
 988                }
 989        }
 990        spin_unlock_irqrestore(&zone->lock, flags);
 991}
 992#endif /* CONFIG_PM */
 993
 994/*
 995 * Free a 0-order page
 996 */
 997static void free_hot_cold_page(struct page *page, int cold)
 998{
 999        struct zone *zone = page_zone(page);
1000        struct per_cpu_pages *pcp;
1001        unsigned long flags;
1002
1003        if (PageAnon(page))
1004                page->mapping = NULL;
1005        if (free_pages_check(page))
1006                return;
1007
1008        if (!PageHighMem(page)) {
1009                debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
1010                debug_check_no_obj_freed(page_address(page), PAGE_SIZE);
1011        }
1012        arch_free_page(page, 0);
1013        kernel_map_pages(page, 1, 0);
1014
1015        pcp = &zone_pcp(zone, get_cpu())->pcp;
1016        local_irq_save(flags);
1017        __count_vm_event(PGFREE);
1018        if (cold)
1019                list_add_tail(&page->lru, &pcp->list);
1020        else
1021                list_add(&page->lru, &pcp->list);
1022        set_page_private(page, get_pageblock_migratetype(page));
1023        pcp->count++;
1024        if (pcp->count >= pcp->high) {
1025                free_pages_bulk(zone, pcp->batch, &pcp->list, 0);
1026                pcp->count -= pcp->batch;
1027        }
1028        local_irq_restore(flags);
1029        put_cpu();
1030}
1031
1032void free_hot_page(struct page *page)
1033{
1034        free_hot_cold_page(page, 0);
1035}
1036        
1037void free_cold_page(struct page *page)
1038{
1039        free_hot_cold_page(page, 1);
1040}
1041
1042/*
1043 * split_page takes a non-compound higher-order page, and splits it into
1044 * n (1<<order) sub-pages: page[0..n]
1045 * Each sub-page must be freed individually.
1046 *
1047 * Note: this is probably too low level an operation for use in drivers.
1048 * Please consult with lkml before using this in your driver.
1049 */
1050void split_page(struct page *page, unsigned int order)
1051{
1052        int i;
1053
1054        VM_BUG_ON(PageCompound(page));
1055        VM_BUG_ON(!page_count(page));
1056        for (i = 1; i < (1 << order); i++)
1057                set_page_refcounted(page + i);
1058}
1059
1060/*
1061 * Really, prep_compound_page() should be called from __rmqueue_bulk().  But
1062 * we cheat by calling it from here, in the order > 0 path.  Saves a branch
1063 * or two.
1064 */
1065static struct page *buffered_rmqueue(struct zone *preferred_zone,
1066                        struct zone *zone, int order, gfp_t gfp_flags)
1067{
1068        unsigned long flags;
1069        struct page *page;
1070        int cold = !!(gfp_flags & __GFP_COLD);
1071        int cpu;
1072        int migratetype = allocflags_to_migratetype(gfp_flags);
1073
1074again:
1075        cpu  = get_cpu();
1076        if (likely(order == 0)) {
1077                struct per_cpu_pages *pcp;
1078
1079                pcp = &zone_pcp(zone, cpu)->pcp;
1080                local_irq_save(flags);
1081                if (!pcp->count) {
1082                        pcp->count = rmqueue_bulk(zone, 0,
1083                                        pcp->batch, &pcp->list,
1084                                        migratetype, cold);
1085                        if (unlikely(!pcp->count))
1086                                goto failed;
1087                }
1088
1089                /* Find a page of the appropriate migrate type */
1090                if (cold) {
1091                        list_for_each_entry_reverse(page, &pcp->list, lru)
1092                                if (page_private(page) == migratetype)
1093                                        break;
1094                } else {
1095                        list_for_each_entry(page, &pcp->list, lru)
1096                                if (page_private(page) == migratetype)
1097                                        break;
1098                }
1099
1100                /* Allocate more to the pcp list if necessary */
1101                if (unlikely(&page->lru == &pcp->list)) {
1102                        pcp->count += rmqueue_bulk(zone, 0,
1103                                        pcp->batch, &pcp->list,
1104                                        migratetype, cold);
1105                        page = list_entry(pcp->list.next, struct page, lru);
1106                }
1107
1108                list_del(&page->lru);
1109                pcp->count--;
1110        } else {
1111                spin_lock_irqsave(&zone->lock, flags);
1112                page = __rmqueue(zone, order, migratetype);
1113                spin_unlock(&zone->lock);
1114                if (!page)
1115                        goto failed;
1116        }
1117
1118        __count_zone_vm_events(PGALLOC, zone, 1 << order);
1119        zone_statistics(preferred_zone, zone);
1120        local_irq_restore(flags);
1121        put_cpu();
1122
1123        VM_BUG_ON(bad_range(zone, page));
1124        if (prep_new_page(page, order, gfp_flags))
1125                goto again;
1126        return page;
1127
1128failed:
1129        local_irq_restore(flags);
1130        put_cpu();
1131        return NULL;
1132}
1133
1134#define ALLOC_NO_WATERMARKS     0x01 /* don't check watermarks at all */
1135#define ALLOC_WMARK_MIN         0x02 /* use pages_min watermark */
1136#define ALLOC_WMARK_LOW         0x04 /* use pages_low watermark */
1137#define ALLOC_WMARK_HIGH        0x08 /* use pages_high watermark */
1138#define ALLOC_HARDER            0x10 /* try to alloc harder */
1139#define ALLOC_HIGH              0x20 /* __GFP_HIGH set */
1140#define ALLOC_CPUSET            0x40 /* check for correct cpuset */
1141
1142#ifdef CONFIG_FAIL_PAGE_ALLOC
1143
1144static struct fail_page_alloc_attr {
1145        struct fault_attr attr;
1146
1147        u32 ignore_gfp_highmem;
1148        u32 ignore_gfp_wait;
1149        u32 min_order;
1150
1151#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1152
1153        struct dentry *ignore_gfp_highmem_file;
1154        struct dentry *ignore_gfp_wait_file;
1155        struct dentry *min_order_file;
1156
1157#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1158
1159} fail_page_alloc = {
1160        .attr = FAULT_ATTR_INITIALIZER,
1161        .ignore_gfp_wait = 1,
1162        .ignore_gfp_highmem = 1,
1163        .min_order = 1,
1164};
1165
1166static int __init setup_fail_page_alloc(char *str)
1167{
1168        return setup_fault_attr(&fail_page_alloc.attr, str);
1169}
1170__setup("fail_page_alloc=", setup_fail_page_alloc);
1171
1172static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1173{
1174        if (order < fail_page_alloc.min_order)
1175                return 0;
1176        if (gfp_mask & __GFP_NOFAIL)
1177                return 0;
1178        if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1179                return 0;
1180        if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1181                return 0;
1182
1183        return should_fail(&fail_page_alloc.attr, 1 << order);
1184}
1185
1186#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1187
1188static int __init fail_page_alloc_debugfs(void)
1189{
1190        mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1191        struct dentry *dir;
1192        int err;
1193
1194        err = init_fault_attr_dentries(&fail_page_alloc.attr,
1195                                       "fail_page_alloc");
1196        if (err)
1197                return err;
1198        dir = fail_page_alloc.attr.dentries.dir;
1199
1200        fail_page_alloc.ignore_gfp_wait_file =
1201                debugfs_create_bool("ignore-gfp-wait", mode, dir,
1202                                      &fail_page_alloc.ignore_gfp_wait);
1203
1204        fail_page_alloc.ignore_gfp_highmem_file =
1205                debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1206                                      &fail_page_alloc.ignore_gfp_highmem);
1207        fail_page_alloc.min_order_file =
1208                debugfs_create_u32("min-order", mode, dir,
1209                                   &fail_page_alloc.min_order);
1210
1211        if (!fail_page_alloc.ignore_gfp_wait_file ||
1212            !fail_page_alloc.ignore_gfp_highmem_file ||
1213            !fail_page_alloc.min_order_file) {
1214                err = -ENOMEM;
1215                debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1216                debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
1217                debugfs_remove(fail_page_alloc.min_order_file);
1218                cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1219        }
1220
1221        return err;
1222}
1223
1224late_initcall(fail_page_alloc_debugfs);
1225
1226#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1227
1228#else /* CONFIG_FAIL_PAGE_ALLOC */
1229
1230static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1231{
1232        return 0;
1233}
1234
1235#endif /* CONFIG_FAIL_PAGE_ALLOC */
1236
1237/*
1238 * Return 1 if free pages are above 'mark'. This takes into account the order
1239 * of the allocation.
1240 */
1241int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1242                      int classzone_idx, int alloc_flags)
1243{
1244        /* free_pages my go negative - that's OK */
1245        long min = mark;
1246        long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
1247        int o;
1248
1249        if (alloc_flags & ALLOC_HIGH)
1250                min -= min / 2;
1251        if (alloc_flags & ALLOC_HARDER)
1252                min -= min / 4;
1253
1254        if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1255                return 0;
1256        for (o = 0; o < order; o++) {
1257                /* At the next order, this order's pages become unavailable */
1258                free_pages -= z->free_area[o].nr_free << o;
1259
1260                /* Require fewer higher order pages to be free */
1261                min >>= 1;
1262
1263                if (free_pages <= min)
1264                        return 0;
1265        }
1266        return 1;
1267}
1268
1269#ifdef CONFIG_NUMA
1270/*
1271 * zlc_setup - Setup for "zonelist cache".  Uses cached zone data to
1272 * skip over zones that are not allowed by the cpuset, or that have
1273 * been recently (in last second) found to be nearly full.  See further
1274 * comments in mmzone.h.  Reduces cache footprint of zonelist scans
1275 * that have to skip over a lot of full or unallowed zones.
1276 *
1277 * If the zonelist cache is present in the passed in zonelist, then
1278 * returns a pointer to the allowed node mask (either the current
1279 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1280 *
1281 * If the zonelist cache is not available for this zonelist, does
1282 * nothing and returns NULL.
1283 *
1284 * If the fullzones BITMAP in the zonelist cache is stale (more than
1285 * a second since last zap'd) then we zap it out (clear its bits.)
1286 *
1287 * We hold off even calling zlc_setup, until after we've checked the
1288 * first zone in the zonelist, on the theory that most allocations will
1289 * be satisfied from that first zone, so best to examine that zone as
1290 * quickly as we can.
1291 */
1292static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1293{
1294        struct zonelist_cache *zlc;     /* cached zonelist speedup info */
1295        nodemask_t *allowednodes;       /* zonelist_cache approximation */
1296
1297        zlc = zonelist->zlcache_ptr;
1298        if (!zlc)
1299                return NULL;
1300
1301        if (time_after(jiffies, zlc->last_full_zap + HZ)) {
1302                bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1303                zlc->last_full_zap = jiffies;
1304        }
1305
1306        allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1307                                        &cpuset_current_mems_allowed :
1308                                        &node_states[N_HIGH_MEMORY];
1309        return allowednodes;
1310}
1311
1312/*
1313 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1314 * if it is worth looking at further for free memory:
1315 *  1) Check that the zone isn't thought to be full (doesn't have its
1316 *     bit set in the zonelist_cache fullzones BITMAP).
1317 *  2) Check that the zones node (obtained from the zonelist_cache
1318 *     z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1319 * Return true (non-zero) if zone is worth looking at further, or
1320 * else return false (zero) if it is not.
1321 *
1322 * This check -ignores- the distinction between various watermarks,
1323 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ...  If a zone is
1324 * found to be full for any variation of these watermarks, it will
1325 * be considered full for up to one second by all requests, unless
1326 * we are so low on memory on all allowed nodes that we are forced
1327 * into the second scan of the zonelist.
1328 *
1329 * In the second scan we ignore this zonelist cache and exactly
1330 * apply the watermarks to all zones, even it is slower to do so.
1331 * We are low on memory in the second scan, and should leave no stone
1332 * unturned looking for a free page.
1333 */
1334static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1335                                                nodemask_t *allowednodes)
1336{
1337        struct zonelist_cache *zlc;     /* cached zonelist speedup info */
1338        int i;                          /* index of *z in zonelist zones */
1339        int n;                          /* node that zone *z is on */
1340
1341        zlc = zonelist->zlcache_ptr;
1342        if (!zlc)
1343                return 1;
1344
1345        i = z - zonelist->_zonerefs;
1346        n = zlc->z_to_n[i];
1347
1348        /* This zone is worth trying if it is allowed but not full */
1349        return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1350}
1351
1352/*
1353 * Given 'z' scanning a zonelist, set the corresponding bit in
1354 * zlc->fullzones, so that subsequent attempts to allocate a page
1355 * from that zone don't waste time re-examining it.
1356 */
1357static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1358{
1359        struct zonelist_cache *zlc;     /* cached zonelist speedup info */
1360        int i;                          /* index of *z in zonelist zones */
1361
1362        zlc = zonelist->zlcache_ptr;
1363        if (!zlc)
1364                return;
1365
1366        i = z - zonelist->_zonerefs;
1367
1368        set_bit(i, zlc->fullzones);
1369}
1370
1371#else   /* CONFIG_NUMA */
1372
1373static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1374{
1375        return NULL;
1376}
1377
1378static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1379                                nodemask_t *allowednodes)
1380{
1381        return 1;
1382}
1383
1384static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1385{
1386}
1387#endif  /* CONFIG_NUMA */
1388
1389/*
1390 * get_page_from_freelist goes through the zonelist trying to allocate
1391 * a page.
1392 */
1393static struct page *
1394get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
1395                struct zonelist *zonelist, int high_zoneidx, int alloc_flags)
1396{
1397        struct zoneref *z;
1398        struct page *page = NULL;
1399        int classzone_idx;
1400        struct zone *zone, *preferred_zone;
1401        nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1402        int zlc_active = 0;             /* set if using zonelist_cache */
1403        int did_zlc_setup = 0;          /* just call zlc_setup() one time */
1404
1405        (void)first_zones_zonelist(zonelist, high_zoneidx, nodemask,
1406                                                        &preferred_zone);
1407        if (!preferred_zone)
1408                return NULL;
1409
1410        classzone_idx = zone_idx(preferred_zone);
1411
1412zonelist_scan:
1413        /*
1414         * Scan zonelist, looking for a zone with enough free.
1415         * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1416         */
1417        for_each_zone_zonelist_nodemask(zone, z, zonelist,
1418                                                high_zoneidx, nodemask) {
1419                if (NUMA_BUILD && zlc_active &&
1420                        !zlc_zone_worth_trying(zonelist, z, allowednodes))
1421                                continue;
1422                if ((alloc_flags & ALLOC_CPUSET) &&
1423                        !cpuset_zone_allowed_softwall(zone, gfp_mask))
1424                                goto try_next_zone;
1425
1426                if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1427                        unsigned long mark;
1428                        int ret;
1429                        if (alloc_flags & ALLOC_WMARK_MIN)
1430                                mark = zone->pages_min;
1431                        else if (alloc_flags & ALLOC_WMARK_LOW)
1432                                mark = zone->pages_low;
1433                        else
1434                                mark = zone->pages_high;
1435
1436                        if (zone_watermark_ok(zone, order, mark,
1437                                    classzone_idx, alloc_flags))
1438                                goto try_this_zone;
1439
1440                        if (zone_reclaim_mode == 0)
1441                                goto this_zone_full;
1442
1443                        ret = zone_reclaim(zone, gfp_mask, order);
1444                        switch (ret) {
1445                        case ZONE_RECLAIM_NOSCAN:
1446                                /* did not scan */
1447                                goto try_next_zone;
1448                        case ZONE_RECLAIM_FULL:
1449                                /* scanned but unreclaimable */
1450                                goto this_zone_full;
1451                        default:
1452                                /* did we reclaim enough */
1453                                if (!zone_watermark_ok(zone, order, mark,
1454                                                classzone_idx, alloc_flags))
1455                                        goto this_zone_full;
1456                        }
1457                }
1458
1459try_this_zone:
1460                page = buffered_rmqueue(preferred_zone, zone, order, gfp_mask);
1461                if (page)
1462                        break;
1463this_zone_full:
1464                if (NUMA_BUILD)
1465                        zlc_mark_zone_full(zonelist, z);
1466try_next_zone:
1467                if (NUMA_BUILD && !did_zlc_setup) {
1468                        /* we do zlc_setup after the first zone is tried */
1469                        allowednodes = zlc_setup(zonelist, alloc_flags);
1470                        zlc_active = 1;
1471                        did_zlc_setup = 1;
1472                }
1473        }
1474
1475        if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1476                /* Disable zlc cache for second zonelist scan */
1477                zlc_active = 0;
1478                goto zonelist_scan;
1479        }
1480        return page;
1481}
1482
1483/*
1484 * This is the 'heart' of the zoned buddy allocator.
