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)
 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                list_add(&page->lru, list);
 881                set_page_private(page, migratetype);
 882                list = &page->lru;
 883        }
 884        spin_unlock(&zone->lock);
 885        return i;
 886}
 887
 888#ifdef CONFIG_NUMA
 889/*
 890 * Called from the vmstat counter updater to drain pagesets of this
 891 * currently executing processor on remote nodes after they have
 892 * expired.
 893 *
 894 * Note that this function must be called with the thread pinned to
 895 * a single processor.
 896 */
 897void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
 898{
 899        unsigned long flags;
 900        int to_drain;
 901
 902        local_irq_save(flags);
 903        if (pcp->count >= pcp->batch)
 904                to_drain = pcp->batch;
 905        else
 906                to_drain = pcp->count;
 907        free_pages_bulk(zone, to_drain, &pcp->list, 0);
 908        pcp->count -= to_drain;
 909        local_irq_restore(flags);
 910}
 911#endif
 912
 913/*
 914 * Drain pages of the indicated processor.
 915 *
 916 * The processor must either be the current processor and the
 917 * thread pinned to the current processor or a processor that
 918 * is not online.
 919 */
 920static void drain_pages(unsigned int cpu)
 921{
 922        unsigned long flags;
 923        struct zone *zone;
 924
 925        for_each_zone(zone) {
 926                struct per_cpu_pageset *pset;
 927                struct per_cpu_pages *pcp;
 928
 929                if (!populated_zone(zone))
 930                        continue;
 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, migratetype);
1084                        if (unlikely(!pcp->count))
1085                                goto failed;
1086                }
1087
1088                /* Find a page of the appropriate migrate type */
1089                if (cold) {
1090                        list_for_each_entry_reverse(page, &pcp->list, lru)
1091                                if (page_private(page) == migratetype)
1092                                        break;
1093                } else {
1094                        list_for_each_entry(page, &pcp->list, lru)
1095                                if (page_private(page) == migratetype)
1096                                        break;
1097                }
1098
1099                /* Allocate more to the pcp list if necessary */
1100                if (unlikely(&page->lru == &pcp->list)) {
1101                        pcp->count += rmqueue_bulk(zone, 0,
1102                                        pcp->batch, &pcp->list, migratetype);
1103                        page = list_entry(pcp->list.next, struct page, lru);
1104                }
1105
1106                list_del(&page->lru);
1107                pcp->count--;
1108        } else {
1109                spin_lock_irqsave(&zone->lock, flags);
1110                page = __rmqueue(zone, order, migratetype);
1111                spin_unlock(&zone->lock);
1112                if (!page)
1113                        goto failed;
1114        }
1115
1116        __count_zone_vm_events(PGALLOC, zone, 1 << order);
1117        zone_statistics(preferred_zone, zone);
1118        local_irq_restore(flags);
1119        put_cpu();
1120
1121        VM_BUG_ON(bad_range(zone, page));
1122        if (prep_new_page(page, order, gfp_flags))
1123                goto again;
1124        return page;
1125
1126failed:
1127        local_irq_restore(flags);
1128        put_cpu();
1129        return NULL;
1130}
1131
1132#define ALLOC_NO_WATERMARKS     0x01 /* don't check watermarks at all */
1133#define ALLOC_WMARK_MIN         0x02 /* use pages_min watermark */
1134#define ALLOC_WMARK_LOW         0x04 /* use pages_low watermark */
1135#define ALLOC_WMARK_HIGH        0x08 /* use pages_high watermark */
1136#define ALLOC_HARDER            0x10 /* try to alloc harder */
1137#define ALLOC_HIGH              0x20 /* __GFP_HIGH set */
1138#define ALLOC_CPUSET            0x40 /* check for correct cpuset */
1139
1140#ifdef CONFIG_FAIL_PAGE_ALLOC
1141
1142static struct fail_page_alloc_attr {
1143        struct fault_attr attr;
1144
1145        u32 ignore_gfp_highmem;
1146        u32 ignore_gfp_wait;
1147        u32 min_order;
1148
1149#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1150
1151        struct dentry *ignore_gfp_highmem_file;
1152        struct dentry *ignore_gfp_wait_file;
1153        struct dentry *min_order_file;
1154
1155#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1156
1157} fail_page_alloc = {
1158        .attr = FAULT_ATTR_INITIALIZER,
1159        .ignore_gfp_wait = 1,
1160        .ignore_gfp_highmem = 1,
1161        .min_order = 1,
1162};
1163
1164static int __init setup_fail_page_alloc(char *str)
1165{
1166        return setup_fault_attr(&fail_page_alloc.attr, str);
1167}
1168__setup("fail_page_alloc=", setup_fail_page_alloc);
1169
1170static int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1171{
1172        if (order < fail_page_alloc.min_order)
1173                return 0;
1174        if (gfp_mask & __GFP_NOFAIL)
1175                return 0;
1176        if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
1177                return 0;
1178        if (fail_page_alloc.ignore_gfp_wait && (gfp_mask & __GFP_WAIT))
1179                return 0;
1180
1181        return should_fail(&fail_page_alloc.attr, 1 << order);
1182}
1183
1184#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1185
1186static int __init fail_page_alloc_debugfs(void)
1187{
1188        mode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
1189        struct dentry *dir;
1190        int err;
1191
1192        err = init_fault_attr_dentries(&fail_page_alloc.attr,
1193                                       "fail_page_alloc");
1194        if (err)
1195                return err;
1196        dir = fail_page_alloc.attr.dentries.dir;
1197
1198        fail_page_alloc.ignore_gfp_wait_file =
1199                debugfs_create_bool("ignore-gfp-wait", mode, dir,
1200                                      &fail_page_alloc.ignore_gfp_wait);
1201
1202        fail_page_alloc.ignore_gfp_highmem_file =
1203                debugfs_create_bool("ignore-gfp-highmem", mode, dir,
1204                                      &fail_page_alloc.ignore_gfp_highmem);
1205        fail_page_alloc.min_order_file =
1206                debugfs_create_u32("min-order", mode, dir,
1207                                   &fail_page_alloc.min_order);
1208
1209        if (!fail_page_alloc.ignore_gfp_wait_file ||
1210            !fail_page_alloc.ignore_gfp_highmem_file ||
1211            !fail_page_alloc.min_order_file) {
1212                err = -ENOMEM;
1213                debugfs_remove(fail_page_alloc.ignore_gfp_wait_file);
1214                debugfs_remove(fail_page_alloc.ignore_gfp_highmem_file);
1215                debugfs_remove(fail_page_alloc.min_order_file);
1216                cleanup_fault_attr_dentries(&fail_page_alloc.attr);
1217        }
1218
1219        return err;
1220}
1221
1222late_initcall(fail_page_alloc_debugfs);
1223
1224#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1225
1226#else /* CONFIG_FAIL_PAGE_ALLOC */
1227
1228static inline int should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
1229{
1230        return 0;
1231}
1232
1233#endif /* CONFIG_FAIL_PAGE_ALLOC */
1234
1235/*
1236 * Return 1 if free pages are above 'mark'. This takes into account the order
1237 * of the allocation.
1238 */
1239int zone_watermark_ok(struct zone *z, int order, unsigned long mark,
1240                      int classzone_idx, int alloc_flags)
1241{
1242        /* free_pages my go negative - that's OK */
1243        long min = mark;
1244        long free_pages = zone_page_state(z, NR_FREE_PAGES) - (1 << order) + 1;
1245        int o;
1246
1247        if (alloc_flags & ALLOC_HIGH)
1248                min -= min / 2;
1249        if (alloc_flags & ALLOC_HARDER)
1250                min -= min / 4;
1251
1252        if (free_pages <= min + z->lowmem_reserve[classzone_idx])
1253                return 0;
1254        for (o = 0; o < order; o++) {
1255                /* At the next order, this order's pages become unavailable */
1256                free_pages -= z->free_area[o].nr_free << o;
1257
1258                /* Require fewer higher order pages to be free */
1259                min >>= 1;
1260
1261                if (free_pages <= min)
1262                        return 0;
1263        }
1264        return 1;
1265}
1266
1267#ifdef CONFIG_NUMA
1268/*
1269 * zlc_setup - Setup for "zonelist cache".  Uses cached zone data to
1270 * skip over zones that are not allowed by the cpuset, or that have
1271 * been recently (in last second) found to be nearly full.  See further
1272 * comments in mmzone.h.  Reduces cache footprint of zonelist scans
1273 * that have to skip over a lot of full or unallowed zones.
1274 *
1275 * If the zonelist cache is present in the passed in zonelist, then
1276 * returns a pointer to the allowed node mask (either the current
1277 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1278 *
1279 * If the zonelist cache is not available for this zonelist, does
1280 * nothing and returns NULL.
1281 *
1282 * If the fullzones BITMAP in the zonelist cache is stale (more than
1283 * a second since last zap'd) then we zap it out (clear its bits.)
1284 *
1285 * We hold off even calling zlc_setup, until after we've checked the
1286 * first zone in the zonelist, on the theory that most allocations will
1287 * be satisfied from that first zone, so best to examine that zone as
1288 * quickly as we can.
1289 */
1290static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1291{
1292        struct zonelist_cache *zlc;     /* cached zonelist speedup info */
1293        nodemask_t *allowednodes;       /* zonelist_cache approximation */
1294
1295        zlc = zonelist->zlcache_ptr;
1296        if (!zlc)
1297                return NULL;
1298
1299        if (time_after(jiffies, zlc->last_full_zap + HZ)) {
1300                bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
1301                zlc->last_full_zap = jiffies;
1302        }
1303
1304        allowednodes = !in_interrupt() && (alloc_flags & ALLOC_CPUSET) ?
1305                                        &cpuset_current_mems_allowed :
1306                                        &node_states[N_HIGH_MEMORY];
1307        return allowednodes;
1308}
1309
1310/*
1311 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1312 * if it is worth looking at further for free memory:
1313 *  1) Check that the zone isn't thought to be full (doesn't have its
1314 *     bit set in the zonelist_cache fullzones BITMAP).
1315 *  2) Check that the zones node (obtained from the zonelist_cache
1316 *     z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1317 * Return true (non-zero) if zone is worth looking at further, or
1318 * else return false (zero) if it is not.
1319 *
1320 * This check -ignores- the distinction between various watermarks,
1321 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ...  If a zone is
1322 * found to be full for any variation of these watermarks, it will
1323 * be considered full for up to one second by all requests, unless
1324 * we are so low on memory on all allowed nodes that we are forced
1325 * into the second scan of the zonelist.
1326 *
1327 * In the second scan we ignore this zonelist cache and exactly
1328 * apply the watermarks to all zones, even it is slower to do so.
1329 * We are low on memory in the second scan, and should leave no stone
1330 * unturned looking for a free page.
1331 */
1332static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1333                                                nodemask_t *allowednodes)
1334{
1335        struct zonelist_cache *zlc;     /* cached zonelist speedup info */
1336        int i;                          /* index of *z in zonelist zones */
1337        int n;                          /* node that zone *z is on */
1338
1339        zlc = zonelist->zlcache_ptr;
1340        if (!zlc)
1341                return 1;
1342
1343        i = z - zonelist->_zonerefs;
1344        n = zlc->z_to_n[i];
1345
1346        /* This zone is worth trying if it is allowed but not full */
1347        return node_isset(n, *allowednodes) && !test_bit(i, zlc->fullzones);
1348}
1349
1350/*
1351 * Given 'z' scanning a zonelist, set the corresponding bit in
1352 * zlc->fullzones, so that subsequent attempts to allocate a page
1353 * from that zone don't waste time re-examining it.
1354 */
1355static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1356{
1357        struct zonelist_cache *zlc;     /* cached zonelist speedup info */
1358        int i;                          /* index of *z in zonelist zones */
1359
1360        zlc = zonelist->zlcache_ptr;
1361        if (!zlc)
1362                return;
1363
1364        i = z - zonelist->_zonerefs;
1365
1366        set_bit(i, zlc->fullzones);
1367}
1368
1369#else   /* CONFIG_NUMA */
1370
1371static nodemask_t *zlc_setup(struct zonelist *zonelist, int alloc_flags)
1372{
1373        return NULL;
1374}
1375
1376static int zlc_zone_worth_trying(struct zonelist *zonelist, struct zoneref *z,
1377                                nodemask_t *allowednodes)
1378{
1379        return 1;
1380}
1381
1382static void zlc_mark_zone_full(struct zonelist *zonelist, struct zoneref *z)
1383{
1384}
1385#endif  /* CONFIG_NUMA */
1386
1387/*
1388 * get_page_from_freelist goes through the zonelist trying to allocate
1389 * a page.
1390 */
1391static struct page *
1392get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
1393                struct zonelist *zonelist, int high_zoneidx, int alloc_flags)
1394{
1395        struct zoneref *z;
1396        struct page *page = NULL;
1397        int classzone_idx;
1398        struct zone *zone, *preferred_zone;
1399        nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
1400        int zlc_active = 0;             /* set if using zonelist_cache */
1401        int did_zlc_setup = 0;          /* just call zlc_setup() one time */
1402
1403        (void)first_zones_zonelist(zonelist, high_zoneidx, nodemask,
1404                                                        &preferred_zone);
1405        if (!preferred_zone)
1406                return NULL;
1407
1408        classzone_idx = zone_idx(preferred_zone);
1409
1410zonelist_scan:
1411        /*
1412         * Scan zonelist, looking for a zone with enough free.
1413         * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1414         */
1415        for_each_zone_zonelist_nodemask(zone, z, zonelist,
1416                                                high_zoneidx, nodemask) {
1417                if (NUMA_BUILD && zlc_active &&
1418                        !zlc_zone_worth_trying(zonelist, z, allowednodes))
1419                                continue;
1420                if ((alloc_flags & ALLOC_CPUSET) &&
1421                        !cpuset_zone_allowed_softwall(zone, gfp_mask))
1422                                goto try_next_zone;
1423
1424                if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
1425                        unsigned long mark;
1426                        if (alloc_flags & ALLOC_WMARK_MIN)
1427                                mark = zone->pages_min;
1428                        else if (alloc_flags & ALLOC_WMARK_LOW)
1429                                mark = zone->pages_low;
1430                        else
1431                                mark = zone->pages_high;
1432                        if (!zone_watermark_ok(zone, order, mark,
1433                                    classzone_idx, alloc_flags)) {
1434                                if (!zone_reclaim_mode ||
1435                                    !zone_reclaim(zone, gfp_mask, order))
1436                                        goto this_zone_full;
1437                        }
1438                }
1439
1440                page = buffered_rmqueue(preferred_zone, zone, order, gfp_mask);
1441                if (page)
1442                        break;
1443this_zone_full:
1444                if (NUMA_BUILD)
1445                        zlc_mark_zone_full(zonelist, z);
1446try_next_zone:
1447                if (NUMA_BUILD && !did_zlc_setup) {
1448                        /* we do zlc_setup after the first zone is tried */
1449                        allowednodes = zlc_setup(zonelist, alloc_flags);
1450                        zlc_active = 1;
1451                        did_zlc_setup = 1;
1452                }
1453        }
1454
1455        if (unlikely(NUMA_BUILD && page == NULL && zlc_active)) {
1456                /* Disable zlc cache for second zonelist scan */
1457                zlc_active = 0;
1458                goto zonelist_scan;
1459        }
1460        return page;
1461}
1462
1463/*
1464 * This is the 'heart' of the zoned buddy allocator.