1485 */
1486struct page *
1487__alloc_pages_internal(gfp_t gfp_mask, unsigned int order,
1488                        struct zonelist *zonelist, nodemask_t *nodemask)
1489{
1490        const gfp_t wait = gfp_mask & __GFP_WAIT;
1491        enum zone_type high_zoneidx = gfp_zone(gfp_mask);
1492        struct zoneref *z;
1493        struct zone *zone;
1494        struct page *page;
1495        struct reclaim_state reclaim_state;
1496        struct task_struct *p = current;
1497        int do_retry;
1498        int alloc_flags;
1499        unsigned long did_some_progress;
1500        unsigned long pages_reclaimed = 0;
1501
1502        lockdep_trace_alloc(gfp_mask);
1503
1504        might_sleep_if(wait);
1505
1506        if (should_fail_alloc_page(gfp_mask, order))
1507                return NULL;
1508
1509restart:
1510        z = zonelist->_zonerefs;  /* the list of zones suitable for gfp_mask */
1511
1512        if (unlikely(!z->zone)) {
1513                /*
1514                 * Happens if we have an empty zonelist as a result of
1515                 * GFP_THISNODE being used on a memoryless node
1516                 */
1517                return NULL;
1518        }
1519
1520        page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
1521                        zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET);
1522        if (page)
1523                goto got_pg;
1524
1525        /*
1526         * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1527         * __GFP_NOWARN set) should not cause reclaim since the subsystem
1528         * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1529         * using a larger set of nodes after it has established that the
1530         * allowed per node queues are empty and that nodes are
1531         * over allocated.
1532         */
1533        if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1534                goto nopage;
1535
1536        for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
1537                wakeup_kswapd(zone, order);
1538
1539        /*
1540         * OK, we're below the kswapd watermark and have kicked background
1541         * reclaim. Now things get more complex, so set up alloc_flags according
1542         * to how we want to proceed.
1543         *
1544         * The caller may dip into page reserves a bit more if the caller
1545         * cannot run direct reclaim, or if the caller has realtime scheduling
1546         * policy or is asking for __GFP_HIGH memory.  GFP_ATOMIC requests will
1547         * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1548         */
1549        alloc_flags = ALLOC_WMARK_MIN;
1550        if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1551                alloc_flags |= ALLOC_HARDER;
1552        if (gfp_mask & __GFP_HIGH)
1553                alloc_flags |= ALLOC_HIGH;
1554        if (wait)
1555                alloc_flags |= ALLOC_CPUSET;
1556
1557        /*
1558         * Go through the zonelist again. Let __GFP_HIGH and allocations
1559         * coming from realtime tasks go deeper into reserves.
1560         *
1561         * This is the last chance, in general, before the goto nopage.
1562         * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1563         * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1564         */
1565        page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
1566                                                high_zoneidx, alloc_flags);
1567        if (page)
1568                goto got_pg;
1569
1570        /* This allocation should allow future memory freeing. */
1571
1572rebalance:
1573        if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1574                        && !in_interrupt()) {
1575                if (!(gfp_mask & __GFP_NOMEMALLOC)) {
1576nofail_alloc:
1577                        /* go through the zonelist yet again, ignoring mins */
1578                        page = get_page_from_freelist(gfp_mask, nodemask, order,
1579                                zonelist, high_zoneidx, ALLOC_NO_WATERMARKS);
1580                        if (page)
1581                                goto got_pg;
1582                        if (gfp_mask & __GFP_NOFAIL) {
1583                                congestion_wait(WRITE, HZ/50);
1584                                goto nofail_alloc;
1585                        }
1586                }
1587                goto nopage;
1588        }
1589
1590        /* Atomic allocations - we can't balance anything */
1591        if (!wait)
1592                goto nopage;
1593
1594        cond_resched();
1595
1596        /* We now go into synchronous reclaim */
1597        cpuset_memory_pressure_bump();
1598        /*
1599         * The task's cpuset might have expanded its set of allowable nodes
1600         */
1601        cpuset_update_task_memory_state();
1602        p->flags |= PF_MEMALLOC;
1603
1604        lockdep_set_current_reclaim_state(gfp_mask);
1605        reclaim_state.reclaimed_slab = 0;
1606        p->reclaim_state = &reclaim_state;
1607
1608        did_some_progress = try_to_free_pages(zonelist, order,
1609                                                gfp_mask, nodemask);
1610
1611        p->reclaim_state = NULL;
1612        lockdep_clear_current_reclaim_state();
1613        p->flags &= ~PF_MEMALLOC;
1614
1615        cond_resched();
1616
1617        if (order != 0)
1618                drain_all_pages();
1619
1620        if (likely(did_some_progress)) {
1621                page = get_page_from_freelist(gfp_mask, nodemask, order,
1622                                        zonelist, high_zoneidx, alloc_flags);
1623                if (page)
1624                        goto got_pg;
1625        } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1626                if (!try_set_zone_oom(zonelist, gfp_mask)) {
1627                        schedule_timeout_uninterruptible(1);
1628                        goto restart;
1629                }
1630
1631                /*
1632                 * Go through the zonelist yet one more time, keep
1633                 * very high watermark here, this is only to catch
1634                 * a parallel oom killing, we must fail if we're still
1635                 * under heavy pressure.
1636                 */
1637                page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1638                        order, zonelist, high_zoneidx,
1639                        ALLOC_WMARK_HIGH|ALLOC_CPUSET);
1640                if (page) {
1641                        clear_zonelist_oom(zonelist, gfp_mask);
1642                        goto got_pg;
1643                }
1644
1645                /* The OOM killer will not help higher order allocs so fail */
1646                if (order > PAGE_ALLOC_COSTLY_ORDER) {
1647                        clear_zonelist_oom(zonelist, gfp_mask);
1648                        goto nopage;
1649                }
1650
1651                out_of_memory(zonelist, gfp_mask, order);
1652                clear_zonelist_oom(zonelist, gfp_mask);
1653                goto restart;
1654        }
1655
1656        /*
1657         * Don't let big-order allocations loop unless the caller explicitly
1658         * requests that.  Wait for some write requests to complete then retry.
1659         *
1660         * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1661         * means __GFP_NOFAIL, but that may not be true in other
1662         * implementations.
1663         *
1664         * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1665         * specified, then we retry until we no longer reclaim any pages
1666         * (above), or we've reclaimed an order of pages at least as
1667         * large as the allocation's order. In both cases, if the
1668         * allocation still fails, we stop retrying.
1669         */
1670        pages_reclaimed += did_some_progress;
1671        do_retry = 0;
1672        if (!(gfp_mask & __GFP_NORETRY)) {
1673                if (order <= PAGE_ALLOC_COSTLY_ORDER) {
1674                        do_retry = 1;
1675                } else {
1676                        if (gfp_mask & __GFP_REPEAT &&
1677                                pages_reclaimed < (1 << order))
1678                                        do_retry = 1;
1679                }
1680                if (gfp_mask & __GFP_NOFAIL)
1681                        do_retry = 1;
1682        }
1683        if (do_retry) {
1684                congestion_wait(WRITE, HZ/50);
1685                goto rebalance;
1686        }
1687
1688nopage:
1689        if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1690                printk(KERN_WARNING "%s: page allocation failure."
1691                        " order:%d, mode:0x%x\n",
1692                        p->comm, order, gfp_mask);
1693                dump_stack();
1694                show_mem();
1695        }
1696got_pg:
1697        return page;
1698}
1699EXPORT_SYMBOL(__alloc_pages_internal);
1700
1701/*
1702 * Common helper functions.
1703 */
1704unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1705{
1706        struct page * page;
1707        page = alloc_pages(gfp_mask, order);
1708        if (!page)
1709                return 0;
1710        return (unsigned long) page_address(page);
1711}
1712
1713EXPORT_SYMBOL(__get_free_pages);
1714
1715unsigned long get_zeroed_page(gfp_t gfp_mask)
1716{
1717        struct page * page;
1718
1719        /*
1720         * get_zeroed_page() returns a 32-bit address, which cannot represent
1721         * a highmem page
1722         */
1723        VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1724
1725        page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1726        if (page)
1727                return (unsigned long) page_address(page);
1728        return 0;
1729}
1730
1731EXPORT_SYMBOL(get_zeroed_page);
1732
1733void __pagevec_free(struct pagevec *pvec)
1734{
1735        int i = pagevec_count(pvec);
1736
1737        while (--i >= 0)
1738                free_hot_cold_page(pvec->pages[i], pvec->cold);
1739}
1740
1741void __free_pages(struct page *page, unsigned int order)
1742{
1743        if (put_page_testzero(page)) {
1744                if (order == 0)
1745                        free_hot_page(page);
1746                else
1747                        __free_pages_ok(page, order);
1748        }
1749}
1750
1751EXPORT_SYMBOL(__free_pages);
1752
1753void free_pages(unsigned long addr, unsigned int order)
1754{
1755        if (addr != 0) {
1756                VM_BUG_ON(!virt_addr_valid((void *)addr));
1757                __free_pages(virt_to_page((void *)addr), order);
1758        }
1759}
1760
1761EXPORT_SYMBOL(free_pages);
1762
1763/**
1764 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1765 * @size: the number of bytes to allocate
1766 * @gfp_mask: GFP flags for the allocation
1767 *
1768 * This function is similar to alloc_pages(), except that it allocates the
1769 * minimum number of pages to satisfy the request.  alloc_pages() can only
1770 * allocate memory in power-of-two pages.
1771 *
1772 * This function is also limited by MAX_ORDER.
1773 *
1774 * Memory allocated by this function must be released by free_pages_exact().
1775 */
1776void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
1777{
1778        unsigned int order = get_order(size);
1779        unsigned long addr;
1780
1781        addr = __get_free_pages(gfp_mask, order);
1782        if (addr) {
1783                unsigned long alloc_end = addr + (PAGE_SIZE << order);
1784                unsigned long used = addr + PAGE_ALIGN(size);
1785
1786                split_page(virt_to_page(addr), order);
1787                while (used < alloc_end) {
1788                        free_page(used);
1789                        used += PAGE_SIZE;
1790                }
1791        }
1792
1793        return (void *)addr;
1794}
1795EXPORT_SYMBOL(alloc_pages_exact);
1796
1797/**
1798 * free_pages_exact - release memory allocated via alloc_pages_exact()
1799 * @virt: the value returned by alloc_pages_exact.
1800 * @size: size of allocation, same value as passed to alloc_pages_exact().
1801 *
1802 * Release the memory allocated by a previous call to alloc_pages_exact.
1803 */
1804void free_pages_exact(void *virt, size_t size)
1805{
1806        unsigned long addr = (unsigned long)virt;
1807        unsigned long end = addr + PAGE_ALIGN(size);
1808
1809        while (addr < end) {
1810                free_page(addr);
1811                addr += PAGE_SIZE;
1812        }
1813}
1814EXPORT_SYMBOL(free_pages_exact);
1815
1816static unsigned int nr_free_zone_pages(int offset)
1817{
1818        struct zoneref *z;
1819        struct zone *zone;
1820
1821        /* Just pick one node, since fallback list is circular */
1822        unsigned int sum = 0;
1823
1824        struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
1825
1826        for_each_zone_zonelist(zone, z, zonelist, offset) {
1827                unsigned long size = zone->present_pages;
1828                unsigned long high = zone->pages_high;
1829                if (size > high)
1830                        sum += size - high;
1831        }
1832
1833        return sum;
1834}
1835
1836/*
1837 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1838 */
1839unsigned int nr_free_buffer_pages(void)
1840{
1841        return nr_free_zone_pages(gfp_zone(GFP_USER));
1842}
1843EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
1844
1845/*
1846 * Amount of free RAM allocatable within all zones
1847 */
1848unsigned int nr_free_pagecache_pages(void)
1849{
1850        return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
1851}
1852
1853static inline void show_node(struct zone *zone)
1854{
1855        if (NUMA_BUILD)
1856                printk("Node %d ", zone_to_nid(zone));
1857}
1858
1859void si_meminfo(struct sysinfo *val)
1860{
1861        val->totalram = totalram_pages;
1862        val->sharedram = 0;
1863        val->freeram = global_page_state(NR_FREE_PAGES);
1864        val->bufferram = nr_blockdev_pages();
1865        val->totalhigh = totalhigh_pages;
1866        val->freehigh = nr_free_highpages();
1867        val->mem_unit = PAGE_SIZE;
1868}
1869
1870EXPORT_SYMBOL(si_meminfo);
1871
1872#ifdef CONFIG_NUMA
1873void si_meminfo_node(struct sysinfo *val, int nid)
1874{
1875        pg_data_t *pgdat = NODE_DATA(nid);
1876
1877        val->totalram = pgdat->node_present_pages;
1878        val->freeram = node_page_state(nid, NR_FREE_PAGES);
1879#ifdef CONFIG_HIGHMEM
1880        val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1881        val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
1882                        NR_FREE_PAGES);
1883#else
1884        val->totalhigh = 0;
1885        val->freehigh = 0;
1886#endif
1887        val->mem_unit = PAGE_SIZE;
1888}
1889#endif
1890
1891#define K(x) ((x) << (PAGE_SHIFT-10))
1892
1893/*
1894 * Show free area list (used inside shift_scroll-lock stuff)
1895 * We also calculate the percentage fragmentation. We do this by counting the
1896 * memory on each free list with the exception of the first item on the list.