1465 */
1466struct page *
1467__alloc_pages_internal(gfp_t gfp_mask, unsigned int order,
1468                        struct zonelist *zonelist, nodemask_t *nodemask)
1469{
1470        const gfp_t wait = gfp_mask & __GFP_WAIT;
1471        enum zone_type high_zoneidx = gfp_zone(gfp_mask);
1472        struct zoneref *z;
1473        struct zone *zone;
1474        struct page *page;
1475        struct reclaim_state reclaim_state;
1476        struct task_struct *p = current;
1477        int do_retry;
1478        int alloc_flags;
1479        unsigned long did_some_progress;
1480        unsigned long pages_reclaimed = 0;
1481
1482        might_sleep_if(wait);
1483
1484        if (should_fail_alloc_page(gfp_mask, order))
1485                return NULL;
1486
1487restart:
1488        z = zonelist->_zonerefs;  /* the list of zones suitable for gfp_mask */
1489
1490        if (unlikely(!z->zone)) {
1491                /*
1492                 * Happens if we have an empty zonelist as a result of
1493                 * GFP_THISNODE being used on a memoryless node
1494                 */
1495                return NULL;
1496        }
1497
1498        page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask, order,
1499                        zonelist, high_zoneidx, ALLOC_WMARK_LOW|ALLOC_CPUSET);
1500        if (page)
1501                goto got_pg;
1502
1503        /*
1504         * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1505         * __GFP_NOWARN set) should not cause reclaim since the subsystem
1506         * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1507         * using a larger set of nodes after it has established that the
1508         * allowed per node queues are empty and that nodes are
1509         * over allocated.
1510         */
1511        if (NUMA_BUILD && (gfp_mask & GFP_THISNODE) == GFP_THISNODE)
1512                goto nopage;
1513
1514        for_each_zone_zonelist(zone, z, zonelist, high_zoneidx)
1515                wakeup_kswapd(zone, order);
1516
1517        /*
1518         * OK, we're below the kswapd watermark and have kicked background
1519         * reclaim. Now things get more complex, so set up alloc_flags according
1520         * to how we want to proceed.
1521         *
1522         * The caller may dip into page reserves a bit more if the caller
1523         * cannot run direct reclaim, or if the caller has realtime scheduling
1524         * policy or is asking for __GFP_HIGH memory.  GFP_ATOMIC requests will
1525         * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1526         */
1527        alloc_flags = ALLOC_WMARK_MIN;
1528        if ((unlikely(rt_task(p)) && !in_interrupt()) || !wait)
1529                alloc_flags |= ALLOC_HARDER;
1530        if (gfp_mask & __GFP_HIGH)
1531                alloc_flags |= ALLOC_HIGH;
1532        if (wait)
1533                alloc_flags |= ALLOC_CPUSET;
1534
1535        /*
1536         * Go through the zonelist again. Let __GFP_HIGH and allocations
1537         * coming from realtime tasks go deeper into reserves.
1538         *
1539         * This is the last chance, in general, before the goto nopage.
1540         * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1541         * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1542         */
1543        page = get_page_from_freelist(gfp_mask, nodemask, order, zonelist,
1544                                                high_zoneidx, alloc_flags);
1545        if (page)
1546                goto got_pg;
1547
1548        /* This allocation should allow future memory freeing. */
1549
1550rebalance:
1551        if (((p->flags & PF_MEMALLOC) || unlikely(test_thread_flag(TIF_MEMDIE)))
1552                        && !in_interrupt()) {
1553                if (!(gfp_mask & __GFP_NOMEMALLOC)) {
1554nofail_alloc:
1555                        /* go through the zonelist yet again, ignoring mins */
1556                        page = get_page_from_freelist(gfp_mask, nodemask, order,
1557                                zonelist, high_zoneidx, ALLOC_NO_WATERMARKS);
1558                        if (page)
1559                                goto got_pg;
1560                        if (gfp_mask & __GFP_NOFAIL) {
1561                                congestion_wait(WRITE, HZ/50);
1562                                goto nofail_alloc;
1563                        }
1564                }
1565                goto nopage;
1566        }
1567
1568        /* Atomic allocations - we can't balance anything */
1569        if (!wait)
1570                goto nopage;
1571
1572        cond_resched();
1573
1574        /* We now go into synchronous reclaim */
1575        cpuset_memory_pressure_bump();
1576        /*
1577         * The task's cpuset might have expanded its set of allowable nodes
1578         */
1579        cpuset_update_task_memory_state();
1580        p->flags |= PF_MEMALLOC;
1581        reclaim_state.reclaimed_slab = 0;
1582        p->reclaim_state = &reclaim_state;
1583
1584        did_some_progress = try_to_free_pages(zonelist, order, gfp_mask);
1585
1586        p->reclaim_state = NULL;
1587        p->flags &= ~PF_MEMALLOC;
1588
1589        cond_resched();
1590
1591        if (order != 0)
1592                drain_all_pages();
1593
1594        if (likely(did_some_progress)) {
1595                page = get_page_from_freelist(gfp_mask, nodemask, order,
1596                                        zonelist, high_zoneidx, alloc_flags);
1597                if (page)
1598                        goto got_pg;
1599        } else if ((gfp_mask & __GFP_FS) && !(gfp_mask & __GFP_NORETRY)) {
1600                if (!try_set_zone_oom(zonelist, gfp_mask)) {
1601                        schedule_timeout_uninterruptible(1);
1602                        goto restart;
1603                }
1604
1605                /*
1606                 * Go through the zonelist yet one more time, keep
1607                 * very high watermark here, this is only to catch
1608                 * a parallel oom killing, we must fail if we're still
1609                 * under heavy pressure.
1610                 */
1611                page = get_page_from_freelist(gfp_mask|__GFP_HARDWALL, nodemask,
1612                        order, zonelist, high_zoneidx,
1613                        ALLOC_WMARK_HIGH|ALLOC_CPUSET);
1614                if (page) {
1615                        clear_zonelist_oom(zonelist, gfp_mask);
1616                        goto got_pg;
1617                }
1618
1619                /* The OOM killer will not help higher order allocs so fail */
1620                if (order > PAGE_ALLOC_COSTLY_ORDER) {
1621                        clear_zonelist_oom(zonelist, gfp_mask);
1622                        goto nopage;
1623                }
1624
1625                out_of_memory(zonelist, gfp_mask, order);
1626                clear_zonelist_oom(zonelist, gfp_mask);
1627                goto restart;
1628        }
1629
1630        /*
1631         * Don't let big-order allocations loop unless the caller explicitly
1632         * requests that.  Wait for some write requests to complete then retry.
1633         *
1634         * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1635         * means __GFP_NOFAIL, but that may not be true in other
1636         * implementations.
1637         *
1638         * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1639         * specified, then we retry until we no longer reclaim any pages
1640         * (above), or we've reclaimed an order of pages at least as
1641         * large as the allocation's order. In both cases, if the
1642         * allocation still fails, we stop retrying.
1643         */
1644        pages_reclaimed += did_some_progress;
1645        do_retry = 0;
1646        if (!(gfp_mask & __GFP_NORETRY)) {
1647                if (order <= PAGE_ALLOC_COSTLY_ORDER) {
1648                        do_retry = 1;
1649                } else {
1650                        if (gfp_mask & __GFP_REPEAT &&
1651                                pages_reclaimed < (1 << order))
1652                                        do_retry = 1;
1653                }
1654                if (gfp_mask & __GFP_NOFAIL)
1655                        do_retry = 1;
1656        }
1657        if (do_retry) {
1658                congestion_wait(WRITE, HZ/50);
1659                goto rebalance;
1660        }
1661
1662nopage:
1663        if (!(gfp_mask & __GFP_NOWARN) && printk_ratelimit()) {
1664                printk(KERN_WARNING "%s: page allocation failure."
1665                        " order:%d, mode:0x%x\n",
1666                        p->comm, order, gfp_mask);
1667                dump_stack();
1668                show_mem();
1669        }
1670got_pg:
1671        return page;
1672}
1673EXPORT_SYMBOL(__alloc_pages_internal);
1674
1675/*
1676 * Common helper functions.
1677 */
1678unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
1679{
1680        struct page * page;
1681        page = alloc_pages(gfp_mask, order);
1682        if (!page)
1683                return 0;
1684        return (unsigned long) page_address(page);
1685}
1686
1687EXPORT_SYMBOL(__get_free_pages);
1688
1689unsigned long get_zeroed_page(gfp_t gfp_mask)
1690{
1691        struct page * page;
1692
1693        /*
1694         * get_zeroed_page() returns a 32-bit address, which cannot represent
1695         * a highmem page
1696         */
1697        VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
1698
1699        page = alloc_pages(gfp_mask | __GFP_ZERO, 0);
1700        if (page)
1701                return (unsigned long) page_address(page);
1702        return 0;
1703}
1704
1705EXPORT_SYMBOL(get_zeroed_page);
1706
1707void __pagevec_free(struct pagevec *pvec)
1708{
1709        int i = pagevec_count(pvec);
1710
1711        while (--i >= 0)
1712                free_hot_cold_page(pvec->pages[i], pvec->cold);
1713}
1714
1715void __free_pages(struct page *page, unsigned int order)
1716{
1717        if (put_page_testzero(page)) {
1718                if (order == 0)
1719                        free_hot_page(page);
1720                else
1721                        __free_pages_ok(page, order);
1722        }
1723}
1724
1725EXPORT_SYMBOL(__free_pages);
1726
1727void free_pages(unsigned long addr, unsigned int order)
1728{
1729        if (addr != 0) {
1730                VM_BUG_ON(!virt_addr_valid((void *)addr));
1731                __free_pages(virt_to_page((void *)addr), order);
1732        }
1733}
1734
1735EXPORT_SYMBOL(free_pages);
1736
1737/**
1738 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1739 * @size: the number of bytes to allocate
1740 * @gfp_mask: GFP flags for the allocation
1741 *
1742 * This function is similar to alloc_pages(), except that it allocates the
1743 * minimum number of pages to satisfy the request.  alloc_pages() can only
1744 * allocate memory in power-of-two pages.
1745 *
1746 * This function is also limited by MAX_ORDER.
1747 *
1748 * Memory allocated by this function must be released by free_pages_exact().
1749 */
1750void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
1751{
1752        unsigned int order = get_order(size);
1753        unsigned long addr;
1754
1755        addr = __get_free_pages(gfp_mask, order);
1756        if (addr) {
1757                unsigned long alloc_end = addr + (PAGE_SIZE << order);
1758                unsigned long used = addr + PAGE_ALIGN(size);
1759
1760                split_page(virt_to_page(addr), order);
1761                while (used < alloc_end) {
1762                        free_page(used);
1763                        used += PAGE_SIZE;
1764                }
1765        }
1766
1767        return (void *)addr;
1768}
1769EXPORT_SYMBOL(alloc_pages_exact);
1770
1771/**
1772 * free_pages_exact - release memory allocated via alloc_pages_exact()
1773 * @virt: the value returned by alloc_pages_exact.
1774 * @size: size of allocation, same value as passed to alloc_pages_exact().
1775 *
1776 * Release the memory allocated by a previous call to alloc_pages_exact.
1777 */
1778void free_pages_exact(void *virt, size_t size)
1779{
1780        unsigned long addr = (unsigned long)virt;
1781        unsigned long end = addr + PAGE_ALIGN(size);
1782
1783        while (addr < end) {
1784                free_page(addr);
1785                addr += PAGE_SIZE;
1786        }
1787}
1788EXPORT_SYMBOL(free_pages_exact);
1789
1790static unsigned int nr_free_zone_pages(int offset)
1791{
1792        struct zoneref *z;
1793        struct zone *zone;
1794
1795        /* Just pick one node, since fallback list is circular */
1796        unsigned int sum = 0;
1797
1798        struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
1799
1800        for_each_zone_zonelist(zone, z, zonelist, offset) {
1801                unsigned long size = zone->present_pages;
1802                unsigned long high = zone->pages_high;
1803                if (size > high)
1804                        sum += size - high;
1805        }
1806
1807        return sum;
1808}
1809
1810/*
1811 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1812 */
1813unsigned int nr_free_buffer_pages(void)
1814{
1815        return nr_free_zone_pages(gfp_zone(GFP_USER));
1816}
1817EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
1818
1819/*
1820 * Amount of free RAM allocatable within all zones
1821 */
1822unsigned int nr_free_pagecache_pages(void)
1823{
1824        return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
1825}
1826
1827static inline void show_node(struct zone *zone)
1828{
1829        if (NUMA_BUILD)
1830                printk("Node %d ", zone_to_nid(zone));
1831}
1832
1833void si_meminfo(struct sysinfo *val)
1834{
1835        val->totalram = totalram_pages;
1836        val->sharedram = 0;
1837        val->freeram = global_page_state(NR_FREE_PAGES);
1838        val->bufferram = nr_blockdev_pages();
1839        val->totalhigh = totalhigh_pages;
1840        val->freehigh = nr_free_highpages();
1841        val->mem_unit = PAGE_SIZE;
1842}
1843
1844EXPORT_SYMBOL(si_meminfo);
1845
1846#ifdef CONFIG_NUMA
1847void si_meminfo_node(struct sysinfo *val, int nid)
1848{
1849        pg_data_t *pgdat = NODE_DATA(nid);
1850
1851        val->totalram = pgdat->node_present_pages;
1852        val->freeram = node_page_state(nid, NR_FREE_PAGES);
1853#ifdef CONFIG_HIGHMEM
1854        val->totalhigh = pgdat->node_zones[ZONE_HIGHMEM].present_pages;
1855        val->freehigh = zone_page_state(&pgdat->node_zones[ZONE_HIGHMEM],
1856                        NR_FREE_PAGES);
1857#else
1858        val->totalhigh = 0;
1859        val->freehigh = 0;
1860#endif
1861        val->mem_unit = PAGE_SIZE;
1862}
1863#endif
1864
1865#define K(x) ((x) << (PAGE_SHIFT-10))
1866
1867/*
1868 * Show free area list (used inside shift_scroll-lock stuff)
1869 * We also calculate the percentage fragmentation. We do this by counting the
1870 * memory on each free list with the exception of the first item on the list.
1871 */
1872void show_free_areas(void)
1873{
1874        int cpu;
1875        struct zone *zone;
1876
1877        for_each_zone(zone) {
1878                if (!populated_zone(zone))
1879                        continue;
1880
1881                show_node(zone);
1882                printk("%s per-cpu:\n", zone->name);
1883
1884                for_each_online_cpu(cpu) {
1885                        struct per_cpu_pageset *pageset;
1886
1887                        pageset = zone_pcp(zone, cpu);
1888
1889                        printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
1890                               cpu, pageset->pcp.high,
1891                               pageset->pcp.batch, pageset->pcp.count);
1892                }
1893        }
1894
1895        printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
1896                " inactive_file:%lu"
1897//TODO:  check/adjust line lengths
1898#ifdef CONFIG_UNEVICTABLE_LRU
1899                " unevictable:%lu"
1900#endif
1901                " dirty:%lu writeback:%lu unstable:%lu\n"
1902                " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1903                global_page_state(NR_ACTIVE_ANON),
1904                global_page_state(NR_ACTIVE_FILE),
1905                global_page_state(NR_INACTIVE_ANON),
1906                global_page_state(NR_INACTIVE_FILE),
1907#ifdef CONFIG_UNEVICTABLE_LRU
1908                global_page_state(NR_UNEVICTABLE),
1909#endif
1910                global_page_state(NR_FILE_DIRTY),
1911                global_page_state(NR_WRITEBACK),
1912                global_page_state(NR_UNSTABLE_NFS),
1913                global_page_state(NR_FREE_PAGES),
1914                global_page_state(NR_SLAB_RECLAIMABLE) +
1915                        global_page_state(NR_SLAB_UNRECLAIMABLE),
1916                global_page_state(NR_FILE_MAPPED),
1917                global_page_state(NR_PAGETABLE),
1918                global_page_state(NR_BOUNCE));
1919
1920        for_each_zone(zone) {
1921                int i;
1922
1923                if (!populated_zone(zone))
1924                        continue;
1925
1926                show_node(zone);
1927                printk("%s"
1928                        " free:%lukB"
1929                        " min:%lukB"
1930                        " low:%lukB"
1931                        " high:%lukB"
1932                        " active_anon:%lukB"
1933                        " inactive_anon:%lukB"
1934                        " active_file:%lukB"
1935                        " inactive_file:%lukB"
1936#ifdef CONFIG_UNEVICTABLE_LRU
1937                        " unevictable:%lukB"
1938#endif
1939                        " present:%lukB"
1940                        " pages_scanned:%lu"
1941                        " all_unreclaimable? %s"
1942                        "\n",
1943                        zone->name,
1944                        K(zone_page_state(zone, NR_FREE_PAGES)),
1945                        K(zone->pages_min),
1946                        K(zone->pages_low),
1947                        K(zone->pages_high),
1948                        K(zone_page_state(zone, NR_ACTIVE_ANON)),
1949                        K(zone_page_state(zone, NR_INACTIVE_ANON)),
1950                        K(zone_page_state(zone, NR_ACTIVE_FILE)),
1951                        K(zone_page_state(zone, NR_INACTIVE_FILE)),
1952#ifdef CONFIG_UNEVICTABLE_LRU
1953                        K(zone_page_state(zone, NR_UNEVICTABLE)),
1954#endif
1955                        K(zone->present_pages),
1956                        zone->pages_scanned,
1957                        (zone_is_all_unreclaimable(zone) ? "yes" : "no")
1958                        );
1959                printk("lowmem_reserve[]:");
1960                for (i = 0; i < MAX_NR_ZONES; i++)
1961                        printk(" %lu", zone->lowmem_reserve[i]);
1962                printk("\n");
1963        }
1964
1965        for_each_zone(zone) {
1966                unsigned long nr[MAX_ORDER], flags, order, total = 0;
1967
1968                if (!populated_zone(zone))
1969                        continue;
1970
1971                show_node(zone);
1972                printk("%s: ", zone->name);
1973
1974                spin_lock_irqsave(&zone->lock, flags);
1975                for (order = 0; order < MAX_ORDER; order++) {
1976                        nr[order] = zone->free_area[order].nr_free;
1977                        total += nr[order] << order;
1978                }
1979                spin_unlock_irqrestore(&zone->lock, flags);
1980                for (order = 0; order < MAX_ORDER; order++)
1981                        printk("%lu*%lukB ", nr[order], K(1UL) << order);
1982                printk("= %lukB\n", K(total));
1983        }
1984
1985        printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
1986
1987        show_swap_cache_info();
1988}
1989
1990static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
1991{
1992        zoneref->zone = zone;
1993        zoneref->zone_idx = zone_idx(zone);
1994}
1995
1996/*
1997 * Builds allocation fallback zone lists.
1998 *
1999 * Add all populated zones of a node to the zonelist.
2000 */
2001static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
2002                                int nr_zones, enum zone_type zone_type)
2003{
2004        struct zone *zone;
2005
2006        BUG_ON(zone_type >= MAX_NR_ZONES);
2007        zone_type++;
2008
2009        do {
2010                zone_type--;
2011                zone = pgdat->node_zones + zone_type;
2012                if (populated_zone(zone)) {
2013                        zoneref_set_zone(zone,
2014                                &zonelist->_zonerefs[nr_zones++]);
2015                        check_highest_zone(zone_type);
2016                }
2017
2018        } while (zone_type);
2019        return nr_zones;
2020}
2021
2022
2023/*
2024 *  zonelist_order:
2025 *  0 = automatic detection of better ordering.
2026 *  1 = order by ([node] distance, -zonetype)
2027 *  2 = order by (-zonetype, [node] distance)
2028 *
2029 *  If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2030 *  the same zonelist. So only NUMA can configure this param.
2031 */
2032#define ZONELIST_ORDER_DEFAULT  0
2033#define ZONELIST_ORDER_NODE     1
2034#define ZONELIST_ORDER_ZONE     2
2035
2036/* zonelist order in the kernel.
2037 * set_zonelist_order() will set this to NODE or ZONE.
2038 */
2039static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
2040static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
2041
2042
2043#ifdef CONFIG_NUMA
2044/* The value user specified ....changed by config */
2045static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2046/* string for sysctl */
2047#define NUMA_ZONELIST_ORDER_LEN 16
2048char numa_zonelist_order[16] = "default";
2049
2050/*
2051 * interface for configure zonelist ordering.
2052 * command line option "numa_zonelist_order"
2053 *      = "[dD]efault   - default, automatic configuration.
2054 *      = "[nN]ode      - order by node locality, then by zone within node
2055 *      = "[zZ]one      - order by zone, then by locality within zone
2056 */
2057
2058static int __parse_numa_zonelist_order(char *s)
2059{
2060        if (*s == 'd' || *s == 'D') {
2061                user_zonelist_order = ZONELIST_ORDER_DEFAULT;
2062        } else if (*s == 'n' || *s == 'N') {
2063                user_zonelist_order = ZONELIST_ORDER_NODE;
2064        } else if (*s == 'z' || *s == 'Z') {
2065                user_zonelist_order = ZONELIST_ORDER_ZONE;
2066        } else {
2067                printk(KERN_WARNING
2068                        "Ignoring invalid numa_zonelist_order value:  "
2069                        "%s\n", s);
2070                return -EINVAL;
2071        }
2072        return 0;
2073}
2074
2075static __init int setup_numa_zonelist_order(char *s)
2076{
2077        if (s)
2078                return __parse_numa_zonelist_order(s);
2079        return 0;
2080}
2081early_param("numa_zonelist_order", setup_numa_zonelist_order);
2082
2083/*
2084 * sysctl handler for numa_zonelist_order
2085 */
2086int numa_zonelist_order_handler(ctl_table *table, int write,
2087                struct file *file, void __user *buffer, size_t *length,
2088                loff_t *ppos)
2089{
2090        char saved_string[NUMA_ZONELIST_ORDER_LEN];
2091        int ret;
2092
2093        if (write)
2094                strncpy(saved_string, (char*)table->data,
2095                        NUMA_ZONELIST_ORDER_LEN);
2096        ret = proc_dostring(table, write, file, buffer, length, ppos);
2097        if (ret)
2098                return ret;
2099        if (write) {
2100                int oldval = user_zonelist_order;
2101                if (__parse_numa_zonelist_order((char*)table->data)) {
2102                        /*
2103                         * bogus value.  restore saved string
2104                         */
2105                        strncpy((char*)table->data, saved_string,
2106                                NUMA_ZONELIST_ORDER_LEN);
2107                        user_zonelist_order = oldval;
2108                } else if (oldval != user_zonelist_order)
2109                        build_all_zonelists();
2110        }
2111        return 0;
2112}
2113
2114
2115#define MAX_NODE_LOAD (num_online_nodes())
2116static int node_load[MAX_NUMNODES];
2117
2118/**
2119 * find_next_best_node - find the next node that should appear in a given node's fallback list
2120 * @node: node whose fallback list we're appending
2121 * @used_node_mask: nodemask_t of already used nodes
2122 *
2123 * We use a number of factors to determine which is the next node that should
2124 * appear on a given node's fallback list.  The node should not have appeared
2125 * already in @node's fallback list, and it should be the next closest node
2126 * according to the distance array (which contains arbitrary distance values
2127 * from each node to each node in the system), and should also prefer nodes
2128 * with no CPUs, since presumably they'll have very little allocation pressure
2129 * on them otherwise.
2130 * It returns -1 if no node is found.
2131 */
2132static int find_next_best_node(int node, nodemask_t *used_node_mask)
2133{
2134        int n, val;
2135        int min_val = INT_MAX;
2136        int best_node = -1;
2137        node_to_cpumask_ptr(tmp, 0);
2138
2139        /* Use the local node if we haven't already */
2140        if (!node_isset(node, *used_node_mask)) {
2141                node_set(node, *used_node_mask);
2142                return node;
2143        }
2144
2145        for_each_node_state(n, N_HIGH_MEMORY) {
2146
2147                /* Don't want a node to appear more than once */
2148                if (node_isset(n, *used_node_mask))
2149                        continue;
2150
2151                /* Use the distance array to find the distance */
2152                val = node_distance(node, n);
2153
2154                /* Penalize nodes under us ("prefer the next node") */
2155                val += (n < node);
2156
2157                /* Give preference to headless and unused nodes */
2158                node_to_cpumask_ptr_next(tmp, n);
2159                if (!cpus_empty(*tmp))
2160                        val += PENALTY_FOR_NODE_WITH_CPUS;
2161
2162                /* Slight preference for less loaded node */
2163                val *= (MAX_NODE_LOAD*MAX_NUMNODES);
2164                val += node_load[n];
2165
2166                if (val < min_val) {
2167                        min_val = val;
2168                        best_node = n;
2169                }
2170        }
2171
2172        if (best_node >= 0)
2173                node_set(best_node, *used_node_mask);
2174
2175        return best_node;
2176}
2177
2178
2179/*
2180 * Build zonelists ordered by node and zones within node.
2181 * This results in maximum locality--normal zone overflows into local
2182 * DMA zone, if any--but risks exhausting DMA zone.
2183 */
2184static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
2185{
2186        int j;
2187        struct zonelist *zonelist;
2188
2189        zonelist = &pgdat->node_zonelists[0];
2190        for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
2191                ;
2192        j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2193                                                        MAX_NR_ZONES - 1);
2194        zonelist->_zonerefs[j].zone = NULL;
2195        zonelist->_zonerefs[j].zone_idx = 0;
2196}
2197
2198/*
2199 * Build gfp_thisnode zonelists
2200 */
2201static void build_thisnode_zonelists(pg_data_t *pgdat)
2202{
2203        int j;
2204        struct zonelist *zonelist;
2205
2206        zonelist = &pgdat->node_zonelists[1];
2207        j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2208        zonelist->_zonerefs[j].zone = NULL;
2209        zonelist->_zonerefs[j].zone_idx = 0;
2210}
2211
2212/*
2213 * Build zonelists ordered by zone and nodes within zones.
2214 * This results in conserving DMA zone[s] until all Normal memory is
2215 * exhausted, but results in overflowing to remote node while memory
2216 * may still exist in local DMA zone.
2217 */
2218static int node_order[MAX_NUMNODES];
2219
2220static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
2221{
2222        int pos, j, node;
2223        int zone_type;          /* needs to be signed */
2224        struct zone *z;
2225        struct zonelist *zonelist;
2226
2227        zonelist = &pgdat->node_zonelists[0];
2228        pos = 0;
2229        for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
2230                for (j = 0; j < nr_nodes; j++) {
2231                        node = node_order[j];
2232                        z = &NODE_DATA(node)->node_zones[zone_type];
2233                        if (populated_zone(z)) {
2234                                zoneref_set_zone(z,
2235                                        &zonelist->_zonerefs[pos++]);
2236                                check_highest_zone(zone_type);
2237                        }
2238                }
2239        }
2240        zonelist->_zonerefs[pos].zone = NULL;
2241        zonelist->_zonerefs[pos].zone_idx = 0;
2242}
2243
2244static int default_zonelist_order(void)
2245{
2246        int nid, zone_type;
2247        unsigned long low_kmem_size,total_size;
2248        struct zone *z;
2249        int average_size;
2250        /*
2251         * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2252         * If they are really small and used heavily, the system can fall
2253         * into OOM very easily.
2254         * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2255         */
2256        /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2257        low_kmem_size = 0;
2258        total_size = 0;
2259        for_each_online_node(nid) {
2260                for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2261                        z = &NODE_DATA(nid)->node_zones[zone_type];
2262                        if (populated_zone(z)) {
2263                                if (zone_type < ZONE_NORMAL)
2264                                        low_kmem_size += z->present_pages;
2265                                total_size += z->present_pages;
2266                        }
2267                }
2268        }
2269        if (!low_kmem_size ||  /* there are no DMA area. */
2270            low_kmem_size > total_size/2) /* DMA/DMA32 is big. */
2271                return ZONELIST_ORDER_NODE;
2272        /*
2273         * look into each node's config.
2274         * If there is a node whose DMA/DMA32 memory is very big area on
2275         * local memory, NODE_ORDER may be suitable.
2276         */
2277        average_size = total_size /
2278                                (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
2279        for_each_online_node(nid) {
2280                low_kmem_size = 0;
2281                total_size = 0;
2282                for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
2283                        z = &NODE_DATA(nid)->node_zones[zone_type];
2284                        if (populated_zone(z)) {
2285                                if (zone_type < ZONE_NORMAL)
2286                                        low_kmem_size += z->present_pages;
2287                                total_size += z->present_pages;
2288                        }
2289                }
2290                if (low_kmem_size &&
2291                    total_size > average_size && /* ignore small node */
2292                    low_kmem_size > total_size * 70/100)
2293                        return ZONELIST_ORDER_NODE;
2294        }
2295        return ZONELIST_ORDER_ZONE;
2296}
2297
2298static void set_zonelist_order(void)
2299{
2300        if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
2301                current_zonelist_order = default_zonelist_order();
2302        else
2303                current_zonelist_order = user_zonelist_order;
2304}
2305
2306static void build_zonelists(pg_data_t *pgdat)
2307{
2308        int j, node, load;
2309        enum zone_type i;
2310        nodemask_t used_mask;
2311        int local_node, prev_node;
2312        struct zonelist *zonelist;
2313        int order = current_zonelist_order;
2314
2315        /* initialize zonelists */
2316        for (i = 0; i < MAX_ZONELISTS; i++) {
2317                zonelist = pgdat->node_zonelists + i;
2318                zonelist->_zonerefs[0].zone = NULL;
2319                zonelist->_zonerefs[0].zone_idx = 0;
2320        }
2321
2322        /* NUMA-aware ordering of nodes */
2323        local_node = pgdat->node_id;
2324        load = num_online_nodes();
2325        prev_node = local_node;
2326        nodes_clear(used_mask);
2327
2328        memset(node_load, 0, sizeof(node_load));
2329        memset(node_order, 0, sizeof(node_order));
2330        j = 0;
2331
2332        while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
2333                int distance = node_distance(local_node, node);
2334
2335                /*
2336                 * If another node is sufficiently far away then it is better
2337                 * to reclaim pages in a zone before going off node.