1897 */
1898void show_free_areas(void)
1899{
1900        int cpu;
1901        struct zone *zone;
1902
1903        for_each_populated_zone(zone) {
1904                show_node(zone);
1905                printk("%s per-cpu:\n", zone->name);
1906
1907                for_each_online_cpu(cpu) {
1908                        struct per_cpu_pageset *pageset;
1909
1910                        pageset = zone_pcp(zone, cpu);
1911
1912                        printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
1913                               cpu, pageset->pcp.high,
1914                               pageset->pcp.batch, pageset->pcp.count);
1915                }
1916        }
1917
1918        printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
1919                " inactive_file:%lu"
1920//TODO:  check/adjust line lengths
1921#ifdef CONFIG_UNEVICTABLE_LRU
1922                " unevictable:%lu"
1923#endif
1924                " dirty:%lu writeback:%lu unstable:%lu\n"
1925                " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1926                global_page_state(NR_ACTIVE_ANON),
1927                global_page_state(NR_ACTIVE_FILE),
1928                global_page_state(NR_INACTIVE_ANON),
1929                global_page_state(NR_INACTIVE_FILE),
1930#ifdef CONFIG_UNEVICTABLE_LRU
1931                global_page_state(NR_UNEVICTABLE),
1932#endif
1933                global_page_state(NR_FILE_DIRTY),
1934                global_page_state(NR_WRITEBACK),
1935                global_page_state(NR_UNSTABLE_NFS),
1936                global_page_state(NR_FREE_PAGES),
1937                global_page_state(NR_SLAB_RECLAIMABLE) +
1938                        global_page_state(NR_SLAB_UNRECLAIMABLE),
1939                global_page_state(NR_FILE_MAPPED),
1940                global_page_state(NR_PAGETABLE),
1941                global_page_state(NR_BOUNCE));
1942
1943        for_each_populated_zone(zone) {
1944                int i;
1945
1946                show_node(zone);
1947                printk("%s"
1948                        " free:%lukB"
1949                        " min:%lukB"
1950                        " low:%lukB"
1951                        " high:%lukB"
1952                        " active_anon:%lukB"
1953                        " inactive_anon:%lukB"
1954                        " active_file:%lukB"
1955                        " inactive_file:%lukB"
1956#ifdef CONFIG_UNEVICTABLE_LRU
1957                        " unevictable:%lukB"
1958#endif
1959                        " present:%lukB"
1960                        " pages_scanned:%lu"
1961                        " all_unreclaimable? %s"
1962                        "\n",
1963                        zone->name,
1964                        K(zone_page_state(zone, NR_FREE_PAGES)),
1965                        K(zone->pages_min),
1966                        K(zone->pages_low),
1967                        K(zone->pages_high),
1968                        K(zone_page_state(zone, NR_ACTIVE_ANON)),
1969                        K(zone_page_state(zone, NR_INACTIVE_ANON)),
1970                        K(zone_page_state(zone, NR_ACTIVE_FILE)),
1971                        K(zone_page_state(zone, NR_INACTIVE_FILE)),
1972#ifdef CONFIG_UNEVICTABLE_LRU
1973                        K(zone_page_state(zone, NR_UNEVICTABLE)),
1974#endif
1975                        K(zone->present_pages),
1976                        zone->pages_scanned,
1977                        (zone_is_all_unreclaimable(zone) ? "yes" : "no")
1978                        );
1979                printk("lowmem_reserve[]:");
1980                for (i = 0; i < MAX_NR_ZONES; i++)
1981                        printk(" %lu", zone->lowmem_reserve[i]);
1982                printk("\n");
1983        }
1984
1985        for_each_populated_zone(zone) {
1986                unsigned long nr[MAX_ORDER], flags, order, total = 0;
1987
1988                show_node(zone);
1989                printk("%s: ", zone->name);
1990
1991                spin_lock_irqsave(&zone->lock, flags);
1992                for (order = 0; order < MAX_ORDER; order++) {
1993                        nr[order] = zone->free_area[order].nr_free;
1994                        total += nr[order] << order;
1995                }
1996                spin_unlock_irqrestore(&zone->lock, flags);
1997                for (order = 0; order < MAX_ORDER; order++)
1998                        printk("%lu*%lukB ", nr[order], K(1UL) << order);
1999                printk("= %lukB\n", K(total));
2000        }
2001
2002        printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
2003
2004        show_swap_cache_info();
2005}
2006
2007static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
2008{
2009        zoneref->zone = zone;
2010        zoneref->zone_idx = zone_idx(zone);
2011}
2012
2013/*
2014 * Builds allocation fallback zone lists.
2015 *
2016 * Add all populated zones of a node to the zonelist.
2017 */
2018static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2019                                int nr_zones, enum zone_type zone_type)
2020{
2021        struct zone *zone;
2022
2023        BUG_ON(zone_type >= MAX_NR_ZONES);
2024        zone_type++;
2025
2026        do {
2027                zone_type--;
2028                zone = pgdat->node_zones + zone_type;
2029                if (populated_zone(zone)) {
2030                        zoneref_set_zone(zone,
2031                                &zonelist->_zonerefs[nr_zones++]);
2032                        check_highest_zone(zone_type);
2033                }
2034
2035        } while (zone_type);
2036        return nr_zones;
2037}
2038
2039
2040/*
2041 *  zonelist_order:
2042 *  0 = automatic detection of better ordering.
2043 *  1 = order by ([node] distance, -zonetype)
2044 *  2 = order by (-zonetype, [node] distance)
2045 *
2046 *  If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2047 *  the same zonelist. So only NUMA can configure this param.
2048 */
2049#define ZONELIST_ORDER_DEFAULT  0
2050#define ZONELIST_ORDER_NODE     1
2051#define ZONELIST_ORDER_ZONE     2
2052
2053/* zonelist order in the kernel.
2054 * set_zonelist_order() will set this to NODE or ZONE.
2055 */
2056static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2057static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2058
2059
2060#ifdef CONFIG_NUMA
2061/* The value user specified ....changed by config */
2062static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2063/* string for sysctl */
2064#define NUMA_ZONELIST_ORDER_LEN 16
2065char numa_zonelist_order[16] = "default";
2066
2067/*
2068 * interface for configure zonelist ordering.
2069 * command line option "numa_zonelist_order"
2070 *      = "[dD]efault   - default, automatic configuration.
2071 *      = "[nN]ode      - order by node locality, then by zone within node
2072 *      = "[zZ]one      - order by zone, then by locality within zone
2073 */
2074
2075static int __parse_numa_zonelist_order(char *s)
2076{
2077        if (*s == 'd' || *s == 'D') {
2078                user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2079        } else if (*s == 'n' || *s == 'N') {
2080                user_zonelist_order = ZONELIST_ORDER_NODE;
2081        } else if (*s == 'z' || *s == 'Z') {
2082                user_zonelist_order = ZONELIST_ORDER_ZONE;
2083        } else {
2084                printk(KERN_WARNING
2085                        "Ignoring invalid numa_zonelist_order value:  "
2086                        "%s\n", s);
2087                return -EINVAL;
2088        }
2089        return 0;
2090}
2091
2092static __init int setup_numa_zonelist_order(char *s)
2093{
2094        if (s)
2095                return __parse_numa_zonelist_order(s);
2096        return 0;
2097}
2098early_param("numa_zonelist_order", setup_numa_zonelist_order);
2099
2100/*
2101 * sysctl handler for numa_zonelist_order
2102 */
2103int numa_zonelist_order_handler(ctl_table *table, int write,
2104                struct file *file, void __user *buffer, size_t *length,
2105                loff_t *ppos)
2106{
2107        char saved_string[NUMA_ZONELIST_ORDER_LEN];
2108        int ret;
2109
2110        if (write)
2111                strncpy(saved_string, (char*)table->data,
2112                        NUMA_ZONELIST_ORDER_LEN);
2113        ret = proc_dostring(table, write, file, buffer, length, ppos);
2114        if (ret)
2115                return ret;
2116        if (write) {
2117                int oldval = user_zonelist_order;
2118                if (__parse_numa_zonelist_order((char*)table->data)) {
2119                        /*
2120                         * bogus value.  restore saved string
2121                         */
2122                        strncpy((char*)table->data, saved_string,
2123                                NUMA_ZONELIST_ORDER_LEN);
2124                        user_zonelist_order = oldval;
2125                } else if (oldval != user_zonelist_order)
2126                        build_all_zonelists();
2127        }
2128        return 0;
2129}
2130
2131
2132#define MAX_NODE_LOAD (num_online_nodes())
2133static int node_load[MAX_NUMNODES];
2134
2135/**
2136 * find_next_best_node - find the next node that should appear in a given node's fallback list
2137 * @node: node whose fallback list we're appending
2138 * @used_node_mask: nodemask_t of already used nodes
2139 *
2140 * We use a number of factors to determine which is the next node that should
2141 * appear on a given node's fallback list.  The node should not have appeared
2142 * already in @node's fallback list, and it should be the next closest node
2143 * according to the distance array (which contains arbitrary distance values
2144 * from each node to each node in the system), and should also prefer nodes
2145 * with no CPUs, since presumably they'll have very little allocation pressure
2146 * on them otherwise.
2147 * It returns -1 if no node is found.
2148 */
2149static int find_next_best_node(int node, nodemask_t *used_node_mask)
2150{
2151        int n, val;
2152        int min_val = INT_MAX;
2153        int best_node = -1;
2154        const struct cpumask *tmp = cpumask_of_node(0);
2155
2156        /* Use the local node if we haven't already */
2157        if (!node_isset(node, *used_node_mask)) {
2158                node_set(node, *used_node_mask);
2159                return node;
2160        }
2161
2162        for_each_node_state(n, N_HIGH_MEMORY) {
2163
2164                /* Don't want a node to appear more than once */
2165                if (node_isset(n, *used_node_mask))
2166                        continue;
2167
2168                /* Use the distance array to find the distance */
2169                val = node_distance(node, n);
2170
2171                /* Penalize nodes under us ("prefer the next node") */
2172                val += (n < node);
2173
2174                /* Give preference to headless and unused nodes */
2175                tmp = cpumask_of_node(n);
2176                if (!cpumask_empty(tmp))
2177                        val += PENALTY_FOR_NODE_WITH_CPUS;
2178
2179                /* Slight preference for less loaded node */
2180                val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2181                val += node_load[n];
2182
2183                if (val < min_val) {
2184                        min_val = val;
2185                        best_node = n;
2186                }
2187        }
2188
2189        if (best_node >= 0)
2190                node_set(best_node, *used_node_mask);
2191
2192        return best_node;
2193}
2194
2195
2196/*
2197 * Build zonelists ordered by node and zones within node.
2198 * This results in maximum locality--normal zone overflows into local
2199 * DMA zone, if any--but risks exhausting DMA zone.
2200 */
2201static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
2202{
2203        int j;
2204        struct zonelist *zonelist;
2205
2206        zonelist = &pgdat->node_zonelists[0];
2207        for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
2208                ;
2209        j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2210                                                        MAX_NR_ZONES - 1);
2211        zonelist->_zonerefs[j].zone = NULL;
2212        zonelist->_zonerefs[j].zone_idx = 0;
2213}
2214
2215/*
2216 * Build gfp_thisnode zonelists
2217 */
2218static void build_thisnode_zonelists(pg_data_t *pgdat)
2219{
2220        int j;
2221        struct zonelist *zonelist;
2222
2223        zonelist = &pgdat->node_zonelists[1];
2224        j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2225        zonelist->_zonerefs[j].zone = NULL;
2226        zonelist->_zonerefs[j].zone_idx = 0;
2227}
2228
2229/*
2230 * Build zonelists ordered by zone and nodes within zones.
2231 * This results in conserving DMA zone[s] until all Normal memory is
2232 * exhausted, but results in overflowing to remote node while memory
2233 * may still exist in local DMA zone.
2234 */
2235static int node_order[MAX_NUMNODES];
2236
2237static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2238{
2239        int pos, j, node;
2240        int zone_type;          /* needs to be signed */
2241        struct zone *z;
2242        struct zonelist *zonelist;
2243
2244        zonelist = &pgdat->node_zonelists[0];
2245        pos = 0;
2246        for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2247                for (j = 0; j < nr_nodes; j++) {
2248                        node = node_order[j];
2249                        z = &NODE_DATA(node)->node_zones[zone_type];
2250                        if (populated_zone(z)) {
2251                                zoneref_set_zone(z,
2252                                        &zonelist->_zonerefs[pos++]);
2253                                check_highest_zone(zone_type);
2254                        }
2255                }
2256        }
2257        zonelist->_zonerefs[pos].zone = NULL;
2258        zonelist->_zonerefs[pos].zone_idx = 0;
2259}
2260
2261static int default_zonelist_order(void)
2262{
2263        int nid, zone_type;
2264        unsigned long low_kmem_size,total_size;
2265        struct zone *z;
2266        int average_size;
2267        /*
2268         * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2269         * If they are really small and used heavily, the system can fall
2270         * into OOM very easily.
2271         * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2272         */
2273        /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2274        low_kmem_size = 0;
2275        total_size = 0;
2276        for_each_online_node(nid) {
2277                for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2278                        z = &NODE_DATA(nid)->node_zones[zone_type];
2279                        if (populated_zone(z)) {
2280                                if (zone_type < ZONE_NORMAL)
2281                                        low_kmem_size += z->present_pages;
2282                                total_size += z->present_pages;
2283                        }
2284                }
2285        }
2286        if (!low_kmem_size ||  /* there are no DMA area. */
2287            low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2288                return ZONELIST_ORDER_NODE;
2289        /*
2290         * look into each node's config.
2291         * If there is a node whose DMA/DMA32 memory is very big area on
2292         * local memory, NODE_ORDER may be suitable.
2293         */
2294        average_size = total_size /
2295                                (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
2296        for_each_online_node(nid) {
2297                low_kmem_size = 0;
2298                total_size = 0;
2299                for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2300                        z = &NODE_DATA(nid)->node_zones[zone_type];
2301                        if (populated_zone(z)) {
2302                                if (zone_type < ZONE_NORMAL)
2303                                        low_kmem_size += z->present_pages;
2304                                total_size += z->present_pages;
2305                        }
2306                }
2307                if (low_kmem_size &&
2308                    total_size > average_size && /* ignore small node */
2309                    low_kmem_size > total_size * 70/100)
2310                        return ZONELIST_ORDER_NODE;
2311        }
2312        return ZONELIST_ORDER_ZONE;
2313}
2314
2315static void set_zonelist_order(void)
2316{
2317        if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2318                current_zonelist_order = default_zonelist_order();
2319        else
2320                current_zonelist_order = user_zonelist_order;
2321}
2322
2323static void build_zonelists(pg_data_t *pgdat)
2324{
2325        int j, node, load;
2326        enum zone_type i;
2327        nodemask_t used_mask;
2328        int local_node, prev_node;
2329        struct zonelist *zonelist;
2330        int order = current_zonelist_order;
2331
2332        /* initialize zonelists */
2333        for (i = 0; i < MAX_ZONELISTS; i++) {
2334                zonelist = pgdat->node_zonelists + i;
2335                zonelist->_zonerefs[0].zone = NULL;
2336                zonelist->_zonerefs[0].zone_idx = 0;
2337        }
2338
2339        /* NUMA-aware ordering of nodes */
2340        local_node = pgdat->node_id;
2341        load = num_online_nodes();
2342        prev_node = local_node;
2343        nodes_clear(used_mask);
2344
2345        memset(node_load, 0, sizeof(node_load));
2346        memset(node_order, 0, sizeof(node_order));
2347        j = 0;
2348
2349        while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
2350                int distance = node_distance(local_node, node);
2351
2352                /*
2353                 * If another node is sufficiently far away then it is better
2354                 * to reclaim pages in a zone before going off node.
2355                 */
2356                if (distance > RECLAIM_DISTANCE)
2357                        zone_reclaim_mode = 1;
2358
2359                /*
2360                 * We don't want to pressure a particular node.
2361                 * So adding penalty to the first node in same
2362                 * distance group to make it round-robin.
2363                 */
2364                if (distance != node_distance(local_node, prev_node))
2365                        node_load[node] = load;
2366
2367                prev_node = node;
2368                load--;
2369                if (order == ZONELIST_ORDER_NODE)
2370                        build_zonelists_in_node_order(pgdat, node);
2371                else
2372                        node_order[j++] = node; /* remember order */
2373        }
2374
2375        if (order == ZONELIST_ORDER_ZONE) {
2376                /* calculate node order -- i.e., DMA last! */
2377                build_zonelists_in_zone_order(pgdat, j);
2378        }
2379
2380        build_thisnode_zonelists(pgdat);
2381}
2382
2383/* Construct the zonelist performance cache - see further mmzone.h */
2384static void build_zonelist_cache(pg_data_t *pgdat)
2385{
2386        struct zonelist *zonelist;
2387        struct zonelist_cache *zlc;
2388        struct zoneref *z;
2389
2390        zonelist = &pgdat->node_zonelists[0];
2391        zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2392        bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2393        for (z = zonelist->_zonerefs; z->zone; z++)
2394                zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
2395}
2396
2397
2398#else   /* CONFIG_NUMA */
2399
2400static void set_zonelist_order(void)
2401{
2402        current_zonelist_order = ZONELIST_ORDER_ZONE;
2403}
2404
2405static void build_zonelists(pg_data_t *pgdat)
2406{
2407        int node, local_node;
2408        enum zone_type j;
2409        struct zonelist *zonelist;
2410
2411        local_node = pgdat->node_id;
2412
2413        zonelist = &pgdat->node_zonelists[0];
2414        j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2415
2416        /*
2417         * Now we build the zonelist so that it contains the zones
2418         * of all the other nodes.
2419         * We don't want to pressure a particular node, so when
2420         * building the zones for node N, we make sure that the
2421         * zones coming right after the local ones are those from
2422         * node N+1 (modulo N)
2423         */
2424        for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2425                if (!node_online(node))
2426                        continue;
2427                j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2428                                                        MAX_NR_ZONES - 1);
2429        }
2430        for (node = 0; node < local_node; node++) {
2431                if (!node_online(node))
2432                        continue;
2433                j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2434                                                        MAX_NR_ZONES - 1);
2435        }
2436
2437        zonelist->_zonerefs[j].zone = NULL;
2438        zonelist->_zonerefs[j].zone_idx = 0;
2439}
2440
2441/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2442static void build_zonelist_cache(pg_data_t *pgdat)
2443{
2444        pgdat->node_zonelists[0].zlcache_ptr = NULL;
2445}
2446
2447#endif  /* CONFIG_NUMA */
2448
2449/* return values int ....just for stop_machine() */
2450static int __build_all_zonelists(void *dummy)
2451{
2452        int nid;
2453
2454        for_each_online_node(nid) {
2455                pg_data_t *pgdat = NODE_DATA(nid);
2456
2457                build_zonelists(pgdat);
2458                build_zonelist_cache(pgdat);
2459        }
2460        return 0;
2461}
2462
2463void build_all_zonelists(void)
2464{
2465        set_zonelist_order();
2466
2467        if (system_state == SYSTEM_BOOTING) {
2468                __build_all_zonelists(NULL);
2469                mminit_verify_zonelist();
2470                cpuset_init_current_mems_allowed();
2471        } else {
2472                /* we have to stop all cpus to guarantee there is no user
2473                   of zonelist */
2474                stop_machine(__build_all_zonelists, NULL, NULL);
2475                /* cpuset refresh routine should be here */
2476        }
2477        vm_total_pages = nr_free_pagecache_pages();
2478        /*
2479         * Disable grouping by mobility if the number of pages in the
2480         * system is too low to allow the mechanism to work. It would be
2481         * more accurate, but expensive to check per-zone. This check is
2482         * made on memory-hotadd so a system can start with mobility
2483         * disabled and enable it later
2484         */
2485        if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
2486                page_group_by_mobility_disabled = 1;
2487        else
2488                page_group_by_mobility_disabled = 0;
2489
2490        printk("Built %i zonelists in %s order, mobility grouping %s.  "
2491                "Total pages: %ld\n",
2492                        num_online_nodes(),
2493                        zonelist_order_name[current_zonelist_order],
2494                        page_group_by_mobility_disabled ? "off" : "on",
2495                        vm_total_pages);
2496#ifdef CONFIG_NUMA
2497        printk("Policy zone: %s\n", zone_names[policy_zone]);
2498#endif
2499}
2500
2501/*
2502 * Helper functions to size the waitqueue hash table.
2503 * Essentially these want to choose hash table sizes sufficiently
2504 * large so that collisions trying to wait on pages are rare.
2505 * But in fact, the number of active page waitqueues on typical
2506 * systems is ridiculously low, less than 200. So this is even
2507 * conservative, even though it seems large.
2508 *
2509 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2510 * waitqueues, i.e. the size of the waitq table given the number of pages.
2511 */
2512#define PAGES_PER_WAITQUEUE     256
2513
2514#ifndef CONFIG_MEMORY_HOTPLUG
2515static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2516{
2517        unsigned long size = 1;
2518
2519        pages /= PAGES_PER_WAITQUEUE;
2520
2521        while (size < pages)
2522                size <<= 1;
2523
2524        /*
2525         * Once we have dozens or even hundreds of threads sleeping
2526         * on IO we've got bigger problems than wait queue collision.
2527         * Limit the size of the wait table to a reasonable size.
2528         */
2529        size = min(size, 4096UL);
2530
2531        return max(size, 4UL);
2532}
2533#else
2534/*
2535 * A zone's size might be changed by hot-add, so it is not possible to determine
2536 * a suitable size for its wait_table.  So we use the maximum size now.
2537 *
2538 * The max wait table size = 4096 x sizeof(wait_queue_head_t).   ie:
2539 *
2540 *    i386 (preemption config)    : 4096 x 16 = 64Kbyte.
2541 *    ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2542 *    ia64, x86-64 (preemption)   : 4096 x 24 = 96Kbyte.
2543 *
2544 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2545 * or more by the traditional way. (See above).  It equals:
2546 *
2547 *    i386, x86-64, powerpc(4K page size) : =  ( 2G + 1M)byte.
2548 *    ia64(16K page size)                 : =  ( 8G + 4M)byte.
2549 *    powerpc (64K page size)             : =  (32G +16M)byte.
2550 */
2551static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2552{
2553        return 4096UL;
2554}
2555#endif
2556
2557/*
2558 * This is an integer logarithm so that shifts can be used later
2559 * to extract the more random high bits from the multiplicative
2560 * hash function before the remainder is taken.
2561 */
2562static inline unsigned long wait_table_bits(unsigned long size)
2563{
2564        return ffz(~size);
2565}
2566
2567#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2568
2569/*
2570 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2571 * of blocks reserved is based on zone->pages_min. The memory within the
2572 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2573 * higher will lead to a bigger reserve which will get freed as contiguous
2574 * blocks as reclaim kicks in
2575 */
2576static void setup_zone_migrate_reserve(struct zone *zone)
2577{
2578        unsigned long start_pfn, pfn, end_pfn;
2579        struct page *page;
2580        unsigned long reserve, block_migratetype;
2581
2582        /* Get the start pfn, end pfn and the number of blocks to reserve */
2583        start_pfn = zone->zone_start_pfn;
2584        end_pfn = start_pfn + zone->spanned_pages;
2585        reserve = roundup(zone->pages_min, pageblock_nr_pages) >>
2586                                                        pageblock_order;
2587
2588        for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
2589                if (!pfn_valid(pfn))
2590                        continue;
2591                page = pfn_to_page(pfn);
2592
2593                /* Watch out for overlapping nodes */
2594                if (page_to_nid(page) != zone_to_nid(zone))
2595                        continue;
2596
2597                /* Blocks with reserved pages will never free, skip them. */
2598                if (PageReserved(page))
2599                        continue;
2600
2601                block_migratetype = get_pageblock_migratetype(page);
2602
2603                /* If this block is reserved, account for it */
2604                if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2605                        reserve--;
2606                        continue;
2607                }
2608
2609                /* Suitable for reserving if this block is movable */
2610                if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2611                        set_pageblock_migratetype(page, MIGRATE_RESERVE);
2612                        move_freepages_block(zone, page, MIGRATE_RESERVE);
2613                        reserve--;
2614                        continue;
2615                }
2616
2617                /*
2618                 * If the reserve is met and this is a previous reserved block,
2619                 * take it back
2620                 */
2621                if (block_migratetype == MIGRATE_RESERVE) {
2622                        set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2623                        move_freepages_block(zone, page, MIGRATE_MOVABLE);
2624                }
2625        }
2626}
2627
2628/*
2629 * Initially all pages are reserved - free ones are freed
2630 * up by free_all_bootmem() once the early boot process is
2631 * done. Non-atomic initialization, single-pass.
2632 */
2633void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
2634                unsigned long start_pfn, enum memmap_context context)
2635{
2636        struct page *page;
2637        unsigned long end_pfn = start_pfn + size;
2638        unsigned long pfn;
2639        struct zone *z;
2640
2641        if (highest_memmap_pfn < end_pfn - 1)
2642                highest_memmap_pfn = end_pfn - 1;
2643
2644        z = &NODE_DATA(nid)->node_zones[zone];
2645        for (pfn = start_pfn; pfn < end_pfn; pfn++) {
2646                /*
2647                 * There can be holes in boot-time mem_map[]s
2648                 * handed to this function.  They do not
2649                 * exist on hotplugged memory.
2650                 */
2651                if (context == MEMMAP_EARLY) {
2652                        if (!early_pfn_valid(pfn))
2653                                continue;
2654                        if (!early_pfn_in_nid(pfn, nid))
2655                                continue;
2656                }
2657                page = pfn_to_page(pfn);
2658                set_page_links(page, zone, nid, pfn);
2659                mminit_verify_page_links(page, zone, nid, pfn);
2660                init_page_count(page);
2661                reset_page_mapcount(page);
2662                SetPageReserved(page);
2663                /*
2664                 * Mark the block movable so that blocks are reserved for
2665                 * movable at startup. This will force kernel allocations
2666                 * to reserve their blocks rather than leaking throughout
2667                 * the address space during boot when many long-lived
2668                 * kernel allocations are made. Later some blocks near
2669                 * the start are marked MIGRATE_RESERVE by
2670                 * setup_zone_migrate_reserve()
2671                 *
2672                 * bitmap is created for zone's valid pfn range. but memmap
2673                 * can be created for invalid pages (for alignment)
2674                 * check here not to call set_pageblock_migratetype() against
2675                 * pfn out of zone.
2676                 */
2677                if ((z->zone_start_pfn <= pfn)
2678                    && (pfn < z->zone_start_pfn + z->spanned_pages)
2679                    && !(pfn & (pageblock_nr_pages - 1)))
2680                        set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2681
2682                INIT_LIST_HEAD(&page->lru);
2683#ifdef WANT_PAGE_VIRTUAL
2684                /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2685                if (!is_highmem_idx(zone))
2686                        set_page_address(page, __va(pfn << PAGE_SHIFT));
2687#endif
2688        }
2689}
2690
2691static void __meminit zone_init_free_lists(struct zone *zone)
2692{
2693        int order, t;
2694        for_each_migratetype_order(order, t) {
2695                INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
2696                zone->free_area[order].nr_free = 0;
2697        }
2698}
2699
2700#ifndef __HAVE_ARCH_MEMMAP_INIT
2701#define memmap_init(size, nid, zone, start_pfn) \
2702        memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2703#endif
2704
2705static int zone_batchsize(struct zone *zone)
2706{
2707#ifdef CONFIG_MMU
2708        int batch;
2709
2710        /*
2711         * The per-cpu-pages pools are set to around 1000th of the
2712         * size of the zone.  But no more than 1/2 of a meg.
2713         *
2714         * OK, so we don't know how big the cache is.  So guess.
2715         */
2716        batch = zone->present_pages / 1024;
2717        if (batch * PAGE_SIZE > 512 * 1024)
2718                batch = (512 * 1024) / PAGE_SIZE;
2719        batch /= 4;             /* We effectively *= 4 below */
2720        if (batch < 1)
2721                batch = 1;
2722
2723        /*
2724         * Clamp the batch to a 2^n - 1 value. Having a power
2725         * of 2 value was found to be more likely to have
2726         * suboptimal cache aliasing properties in some cases.
2727         *
2728         * For example if 2 tasks are alternately allocating
2729         * batches of pages, one task can end up with a lot
2730         * of pages of one half of the possible page colors
2731         * and the other with pages of the other colors.
2732         */
2733        batch = rounddown_pow_of_two(batch + batch/2) - 1;
2734
2735        return batch;
2736
2737#else
2738        /* The deferral and batching of frees should be suppressed under NOMMU
2739         * conditions.
2740         *
2741         * The problem is that NOMMU needs to be able to allocate large chunks
2742         * of contiguous memory as there's no hardware page translation to
2743         * assemble apparent contiguous memory from discontiguous pages.
2744         *
2745         * Queueing large contiguous runs of pages for batching, however,
2746         * causes the pages to actually be freed in smaller chunks.  As there
2747         * can be a significant delay between the individual batches being
2748         * recycled, this leads to the once large chunks of space being
2749         * fragmented and becoming unavailable for high-order allocations.
2750         */
2751        return 0;
2752#endif
2753}
2754
2755static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2756{
2757        struct per_cpu_pages *pcp;
2758
2759        memset(p, 0, sizeof(*p));
2760
2761        pcp = &p->pcp;
2762        pcp->count = 0;
2763        pcp->high = 6 * batch;
2764        pcp->batch = max(1UL, 1 * batch);
2765        INIT_LIST_HEAD(&pcp->list);
2766}
2767
2768/*
2769 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2770 * to the value high for the pageset p.
2771 */
2772
2773static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2774                                unsigned long high)
2775{
2776        struct per_cpu_pages *pcp;
2777
2778        pcp = &p->pcp;
2779        pcp->high = high;
2780        pcp->batch = max(1UL, high/4);
2781        if ((high/4) > (PAGE_SHIFT * 8))
2782                pcp->batch = PAGE_SHIFT * 8;
2783}
2784
2785
2786#ifdef CONFIG_NUMA
2787/*
2788 * Boot pageset table. One per cpu which is going to be used for all
2789 * zones and all nodes. The parameters will be set in such a way
2790 * that an item put on a list will immediately be handed over to
2791 * the buddy list. This is safe since pageset manipulation is done
2792 * with interrupts disabled.
2793 *
2794 * Some NUMA counter updates may also be caught by the boot pagesets.
2795 *
2796 * The boot_pagesets must be kept even after bootup is complete for
2797 * unused processors and/or zones. They do play a role for bootstrapping
2798 * hotplugged processors.
2799 *
2800 * zoneinfo_show() and maybe other functions do
2801 * not check if the processor is online before following the pageset pointer.
2802 * Other parts of the kernel may not check if the zone is available.
2803 */
2804static struct per_cpu_pageset boot_pageset[NR_CPUS];
2805
2806/*
2807 * Dynamically allocate memory for the
2808 * per cpu pageset array in struct zone.
2809 */
2810static int __cpuinit process_zones(int cpu)
2811{
2812        struct zone *zone, *dzone;
2813        int node = cpu_to_node(cpu);
2814
2815        node_set_state(node, N_CPU);    /* this node has a cpu */
2816
2817        for_each_populated_zone(zone) {
2818                zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
2819                                         GFP_KERNEL, node);
2820                if (!zone_pcp(zone, cpu))
2821                        goto bad;
2822
2823                setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
2824
2825                if (percpu_pagelist_fraction)
2826                        setup_pagelist_highmark(zone_pcp(zone, cpu),
2827                                (zone->present_pages / percpu_pagelist_fraction));
2828        }
2829
2830        return 0;
2831bad:
2832        for_each_zone(dzone) {
2833                if (!populated_zone(dzone))
2834                        continue;
2835                if (dzone == zone)
2836                        break;
2837                kfree(zone_pcp(dzone, cpu));
2838                zone_pcp(dzone, cpu) = &boot_pageset[cpu];
2839        }
2840        return -ENOMEM;
2841}
2842
2843static inline void free_zone_pagesets(int cpu)
2844{
2845        struct zone *zone;
2846
2847        for_each_zone(zone) {
2848                struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2849
2850                /* Free per_cpu_pageset if it is slab allocated */
2851                if (pset != &boot_pageset[cpu])
2852                        kfree(pset);
2853                zone_pcp(zone, cpu) = &boot_pageset[cpu];
2854        }
2855}
2856
2857static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
2858                unsigned long action,
2859                void *hcpu)
2860{
2861        int cpu = (long)hcpu;
2862        int ret = NOTIFY_OK;
2863
2864        switch (action) {
2865        case CPU_UP_PREPARE:
2866        case CPU_UP_PREPARE_FROZEN:
2867                if (process_zones(cpu))
2868                        ret = NOTIFY_BAD;
2869                break;
2870        case CPU_UP_CANCELED:
2871        case CPU_UP_CANCELED_FROZEN:
2872        case CPU_DEAD:
2873        case CPU_DEAD_FROZEN:
2874                free_zone_pagesets(cpu);
2875                break;
2876        default:
2877                break;
2878        }
2879        return ret;
2880}
2881
2882static struct notifier_block __cpuinitdata pageset_notifier =
2883        { &pageset_cpuup_callback, NULL, 0 };
2884
2885void __init setup_per_cpu_pageset(void)
2886{
2887        int err;
2888
2889        /* Initialize per_cpu_pageset for cpu 0.
2890         * A cpuup callback will do this for every cpu
2891         * as it comes online
2892         */
2893        err = process_zones(smp_processor_id());
2894        BUG_ON(err);
2895        register_cpu_notifier(&pageset_notifier);
2896}
2897
2898#endif
2899
2900static noinline __init_refok
2901int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
2902{
2903        int i;
2904        struct pglist_data *pgdat = zone->zone_pgdat;
2905        size_t alloc_size;
2906
2907        /*
2908         * The per-page waitqueue mechanism uses hashed waitqueues
2909         * per zone.