2338                 */
2339                if (distance > RECLAIM_DISTANCE)
2340                        zone_reclaim_mode = 1;
2341
2342                /*
2343                 * We don't want to pressure a particular node.
2344                 * So adding penalty to the first node in same
2345                 * distance group to make it round-robin.
2346                 */
2347                if (distance != node_distance(local_node, prev_node))
2348                        node_load[node] = load;
2349
2350                prev_node = node;
2351                load--;
2352                if (order == ZONELIST_ORDER_NODE)
2353                        build_zonelists_in_node_order(pgdat, node);
2354                else
2355                        node_order[j++] = node; /* remember order */
2356        }
2357
2358        if (order == ZONELIST_ORDER_ZONE) {
2359                /* calculate node order -- i.e., DMA last! */
2360                build_zonelists_in_zone_order(pgdat, j);
2361        }
2362
2363        build_thisnode_zonelists(pgdat);
2364}
2365
2366/* Construct the zonelist performance cache - see further mmzone.h */
2367static void build_zonelist_cache(pg_data_t *pgdat)
2368{
2369        struct zonelist *zonelist;
2370        struct zonelist_cache *zlc;
2371        struct zoneref *z;
2372
2373        zonelist = &pgdat->node_zonelists[0];
2374        zonelist->zlcache_ptr = zlc = &zonelist->zlcache;
2375        bitmap_zero(zlc->fullzones, MAX_ZONES_PER_ZONELIST);
2376        for (z = zonelist->_zonerefs; z->zone; z++)
2377                zlc->z_to_n[z - zonelist->_zonerefs] = zonelist_node_idx(z);
2378}
2379
2380
2381#else   /* CONFIG_NUMA */
2382
2383static void set_zonelist_order(void)
2384{
2385        current_zonelist_order = ZONELIST_ORDER_ZONE;
2386}
2387
2388static void build_zonelists(pg_data_t *pgdat)
2389{
2390        int node, local_node;
2391        enum zone_type j;
2392        struct zonelist *zonelist;
2393
2394        local_node = pgdat->node_id;
2395
2396        zonelist = &pgdat->node_zonelists[0];
2397        j = build_zonelists_node(pgdat, zonelist, 0, MAX_NR_ZONES - 1);
2398
2399        /*
2400         * Now we build the zonelist so that it contains the zones
2401         * of all the other nodes.
2402         * We don't want to pressure a particular node, so when
2403         * building the zones for node N, we make sure that the
2404         * zones coming right after the local ones are those from
2405         * node N+1 (modulo N)
2406         */
2407        for (node = local_node + 1; node < MAX_NUMNODES; node++) {
2408                if (!node_online(node))
2409                        continue;
2410                j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2411                                                        MAX_NR_ZONES - 1);
2412        }
2413        for (node = 0; node < local_node; node++) {
2414                if (!node_online(node))
2415                        continue;
2416                j = build_zonelists_node(NODE_DATA(node), zonelist, j,
2417                                                        MAX_NR_ZONES - 1);
2418        }
2419
2420        zonelist->_zonerefs[j].zone = NULL;
2421        zonelist->_zonerefs[j].zone_idx = 0;
2422}
2423
2424/* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2425static void build_zonelist_cache(pg_data_t *pgdat)
2426{
2427        pgdat->node_zonelists[0].zlcache_ptr = NULL;
2428}
2429
2430#endif  /* CONFIG_NUMA */
2431
2432/* return values int ....just for stop_machine() */
2433static int __build_all_zonelists(void *dummy)
2434{
2435        int nid;
2436
2437        for_each_online_node(nid) {
2438                pg_data_t *pgdat = NODE_DATA(nid);
2439
2440                build_zonelists(pgdat);
2441                build_zonelist_cache(pgdat);
2442        }
2443        return 0;
2444}
2445
2446void build_all_zonelists(void)
2447{
2448        set_zonelist_order();
2449
2450        if (system_state == SYSTEM_BOOTING) {
2451                __build_all_zonelists(NULL);
2452                mminit_verify_zonelist();
2453                cpuset_init_current_mems_allowed();
2454        } else {
2455                /* we have to stop all cpus to guarantee there is no user
2456                   of zonelist */
2457                stop_machine(__build_all_zonelists, NULL, NULL);
2458                /* cpuset refresh routine should be here */
2459        }
2460        vm_total_pages = nr_free_pagecache_pages();
2461        /*
2462         * Disable grouping by mobility if the number of pages in the
2463         * system is too low to allow the mechanism to work. It would be
2464         * more accurate, but expensive to check per-zone. This check is
2465         * made on memory-hotadd so a system can start with mobility
2466         * disabled and enable it later
2467         */
2468        if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
2469                page_group_by_mobility_disabled = 1;
2470        else
2471                page_group_by_mobility_disabled = 0;
2472
2473        printk("Built %i zonelists in %s order, mobility grouping %s.  "
2474                "Total pages: %ld\n",
2475                        num_online_nodes(),
2476                        zonelist_order_name[current_zonelist_order],
2477                        page_group_by_mobility_disabled ? "off" : "on",
2478                        vm_total_pages);
2479#ifdef CONFIG_NUMA
2480        printk("Policy zone: %s\n", zone_names[policy_zone]);
2481#endif
2482}
2483
2484/*
2485 * Helper functions to size the waitqueue hash table.
2486 * Essentially these want to choose hash table sizes sufficiently
2487 * large so that collisions trying to wait on pages are rare.
2488 * But in fact, the number of active page waitqueues on typical
2489 * systems is ridiculously low, less than 200. So this is even
2490 * conservative, even though it seems large.
2491 *
2492 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2493 * waitqueues, i.e. the size of the waitq table given the number of pages.
2494 */
2495#define PAGES_PER_WAITQUEUE     256
2496
2497#ifndef CONFIG_MEMORY_HOTPLUG
2498static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2499{
2500        unsigned long size = 1;
2501
2502        pages /= PAGES_PER_WAITQUEUE;
2503
2504        while (size < pages)
2505                size <<= 1;
2506
2507        /*
2508         * Once we have dozens or even hundreds of threads sleeping
2509         * on IO we've got bigger problems than wait queue collision.
2510         * Limit the size of the wait table to a reasonable size.
2511         */
2512        size = min(size, 4096UL);
2513
2514        return max(size, 4UL);
2515}
2516#else
2517/*
2518 * A zone's size might be changed by hot-add, so it is not possible to determine
2519 * a suitable size for its wait_table.  So we use the maximum size now.
2520 *
2521 * The max wait table size = 4096 x sizeof(wait_queue_head_t).   ie:
2522 *
2523 *    i386 (preemption config)    : 4096 x 16 = 64Kbyte.
2524 *    ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2525 *    ia64, x86-64 (preemption)   : 4096 x 24 = 96Kbyte.
2526 *
2527 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2528 * or more by the traditional way. (See above).  It equals:
2529 *
2530 *    i386, x86-64, powerpc(4K page size) : =  ( 2G + 1M)byte.
2531 *    ia64(16K page size)                 : =  ( 8G + 4M)byte.
2532 *    powerpc (64K page size)             : =  (32G +16M)byte.
2533 */
2534static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
2535{
2536        return 4096UL;
2537}
2538#endif
2539
2540/*
2541 * This is an integer logarithm so that shifts can be used later
2542 * to extract the more random high bits from the multiplicative
2543 * hash function before the remainder is taken.
2544 */
2545static inline unsigned long wait_table_bits(unsigned long size)
2546{
2547        return ffz(~size);
2548}
2549
2550#define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2551
2552/*
2553 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2554 * of blocks reserved is based on zone->pages_min. The memory within the
2555 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2556 * higher will lead to a bigger reserve which will get freed as contiguous
2557 * blocks as reclaim kicks in
2558 */
2559static void setup_zone_migrate_reserve(struct zone *zone)
2560{
2561        unsigned long start_pfn, pfn, end_pfn;
2562        struct page *page;
2563        unsigned long reserve, block_migratetype;
2564
2565        /* Get the start pfn, end pfn and the number of blocks to reserve */
2566        start_pfn = zone->zone_start_pfn;
2567        end_pfn = start_pfn + zone->spanned_pages;
2568        reserve = roundup(zone->pages_min, pageblock_nr_pages) >>
2569                                                        pageblock_order;
2570
2571        for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
2572                if (!pfn_valid(pfn))
2573                        continue;
2574                page = pfn_to_page(pfn);
2575
2576                /* Watch out for overlapping nodes */
2577                if (page_to_nid(page) != zone_to_nid(zone))
2578                        continue;
2579
2580                /* Blocks with reserved pages will never free, skip them. */
2581                if (PageReserved(page))
2582                        continue;
2583
2584                block_migratetype = get_pageblock_migratetype(page);
2585
2586                /* If this block is reserved, account for it */
2587                if (reserve > 0 && block_migratetype == MIGRATE_RESERVE) {
2588                        reserve--;
2589                        continue;
2590                }
2591
2592                /* Suitable for reserving if this block is movable */
2593                if (reserve > 0 && block_migratetype == MIGRATE_MOVABLE) {
2594                        set_pageblock_migratetype(page, MIGRATE_RESERVE);
2595                        move_freepages_block(zone, page, MIGRATE_RESERVE);
2596                        reserve--;
2597                        continue;
2598                }
2599
2600                /*
2601                 * If the reserve is met and this is a previous reserved block,
2602                 * take it back
2603                 */
2604                if (block_migratetype == MIGRATE_RESERVE) {
2605                        set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2606                        move_freepages_block(zone, page, MIGRATE_MOVABLE);
2607                }
2608        }
2609}
2610
2611/*
2612 * Initially all pages are reserved - free ones are freed
2613 * up by free_all_bootmem() once the early boot process is
2614 * done. Non-atomic initialization, single-pass.
2615 */
2616void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
2617                unsigned long start_pfn, enum memmap_context context)
2618{
2619        struct page *page;
2620        unsigned long end_pfn = start_pfn + size;
2621        unsigned long pfn;
2622        struct zone *z;
2623
2624        if (highest_memmap_pfn < end_pfn - 1)
2625                highest_memmap_pfn = end_pfn - 1;
2626
2627        z = &NODE_DATA(nid)->node_zones[zone];
2628        for (pfn = start_pfn; pfn < end_pfn; pfn++) {
2629                /*
2630                 * There can be holes in boot-time mem_map[]s
2631                 * handed to this function.  They do not
2632                 * exist on hotplugged memory.
2633                 */
2634                if (context == MEMMAP_EARLY) {
2635                        if (!early_pfn_valid(pfn))
2636                                continue;
2637                        if (!early_pfn_in_nid(pfn, nid))
2638                                continue;
2639                }
2640                page = pfn_to_page(pfn);
2641                set_page_links(page, zone, nid, pfn);
2642                mminit_verify_page_links(page, zone, nid, pfn);
2643                init_page_count(page);
2644                reset_page_mapcount(page);
2645                SetPageReserved(page);
2646                /*
2647                 * Mark the block movable so that blocks are reserved for
2648                 * movable at startup. This will force kernel allocations
2649                 * to reserve their blocks rather than leaking throughout
2650                 * the address space during boot when many long-lived
2651                 * kernel allocations are made. Later some blocks near
2652                 * the start are marked MIGRATE_RESERVE by
2653                 * setup_zone_migrate_reserve()
2654                 *
2655                 * bitmap is created for zone's valid pfn range. but memmap
2656                 * can be created for invalid pages (for alignment)
2657                 * check here not to call set_pageblock_migratetype() against
2658                 * pfn out of zone.
2659                 */
2660                if ((z->zone_start_pfn <= pfn)
2661                    && (pfn < z->zone_start_pfn + z->spanned_pages)
2662                    && !(pfn & (pageblock_nr_pages - 1)))
2663                        set_pageblock_migratetype(page, MIGRATE_MOVABLE);
2664
2665                INIT_LIST_HEAD(&page->lru);
2666#ifdef WANT_PAGE_VIRTUAL
2667                /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2668                if (!is_highmem_idx(zone))
2669                        set_page_address(page, __va(pfn << PAGE_SHIFT));
2670#endif
2671        }
2672}
2673
2674static void __meminit zone_init_free_lists(struct zone *zone)
2675{
2676        int order, t;
2677        for_each_migratetype_order(order, t) {
2678                INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
2679                zone->free_area[order].nr_free = 0;
2680        }
2681}
2682
2683#ifndef __HAVE_ARCH_MEMMAP_INIT
2684#define memmap_init(size, nid, zone, start_pfn) \
2685        memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2686#endif
2687
2688static int zone_batchsize(struct zone *zone)
2689{
2690        int batch;
2691
2692        /*
2693         * The per-cpu-pages pools are set to around 1000th of the
2694         * size of the zone.  But no more than 1/2 of a meg.
2695         *
2696         * OK, so we don't know how big the cache is.  So guess.
2697         */
2698        batch = zone->present_pages / 1024;
2699        if (batch * PAGE_SIZE > 512 * 1024)
2700                batch = (512 * 1024) / PAGE_SIZE;
2701        batch /= 4;             /* We effectively *= 4 below */
2702        if (batch < 1)
2703                batch = 1;
2704
2705        /*
2706         * Clamp the batch to a 2^n - 1 value. Having a power
2707         * of 2 value was found to be more likely to have
2708         * suboptimal cache aliasing properties in some cases.
2709         *
2710         * For example if 2 tasks are alternately allocating
2711         * batches of pages, one task can end up with a lot
2712         * of pages of one half of the possible page colors
2713         * and the other with pages of the other colors.
2714         */
2715        batch = (1 << (fls(batch + batch/2)-1)) - 1;
2716
2717        return batch;
2718}
2719
2720static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
2721{
2722        struct per_cpu_pages *pcp;
2723
2724        memset(p, 0, sizeof(*p));
2725
2726        pcp = &p->pcp;
2727        pcp->count = 0;
2728        pcp->high = 6 * batch;
2729        pcp->batch = max(1UL, 1 * batch);
2730        INIT_LIST_HEAD(&pcp->list);
2731}
2732
2733/*
2734 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2735 * to the value high for the pageset p.