2910         */
2911        zone->wait_table_hash_nr_entries =
2912                 wait_table_hash_nr_entries(zone_size_pages);
2913        zone->wait_table_bits =
2914                wait_table_bits(zone->wait_table_hash_nr_entries);
2915        alloc_size = zone->wait_table_hash_nr_entries
2916                                        * sizeof(wait_queue_head_t);
2917
2918        if (!slab_is_available()) {
2919                zone->wait_table = (wait_queue_head_t *)
2920                        alloc_bootmem_node(pgdat, alloc_size);
2921        } else {
2922                /*
2923                 * This case means that a zone whose size was 0 gets new memory
2924                 * via memory hot-add.
2925                 * But it may be the case that a new node was hot-added.  In
2926                 * this case vmalloc() will not be able to use this new node's
2927                 * memory - this wait_table must be initialized to use this new
2928                 * node itself as well.
2929                 * To use this new node's memory, further consideration will be
2930                 * necessary.
2931                 */
2932                zone->wait_table = vmalloc(alloc_size);
2933        }
2934        if (!zone->wait_table)
2935                return -ENOMEM;
2936
2937        for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
2938                init_waitqueue_head(zone->wait_table + i);
2939
2940        return 0;
2941}
2942
2943static __meminit void zone_pcp_init(struct zone *zone)
2944{
2945        int cpu;
2946        unsigned long batch = zone_batchsize(zone);
2947
2948        for (cpu = 0; cpu < NR_CPUS; cpu++) {
2949#ifdef CONFIG_NUMA
2950                /* Early boot. Slab allocator not functional yet */
2951                zone_pcp(zone, cpu) = &boot_pageset[cpu];
2952                setup_pageset(&boot_pageset[cpu],0);
2953#else
2954                setup_pageset(zone_pcp(zone,cpu), batch);
2955#endif
2956        }
2957        if (zone->present_pages)
2958                printk(KERN_DEBUG "  %s zone: %lu pages, LIFO batch:%lu\n",
2959                        zone->name, zone->present_pages, batch);
2960}
2961
2962__meminit int init_currently_empty_zone(struct zone *zone,
2963                                        unsigned long zone_start_pfn,
2964                                        unsigned long size,
2965                                        enum memmap_context context)
2966{
2967        struct pglist_data *pgdat = zone->zone_pgdat;
2968        int ret;
2969        ret = zone_wait_table_init(zone, size);
2970        if (ret)
2971                return ret;
2972        pgdat->nr_zones = zone_idx(zone) + 1;
2973
2974        zone->zone_start_pfn = zone_start_pfn;
2975
2976        mminit_dprintk(MMINIT_TRACE, "memmap_init",
2977                        "Initialising map node %d zone %lu pfns %lu -> %lu\n",
2978                        pgdat->node_id,
2979                        (unsigned long)zone_idx(zone),
2980                        zone_start_pfn, (zone_start_pfn + size));
2981
2982        zone_init_free_lists(zone);
2983
2984        return 0;
2985}
2986
2987#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2988/*
2989 * Basic iterator support. Return the first range of PFNs for a node
2990 * Note: nid == MAX_NUMNODES returns first region regardless of node
2991 */
2992static int __meminit first_active_region_index_in_nid(int nid)
2993{
2994        int i;
2995
2996        for (i = 0; i < nr_nodemap_entries; i++)
2997                if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2998                        return i;
2999
3000        return -1;
3001}
3002
3003/*
3004 * Basic iterator support. Return the next active range of PFNs for a node
3005 * Note: nid == MAX_NUMNODES returns next region regardless of node
3006 */
3007static int __meminit next_active_region_index_in_nid(int index, int nid)
3008{
3009        for (index = index + 1; index < nr_nodemap_entries; index++)
3010                if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
3011                        return index;
3012
3013        return -1;
3014}
3015
3016#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3017/*
3018 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3019 * Architectures may implement their own version but if add_active_range()
3020 * was used and there are no special requirements, this is a convenient
3021 * alternative
3022 */
3023int __meminit __early_pfn_to_nid(unsigned long pfn)
3024{
3025        int i;
3026
3027        for (i = 0; i < nr_nodemap_entries; i++) {
3028                unsigned long start_pfn = early_node_map[i].start_pfn;
3029                unsigned long end_pfn = early_node_map[i].end_pfn;
3030
3031                if (start_pfn <= pfn && pfn < end_pfn)
3032                        return early_node_map[i].nid;
3033        }
3034        /* This is a memory hole */
3035        return -1;
3036}
3037#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3038
3039int __meminit early_pfn_to_nid(unsigned long pfn)
3040{
3041        int nid;
3042
3043        nid = __early_pfn_to_nid(pfn);
3044        if (nid >= 0)
3045                return nid;
3046        /* just returns 0 */
3047        return 0;
3048}
3049
3050#ifdef CONFIG_NODES_SPAN_OTHER_NODES
3051bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3052{
3053        int nid;
3054
3055        nid = __early_pfn_to_nid(pfn);
3056        if (nid >= 0 && nid != node)
3057                return false;
3058        return true;
3059}
3060#endif
3061
3062/* Basic iterator support to walk early_node_map[] */
3063#define for_each_active_range_index_in_nid(i, nid) \
3064        for (i = first_active_region_index_in_nid(nid); i != -1; \
3065                                i = next_active_region_index_in_nid(i, nid))
3066
3067/**
3068 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3069 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3070 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3071 *
3072 * If an architecture guarantees that all ranges registered with
3073 * add_active_ranges() contain no holes and may be freed, this
3074 * this function may be used instead of calling free_bootmem() manually.
3075 */
3076void __init free_bootmem_with_active_regions(int nid,
3077                                                unsigned long max_low_pfn)
3078{
3079        int i;
3080
3081        for_each_active_range_index_in_nid(i, nid) {
3082                unsigned long size_pages = 0;
3083                unsigned long end_pfn = early_node_map[i].end_pfn;
3084
3085                if (early_node_map[i].start_pfn >= max_low_pfn)
3086                        continue;
3087
3088                if (end_pfn > max_low_pfn)
3089                        end_pfn = max_low_pfn;
3090
3091                size_pages = end_pfn - early_node_map[i].start_pfn;
3092                free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3093                                PFN_PHYS(early_node_map[i].start_pfn),
3094                                size_pages << PAGE_SHIFT);
3095        }
3096}
3097
3098void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3099{
3100        int i;
3101        int ret;
3102
3103        for_each_active_range_index_in_nid(i, nid) {
3104                ret = work_fn(early_node_map[i].start_pfn,
3105                              early_node_map[i].end_pfn, data);
3106                if (ret)
3107                        break;
3108        }
3109}
3110/**
3111 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3112 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3113 *
3114 * If an architecture guarantees that all ranges registered with
3115 * add_active_ranges() contain no holes and may be freed, this
3116 * function may be used instead of calling memory_present() manually.
3117 */
3118void __init sparse_memory_present_with_active_regions(int nid)
3119{
3120        int i;
3121
3122        for_each_active_range_index_in_nid(i, nid)
3123                memory_present(early_node_map[i].nid,
3124                                early_node_map[i].start_pfn,
3125                                early_node_map[i].end_pfn);
3126}
3127
3128/**
3129 * push_node_boundaries - Push node boundaries to at least the requested boundary
3130 * @nid: The nid of the node to push the boundary for
3131 * @start_pfn: The start pfn of the node
3132 * @end_pfn: The end pfn of the node
3133 *
3134 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
3135 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
3136 * be hotplugged even though no physical memory exists. This function allows
3137 * an arch to push out the node boundaries so mem_map is allocated that can
3138 * be used later.
3139 */
3140#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3141void __init push_node_boundaries(unsigned int nid,
3142                unsigned long start_pfn, unsigned long end_pfn)
3143{
3144        mminit_dprintk(MMINIT_TRACE, "zoneboundary",
3145                        "Entering push_node_boundaries(%u, %lu, %lu)\n",
3146                        nid, start_pfn, end_pfn);
3147
3148        /* Initialise the boundary for this node if necessary */
3149        if (node_boundary_end_pfn[nid] == 0)
3150                node_boundary_start_pfn[nid] = -1UL;
3151
3152        /* Update the boundaries */
3153        if (node_boundary_start_pfn[nid] > start_pfn)
3154                node_boundary_start_pfn[nid] = start_pfn;
3155        if (node_boundary_end_pfn[nid] < end_pfn)
3156                node_boundary_end_pfn[nid] = end_pfn;
3157}
3158
3159/* If necessary, push the node boundary out for reserve hotadd */
3160static void __meminit account_node_boundary(unsigned int nid,
3161                unsigned long *start_pfn, unsigned long *end_pfn)
3162{
3163        mminit_dprintk(MMINIT_TRACE, "zoneboundary",
3164                        "Entering account_node_boundary(%u, %lu, %lu)\n",
3165                        nid, *start_pfn, *end_pfn);
3166
3167        /* Return if boundary information has not been provided */
3168        if (node_boundary_end_pfn[nid] == 0)
3169                return;
3170
3171        /* Check the boundaries and update if necessary */
3172        if (node_boundary_start_pfn[nid] < *start_pfn)
3173                *start_pfn = node_boundary_start_pfn[nid];
3174        if (node_boundary_end_pfn[nid] > *end_pfn)
3175                *end_pfn = node_boundary_end_pfn[nid];
3176}
3177#else
3178void __init push_node_boundaries(unsigned int nid,
3179                unsigned long start_pfn, unsigned long end_pfn) {}
3180
3181static void __meminit account_node_boundary(unsigned int nid,
3182                unsigned long *start_pfn, unsigned long *end_pfn) {}
3183#endif
3184
3185
3186/**
3187 * get_pfn_range_for_nid - Return the start and end page frames for a node
3188 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3189 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3190 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3191 *
3192 * It returns the start and end page frame of a node based on information
3193 * provided by an arch calling add_active_range(). If called for a node
3194 * with no available memory, a warning is printed and the start and end
3195 * PFNs will be 0.
3196 */
3197void __meminit get_pfn_range_for_nid(unsigned int nid,
3198                        unsigned long *start_pfn, unsigned long *end_pfn)
3199{
3200        int i;
3201        *start_pfn = -1UL;
3202        *end_pfn = 0;
3203
3204        for_each_active_range_index_in_nid(i, nid) {
3205                *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3206                *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3207        }
3208
3209        if (*start_pfn == -1UL)
3210                *start_pfn = 0;
3211
3212        /* Push the node boundaries out if requested */
3213        account_node_boundary(nid, start_pfn, end_pfn);
3214}
3215
3216/*
3217 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3218 * assumption is made that zones within a node are ordered in monotonic
3219 * increasing memory addresses so that the "highest" populated zone is used
3220 */
3221static void __init find_usable_zone_for_movable(void)
3222{
3223        int zone_index;
3224        for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3225                if (zone_index == ZONE_MOVABLE)
3226                        continue;
3227
3228                if (arch_zone_highest_possible_pfn[zone_index] >
3229                                arch_zone_lowest_possible_pfn[zone_index])
3230                        break;
3231        }
3232
3233        VM_BUG_ON(zone_index == -1);
3234        movable_zone = zone_index;
3235}
3236
3237/*
3238 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3239 * because it is sized independant of architecture. Unlike the other zones,
3240 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3241 * in each node depending on the size of each node and how evenly kernelcore
3242 * is distributed. This helper function adjusts the zone ranges
3243 * provided by the architecture for a given node by using the end of the
3244 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3245 * zones within a node are in order of monotonic increases memory addresses
3246 */
3247static void __meminit adjust_zone_range_for_zone_movable(int nid,
3248                                        unsigned long zone_type,
3249                                        unsigned long node_start_pfn,
3250                                        unsigned long node_end_pfn,
3251                                        unsigned long *zone_start_pfn,
3252                                        unsigned long *zone_end_pfn)
3253{
3254        /* Only adjust if ZONE_MOVABLE is on this node */
3255        if (zone_movable_pfn[nid]) {
3256                /* Size ZONE_MOVABLE */
3257                if (zone_type == ZONE_MOVABLE) {
3258                        *zone_start_pfn = zone_movable_pfn[nid];
3259                        *zone_end_pfn = min(node_end_pfn,
3260                                arch_zone_highest_possible_pfn[movable_zone]);
3261
3262                /* Adjust for ZONE_MOVABLE starting within this range */
3263                } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3264                                *zone_end_pfn > zone_movable_pfn[nid]) {
3265                        *zone_end_pfn = zone_movable_pfn[nid];
3266
3267                /* Check if this whole range is within ZONE_MOVABLE */
3268                } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3269                        *zone_start_pfn = *zone_end_pfn;
3270        }
3271}
3272
3273/*
3274 * Return the number of pages a zone spans in a node, including holes
3275 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3276 */
3277static unsigned long __meminit zone_spanned_pages_in_node(int nid,
3278                                        unsigned long zone_type,
3279                                        unsigned long *ignored)
3280{
3281        unsigned long node_start_pfn, node_end_pfn;
3282        unsigned long zone_start_pfn, zone_end_pfn;
3283
3284        /* Get the start and end of the node and zone */
3285        get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3286        zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3287        zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
3288        adjust_zone_range_for_zone_movable(nid, zone_type,
3289                                node_start_pfn, node_end_pfn,
3290                                &zone_start_pfn, &zone_end_pfn);
3291
3292        /* Check that this node has pages within the zone's required range */
3293        if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3294                return 0;
3295
3296        /* Move the zone boundaries inside the node if necessary */
3297        zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3298        zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3299
3300        /* Return the spanned pages */
3301        return zone_end_pfn - zone_start_pfn;
3302}
3303
3304/*
3305 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3306 * then all holes in the requested range will be accounted for.
3307 */
3308static unsigned long __meminit __absent_pages_in_range(int nid,
3309                                unsigned long range_start_pfn,
3310                                unsigned long range_end_pfn)
3311{
3312        int i = 0;
3313        unsigned long prev_end_pfn = 0, hole_pages = 0;
3314        unsigned long start_pfn;
3315
3316        /* Find the end_pfn of the first active range of pfns in the node */
3317        i = first_active_region_index_in_nid(nid);
3318        if (i == -1)
3319                return 0;
3320
3321        prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3322
3323        /* Account for ranges before physical memory on this node */
3324        if (early_node_map[i].start_pfn > range_start_pfn)
3325                hole_pages = prev_end_pfn - range_start_pfn;
3326
3327        /* Find all holes for the zone within the node */
3328        for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3329
3330                /* No need to continue if prev_end_pfn is outside the zone */
3331                if (prev_end_pfn >= range_end_pfn)
3332                        break;
3333
3334                /* Make sure the end of the zone is not within the hole */
3335                start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3336                prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3337
3338                /* Update the hole size cound and move on */
3339                if (start_pfn > range_start_pfn) {
3340                        BUG_ON(prev_end_pfn > start_pfn);
3341                        hole_pages += start_pfn - prev_end_pfn;
3342                }
3343                prev_end_pfn = early_node_map[i].end_pfn;
3344        }
3345
3346        /* Account for ranges past physical memory on this node */
3347        if (range_end_pfn > prev_end_pfn)
3348                hole_pages += range_end_pfn -
3349                                max(range_start_pfn, prev_end_pfn);
3350
3351        return hole_pages;
3352}
3353
3354/**
3355 * absent_pages_in_range - Return number of page frames in holes within a range
3356 * @start_pfn: The start PFN to start searching for holes
3357 * @end_pfn: The end PFN to stop searching for holes
3358 *
3359 * It returns the number of pages frames in memory holes within a range.