2736 */
2737
2738static void setup_pagelist_highmark(struct per_cpu_pageset *p,
2739                                unsigned long high)
2740{
2741        struct per_cpu_pages *pcp;
2742
2743        pcp = &p->pcp;
2744        pcp->high = high;
2745        pcp->batch = max(1UL, high/4);
2746        if ((high/4) > (PAGE_SHIFT * 8))
2747                pcp->batch = PAGE_SHIFT * 8;
2748}
2749
2750
2751#ifdef CONFIG_NUMA
2752/*
2753 * Boot pageset table. One per cpu which is going to be used for all
2754 * zones and all nodes. The parameters will be set in such a way
2755 * that an item put on a list will immediately be handed over to
2756 * the buddy list. This is safe since pageset manipulation is done
2757 * with interrupts disabled.
2758 *
2759 * Some NUMA counter updates may also be caught by the boot pagesets.
2760 *
2761 * The boot_pagesets must be kept even after bootup is complete for
2762 * unused processors and/or zones. They do play a role for bootstrapping
2763 * hotplugged processors.
2764 *
2765 * zoneinfo_show() and maybe other functions do
2766 * not check if the processor is online before following the pageset pointer.
2767 * Other parts of the kernel may not check if the zone is available.
2768 */
2769static struct per_cpu_pageset boot_pageset[NR_CPUS];
2770
2771/*
2772 * Dynamically allocate memory for the
2773 * per cpu pageset array in struct zone.
2774 */
2775static int __cpuinit process_zones(int cpu)
2776{
2777        struct zone *zone, *dzone;
2778        int node = cpu_to_node(cpu);
2779
2780        node_set_state(node, N_CPU);    /* this node has a cpu */
2781
2782        for_each_zone(zone) {
2783
2784                if (!populated_zone(zone))
2785                        continue;
2786
2787                zone_pcp(zone, cpu) = kmalloc_node(sizeof(struct per_cpu_pageset),
2788                                         GFP_KERNEL, node);
2789                if (!zone_pcp(zone, cpu))
2790                        goto bad;
2791
2792                setup_pageset(zone_pcp(zone, cpu), zone_batchsize(zone));
2793
2794                if (percpu_pagelist_fraction)
2795                        setup_pagelist_highmark(zone_pcp(zone, cpu),
2796                                (zone->present_pages / percpu_pagelist_fraction));
2797        }
2798
2799        return 0;
2800bad:
2801        for_each_zone(dzone) {
2802                if (!populated_zone(dzone))
2803                        continue;
2804                if (dzone == zone)
2805                        break;
2806                kfree(zone_pcp(dzone, cpu));
2807                zone_pcp(dzone, cpu) = NULL;
2808        }
2809        return -ENOMEM;
2810}
2811
2812static inline void free_zone_pagesets(int cpu)
2813{
2814        struct zone *zone;
2815
2816        for_each_zone(zone) {
2817                struct per_cpu_pageset *pset = zone_pcp(zone, cpu);
2818
2819                /* Free per_cpu_pageset if it is slab allocated */
2820                if (pset != &boot_pageset[cpu])
2821                        kfree(pset);
2822                zone_pcp(zone, cpu) = NULL;
2823        }
2824}
2825
2826static int __cpuinit pageset_cpuup_callback(struct notifier_block *nfb,
2827                unsigned long action,
2828                void *hcpu)
2829{
2830        int cpu = (long)hcpu;
2831        int ret = NOTIFY_OK;
2832
2833        switch (action) {
2834        case CPU_UP_PREPARE:
2835        case CPU_UP_PREPARE_FROZEN:
2836                if (process_zones(cpu))
2837                        ret = NOTIFY_BAD;
2838                break;
2839        case CPU_UP_CANCELED:
2840        case CPU_UP_CANCELED_FROZEN:
2841        case CPU_DEAD:
2842        case CPU_DEAD_FROZEN:
2843                free_zone_pagesets(cpu);
2844                break;
2845        default:
2846                break;
2847        }
2848        return ret;
2849}
2850
2851static struct notifier_block __cpuinitdata pageset_notifier =
2852        { &pageset_cpuup_callback, NULL, 0 };
2853
2854void __init setup_per_cpu_pageset(void)
2855{
2856        int err;
2857
2858        /* Initialize per_cpu_pageset for cpu 0.
2859         * A cpuup callback will do this for every cpu
2860         * as it comes online
2861         */
2862        err = process_zones(smp_processor_id());
2863        BUG_ON(err);
2864        register_cpu_notifier(&pageset_notifier);
2865}
2866
2867#endif
2868
2869static noinline __init_refok
2870int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
2871{
2872        int i;
2873        struct pglist_data *pgdat = zone->zone_pgdat;
2874        size_t alloc_size;
2875
2876        /*
2877         * The per-page waitqueue mechanism uses hashed waitqueues
2878         * per zone.
2879         */
2880        zone->wait_table_hash_nr_entries =
2881                 wait_table_hash_nr_entries(zone_size_pages);
2882        zone->wait_table_bits =
2883                wait_table_bits(zone->wait_table_hash_nr_entries);
2884        alloc_size = zone->wait_table_hash_nr_entries
2885                                        * sizeof(wait_queue_head_t);
2886
2887        if (!slab_is_available()) {
2888                zone->wait_table = (wait_queue_head_t *)
2889                        alloc_bootmem_node(pgdat, alloc_size);
2890        } else {
2891                /*
2892                 * This case means that a zone whose size was 0 gets new memory
2893                 * via memory hot-add.
2894                 * But it may be the case that a new node was hot-added.  In
2895                 * this case vmalloc() will not be able to use this new node's
2896                 * memory - this wait_table must be initialized to use this new
2897                 * node itself as well.
2898                 * To use this new node's memory, further consideration will be
2899                 * necessary.
2900                 */
2901                zone->wait_table = vmalloc(alloc_size);
2902        }
2903        if (!zone->wait_table)
2904                return -ENOMEM;
2905
2906        for(i = 0; i < zone->wait_table_hash_nr_entries; ++i)
2907                init_waitqueue_head(zone->wait_table + i);
2908
2909        return 0;
2910}
2911
2912static __meminit void zone_pcp_init(struct zone *zone)
2913{
2914        int cpu;
2915        unsigned long batch = zone_batchsize(zone);
2916
2917        for (cpu = 0; cpu < NR_CPUS; cpu++) {
2918#ifdef CONFIG_NUMA
2919                /* Early boot. Slab allocator not functional yet */
2920                zone_pcp(zone, cpu) = &boot_pageset[cpu];
2921                setup_pageset(&boot_pageset[cpu],0);
2922#else
2923                setup_pageset(zone_pcp(zone,cpu), batch);
2924#endif
2925        }
2926        if (zone->present_pages)
2927                printk(KERN_DEBUG "  %s zone: %lu pages, LIFO batch:%lu\n",
2928                        zone->name, zone->present_pages, batch);
2929}
2930
2931__meminit int init_currently_empty_zone(struct zone *zone,
2932                                        unsigned long zone_start_pfn,
2933                                        unsigned long size,
2934                                        enum memmap_context context)
2935{
2936        struct pglist_data *pgdat = zone->zone_pgdat;
2937        int ret;
2938        ret = zone_wait_table_init(zone, size);
2939        if (ret)
2940                return ret;
2941        pgdat->nr_zones = zone_idx(zone) + 1;
2942
2943        zone->zone_start_pfn = zone_start_pfn;
2944
2945        mminit_dprintk(MMINIT_TRACE, "memmap_init",
2946                        "Initialising map node %d zone %lu pfns %lu -> %lu\n",
2947                        pgdat->node_id,
2948                        (unsigned long)zone_idx(zone),
2949                        zone_start_pfn, (zone_start_pfn + size));
2950
2951        zone_init_free_lists(zone);
2952
2953        return 0;
2954}
2955
2956#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2957/*
2958 * Basic iterator support. Return the first range of PFNs for a node
2959 * Note: nid == MAX_NUMNODES returns first region regardless of node
2960 */
2961static int __meminit first_active_region_index_in_nid(int nid)
2962{
2963        int i;
2964
2965        for (i = 0; i < nr_nodemap_entries; i++)
2966                if (nid == MAX_NUMNODES || early_node_map[i].nid == nid)
2967                        return i;
2968
2969        return -1;
2970}
2971
2972/*
2973 * Basic iterator support. Return the next active range of PFNs for a node
2974 * Note: nid == MAX_NUMNODES returns next region regardless of node
2975 */
2976static int __meminit next_active_region_index_in_nid(int index, int nid)
2977{
2978        for (index = index + 1; index < nr_nodemap_entries; index++)
2979                if (nid == MAX_NUMNODES || early_node_map[index].nid == nid)
2980                        return index;
2981
2982        return -1;
2983}
2984
2985#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2986/*
2987 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2988 * Architectures may implement their own version but if add_active_range()
2989 * was used and there are no special requirements, this is a convenient
2990 * alternative
2991 */
2992int __meminit __early_pfn_to_nid(unsigned long pfn)
2993{
2994        int i;
2995
2996        for (i = 0; i < nr_nodemap_entries; i++) {
2997                unsigned long start_pfn = early_node_map[i].start_pfn;
2998                unsigned long end_pfn = early_node_map[i].end_pfn;
2999
3000                if (start_pfn <= pfn && pfn < end_pfn)
3001                        return early_node_map[i].nid;
3002        }
3003        /* This is a memory hole */
3004        return -1;
3005}
3006#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3007
3008int __meminit early_pfn_to_nid(unsigned long pfn)
3009{
3010        int nid;
3011
3012        nid = __early_pfn_to_nid(pfn);
3013        if (nid >= 0)
3014                return nid;
3015        /* just returns 0 */
3016        return 0;
3017}
3018
3019#ifdef CONFIG_NODES_SPAN_OTHER_NODES
3020bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
3021{
3022        int nid;
3023
3024        nid = __early_pfn_to_nid(pfn);
3025        if (nid >= 0 && nid != node)
3026                return false;
3027        return true;
3028}
3029#endif
3030
3031/* Basic iterator support to walk early_node_map[] */
3032#define for_each_active_range_index_in_nid(i, nid) \
3033        for (i = first_active_region_index_in_nid(nid); i != -1; \
3034                                i = next_active_region_index_in_nid(i, nid))
3035
3036/**
3037 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3038 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3039 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3040 *
3041 * If an architecture guarantees that all ranges registered with
3042 * add_active_ranges() contain no holes and may be freed, this
3043 * this function may be used instead of calling free_bootmem() manually.
3044 */
3045void __init free_bootmem_with_active_regions(int nid,
3046                                                unsigned long max_low_pfn)
3047{
3048        int i;
3049
3050        for_each_active_range_index_in_nid(i, nid) {
3051                unsigned long size_pages = 0;
3052                unsigned long end_pfn = early_node_map[i].end_pfn;
3053
3054                if (early_node_map[i].start_pfn >= max_low_pfn)
3055                        continue;
3056
3057                if (end_pfn > max_low_pfn)
3058                        end_pfn = max_low_pfn;
3059
3060                size_pages = end_pfn - early_node_map[i].start_pfn;
3061                free_bootmem_node(NODE_DATA(early_node_map[i].nid),
3062                                PFN_PHYS(early_node_map[i].start_pfn),
3063                                size_pages << PAGE_SHIFT);
3064        }
3065}
3066
3067void __init work_with_active_regions(int nid, work_fn_t work_fn, void *data)
3068{
3069        int i;
3070        int ret;
3071
3072        for_each_active_range_index_in_nid(i, nid) {
3073                ret = work_fn(early_node_map[i].start_pfn,
3074                              early_node_map[i].end_pfn, data);
3075                if (ret)
3076                        break;
3077        }
3078}
3079/**
3080 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3081 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3082 *
3083 * If an architecture guarantees that all ranges registered with
3084 * add_active_ranges() contain no holes and may be freed, this
3085 * function may be used instead of calling memory_present() manually.
3086 */
3087void __init sparse_memory_present_with_active_regions(int nid)
3088{
3089        int i;
3090
3091        for_each_active_range_index_in_nid(i, nid)
3092                memory_present(early_node_map[i].nid,
3093                                early_node_map[i].start_pfn,
3094                                early_node_map[i].end_pfn);
3095}
3096
3097/**
3098 * push_node_boundaries - Push node boundaries to at least the requested boundary
3099 * @nid: The nid of the node to push the boundary for
3100 * @start_pfn: The start pfn of the node
3101 * @end_pfn: The end pfn of the node
3102 *
3103 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
3104 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
3105 * be hotplugged even though no physical memory exists. This function allows
3106 * an arch to push out the node boundaries so mem_map is allocated that can
3107 * be used later.
3108 */
3109#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3110void __init push_node_boundaries(unsigned int nid,
3111                unsigned long start_pfn, unsigned long end_pfn)
3112{
3113        mminit_dprintk(MMINIT_TRACE, "zoneboundary",
3114                        "Entering push_node_boundaries(%u, %lu, %lu)\n",
3115                        nid, start_pfn, end_pfn);
3116
3117        /* Initialise the boundary for this node if necessary */
3118        if (node_boundary_end_pfn[nid] == 0)
3119                node_boundary_start_pfn[nid] = -1UL;
3120
3121        /* Update the boundaries */
3122        if (node_boundary_start_pfn[nid] > start_pfn)
3123                node_boundary_start_pfn[nid] = start_pfn;
3124        if (node_boundary_end_pfn[nid] < end_pfn)
3125                node_boundary_end_pfn[nid] = end_pfn;
3126}
3127
3128/* If necessary, push the node boundary out for reserve hotadd */
3129static void __meminit account_node_boundary(unsigned int nid,
3130                unsigned long *start_pfn, unsigned long *end_pfn)
3131{
3132        mminit_dprintk(MMINIT_TRACE, "zoneboundary",
3133                        "Entering account_node_boundary(%u, %lu, %lu)\n",
3134                        nid, *start_pfn, *end_pfn);
3135
3136        /* Return if boundary information has not been provided */
3137        if (node_boundary_end_pfn[nid] == 0)
3138                return;
3139
3140        /* Check the boundaries and update if necessary */
3141        if (node_boundary_start_pfn[nid] < *start_pfn)
3142                *start_pfn = node_boundary_start_pfn[nid];
3143        if (node_boundary_end_pfn[nid] > *end_pfn)
3144                *end_pfn = node_boundary_end_pfn[nid];
3145}
3146#else
3147void __init push_node_boundaries(unsigned int nid,
3148                unsigned long start_pfn, unsigned long end_pfn) {}
3149
3150static void __meminit account_node_boundary(unsigned int nid,
3151                unsigned long *start_pfn, unsigned long *end_pfn) {}
3152#endif
3153
3154
3155/**
3156 * get_pfn_range_for_nid - Return the start and end page frames for a node
3157 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3158 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3159 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3160 *
3161 * It returns the start and end page frame of a node based on information
3162 * provided by an arch calling add_active_range(). If called for a node
3163 * with no available memory, a warning is printed and the start and end
3164 * PFNs will be 0.
3165 */
3166void __meminit get_pfn_range_for_nid(unsigned int nid,
3167                        unsigned long *start_pfn, unsigned long *end_pfn)
3168{
3169        int i;
3170        *start_pfn = -1UL;
3171        *end_pfn = 0;
3172
3173        for_each_active_range_index_in_nid(i, nid) {
3174                *start_pfn = min(*start_pfn, early_node_map[i].start_pfn);
3175                *end_pfn = max(*end_pfn, early_node_map[i].end_pfn);
3176        }
3177
3178        if (*start_pfn == -1UL)
3179                *start_pfn = 0;
3180
3181        /* Push the node boundaries out if requested */
3182        account_node_boundary(nid, start_pfn, end_pfn);
3183}
3184
3185/*
3186 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3187 * assumption is made that zones within a node are ordered in monotonic
3188 * increasing memory addresses so that the "highest" populated zone is used
3189 */
3190static void __init find_usable_zone_for_movable(void)
3191{
3192        int zone_index;
3193        for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
3194                if (zone_index == ZONE_MOVABLE)
3195                        continue;
3196
3197                if (arch_zone_highest_possible_pfn[zone_index] >
3198                                arch_zone_lowest_possible_pfn[zone_index])
3199                        break;
3200        }
3201
3202        VM_BUG_ON(zone_index == -1);
3203        movable_zone = zone_index;
3204}
3205
3206/*
3207 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3208 * because it is sized independant of architecture. Unlike the other zones,
3209 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3210 * in each node depending on the size of each node and how evenly kernelcore
3211 * is distributed. This helper function adjusts the zone ranges
3212 * provided by the architecture for a given node by using the end of the
3213 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3214 * zones within a node are in order of monotonic increases memory addresses
3215 */
3216static void __meminit adjust_zone_range_for_zone_movable(int nid,
3217                                        unsigned long zone_type,
3218                                        unsigned long node_start_pfn,
3219                                        unsigned long node_end_pfn,
3220                                        unsigned long *zone_start_pfn,
3221                                        unsigned long *zone_end_pfn)
3222{
3223        /* Only adjust if ZONE_MOVABLE is on this node */
3224        if (zone_movable_pfn[nid]) {
3225                /* Size ZONE_MOVABLE */
3226                if (zone_type == ZONE_MOVABLE) {
3227                        *zone_start_pfn = zone_movable_pfn[nid];
3228                        *zone_end_pfn = min(node_end_pfn,
3229                                arch_zone_highest_possible_pfn[movable_zone]);
3230
3231                /* Adjust for ZONE_MOVABLE starting within this range */
3232                } else if (*zone_start_pfn < zone_movable_pfn[nid] &&
3233                                *zone_end_pfn > zone_movable_pfn[nid]) {
3234                        *zone_end_pfn = zone_movable_pfn[nid];
3235
3236                /* Check if this whole range is within ZONE_MOVABLE */
3237                } else if (*zone_start_pfn >= zone_movable_pfn[nid])
3238                        *zone_start_pfn = *zone_end_pfn;
3239        }
3240}
3241
3242/*
3243 * Return the number of pages a zone spans in a node, including holes
3244 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3245 */
3246static unsigned long __meminit zone_spanned_pages_in_node(int nid,
3247                                        unsigned long zone_type,
3248                                        unsigned long *ignored)
3249{
3250        unsigned long node_start_pfn, node_end_pfn;
3251        unsigned long zone_start_pfn, zone_end_pfn;
3252
3253        /* Get the start and end of the node and zone */
3254        get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3255        zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
3256        zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
3257        adjust_zone_range_for_zone_movable(nid, zone_type,
3258                                node_start_pfn, node_end_pfn,
3259                                &zone_start_pfn, &zone_end_pfn);
3260
3261        /* Check that this node has pages within the zone's required range */
3262        if (zone_end_pfn < node_start_pfn || zone_start_pfn > node_end_pfn)
3263                return 0;
3264
3265        /* Move the zone boundaries inside the node if necessary */
3266        zone_end_pfn = min(zone_end_pfn, node_end_pfn);
3267        zone_start_pfn = max(zone_start_pfn, node_start_pfn);
3268
3269        /* Return the spanned pages */
3270        return zone_end_pfn - zone_start_pfn;
3271}
3272
3273/*
3274 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3275 * then all holes in the requested range will be accounted for.
3276 */
3277static unsigned long __meminit __absent_pages_in_range(int nid,
3278                                unsigned long range_start_pfn,
3279                                unsigned long range_end_pfn)
3280{
3281        int i = 0;
3282        unsigned long prev_end_pfn = 0, hole_pages = 0;
3283        unsigned long start_pfn;
3284
3285        /* Find the end_pfn of the first active range of pfns in the node */
3286        i = first_active_region_index_in_nid(nid);
3287        if (i == -1)
3288                return 0;
3289
3290        prev_end_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3291
3292        /* Account for ranges before physical memory on this node */
3293        if (early_node_map[i].start_pfn > range_start_pfn)
3294                hole_pages = prev_end_pfn - range_start_pfn;
3295
3296        /* Find all holes for the zone within the node */
3297        for (; i != -1; i = next_active_region_index_in_nid(i, nid)) {
3298
3299                /* No need to continue if prev_end_pfn is outside the zone */
3300                if (prev_end_pfn >= range_end_pfn)
3301                        break;
3302
3303                /* Make sure the end of the zone is not within the hole */
3304                start_pfn = min(early_node_map[i].start_pfn, range_end_pfn);
3305                prev_end_pfn = max(prev_end_pfn, range_start_pfn);
3306
3307                /* Update the hole size cound and move on */
3308                if (start_pfn > range_start_pfn) {
3309                        BUG_ON(prev_end_pfn > start_pfn);
3310                        hole_pages += start_pfn - prev_end_pfn;
3311                }
3312                prev_end_pfn = early_node_map[i].end_pfn;
3313        }
3314
3315        /* Account for ranges past physical memory on this node */
3316        if (range_end_pfn > prev_end_pfn)
3317                hole_pages += range_end_pfn -
3318                                max(range_start_pfn, prev_end_pfn);
3319
3320        return hole_pages;
3321}
3322
3323/**
3324 * absent_pages_in_range - Return number of page frames in holes within a range
3325 * @start_pfn: The start PFN to start searching for holes
3326 * @end_pfn: The end PFN to stop searching for holes
3327 *
3328 * It returns the number of pages frames in memory holes within a range.
3329 */
3330unsigned long __init absent_pages_in_range(unsigned long start_pfn,
3331                                                        unsigned long end_pfn)
3332{
3333        return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
3334}
3335
3336/* Return the number of page frames in holes in a zone on a node */
3337static unsigned long __meminit zone_absent_pages_in_node(int nid,
3338                                        unsigned long zone_type,
3339                                        unsigned long *ignored)
3340{
3341        unsigned long node_start_pfn, node_end_pfn;
3342        unsigned long zone_start_pfn, zone_end_pfn;
3343
3344        get_pfn_range_for_nid(nid, &node_start_pfn, &node_end_pfn);
3345        zone_start_pfn = max(arch_zone_lowest_possible_pfn[zone_type],
3346                                                        node_start_pfn);
3347        zone_end_pfn = min(arch_zone_highest_possible_pfn[zone_type],
3348                                                        node_end_pfn);
3349
3350        adjust_zone_range_for_zone_movable(nid, zone_type,
3351                        node_start_pfn, node_end_pfn,
3352                        &zone_start_pfn, &zone_end_pfn);
3353        return __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
3354}
3355
3356#else
3357static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
3358                                        unsigned long zone_type,
3359                                        unsigned long *zones_size)
3360{
3361        return zones_size[zone_type];
3362}
3363
3364static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
3365                                                unsigned long zone_type,
3366                                                unsigned long *zholes_size)
3367{
3368        if (!zholes_size)
3369                return 0;
3370
3371        return zholes_size[zone_type];
3372}
3373
3374#endif
3375
3376static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
3377                unsigned long *zones_size, unsigned long *zholes_size)
3378{
3379        unsigned long realtotalpages, totalpages = 0;
3380        enum zone_type i;
3381
3382        for (i = 0; i < MAX_NR_ZONES; i++)
3383                totalpages += zone_spanned_pages_in_node(pgdat->node_id, i,
3384                                                                zones_size);
3385        pgdat->node_spanned_pages = totalpages;
3386
3387        realtotalpages = totalpages;
3388        for (i = 0; i < MAX_NR_ZONES; i++)
3389                realtotalpages -=
3390                        zone_absent_pages_in_node(pgdat->node_id, i,
3391                                                                zholes_size);
3392        pgdat->node_present_pages = realtotalpages;
3393        printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
3394                                                        realtotalpages);
3395}
3396
3397#ifndef CONFIG_SPARSEMEM
3398/*
3399 * Calculate the size of the zone->blockflags rounded to an unsigned long
3400 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3401 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3402 * round what is now in bits to nearest long in bits, then return it in
3403 * bytes.
3404 */
3405static unsigned long __init usemap_size(unsigned long zonesize)
3406{
3407        unsigned long usemapsize;
3408
3409        usemapsize = roundup(zonesize, pageblock_nr_pages);
3410        usemapsize = usemapsize >> pageblock_order;
3411        usemapsize *= NR_PAGEBLOCK_BITS;
3412        usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
3413
3414        return usemapsize / 8;
3415}
3416
3417static void __init setup_usemap(struct pglist_data *pgdat,
3418                                struct zone *zone, unsigned long zonesize)
3419{
3420        unsigned long usemapsize = usemap_size(zonesize);
3421        zone->pageblock_flags = NULL;
3422        if (usemapsize)
3423                zone->pageblock_flags = alloc_bootmem_node(pgdat, usemapsize);
3424}
3425#else
3426static void inline setup_usemap(struct pglist_data *pgdat,
3427                                struct zone *zone, unsigned long zonesize) {}
3428#endif /* CONFIG_SPARSEMEM */
3429
3430#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3431
3432/* Return a sensible default order for the pageblock size. */
3433static inline int pageblock_default_order(void)
3434{
3435        if (HPAGE_SHIFT > PAGE_SHIFT)
3436                return HUGETLB_PAGE_ORDER;
3437
3438        return MAX_ORDER-1;
3439}
3440
3441/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3442static inline void __init set_pageblock_order(unsigned int order)
3443{
3444        /* Check that pageblock_nr_pages has not already been setup */
3445        if (pageblock_order)
3446                return;
3447
3448        /*
3449         * Assume the largest contiguous order of interest is a huge page.
3450         * This value may be variable depending on boot parameters on IA64
3451         */
3452        pageblock_order = order;
3453}
3454#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3455
3456/*
3457 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3458 * and pageblock_default_order() are unused as pageblock_order is set
3459 * at compile-time. See include/linux/pageblock-flags.h for the values of
3460 * pageblock_order based on the kernel config
3461 */
3462static inline int pageblock_default_order(unsigned int order)
3463{
3464        return MAX_ORDER-1;
3465}
3466#define set_pageblock_order(x)  do {} while (0)
3467
3468#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3469
3470/*
3471 * Set up the zone data structures:
3472 *   - mark all pages reserved
3473 *   - mark all memory queues empty
3474 *   - clear the memory bitmaps
3475 */
3476static void __paginginit free_area_init_core(struct pglist_data *pgdat,
3477                unsigned long *zones_size, unsigned long *zholes_size)
3478{
3479        enum zone_type j;
3480        int nid = pgdat->node_id;
3481        unsigned long zone_start_pfn = pgdat->node_start_pfn;
3482        int ret;
3483
3484        pgdat_resize_init(pgdat);
3485        pgdat->nr_zones = 0;
3486        init_waitqueue_head(&pgdat->kswapd_wait);
3487        pgdat->kswapd_max_order = 0;
3488        pgdat_page_cgroup_init(pgdat);
3489        
3490        for (j = 0; j < MAX_NR_ZONES; j++) {
3491                struct zone *zone = pgdat->node_zones + j;
3492                unsigned long size, realsize, memmap_pages;
3493                enum lru_list l;
3494
3495                size = zone_spanned_pages_in_node(nid, j, zones_size);
3496                realsize = size - zone_absent_pages_in_node(nid, j,
3497                                                                zholes_size);
3498
3499                /*
3500                 * Adjust realsize so that it accounts for how much memory
3501                 * is used by this zone for memmap. This affects the watermark
3502                 * and per-cpu initialisations
3503                 */
3504                memmap_pages =
3505                        PAGE_ALIGN(size * sizeof(struct page)) >> PAGE_SHIFT;
3506                if (realsize >= memmap_pages) {
3507                        realsize -= memmap_pages;
3508                        if (memmap_pages)
3509                                printk(KERN_DEBUG
3510                                       "  %s zone: %lu pages used for memmap\n",
3511                                       zone_names[j], memmap_pages);
3512                } else
3513                        printk(KERN_WARNING
3514                                "  %s zone: %lu pages exceeds realsize %lu\n",
3515                                zone_names[j], memmap_pages, realsize);
3516
3517                /* Account for reserved pages */
3518                if (j == 0 && realsize > dma_reserve) {
3519                        realsize -= dma_reserve;
3520                        printk(KERN_DEBUG "  %s zone: %lu pages reserved\n",
3521                                        zone_names[0], dma_reserve);
3522                }
3523
3524                if (!is_highmem_idx(j))
3525                        nr_kernel_pages += realsize;
3526                nr_all_pages += realsize;
3527
3528                zone->spanned_pages = size;
3529                zone->present_pages = realsize;
3530#ifdef CONFIG_NUMA
3531                zone->node = nid;
3532                zone->min_unmapped_pages = (realsize*sysctl_min_unmapped_ratio)
3533                                                / 100;
3534                zone->min_slab_pages = (realsize * sysctl_min_slab_ratio) / 100;
3535#endif
3536                zone->name = zone_names[j];
3537                spin_lock_init(&zone->lock);
3538                spin_lock_init(&zone->lru_lock);
3539                zone_seqlock_init(zone);
3540                zone->zone_pgdat = pgdat;
3541
3542                zone->prev_priority = DEF_PRIORITY;
3543
3544                zone_pcp_init(zone);
3545                for_each_lru(l) {
3546                        INIT_LIST_HEAD(&zone->lru[l].list);
3547                        zone->lru[l].nr_scan = 0;
3548                }
3549                zone->reclaim_stat.recent_rotated[0] = 0;
3550                zone->reclaim_stat.recent_rotated[1] = 0;
3551                zone->reclaim_stat.recent_scanned[0] = 0;
3552                zone->reclaim_stat.recent_scanned[1] = 0;
3553                zap_zone_vm_stats(zone);
3554                zone->flags = 0;
3555                if (!size)
3556                        continue;
3557
3558                set_pageblock_order(pageblock_default_order());
3559                setup_usemap(pgdat, zone, size);
3560                ret = init_currently_empty_zone(zone, zone_start_pfn,
3561                                                size, MEMMAP_EARLY);
3562                BUG_ON(ret);
3563                memmap_init(size, nid, j, zone_start_pfn);
3564                zone_start_pfn += size;
3565        }
3566}
3567
3568static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
3569{
3570        /* Skip empty nodes */
3571        if (!pgdat->node_spanned_pages)
3572                return;
3573
3574#ifdef CONFIG_FLAT_NODE_MEM_MAP
3575        /* ia64 gets its own node_mem_map, before this, without bootmem */
3576        if (!pgdat->node_mem_map) {
3577                unsigned long size, start, end;
3578                struct page *map;
3579
3580                /*
3581                 * The zone's endpoints aren't required to be MAX_ORDER
3582                 * aligned but the node_mem_map endpoints must be in order
3583                 * for the buddy allocator to function correctly.