3360 */
3361unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3362                                                        unsigned long end_pfn)
3363{
3364        return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3365}
3366
3367/* Return the number of page frames in holes in a zone on a node */
3368static unsigned long __meminit zone_absent_pages_in_node(int nid,
3369                                        unsigned long zone_type,
3370                                        unsigned long *ignored)
3371{
3372        unsigned long node_start_pfn, node_end_pfn;
3373        unsigned long zone_start_pfn, zone_end_pfn;
3374
3375        get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3376        zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3377                                                        node_start_pfn);
3378        zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3379                                                        node_end_pfn);
3380
3381        adjust_zone_range_for_zone_movable(nid, zone_type,
3382                        node_start_pfn, node_end_pfn,
3383                        &zone_start_pfn, &zone_end_pfn);
3384        return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
3385}
3386
3387#else
3388static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
3389                                        unsigned long zone_type,
3390                                        unsigned long *zones_size)
3391{
3392        return zones_size[zone_type];
3393}
3394
3395static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
3396                                                unsigned long zone_type,
3397                                                unsigned long *zholes_size)
3398{
3399        if (!zholes_size)
3400                return 0;
3401
3402        return zholes_size[zone_type];
3403}
3404
3405#endif
3406
3407static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
3408                unsigned long *zones_size, unsigned long *zholes_size)
3409{
3410        unsigned long realtotalpages, totalpages = 0;
3411        enum zone_type i;
3412
3413        for (i = 0; i < MAX_NR_ZONES; i++)
3414                totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3415                                                                zones_size);
3416        pgdat->node_spanned_pages = totalpages;
3417
3418        realtotalpages = totalpages;
3419        for (i = 0; i < MAX_NR_ZONES; i++)
3420                realtotalpages -=
3421                        zone_absent_pages_in_node(pgdat->node_id, i,
3422                                                                zholes_size);
3423        pgdat->node_present_pages = realtotalpages;
3424        printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3425                                                        realtotalpages);
3426}
3427
3428#ifndef CONFIG_SPARSEMEM
3429/*
3430 * Calculate the size of the zone->blockflags rounded to an unsigned long
3431 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3432 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3433 * round what is now in bits to nearest long in bits, then return it in
3434 * bytes.
3435 */
3436static unsigned long __init usemap_size(unsigned long zonesize)
3437{
3438        unsigned long usemapsize;
3439
3440        usemapsize = roundup(zonesize, pageblock_nr_pages);
3441        usemapsize = usemapsize >> pageblock_order;
3442        usemapsize *= NR_PAGEBLOCK_BITS;
3443        usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3444
3445        return usemapsize / 8;
3446}
3447
3448static void __init setup_usemap(struct pglist_data *pgdat,
3449                                struct zone *zone, unsigned long zonesize)
3450{
3451        unsigned long usemapsize = usemap_size(zonesize);
3452        zone->pageblock_flags = NULL;
3453        if (usemapsize)
3454                zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
3455}
3456#else
3457static void inline setup_usemap(struct pglist_data *pgdat,
3458                                struct zone *zone, unsigned long zonesize) {}
3459#endif /* CONFIG_SPARSEMEM */
3460
3461#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3462
3463/* Return a sensible default order for the pageblock size. */
3464static inline int pageblock_default_order(void)
3465{
3466        if (HPAGE_SHIFT > PAGE_SHIFT)
3467                return HUGETLB_PAGE_ORDER;
3468
3469        return MAX_ORDER-1;
3470}
3471
3472/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3473static inline void __init set_pageblock_order(unsigned int order)
3474{
3475        /* Check that pageblock_nr_pages has not already been setup */
3476        if (pageblock_order)
3477                return;
3478
3479        /*
3480         * Assume the largest contiguous order of interest is a huge page.
3481         * This value may be variable depending on boot parameters on IA64
3482         */
3483        pageblock_order = order;
3484}
3485#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3486
3487/*
3488 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3489 * and pageblock_default_order() are unused as pageblock_order is set
3490 * at compile-time. See include/linux/pageblock-flags.h for the values of
3491 * pageblock_order based on the kernel config
3492 */
3493static inline int pageblock_default_order(unsigned int order)
3494{
3495        return MAX_ORDER-1;
3496}
3497#define set_pageblock_order(x)  do {} while (0)
3498
3499#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3500
3501/*
3502 * Set up the zone data structures:
3503 *   - mark all pages reserved
3504 *   - mark all memory queues empty
3505 *   - clear the memory bitmaps
3506 */
3507static void __paginginit free_area_init_core(struct pglist_data *pgdat,
3508                unsigned long *zones_size, unsigned long *zholes_size)
3509{
3510        enum zone_type j;
3511        int nid = pgdat->node_id;
3512        unsigned long zone_start_pfn = pgdat->node_start_pfn;
3513        int ret;
3514
3515        pgdat_resize_init(pgdat);
3516        pgdat->nr_zones = 0;
3517        init_waitqueue_head(&pgdat->kswapd_wait);
3518        pgdat->kswapd_max_order = 0;
3519        pgdat_page_cgroup_init(pgdat);
3520        
3521        for (j = 0; j < MAX_NR_ZONES; j++) {
3522                struct zone *zone = pgdat->node_zones + j;
3523                unsigned long size, realsize, memmap_pages;
3524                enum lru_list l;
3525
3526                size = zone_spanned_pages_in_node(nid, j, zones_size);
3527                realsize = size - zone_absent_pages_in_node(nid, j,
3528                                                                zholes_size);
3529
3530                /*
3531                 * Adjust realsize so that it accounts for how much memory
3532                 * is used by this zone for memmap. This affects the watermark
3533                 * and per-cpu initialisations
3534                 */
3535                memmap_pages =
3536                        PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
3537                if (realsize >= memmap_pages) {
3538                        realsize -= memmap_pages;
3539                        if (memmap_pages)
3540                                printk(KERN_DEBUG
3541                                       "  %s zone: %lu pages used for memmap\n",
3542                                       zone_names[j], memmap_pages);
3543                } else
3544                        printk(KERN_WARNING
3545                                "  %s zone: %lu pages exceeds realsize %lu\n",
3546                                zone_names[j], memmap_pages, realsize);
3547
3548                /* Account for reserved pages */
3549                if (j == 0 && realsize > dma_reserve) {
3550                        realsize -= dma_reserve;
3551                        printk(KERN_DEBUG "  %s zone: %lu pages reserved\n",
3552                                        zone_names[0], dma_reserve);
3553                }
3554
3555                if (!is_highmem_idx(j))
3556                        nr_kernel_pages += realsize;
3557                nr_all_pages += realsize;
3558
3559                zone->spanned_pages = size;
3560                zone->present_pages = realsize;
3561#ifdef CONFIG_NUMA
3562                zone->node = nid;
3563                zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
3564                                                / 100;
3565                zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
3566#endif
3567                zone->name = zone_names[j];
3568                spin_lock_init(&zone->lock);
3569                spin_lock_init(&zone->lru_lock);
3570                zone_seqlock_init(zone);
3571                zone->zone_pgdat = pgdat;
3572
3573                zone->prev_priority = DEF_PRIORITY;
3574
3575                zone_pcp_init(zone);
3576                for_each_lru(l) {
3577                        INIT_LIST_HEAD(&zone->lru[l].list);
3578                        zone->lru[l].nr_scan = 0;
3579                }
3580                zone->reclaim_stat.recent_rotated[0] = 0;
3581                zone->reclaim_stat.recent_rotated[1] = 0;
3582                zone->reclaim_stat.recent_scanned[0] = 0;
3583                zone->reclaim_stat.recent_scanned[1] = 0;
3584                zap_zone_vm_stats(zone);
3585                zone->flags = 0;
3586                if (!size)
3587                        continue;
3588
3589                set_pageblock_order(pageblock_default_order());
3590                setup_usemap(pgdat, zone, size);
3591                ret = init_currently_empty_zone(zone, zone_start_pfn,
3592                                                size, MEMMAP_EARLY);
3593                BUG_ON(ret);
3594                memmap_init(size, nid, j, zone_start_pfn);
3595                zone_start_pfn += size;
3596        }
3597}
3598
3599static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
3600{
3601        /* Skip empty nodes */
3602        if (!pgdat->node_spanned_pages)
3603                return;
3604
3605#ifdef CONFIG_FLAT_NODE_MEM_MAP
3606        /* ia64 gets its own node_mem_map, before this, without bootmem */
3607        if (!pgdat->node_mem_map) {
3608                unsigned long size, start, end;
3609                struct page *map;
3610
3611                /*
3612                 * The zone's endpoints aren't required to be MAX_ORDER
3613                 * aligned but the node_mem_map endpoints must be in order
3614                 * for the buddy allocator to function correctly.
3615                 */
3616                start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3617                end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3618                end = ALIGN(end, MAX_ORDER_NR_PAGES);
3619                size =  (end - start) * sizeof(struct page);
3620                map = alloc_remap(pgdat->node_id, size);
3621                if (!map)
3622                        map = alloc_bootmem_node(pgdat, size);
3623                pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
3624        }
3625#ifndef CONFIG_NEED_MULTIPLE_NODES
3626        /*
3627         * With no DISCONTIG, the global mem_map is just set as node 0's
3628         */
3629        if (pgdat == NODE_DATA(0)) {
3630                mem_map = NODE_DATA(0)->node_mem_map;
3631#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3632                if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
3633                        mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
3634#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3635        }
3636#endif
3637#endif /* CONFIG_FLAT_NODE_MEM_MAP */
3638}
3639
3640void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
3641                unsigned long node_start_pfn, unsigned long *zholes_size)
3642{
3643        pg_data_t *pgdat = NODE_DATA(nid);
3644
3645        pgdat->node_id = nid;
3646        pgdat->node_start_pfn = node_start_pfn;
3647        calculate_node_totalpages(pgdat, zones_size, zholes_size);
3648
3649        alloc_node_mem_map(pgdat);
3650#ifdef CONFIG_FLAT_NODE_MEM_MAP
3651        printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3652                nid, (unsigned long)pgdat,
3653                (unsigned long)pgdat->node_mem_map);
3654#endif
3655
3656        free_area_init_core(pgdat, zones_size, zholes_size);
3657}
3658
3659#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3660
3661#if MAX_NUMNODES > 1
3662/*
3663 * Figure out the number of possible node ids.
3664 */
3665static void __init setup_nr_node_ids(void)
3666{
3667        unsigned int node;
3668        unsigned int highest = 0;
3669
3670        for_each_node_mask(node, node_possible_map)
3671                highest = node;
3672        nr_node_ids = highest + 1;
3673}
3674#else
3675static inline void setup_nr_node_ids(void)
3676{
3677}
3678#endif
3679
3680/**
3681 * add_active_range - Register a range of PFNs backed by physical memory
3682 * @nid: The node ID the range resides on
3683 * @start_pfn: The start PFN of the available physical memory
3684 * @end_pfn: The end PFN of the available physical memory
3685 *
3686 * These ranges are stored in an early_node_map[] and later used by
3687 * free_area_init_nodes() to calculate zone sizes and holes. If the
3688 * range spans a memory hole, it is up to the architecture to ensure
3689 * the memory is not freed by the bootmem allocator. If possible
3690 * the range being registered will be merged with existing ranges.
3691 */
3692void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3693                                                unsigned long end_pfn)
3694{
3695        int i;
3696
3697        mminit_dprintk(MMINIT_TRACE, "memory_register",
3698                        "Entering add_active_range(%d, %#lx, %#lx) "
3699                        "%d entries of %d used\n",
3700                        nid, start_pfn, end_pfn,
3701                        nr_nodemap_entries, MAX_ACTIVE_REGIONS);
3702
3703        mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
3704
3705        /* Merge with existing active regions if possible */
3706        for (i = 0; i < nr_nodemap_entries; i++) {
3707                if (early_node_map[i].nid != nid)
3708                        continue;
3709
3710                /* Skip if an existing region covers this new one */
3711                if (start_pfn >= early_node_map[i].start_pfn &&
3712                                end_pfn <= early_node_map[i].end_pfn)
3713                        return;
3714
3715                /* Merge forward if suitable */
3716                if (start_pfn <= early_node_map[i].end_pfn &&
3717                                end_pfn > early_node_map[i].end_pfn) {
3718                        early_node_map[i].end_pfn = end_pfn;
3719                        return;
3720                }
3721
3722                /* Merge backward if suitable */
3723                if (start_pfn < early_node_map[i].end_pfn &&
3724                                end_pfn >= early_node_map[i].start_pfn) {
3725                        early_node_map[i].start_pfn = start_pfn;
3726                        return;
3727                }
3728        }
3729
3730        /* Check that early_node_map is large enough */
3731        if (i >= MAX_ACTIVE_REGIONS) {
3732                printk(KERN_CRIT "More than %d memory regions, truncating\n",
3733                                                        MAX_ACTIVE_REGIONS);
3734                return;
3735        }
3736
3737        early_node_map[i].nid = nid;
3738        early_node_map[i].start_pfn = start_pfn;
3739        early_node_map[i].end_pfn = end_pfn;
3740        nr_nodemap_entries = i + 1;
3741}
3742
3743/**
3744 * remove_active_range - Shrink an existing registered range of PFNs
3745 * @nid: The node id the range is on that should be shrunk
3746 * @start_pfn: The new PFN of the range
3747 * @end_pfn: The new PFN of the range
3748 *
3749 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3750 * The map is kept near the end physical page range that has already been
3751 * registered. This function allows an arch to shrink an existing registered
3752 * range.
3753 */
3754void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
3755                                unsigned long end_pfn)
3756{
3757        int i, j;
3758        int removed = 0;
3759
3760        printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
3761                          nid, start_pfn, end_pfn);
3762
3763        /* Find the old active region end and shrink */
3764        for_each_active_range_index_in_nid(i, nid) {
3765                if (early_node_map[i].start_pfn >= start_pfn &&
3766                    early_node_map[i].end_pfn <= end_pfn) {
3767                        /* clear it */
3768                        early_node_map[i].start_pfn = 0;
3769                        early_node_map[i].end_pfn = 0;
3770                        removed = 1;
3771                        continue;
3772                }
3773                if (early_node_map[i].start_pfn < start_pfn &&
3774                    early_node_map[i].end_pfn > start_pfn) {
3775                        unsigned long temp_end_pfn = early_node_map[i].end_pfn;
3776                        early_node_map[i].end_pfn = start_pfn;
3777                        if (temp_end_pfn > end_pfn)
3778                                add_active_range(nid, end_pfn, temp_end_pfn);
3779                        continue;
3780                }
3781                if (early_node_map[i].start_pfn >= start_pfn &&
3782                    early_node_map[i].end_pfn > end_pfn &&
3783                    early_node_map[i].start_pfn < end_pfn) {
3784                        early_node_map[i].start_pfn = end_pfn;
3785                        continue;
3786                }
3787        }
3788
3789        if (!removed)
3790                return;
3791
3792        /* remove the blank ones */
3793        for (i = nr_nodemap_entries - 1; i > 0; i--) {
3794                if (early_node_map[i].nid != nid)
3795                        continue;
3796                if (early_node_map[i].end_pfn)
3797                        continue;
3798                /* we found it, get rid of it */
3799                for (j = i; j < nr_nodemap_entries - 1; j++)
3800                        memcpy(&early_node_map[j], &early_node_map[j+1],
3801                                sizeof(early_node_map[j]));
3802                j = nr_nodemap_entries - 1;
3803                memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
3804                nr_nodemap_entries--;
3805        }
3806}
3807
3808/**
3809 * remove_all_active_ranges - Remove all currently registered regions
3810 *
3811 * During discovery, it may be found that a table like SRAT is invalid
3812 * and an alternative discovery method must be used. This function removes
3813 * all currently registered regions.