3584                 */
3585                start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
3586                end = pgdat->node_start_pfn + pgdat->node_spanned_pages;
3587                end = ALIGN(end, MAX_ORDER_NR_PAGES);
3588                size =  (end - start) * sizeof(struct page);
3589                map = alloc_remap(pgdat->node_id, size);
3590                if (!map)
3591                        map = alloc_bootmem_node(pgdat, size);
3592                pgdat->node_mem_map = map + (pgdat->node_start_pfn - start);
3593        }
3594#ifndef CONFIG_NEED_MULTIPLE_NODES
3595        /*
3596         * With no DISCONTIG, the global mem_map is just set as node 0's
3597         */
3598        if (pgdat == NODE_DATA(0)) {
3599                mem_map = NODE_DATA(0)->node_mem_map;
3600#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3601                if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
3602                        mem_map -= (pgdat->node_start_pfn - ARCH_PFN_OFFSET);
3603#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3604        }
3605#endif
3606#endif /* CONFIG_FLAT_NODE_MEM_MAP */
3607}
3608
3609void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
3610                unsigned long node_start_pfn, unsigned long *zholes_size)
3611{
3612        pg_data_t *pgdat = NODE_DATA(nid);
3613
3614        pgdat->node_id = nid;
3615        pgdat->node_start_pfn = node_start_pfn;
3616        calculate_node_totalpages(pgdat, zones_size, zholes_size);
3617
3618        alloc_node_mem_map(pgdat);
3619#ifdef CONFIG_FLAT_NODE_MEM_MAP
3620        printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3621                nid, (unsigned long)pgdat,
3622                (unsigned long)pgdat->node_mem_map);
3623#endif
3624
3625        free_area_init_core(pgdat, zones_size, zholes_size);
3626}
3627
3628#ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3629
3630#if MAX_NUMNODES > 1
3631/*
3632 * Figure out the number of possible node ids.
3633 */
3634static void __init setup_nr_node_ids(void)
3635{
3636        unsigned int node;
3637        unsigned int highest = 0;
3638
3639        for_each_node_mask(node, node_possible_map)
3640                highest = node;
3641        nr_node_ids = highest + 1;
3642}
3643#else
3644static inline void setup_nr_node_ids(void)
3645{
3646}
3647#endif
3648
3649/**
3650 * add_active_range - Register a range of PFNs backed by physical memory
3651 * @nid: The node ID the range resides on
3652 * @start_pfn: The start PFN of the available physical memory
3653 * @end_pfn: The end PFN of the available physical memory
3654 *
3655 * These ranges are stored in an early_node_map[] and later used by
3656 * free_area_init_nodes() to calculate zone sizes and holes. If the
3657 * range spans a memory hole, it is up to the architecture to ensure
3658 * the memory is not freed by the bootmem allocator. If possible
3659 * the range being registered will be merged with existing ranges.
3660 */
3661void __init add_active_range(unsigned int nid, unsigned long start_pfn,
3662                                                unsigned long end_pfn)
3663{
3664        int i;
3665
3666        mminit_dprintk(MMINIT_TRACE, "memory_register",
3667                        "Entering add_active_range(%d, %#lx, %#lx) "
3668                        "%d entries of %d used\n",
3669                        nid, start_pfn, end_pfn,
3670                        nr_nodemap_entries, MAX_ACTIVE_REGIONS);
3671
3672        mminit_validate_memmodel_limits(&start_pfn, &end_pfn);
3673
3674        /* Merge with existing active regions if possible */
3675        for (i = 0; i < nr_nodemap_entries; i++) {
3676                if (early_node_map[i].nid != nid)
3677                        continue;
3678
3679                /* Skip if an existing region covers this new one */
3680                if (start_pfn >= early_node_map[i].start_pfn &&
3681                                end_pfn <= early_node_map[i].end_pfn)
3682                        return;
3683
3684                /* Merge forward if suitable */
3685                if (start_pfn <= early_node_map[i].end_pfn &&
3686                                end_pfn > early_node_map[i].end_pfn) {
3687                        early_node_map[i].end_pfn = end_pfn;
3688                        return;
3689                }
3690
3691                /* Merge backward if suitable */
3692                if (start_pfn < early_node_map[i].end_pfn &&
3693                                end_pfn >= early_node_map[i].start_pfn) {
3694                        early_node_map[i].start_pfn = start_pfn;
3695                        return;
3696                }
3697        }
3698
3699        /* Check that early_node_map is large enough */
3700        if (i >= MAX_ACTIVE_REGIONS) {
3701                printk(KERN_CRIT "More than %d memory regions, truncating\n",
3702                                                        MAX_ACTIVE_REGIONS);
3703                return;
3704        }
3705
3706        early_node_map[i].nid = nid;
3707        early_node_map[i].start_pfn = start_pfn;
3708        early_node_map[i].end_pfn = end_pfn;
3709        nr_nodemap_entries = i + 1;
3710}
3711
3712/**
3713 * remove_active_range - Shrink an existing registered range of PFNs
3714 * @nid: The node id the range is on that should be shrunk
3715 * @start_pfn: The new PFN of the range
3716 * @end_pfn: The new PFN of the range
3717 *
3718 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3719 * The map is kept near the end physical page range that has already been
3720 * registered. This function allows an arch to shrink an existing registered
3721 * range.
3722 */
3723void __init remove_active_range(unsigned int nid, unsigned long start_pfn,
3724                                unsigned long end_pfn)
3725{
3726        int i, j;
3727        int removed = 0;
3728
3729        printk(KERN_DEBUG "remove_active_range (%d, %lu, %lu)\n",
3730                          nid, start_pfn, end_pfn);
3731
3732        /* Find the old active region end and shrink */
3733        for_each_active_range_index_in_nid(i, nid) {
3734                if (early_node_map[i].start_pfn >= start_pfn &&
3735                    early_node_map[i].end_pfn <= end_pfn) {
3736                        /* clear it */
3737                        early_node_map[i].start_pfn = 0;
3738                        early_node_map[i].end_pfn = 0;
3739                        removed = 1;
3740                        continue;
3741                }
3742                if (early_node_map[i].start_pfn < start_pfn &&
3743                    early_node_map[i].end_pfn > start_pfn) {
3744                        unsigned long temp_end_pfn = early_node_map[i].end_pfn;
3745                        early_node_map[i].end_pfn = start_pfn;
3746                        if (temp_end_pfn > end_pfn)
3747                                add_active_range(nid, end_pfn, temp_end_pfn);
3748                        continue;
3749                }
3750                if (early_node_map[i].start_pfn >= start_pfn &&
3751                    early_node_map[i].end_pfn > end_pfn &&
3752                    early_node_map[i].start_pfn < end_pfn) {
3753                        early_node_map[i].start_pfn = end_pfn;
3754                        continue;
3755                }
3756        }
3757
3758        if (!removed)
3759                return;
3760
3761        /* remove the blank ones */
3762        for (i = nr_nodemap_entries - 1; i > 0; i--) {
3763                if (early_node_map[i].nid != nid)
3764                        continue;
3765                if (early_node_map[i].end_pfn)
3766                        continue;
3767                /* we found it, get rid of it */
3768                for (j = i; j < nr_nodemap_entries - 1; j++)
3769                        memcpy(&early_node_map[j], &early_node_map[j+1],
3770                                sizeof(early_node_map[j]));
3771                j = nr_nodemap_entries - 1;
3772                memset(&early_node_map[j], 0, sizeof(early_node_map[j]));
3773                nr_nodemap_entries--;
3774        }
3775}
3776
3777/**
3778 * remove_all_active_ranges - Remove all currently registered regions
3779 *
3780 * During discovery, it may be found that a table like SRAT is invalid
3781 * and an alternative discovery method must be used. This function removes
3782 * all currently registered regions.
3783 */
3784void __init remove_all_active_ranges(void)
3785{
3786        memset(early_node_map, 0, sizeof(early_node_map));
3787        nr_nodemap_entries = 0;
3788#ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3789        memset(node_boundary_start_pfn, 0, sizeof(node_boundary_start_pfn));
3790        memset(node_boundary_end_pfn, 0, sizeof(node_boundary_end_pfn));
3791#endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3792}
3793
3794/* Compare two active node_active_regions */
3795static int __init cmp_node_active_region(const void *a, const void *b)
3796{
3797        struct node_active_region *arange = (struct node_active_region *)a;
3798        struct node_active_region *brange = (struct node_active_region *)b;
3799
3800        /* Done this way to avoid overflows */
3801        if (arange->start_pfn > brange->start_pfn)
3802                return 1;
3803        if (arange->start_pfn < brange->start_pfn)
3804                return -1;
3805
3806        return 0;
3807}
3808
3809/* sort the node_map by start_pfn */
3810static void __init sort_node_map(void)
3811{
3812        sort(early_node_map, (size_t)nr_nodemap_entries,
3813                        sizeof(struct node_active_region),
3814                        cmp_node_active_region, NULL);
3815}
3816
3817/* Find the lowest pfn for a node */
3818static unsigned long __init find_min_pfn_for_node(int nid)
3819{
3820        int i;
3821        unsigned long min_pfn = ULONG_MAX;
3822
3823        /* Assuming a sorted map, the first range found has the starting pfn */
3824        for_each_active_range_index_in_nid(i, nid)
3825                min_pfn = min(min_pfn, early_node_map[i].start_pfn);
3826
3827        if (min_pfn == ULONG_MAX) {
3828                printk(KERN_WARNING
3829                        "Could not find start_pfn for node %d\n", nid);
3830                return 0;
3831        }
3832
3833        return min_pfn;
3834}
3835
3836/**
3837 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3838 *
3839 * It returns the minimum PFN based on information provided via
3840 * add_active_range().
3841 */
3842unsigned long __init find_min_pfn_with_active_regions(void)
3843{
3844        return find_min_pfn_for_node(MAX_NUMNODES);
3845}
3846
3847/*
3848 * early_calculate_totalpages()
3849 * Sum pages in active regions for movable zone.
3850 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3851 */
3852static unsigned long __init early_calculate_totalpages(void)
3853{
3854        int i;
3855        unsigned long totalpages = 0;
3856
3857        for (i = 0; i < nr_nodemap_entries; i++) {
3858                unsigned long pages = early_node_map[i].end_pfn -
3859                                                early_node_map[i].start_pfn;
3860                totalpages += pages;
3861                if (pages)
3862                        node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
3863        }
3864        return totalpages;
3865}
3866
3867/*
3868 * Find the PFN the Movable zone begins in each node. Kernel memory
3869 * is spread evenly between nodes as long as the nodes have enough
3870 * memory. When they don't, some nodes will have more kernelcore than
3871 * others
3872 */
3873static void __init find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn)
3874{
3875        int i, nid;
3876        unsigned long usable_startpfn;
3877        unsigned long kernelcore_node, kernelcore_remaining;
3878        unsigned long totalpages = early_calculate_totalpages();
3879        int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
3880
3881        /*
3882         * If movablecore was specified, calculate what size of
3883         * kernelcore that corresponds so that memory usable for
3884         * any allocation type is evenly spread. If both kernelcore
3885         * and movablecore are specified, then the value of kernelcore
3886         * will be used for required_kernelcore if it's greater than
3887         * what movablecore would have allowed.
3888         */
3889        if (required_movablecore) {
3890                unsigned long corepages;
3891
3892                /*
3893                 * Round-up so that ZONE_MOVABLE is at least as large as what
3894                 * was requested by the user
3895                 */
3896                required_movablecore =
3897                        roundup(required_movablecore, MAX_ORDER_NR_PAGES);
3898                corepages = totalpages - required_movablecore;
3899
3900                required_kernelcore = max(required_kernelcore, corepages);
3901        }
3902
3903        /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3904        if (!required_kernelcore)
3905                return;
3906
3907        /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3908        find_usable_zone_for_movable();
3909        usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
3910
3911restart:
3912        /* Spread kernelcore memory as evenly as possible throughout nodes */
3913        kernelcore_node = required_kernelcore / usable_nodes;
3914        for_each_node_state(nid, N_HIGH_MEMORY) {
3915                /*
3916                 * Recalculate kernelcore_node if the division per node
3917                 * now exceeds what is necessary to satisfy the requested
3918                 * amount of memory for the kernel
3919                 */
3920                if (required_kernelcore < kernelcore_node)
3921                        kernelcore_node = required_kernelcore / usable_nodes;
3922
3923                /*
3924                 * As the map is walked, we track how much memory is usable
3925                 * by the kernel using kernelcore_remaining. When it is
3926                 * 0, the rest of the node is usable by ZONE_MOVABLE
3927                 */
3928                kernelcore_remaining = kernelcore_node;
3929
3930                /* Go through each range of PFNs within this node */
3931                for_each_active_range_index_in_nid(i, nid) {
3932                        unsigned long start_pfn, end_pfn;
3933                        unsigned long size_pages;
3934
3935                        start_pfn = max(early_node_map[i].start_pfn,
3936                                                zone_movable_pfn[nid]);
3937                        end_pfn = early_node_map[i].end_pfn;
3938                        if (start_pfn >= end_pfn)
3939                                continue;
3940
3941                        /* Account for what is only usable for kernelcore */
3942                        if (start_pfn < usable_startpfn) {
3943                                unsigned long kernel_pages;
3944                                kernel_pages = min(end_pfn, usable_startpfn)
3945                                                                - start_pfn;
3946
3947                                kernelcore_remaining -= min(kernel_pages,
3948                                                        kernelcore_remaining);
3949                                required_kernelcore -= min(kernel_pages,
3950                                                        required_kernelcore);
3951
3952                                /* Continue if range is now fully accounted */
3953                                if (end_pfn <= usable_startpfn) {
3954
3955                                        /*
3956                                         * Push zone_movable_pfn to the end so
3957                                         * that if we have to rebalance
3958                                         * kernelcore across nodes, we will
3959                                         * not double account here
3960                                         */
3961                                        zone_movable_pfn[nid] = end_pfn;
3962                                        continue;
3963                                }
3964                                start_pfn = usable_startpfn;
3965                        }
3966
3967                        /*
3968                         * The usable PFN range for ZONE_MOVABLE is from
3969                         * start_pfn->end_pfn. Calculate size_pages as the
3970                         * number of pages used as kernelcore
3971                         */
3972                        size_pages = end_pfn - start_pfn;
3973                        if (size_pages > kernelcore_remaining)
3974                                size_pages = kernelcore_remaining;
3975                        zone_movable_pfn[nid] = start_pfn + size_pages;
3976
3977                        /*
3978                         * Some kernelcore has been met, update counts and
3979                         * break if the kernelcore for this node has been
3980                         * satisified
3981                         */
3982                        required_kernelcore -= min(required_kernelcore,
3983                                                                size_pages);
3984                        kernelcore_remaining -= size_pages;
3985                        if (!kernelcore_remaining)
3986                                break;
3987                }
3988        }
3989
3990        /*
3991         * If there is still required_kernelcore, we do another pass with one
3992         * less node in the count. This will push zone_movable_pfn[nid] further
3993         * along on the nodes that still have memory until kernelcore is
3994         * satisified
3995         */
3996        usable_nodes--;
3997        if (usable_nodes && required_kernelcore > usable_nodes)
3998                goto restart;
3999
4000        /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4001        for (nid = 0; nid < MAX_NUMNODES; nid++)
4002                zone_movable_pfn[nid] =
4003                        roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
4004}
4005
4006/* Any regular memory on that node ? */
4007static void check_for_regular_memory(pg_data_t *pgdat)
4008{
4009#ifdef CONFIG_HIGHMEM
4010        enum zone_type zone_type;
4011
4012        for (zone_type = 0; zone_type <= ZONE_NORMAL; zone_type++) {
4013                struct zone *zone = &pgdat->node_zones[zone_type];
4014                if (zone->present_pages)
4015                        node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
4016        }
4017#endif
4018}
4019
4020/**
4021 * free_area_init_nodes - Initialise all pg_data_t and zone data
4022 * @max_zone_pfn: an array of max PFNs for each zone
4023 *
4024 * This will call free_area_init_node() for each active node in the system.