3814 */
3815void __init remove_all_active_ranges(void)
3816{
3817        memset(early_node_map, 0, sizeof(early_node_map));
3818        nr_nodemap_entries = 0;
3819#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3820        memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
3821        memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
3822#endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3823}
3824
3825/* Compare two active node_active_regions */
3826static int __init cmp_node_active_region(const void *a, const void *b)
3827{
3828        struct node_active_region *arange = (struct node_active_region *)a;
3829        struct node_active_region *brange = (struct node_active_region *)b;
3830
3831        /* Done this way to avoid overflows */
3832        if (arange->start_pfn > brange->start_pfn)
3833                return 1;
3834        if (arange->start_pfn < brange->start_pfn)
3835                return -1;
3836
3837        return 0;
3838}
3839
3840/* sort the node_map by start_pfn */
3841static void __init sort_node_map(void)
3842{
3843        sort(early_node_map, (size_t)nr_nodemap_entries,
3844                        sizeof(struct node_active_region),
3845                        cmp_node_active_region, NULL);
3846}
3847
3848/* Find the lowest pfn for a node */
3849static unsigned long __init find_min_pfn_for_node(int nid)
3850{
3851        int i;
3852        unsigned long min_pfn = ULONG_MAX;
3853
3854        /* Assuming a sorted map, the first range found has the starting pfn */
3855        for_each_active_range_index_in_nid(i, nid)
3856                min_pfn = min(min_pfn, early_node_map[i].start_pfn);
3857
3858        if (min_pfn == ULONG_MAX) {
3859                printk(KERN_WARNING
3860                        "Could not find start_pfn for node %d\n", nid);
3861                return 0;
3862        }
3863
3864        return min_pfn;
3865}
3866
3867/**
3868 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3869 *
3870 * It returns the minimum PFN based on information provided via
3871 * add_active_range().
3872 */
3873unsigned long __init find_min_pfn_with_active_regions(void)
3874{
3875        return find_min_pfn_for_node(MAX_NUMNODES);
3876}
3877
3878/*
3879 * early_calculate_totalpages()
3880 * Sum pages in active regions for movable zone.
3881 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3882 */
3883static unsigned long __init early_calculate_totalpages(void)
3884{
3885        int i;
3886        unsigned long totalpages = 0;
3887
3888        for (i = 0; i < nr_nodemap_entries; i++) {
3889                unsigned long pages = early_node_map[i].end_pfn -
3890                                                early_node_map[i].start_pfn;
3891                totalpages += pages;
3892                if (pages)
3893                        node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3894        }
3895        return totalpages;
3896}
3897
3898/*
3899 * Find the PFN the Movable zone begins in each node. Kernel memory
3900 * is spread evenly between nodes as long as the nodes have enough
3901 * memory. When they don't, some nodes will have more kernelcore than
3902 * others
3903 */
3904static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
3905{
3906        int i, nid;
3907        unsigned long usable_startpfn;
3908        unsigned long kernelcore_node, kernelcore_remaining;
3909        unsigned long totalpages = early_calculate_totalpages();
3910        int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
3911
3912        /*
3913         * If movablecore was specified, calculate what size of
3914         * kernelcore that corresponds so that memory usable for
3915         * any allocation type is evenly spread. If both kernelcore
3916         * and movablecore are specified, then the value of kernelcore
3917         * will be used for required_kernelcore if it's greater than
3918         * what movablecore would have allowed.
3919         */
3920        if (required_movablecore) {
3921                unsigned long corepages;
3922
3923                /*
3924                 * Round-up so that ZONE_MOVABLE is at least as large as what
3925                 * was requested by the user
3926                 */
3927                required_movablecore =
3928                        roundup(required_movablecore, MAX_ORDER_NR_PAGES);
3929                corepages = totalpages - required_movablecore;
3930
3931                required_kernelcore = max(required_kernelcore, corepages);
3932        }
3933
3934        /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3935        if (!required_kernelcore)
3936                return;
3937
3938        /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3939        find_usable_zone_for_movable();
3940        usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
3941
3942restart:
3943        /* Spread kernelcore memory as evenly as possible throughout nodes */
3944        kernelcore_node = required_kernelcore / usable_nodes;
3945        for_each_node_state(nid, N_HIGH_MEMORY) {
3946                /*
3947                 * Recalculate kernelcore_node if the division per node
3948                 * now exceeds what is necessary to satisfy the requested
3949                 * amount of memory for the kernel
3950                 */
3951                if (required_kernelcore < kernelcore_node)
3952                        kernelcore_node = required_kernelcore / usable_nodes;
3953
3954                /*
3955                 * As the map is walked, we track how much memory is usable
3956                 * by the kernel using kernelcore_remaining. When it is
3957                 * 0, the rest of the node is usable by ZONE_MOVABLE
3958                 */
3959                kernelcore_remaining = kernelcore_node;
3960
3961                /* Go through each range of PFNs within this node */
3962                for_each_active_range_index_in_nid(i, nid) {
3963                        unsigned long start_pfn, end_pfn;
3964                        unsigned long size_pages;
3965
3966                        start_pfn = max(early_node_map[i].start_pfn,
3967                                                zone_movable_pfn[nid]);
3968                        end_pfn = early_node_map[i].end_pfn;
3969                        if (start_pfn >= end_pfn)
3970                                continue;
3971
3972                        /* Account for what is only usable for kernelcore */
3973                        if (start_pfn < usable_startpfn) {
3974                                unsigned long kernel_pages;
3975                                kernel_pages = min(end_pfn, usable_startpfn)
3976                                                                - start_pfn;
3977
3978                                kernelcore_remaining -= min(kernel_pages,
3979                                                        kernelcore_remaining);
3980                                required_kernelcore -= min(kernel_pages,
3981                                                        required_kernelcore);
3982
3983                                /* Continue if range is now fully accounted */
3984                                if (end_pfn <= usable_startpfn) {
3985
3986                                        /*
3987                                         * Push zone_movable_pfn to the end so
3988                                         * that if we have to rebalance
3989                                         * kernelcore across nodes, we will
3990                                         * not double account here
3991                                         */
3992                                        zone_movable_pfn[nid] = end_pfn;
3993                                        continue;
3994                                }
3995                                start_pfn = usable_startpfn;
3996                        }
3997
3998                        /*
3999                         * The usable PFN range for ZONE_MOVABLE is from
4000                         * start_pfn->end_pfn. Calculate size_pages as the
4001                         * number of pages used as kernelcore
4002                         */
4003                        size_pages = end_pfn - start_pfn;
4004                        if (size_pages > kernelcore_remaining)
4005                                size_pages = kernelcore_remaining;
4006                        zone_movable_pfn[nid] = start_pfn + size_pages;
4007
4008                        /*
4009                         * Some kernelcore has been met, update counts and
4010                         * break if the kernelcore for this node has been
4011                         * satisified
4012                         */
4013                        required_kernelcore -= min(required_kernelcore,
4014                                                                size_pages);
4015                        kernelcore_remaining -= size_pages;
4016                        if (!kernelcore_remaining)
4017                                break;
4018                }
4019        }
4020
4021        /*
4022         * If there is still required_kernelcore, we do another pass with one
4023         * less node in the count. This will push zone_movable_pfn[nid] further
4024         * along on the nodes that still have memory until kernelcore is
4025         * satisified
4026         */
4027        usable_nodes--;
4028        if (usable_nodes && required_kernelcore > usable_nodes)
4029                goto restart;
4030
4031        /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4032        for (nid = 0; nid < MAX_NUMNODES; nid++)
4033                zone_movable_pfn[nid] =
4034                        roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
4035}
4036
4037/* Any regular memory on that node ? */
4038static void check_for_regular_memory(pg_data_t *pgdat)
4039{
4040#ifdef CONFIG_HIGHMEM
4041        enum zone_type zone_type;
4042
4043        for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4044                struct zone *zone = &pgdat->node_zones[zone_type];
4045                if (zone->present_pages)
4046                        node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4047        }
4048#endif
4049}
4050
4051/**
4052 * free_area_init_nodes - Initialise all pg_data_t and zone data
4053 * @max_zone_pfn: an array of max PFNs for each zone
4054 *
4055 * This will call free_area_init_node() for each active node in the system.
4056 * Using the page ranges provided by add_active_range(), the size of each
4057 * zone in each node and their holes is calculated. If the maximum PFN
4058 * between two adjacent zones match, it is assumed that the zone is empty.
4059 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4060 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4061 * starts where the previous one ended. For example, ZONE_DMA32 starts
4062 * at arch_max_dma_pfn.
4063 */
4064void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4065{
4066        unsigned long nid;
4067        int i;
4068
4069        /* Sort early_node_map as initialisation assumes it is sorted */
4070        sort_node_map();
4071
4072        /* Record where the zone boundaries are */
4073        memset(arch_zone_lowest_possible_pfn, 0,
4074                                sizeof(arch_zone_lowest_possible_pfn));
4075        memset(arch_zone_highest_possible_pfn, 0,
4076                                sizeof(arch_zone_highest_possible_pfn));
4077        arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4078        arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4079        for (i = 1; i < MAX_NR_ZONES; i++) {
4080                if (i == ZONE_MOVABLE)
4081                        continue;
4082                arch_zone_lowest_possible_pfn[i] =
4083                        arch_zone_highest_possible_pfn[i-1];
4084                arch_zone_highest_possible_pfn[i] =
4085                        max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4086        }
4087        arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4088        arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4089
4090        /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4091        memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4092        find_zone_movable_pfns_for_nodes(zone_movable_pfn);
4093
4094        /* Print out the zone ranges */
4095        printk("Zone PFN ranges:\n");
4096        for (i = 0; i < MAX_NR_ZONES; i++) {
4097                if (i == ZONE_MOVABLE)
4098                        continue;
4099                printk("  %-8s %0#10lx -> %0#10lx\n",
4100                                zone_names[i],
4101                                arch_zone_lowest_possible_pfn[i],
4102                                arch_zone_highest_possible_pfn[i]);
4103        }
4104
4105        /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4106        printk("Movable zone start PFN for each node\n");
4107        for (i = 0; i < MAX_NUMNODES; i++) {
4108                if (zone_movable_pfn[i])
4109                        printk("  Node %d: %lu\n", i, zone_movable_pfn[i]);
4110        }
4111
4112        /* Print out the early_node_map[] */
4113        printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4114        for (i = 0; i < nr_nodemap_entries; i++)
4115                printk("  %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
4116                                                early_node_map[i].start_pfn,
4117                                                early_node_map[i].end_pfn);
4118
4119        /* Initialise every node */
4120        mminit_verify_pageflags_layout();
4121        setup_nr_node_ids();
4122        for_each_online_node(nid) {
4123                pg_data_t *pgdat = NODE_DATA(nid);
4124                free_area_init_node(nid, NULL,
4125                                find_min_pfn_for_node(nid), NULL);
4126
4127                /* Any memory on that node */
4128                if (pgdat->node_present_pages)
4129                        node_set_state(nid, N_HIGH_MEMORY);
4130                check_for_regular_memory(pgdat);
4131        }
4132}
4133
4134static int __init cmdline_parse_core(char *p, unsigned long *core)
4135{
4136        unsigned long long coremem;
4137        if (!p)
4138                return -EINVAL;
4139
4140        coremem = memparse(p, &p);
4141        *core = coremem >> PAGE_SHIFT;
4142
4143        /* Paranoid check that UL is enough for the coremem value */
4144        WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4145
4146        return 0;
4147}
4148
4149/*
4150 * kernelcore=size sets the amount of memory for use for allocations that
4151 * cannot be reclaimed or migrated.
4152 */
4153static int __init cmdline_parse_kernelcore(char *p)
4154{
4155        return cmdline_parse_core(p, &required_kernelcore);
4156}
4157
4158/*
4159 * movablecore=size sets the amount of memory for use for allocations that
4160 * can be reclaimed or migrated.
4161 */
4162static int __init cmdline_parse_movablecore(char *p)
4163{
4164        return cmdline_parse_core(p, &required_movablecore);
4165}
4166
4167early_param("kernelcore", cmdline_parse_kernelcore);
4168early_param("movablecore", cmdline_parse_movablecore);
4169
4170#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4171
4172/**
4173 * set_dma_reserve - set the specified number of pages reserved in the first zone
4174 * @new_dma_reserve: The number of pages to mark reserved
4175 *
4176 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4177 * In the DMA zone, a significant percentage may be consumed by kernel image
4178 * and other unfreeable allocations which can skew the watermarks badly. This
4179 * function may optionally be used to account for unfreeable pages in the
4180 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4181 * smaller per-cpu batchsize.
4182 */
4183void __init set_dma_reserve(unsigned long new_dma_reserve)
4184{
4185        dma_reserve = new_dma_reserve;
4186}
4187
4188#ifndef CONFIG_NEED_MULTIPLE_NODES
4189struct pglist_data __refdata contig_page_data = { .bdata = &bootmem_node_data[0] };
4190EXPORT_SYMBOL(contig_page_data);
4191#endif
4192
4193void __init free_area_init(unsigned long *zones_size)
4194{
4195        free_area_init_node(0, zones_size,
4196                        __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4197}
4198
4199static int page_alloc_cpu_notify(struct notifier_block *self,
4200                                 unsigned long action, void *hcpu)
4201{
4202        int cpu = (unsigned long)hcpu;
4203
4204        if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
4205                drain_pages(cpu);
4206
4207                /*
4208                 * Spill the event counters of the dead processor
4209                 * into the current processors event counters.
4210                 * This artificially elevates the count of the current
4211                 * processor.
4212                 */
4213                vm_events_fold_cpu(cpu);
4214
4215                /*
4216                 * Zero the differential counters of the dead processor
4217                 * so that the vm statistics are consistent.
4218                 *
4219                 * This is only okay since the processor is dead and cannot
4220                 * race with what we are doing.
4221                 */
4222                refresh_cpu_vm_stats(cpu);
4223        }
4224        return NOTIFY_OK;
4225}
4226
4227void __init page_alloc_init(void)
4228{
4229        hotcpu_notifier(page_alloc_cpu_notify, 0);
4230}
4231
4232/*
4233 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4234 *      or min_free_kbytes changes.
4235 */
4236static void calculate_totalreserve_pages(void)
4237{
4238        struct pglist_data *pgdat;
4239        unsigned long reserve_pages = 0;
4240        enum zone_type i, j;
4241
4242        for_each_online_pgdat(pgdat) {
4243                for (i = 0; i < MAX_NR_ZONES; i++) {
4244                        struct zone *zone = pgdat->node_zones + i;
4245                        unsigned long max = 0;
4246
4247                        /* Find valid and maximum lowmem_reserve in the zone */
4248                        for (j = i; j < MAX_NR_ZONES; j++) {
4249                                if (zone->lowmem_reserve[j] > max)
4250                                        max = zone->lowmem_reserve[j];
4251                        }
4252
4253                        /* we treat pages_high as reserved pages. */
4254                        max += zone->pages_high;
4255
4256                        if (max > zone->present_pages)
4257                                max = zone->present_pages;
4258                        reserve_pages += max;
4259                }
4260        }
4261        totalreserve_pages = reserve_pages;
4262}
4263
4264/*
4265 * setup_per_zone_lowmem_reserve - called whenever
4266 *      sysctl_lower_zone_reserve_ratio changes.  Ensures that each zone
4267 *      has a correct pages reserved value, so an adequate number of
4268 *      pages are left in the zone after a successful __alloc_pages().
4269 */
4270static void setup_per_zone_lowmem_reserve(void)
4271{
4272        struct pglist_data *pgdat;
4273        enum zone_type j, idx;
4274
4275        for_each_online_pgdat(pgdat) {
4276                for (j = 0; j < MAX_NR_ZONES; j++) {
4277                        struct zone *zone = pgdat->node_zones + j;
4278                        unsigned long present_pages = zone->present_pages;
4279
4280                        zone->lowmem_reserve[j] = 0;
4281
4282                        idx = j;
4283                        while (idx) {
4284                                struct zone *lower_zone;
4285
4286                                idx--;
4287
4288                                if (sysctl_lowmem_reserve_ratio[idx] < 1)
4289                                        sysctl_lowmem_reserve_ratio[idx] = 1;
4290
4291                                lower_zone = pgdat->node_zones + idx;
4292                                lower_zone->lowmem_reserve[j] = present_pages /
4293                                        sysctl_lowmem_reserve_ratio[idx];
4294                                present_pages += lower_zone->present_pages;
4295                        }
4296                }
4297        }
4298
4299        /* update totalreserve_pages */
4300        calculate_totalreserve_pages();
4301}
4302
4303/**
4304 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4305 *
4306 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4307 * with respect to min_free_kbytes.
4308 */
4309void setup_per_zone_pages_min(void)
4310{
4311        unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4312        unsigned long lowmem_pages = 0;
4313        struct zone *zone;
4314        unsigned long flags;
4315
4316        /* Calculate total number of !ZONE_HIGHMEM pages */
4317        for_each_zone(zone) {
4318                if (!is_highmem(zone))
4319                        lowmem_pages += zone->present_pages;
4320        }
4321
4322        for_each_zone(zone) {
4323                u64 tmp;
4324
4325                spin_lock_irqsave(&zone->lock, flags);
4326                tmp = (u64)pages_min * zone->present_pages;
4327                do_div(tmp, lowmem_pages);
4328                if (is_highmem(zone)) {
4329                        /*
4330                         * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4331                         * need highmem pages, so cap pages_min to a small
4332                         * value here.
4333                         *
4334                         * The (pages_high-pages_low) and (pages_low-pages_min)
4335                         * deltas controls asynch page reclaim, and so should
4336                         * not be capped for highmem.
4337                         */
4338                        int min_pages;
4339
4340                        min_pages = zone->present_pages / 1024;
4341                        if (min_pages < SWAP_CLUSTER_MAX)
4342                                min_pages = SWAP_CLUSTER_MAX;
4343                        if (min_pages > 128)
4344                                min_pages = 128;
4345                        zone->pages_min = min_pages;
4346                } else {
4347                        /*
4348                         * If it's a lowmem zone, reserve a number of pages
4349                         * proportionate to the zone's size.
4350                         */
4351                        zone->pages_min = tmp;
4352                }
4353
4354                zone->pages_low   = zone->pages_min + (tmp >> 2);
4355                zone->pages_high  = zone->pages_min + (tmp >> 1);
4356                setup_zone_migrate_reserve(zone);
4357                spin_unlock_irqrestore(&zone->lock, flags);
4358        }
4359
4360        /* update totalreserve_pages */
4361        calculate_totalreserve_pages();
4362}
4363
4364/**
4365 * setup_per_zone_inactive_ratio - called when min_free_kbytes changes.
4366 *
4367 * The inactive anon list should be small enough that the VM never has to
4368 * do too much work, but large enough that each inactive page has a chance
4369 * to be referenced again before it is swapped out.
4370 *
4371 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4372 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4373 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4374 * the anonymous pages are kept on the inactive list.
4375 *
4376 * total     target    max
4377 * memory    ratio     inactive anon
4378 * -------------------------------------
4379 *   10MB       1         5MB
4380 *  100MB       1        50MB
4381 *    1GB       3       250MB
4382 *   10GB      10       0.9GB
4383 *  100GB      31         3GB
4384 *    1TB     101        10GB
4385 *   10TB     320        32GB
4386 */
4387static void setup_per_zone_inactive_ratio(void)
4388{
4389        struct zone *zone;
4390
4391        for_each_zone(zone) {
4392                unsigned int gb, ratio;
4393
4394                /* Zone size in gigabytes */
4395                gb = zone->present_pages >> (30 - PAGE_SHIFT);
4396                ratio = int_sqrt(10 * gb);
4397                if (!ratio)
4398                        ratio = 1;
4399
4400                zone->inactive_ratio = ratio;
4401        }
4402}
4403
4404/*
4405 * Initialise min_free_kbytes.
4406 *
4407 * For small machines we want it small (128k min).  For large machines
4408 * we want it large (64MB max).  But it is not linear, because network
4409 * bandwidth does not increase linearly with machine size.  We use
4410 *
4411 *      min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4412 *      min_free_kbytes = sqrt(lowmem_kbytes * 16)
4413 *
4414 * which yields
4415 *
4416 * 16MB:        512k
4417 * 32MB:        724k
4418 * 64MB:        1024k
4419 * 128MB:       1448k
4420 * 256MB:       2048k
4421 * 512MB:       2896k
4422 * 1024MB:      4096k
4423 * 2048MB:      5792k
4424 * 4096MB:      8192k
4425 * 8192MB:      11584k
4426 * 16384MB:     16384k
4427 */
4428static int __init init_per_zone_pages_min(void)
4429{
4430        unsigned long lowmem_kbytes;
4431
4432        lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4433
4434        min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4435        if (min_free_kbytes < 128)
4436                min_free_kbytes = 128;
4437        if (min_free_kbytes > 65536)
4438                min_free_kbytes = 65536;
4439        setup_per_zone_pages_min();
4440        setup_per_zone_lowmem_reserve();
4441        setup_per_zone_inactive_ratio();
4442        return 0;
4443}
4444module_init(init_per_zone_pages_min)
4445
4446/*
4447 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so 
4448 *      that we can call two helper functions whenever min_free_kbytes
4449 *      changes.
4450 */
4451int min_free_kbytes_sysctl_handler(ctl_table *table, int write, 
4452        struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4453{
4454        proc_dointvec(table, write, file, buffer, length, ppos);
4455        if (write)
4456                setup_per_zone_pages_min();
4457        return 0;
4458}
4459
4460#ifdef CONFIG_NUMA
4461int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4462        struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4463{
4464        struct zone *zone;
4465        int rc;
4466
4467        rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4468        if (rc)
4469                return rc;
4470
4471        for_each_zone(zone)
4472                zone->min_unmapped_pages = (zone->present_pages *
4473                                sysctl_min_unmapped_ratio) / 100;
4474        return 0;
4475}
4476
4477int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4478        struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4479{
4480        struct zone *zone;
4481        int rc;
4482
4483        rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4484        if (rc)
4485                return rc;
4486
4487        for_each_zone(zone)
4488                zone->min_slab_pages = (zone->present_pages *
4489                                sysctl_min_slab_ratio) / 100;
4490        return 0;
4491}
4492#endif
4493
4494/*
4495 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4496 *      proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4497 *      whenever sysctl_lowmem_reserve_ratio changes.
4498 *
4499 * The reserve ratio obviously has absolutely no relation with the
4500 * pages_min watermarks. The lowmem reserve ratio can only make sense
4501 * if in function of the boot time zone sizes.
4502 */
4503int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4504        struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4505{
4506        proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4507        setup_per_zone_lowmem_reserve();
4508        return 0;
4509}
4510
4511/*
4512 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4513 * cpu.  It is the fraction of total pages in each zone that a hot per cpu pagelist
4514 * can have before it gets flushed back to buddy allocator.
4515 */
4516
4517int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4518        struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4519{
4520        struct zone *zone;
4521        unsigned int cpu;
4522        int ret;
4523
4524        ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4525        if (!write || (ret == -EINVAL))
4526                return ret;
4527        for_each_populated_zone(zone) {
4528                for_each_online_cpu(cpu) {
4529                        unsigned long  high;
4530                        high = zone->present_pages / percpu_pagelist_fraction;
4531                        setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4532                }
4533        }
4534        return 0;
4535}
4536
4537int hashdist = HASHDIST_DEFAULT;
4538
4539#ifdef CONFIG_NUMA
4540static int __init set_hashdist(char *str)
4541{
4542        if (!str)
4543                return 0;
4544        hashdist = simple_strtoul(str, &str, 0);
4545        return 1;
4546}
4547__setup("hashdist=", set_hashdist);
4548#endif
4549
4550/*
4551 * allocate a large system hash table from bootmem
4552 * - it is assumed that the hash table must contain an exact power-of-2
4553 *   quantity of entries
4554 * - limit is the number of hash buckets, not the total allocation size
4555 */
4556void *__init alloc_large_system_hash(const char *tablename,
4557                                     unsigned long bucketsize,
4558                                     unsigned long numentries,
4559                                     int scale,
4560                                     int flags,
4561                                     unsigned int *_hash_shift,
4562                                     unsigned int *_hash_mask,
4563                                     unsigned long limit)
4564{
4565        unsigned long long max = limit;
4566        unsigned long log2qty, size;
4567        void *table = NULL;
4568
4569        /* allow the kernel cmdline to have a say */
4570        if (!numentries) {
4571                /* round applicable memory size up to nearest megabyte */
4572                numentries = nr_kernel_pages;
4573                numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4574                numentries >>= 20 - PAGE_SHIFT;
4575                numentries <<= 20 - PAGE_SHIFT;
4576
4577                /* limit to 1 bucket per 2^scale bytes of low memory */
4578                if (scale > PAGE_SHIFT)
4579                        numentries >>= (scale - PAGE_SHIFT);
4580                else
4581                        numentries <<= (PAGE_SHIFT - scale);
4582
4583                /* Make sure we've got at least a 0-order allocation.. */
4584                if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4585                        numentries = PAGE_SIZE / bucketsize;
4586        }
4587        numentries = roundup_pow_of_two(numentries);
4588
4589        /* limit allocation size to 1/16 total memory by default */
4590        if (max == 0) {
4591                max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4592                do_div(max, bucketsize);
4593        }
4594
4595        if (numentries > max)
4596                numentries = max;
4597
4598        log2qty = ilog2(numentries);
4599
4600        do {
4601                size = bucketsize << log2qty;
4602                if (flags & HASH_EARLY)
4603                        table = alloc_bootmem_nopanic(size);
4604                else if (hashdist)
4605                        table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4606                else {
4607                        unsigned long order = get_order(size);
4608                        table = (void*) __get_free_pages(GFP_ATOMIC, order);
4609                        /*
4610                         * If bucketsize is not a power-of-two, we may free
4611                         * some pages at the end of hash table.
4612                         */
4613                        if (table) {
4614                                unsigned long alloc_end = (unsigned long)table +
4615                                                (PAGE_SIZE << order);
4616                                unsigned long used = (unsigned long)table +
4617                                                PAGE_ALIGN(size);
4618                                split_page(virt_to_page(table), order);
4619                                while (used < alloc_end) {
4620                                        free_page(used);
4621                                        used += PAGE_SIZE;
4622                                }
4623                        }
4624                }
4625        } while (!table && size > PAGE_SIZE && --log2qty);
4626
4627        if (!table)
4628                panic("Failed to allocate %s hash table\n", tablename);
4629
4630        printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
4631               tablename,
4632               (1U << log2qty),
4633               ilog2(size) - PAGE_SHIFT,
4634               size);
4635
4636        if (_hash_shift)
4637                *_hash_shift = log2qty;
4638        if (_hash_mask)
4639                *_hash_mask = (1 << log2qty) - 1;
4640
4641        return table;
4642}
4643
4644/* Return a pointer to the bitmap storing bits affecting a block of pages */
4645static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4646                                                        unsigned long pfn)
4647{
4648#ifdef CONFIG_SPARSEMEM
4649        return __pfn_to_section(pfn)->pageblock_flags;
4650#else
4651        return zone->pageblock_flags;
4652#endif /* CONFIG_SPARSEMEM */
4653}
4654
4655static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4656{
4657#ifdef CONFIG_SPARSEMEM
4658        pfn &= (PAGES_PER_SECTION-1);
4659        return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4660#else
4661        pfn = pfn - zone->zone_start_pfn;
4662        return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4663#endif /* CONFIG_SPARSEMEM */
4664}
4665
4666/**
4667 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4668 * @page: The page within the block of interest
4669 * @start_bitidx: The first bit of interest to retrieve
4670 * @end_bitidx: The last bit of interest
4671 * returns pageblock_bits flags
4672 */
4673unsigned long get_pageblock_flags_group(struct page *page,
4674                                        int start_bitidx, int end_bitidx)
4675{
4676        struct zone *zone;
4677        unsigned long *bitmap;
4678        unsigned long pfn, bitidx;
4679        unsigned long flags = 0;
4680        unsigned long value = 1;
4681
4682        zone = page_zone(page);
4683        pfn = page_to_pfn(page);
4684        bitmap = get_pageblock_bitmap(zone, pfn);
4685        bitidx = pfn_to_bitidx(zone, pfn);
4686
4687        for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4688                if (test_bit(bitidx + start_bitidx, bitmap))
4689                        flags |= value;
4690
4691        return flags;
4692}
4693
4694/**
4695 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4696 * @page: The page within the block of interest
4697 * @start_bitidx: The first bit of interest
4698 * @end_bitidx: The last bit of interest
4699 * @flags: The flags to set
4700 */
4701void set_pageblock_flags_group(struct page *page, unsigned long flags,
4702                                        int start_bitidx, int end_bitidx)
4703{
4704        struct zone *zone;
4705        unsigned long *bitmap;
4706        unsigned long pfn, bitidx;
4707        unsigned long value = 1;
4708
4709        zone = page_zone(page);
4710        pfn = page_to_pfn(page);
4711        bitmap = get_pageblock_bitmap(zone, pfn);
4712        bitidx = pfn_to_bitidx(zone, pfn);
4713        VM_BUG_ON(pfn < zone->zone_start_pfn);
4714        VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
4715
4716        for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4717                if (flags & value)
4718                        __set_bit(bitidx + start_bitidx, bitmap);
4719                else
4720                        __clear_bit(bitidx + start_bitidx, bitmap);
4721}
4722
4723/*
4724 * This is designed as sub function...plz see page_isolation.c also.
4725 * set/clear page block's type to be ISOLATE.
4726 * page allocater never alloc memory from ISOLATE block.
4727 */
4728
4729int set_migratetype_isolate(struct page *page)
4730{
4731        struct zone *zone;
4732        unsigned long flags;
4733        int ret = -EBUSY;
4734
4735        zone = page_zone(page);
4736        spin_lock_irqsave(&zone->lock, flags);
4737        /*
4738         * In future, more migrate types will be able to be isolation target.
4739         */
4740        if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
4741                goto out;
4742        set_pageblock_migratetype(page, MIGRATE_ISOLATE);
4743        move_freepages_block(zone, page, MIGRATE_ISOLATE);
4744        ret = 0;
4745out:
4746        spin_unlock_irqrestore(&zone->lock, flags);
4747        if (!ret)
4748                drain_all_pages();
4749        return ret;
4750}
4751
4752void unset_migratetype_isolate(struct page *page)
4753{
4754        struct zone *zone;
4755        unsigned long flags;
4756        zone = page_zone(page);
4757        spin_lock_irqsave(&zone->lock, flags);
4758        if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
4759                goto out;
4760        set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4761        move_freepages_block(zone, page, MIGRATE_MOVABLE);
4762out:
4763        spin_unlock_irqrestore(&zone->lock, flags);
4764}
4765
4766#ifdef CONFIG_MEMORY_HOTREMOVE
4767/*
4768 * All pages in the range must be isolated before calling this.
4769 */
4770void
4771__offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
4772{
4773        struct page *page;
4774        struct zone *zone;
4775        int order, i;
4776        unsigned long pfn;
4777        unsigned long flags;
4778        /* find the first valid pfn */
4779        for (pfn = start_pfn; pfn < end_pfn; pfn++)
4780                if (pfn_valid(pfn))
4781                        break;
4782        if (pfn == end_pfn)
4783                return;
4784        zone = page_zone(pfn_to_page(pfn));
4785        spin_lock_irqsave(&zone->lock, flags);
4786        pfn = start_pfn;
4787        while (pfn < end_pfn) {
4788                if (!pfn_valid(pfn)) {
4789                        pfn++;
4790                        continue;
4791                }
4792                page = pfn_to_page(pfn);
4793                BUG_ON(page_count(page));
4794                BUG_ON(!PageBuddy(page));
4795                order = page_order(page);
4796#ifdef CONFIG_DEBUG_VM
4797                printk(KERN_INFO "remove from free list %lx %d %lx\n",
4798                       pfn, 1 << order, end_pfn);
4799#endif
4800                list_del(&page->lru);
4801                rmv_page_order(page);
4802                zone->free_area[order].nr_free--;
4803                __mod_zone_page_state(zone, NR_FREE_PAGES,
4804                                      - (1UL << order));
4805                for (i = 0; i < (1 << order); i++)
4806                        SetPageReserved((page+i));
4807                pfn += (1 << order);
4808        }
4809        spin_unlock_irqrestore(&zone->lock, flags);
4810}
4811#endif
4812