4025 * Using the page ranges provided by add_active_range(), the size of each
4026 * zone in each node and their holes is calculated. If the maximum PFN
4027 * between two adjacent zones match, it is assumed that the zone is empty.
4028 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4029 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4030 * starts where the previous one ended. For example, ZONE_DMA32 starts
4031 * at arch_max_dma_pfn.
4032 */
4033void __init free_area_init_nodes(unsigned long *max_zone_pfn)
4034{
4035        unsigned long nid;
4036        int i;
4037
4038        /* Sort early_node_map as initialisation assumes it is sorted */
4039        sort_node_map();
4040
4041        /* Record where the zone boundaries are */
4042        memset(arch_zone_lowest_possible_pfn, 0,
4043                                sizeof(arch_zone_lowest_possible_pfn));
4044        memset(arch_zone_highest_possible_pfn, 0,
4045                                sizeof(arch_zone_highest_possible_pfn));
4046        arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
4047        arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
4048        for (i = 1; i < MAX_NR_ZONES; i++) {
4049                if (i == ZONE_MOVABLE)
4050                        continue;
4051                arch_zone_lowest_possible_pfn[i] =
4052                        arch_zone_highest_possible_pfn[i-1];
4053                arch_zone_highest_possible_pfn[i] =
4054                        max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
4055        }
4056        arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
4057        arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
4058
4059        /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4060        memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
4061        find_zone_movable_pfns_for_nodes(zone_movable_pfn);
4062
4063        /* Print out the zone ranges */
4064        printk("Zone PFN ranges:\n");
4065        for (i = 0; i < MAX_NR_ZONES; i++) {
4066                if (i == ZONE_MOVABLE)
4067                        continue;
4068                printk("  %-8s %0#10lx -> %0#10lx\n",
4069                                zone_names[i],
4070                                arch_zone_lowest_possible_pfn[i],
4071                                arch_zone_highest_possible_pfn[i]);
4072        }
4073
4074        /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4075        printk("Movable zone start PFN for each node\n");
4076        for (i = 0; i < MAX_NUMNODES; i++) {
4077                if (zone_movable_pfn[i])
4078                        printk("  Node %d: %lu\n", i, zone_movable_pfn[i]);
4079        }
4080
4081        /* Print out the early_node_map[] */
4082        printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries);
4083        for (i = 0; i < nr_nodemap_entries; i++)
4084                printk("  %3d: %0#10lx -> %0#10lx\n", early_node_map[i].nid,
4085                                                early_node_map[i].start_pfn,
4086                                                early_node_map[i].end_pfn);
4087
4088        /* Initialise every node */
4089        mminit_verify_pageflags_layout();
4090        setup_nr_node_ids();
4091        for_each_online_node(nid) {
4092                pg_data_t *pgdat = NODE_DATA(nid);
4093                free_area_init_node(nid, NULL,
4094                                find_min_pfn_for_node(nid), NULL);
4095
4096                /* Any memory on that node */
4097                if (pgdat->node_present_pages)
4098                        node_set_state(nid, N_HIGH_MEMORY);
4099                check_for_regular_memory(pgdat);
4100        }
4101}
4102
4103static int __init cmdline_parse_core(char *p, unsigned long *core)
4104{
4105        unsigned long long coremem;
4106        if (!p)
4107                return -EINVAL;
4108
4109        coremem = memparse(p, &p);
4110        *core = coremem >> PAGE_SHIFT;
4111
4112        /* Paranoid check that UL is enough for the coremem value */
4113        WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
4114
4115        return 0;
4116}
4117
4118/*
4119 * kernelcore=size sets the amount of memory for use for allocations that
4120 * cannot be reclaimed or migrated.
4121 */
4122static int __init cmdline_parse_kernelcore(char *p)
4123{
4124        return cmdline_parse_core(p, &required_kernelcore);
4125}
4126
4127/*
4128 * movablecore=size sets the amount of memory for use for allocations that
4129 * can be reclaimed or migrated.
4130 */
4131static int __init cmdline_parse_movablecore(char *p)
4132{
4133        return cmdline_parse_core(p, &required_movablecore);
4134}
4135
4136early_param("kernelcore", cmdline_parse_kernelcore);
4137early_param("movablecore", cmdline_parse_movablecore);
4138
4139#endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4140
4141/**
4142 * set_dma_reserve - set the specified number of pages reserved in the first zone
4143 * @new_dma_reserve: The number of pages to mark reserved
4144 *
4145 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4146 * In the DMA zone, a significant percentage may be consumed by kernel image
4147 * and other unfreeable allocations which can skew the watermarks badly. This
4148 * function may optionally be used to account for unfreeable pages in the
4149 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4150 * smaller per-cpu batchsize.
4151 */
4152void __init set_dma_reserve(unsigned long new_dma_reserve)
4153{
4154        dma_reserve = new_dma_reserve;
4155}
4156
4157#ifndef CONFIG_NEED_MULTIPLE_NODES
4158struct pglist_data __refdata contig_page_data = { .bdata = &bootmem_node_data[0] };
4159EXPORT_SYMBOL(contig_page_data);
4160#endif
4161
4162void __init free_area_init(unsigned long *zones_size)
4163{
4164        free_area_init_node(0, zones_size,
4165                        __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
4166}
4167
4168static int page_alloc_cpu_notify(struct notifier_block *self,
4169                                 unsigned long action, void *hcpu)
4170{
4171        int cpu = (unsigned long)hcpu;
4172
4173        if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
4174                drain_pages(cpu);
4175
4176                /*
4177                 * Spill the event counters of the dead processor
4178                 * into the current processors event counters.
4179                 * This artificially elevates the count of the current
4180                 * processor.
4181                 */
4182                vm_events_fold_cpu(cpu);
4183
4184                /*
4185                 * Zero the differential counters of the dead processor
4186                 * so that the vm statistics are consistent.
4187                 *
4188                 * This is only okay since the processor is dead and cannot
4189                 * race with what we are doing.
4190                 */
4191                refresh_cpu_vm_stats(cpu);
4192        }
4193        return NOTIFY_OK;
4194}
4195
4196void __init page_alloc_init(void)
4197{
4198        hotcpu_notifier(page_alloc_cpu_notify, 0);
4199}
4200
4201/*
4202 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4203 *      or min_free_kbytes changes.
4204 */
4205static void calculate_totalreserve_pages(void)
4206{
4207        struct pglist_data *pgdat;
4208        unsigned long reserve_pages = 0;
4209        enum zone_type i, j;
4210
4211        for_each_online_pgdat(pgdat) {
4212                for (i = 0; i < MAX_NR_ZONES; i++) {
4213                        struct zone *zone = pgdat->node_zones + i;
4214                        unsigned long max = 0;
4215
4216                        /* Find valid and maximum lowmem_reserve in the zone */
4217                        for (j = i; j < MAX_NR_ZONES; j++) {
4218                                if (zone->lowmem_reserve[j] > max)
4219                                        max = zone->lowmem_reserve[j];
4220                        }
4221
4222                        /* we treat pages_high as reserved pages. */
4223                        max += zone->pages_high;
4224
4225                        if (max > zone->present_pages)
4226                                max = zone->present_pages;
4227                        reserve_pages += max;
4228                }
4229        }
4230        totalreserve_pages = reserve_pages;
4231}
4232
4233/*
4234 * setup_per_zone_lowmem_reserve - called whenever
4235 *      sysctl_lower_zone_reserve_ratio changes.  Ensures that each zone
4236 *      has a correct pages reserved value, so an adequate number of
4237 *      pages are left in the zone after a successful __alloc_pages().
4238 */
4239static void setup_per_zone_lowmem_reserve(void)
4240{
4241        struct pglist_data *pgdat;
4242        enum zone_type j, idx;
4243
4244        for_each_online_pgdat(pgdat) {
4245                for (j = 0; j < MAX_NR_ZONES; j++) {
4246                        struct zone *zone = pgdat->node_zones + j;
4247                        unsigned long present_pages = zone->present_pages;
4248
4249                        zone->lowmem_reserve[j] = 0;
4250
4251                        idx = j;
4252                        while (idx) {
4253                                struct zone *lower_zone;
4254
4255                                idx--;
4256
4257                                if (sysctl_lowmem_reserve_ratio[idx] < 1)
4258                                        sysctl_lowmem_reserve_ratio[idx] = 1;
4259
4260                                lower_zone = pgdat->node_zones + idx;
4261                                lower_zone->lowmem_reserve[j] = present_pages /
4262                                        sysctl_lowmem_reserve_ratio[idx];
4263                                present_pages += lower_zone->present_pages;
4264                        }
4265                }
4266        }
4267
4268        /* update totalreserve_pages */
4269        calculate_totalreserve_pages();
4270}
4271
4272/**
4273 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4274 *
4275 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4276 * with respect to min_free_kbytes.
4277 */
4278void setup_per_zone_pages_min(void)
4279{
4280        unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
4281        unsigned long lowmem_pages = 0;
4282        struct zone *zone;
4283        unsigned long flags;
4284
4285        /* Calculate total number of !ZONE_HIGHMEM pages */
4286        for_each_zone(zone) {
4287                if (!is_highmem(zone))
4288                        lowmem_pages += zone->present_pages;
4289        }
4290
4291        for_each_zone(zone) {
4292                u64 tmp;
4293
4294                spin_lock_irqsave(&zone->lock, flags);
4295                tmp = (u64)pages_min * zone->present_pages;
4296                do_div(tmp, lowmem_pages);
4297                if (is_highmem(zone)) {
4298                        /*
4299                         * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4300                         * need highmem pages, so cap pages_min to a small
4301                         * value here.
4302                         *
4303                         * The (pages_high-pages_low) and (pages_low-pages_min)
4304                         * deltas controls asynch page reclaim, and so should
4305                         * not be capped for highmem.
4306                         */
4307                        int min_pages;
4308
4309                        min_pages = zone->present_pages / 1024;
4310                        if (min_pages < SWAP_CLUSTER_MAX)
4311                                min_pages = SWAP_CLUSTER_MAX;
4312                        if (min_pages > 128)
4313                                min_pages = 128;
4314                        zone->pages_min = min_pages;
4315                } else {
4316                        /*
4317                         * If it's a lowmem zone, reserve a number of pages
4318                         * proportionate to the zone's size.
4319                         */
4320                        zone->pages_min = tmp;
4321                }
4322
4323                zone->pages_low   = zone->pages_min + (tmp >> 2);
4324                zone->pages_high  = zone->pages_min + (tmp >> 1);
4325                setup_zone_migrate_reserve(zone);
4326                spin_unlock_irqrestore(&zone->lock, flags);
4327        }
4328
4329        /* update totalreserve_pages */
4330        calculate_totalreserve_pages();
4331}
4332
4333/**
4334 * setup_per_zone_inactive_ratio - called when min_free_kbytes changes.
4335 *
4336 * The inactive anon list should be small enough that the VM never has to
4337 * do too much work, but large enough that each inactive page has a chance
4338 * to be referenced again before it is swapped out.
4339 *
4340 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4341 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4342 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4343 * the anonymous pages are kept on the inactive list.
4344 *
4345 * total     target    max
4346 * memory    ratio     inactive anon
4347 * -------------------------------------
4348 *   10MB       1         5MB
4349 *  100MB       1        50MB
4350 *    1GB       3       250MB
4351 *   10GB      10       0.9GB
4352 *  100GB      31         3GB
4353 *    1TB     101        10GB
4354 *   10TB     320        32GB
4355 */
4356static void setup_per_zone_inactive_ratio(void)
4357{
4358        struct zone *zone;
4359
4360        for_each_zone(zone) {
4361                unsigned int gb, ratio;
4362
4363                /* Zone size in gigabytes */
4364                gb = zone->present_pages >> (30 - PAGE_SHIFT);
4365                ratio = int_sqrt(10 * gb);
4366                if (!ratio)
4367                        ratio = 1;
4368
4369                zone->inactive_ratio = ratio;
4370        }
4371}
4372
4373/*
4374 * Initialise min_free_kbytes.
4375 *
4376 * For small machines we want it small (128k min).  For large machines
4377 * we want it large (64MB max).  But it is not linear, because network
4378 * bandwidth does not increase linearly with machine size.  We use
4379 *
4380 *      min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4381 *      min_free_kbytes = sqrt(lowmem_kbytes * 16)
4382 *
4383 * which yields
4384 *
4385 * 16MB:        512k
4386 * 32MB:        724k
4387 * 64MB:        1024k
4388 * 128MB:       1448k
4389 * 256MB:       2048k
4390 * 512MB:       2896k
4391 * 1024MB:      4096k
4392 * 2048MB:      5792k
4393 * 4096MB:      8192k
4394 * 8192MB:      11584k
4395 * 16384MB:     16384k
4396 */
4397static int __init init_per_zone_pages_min(void)
4398{
4399        unsigned long lowmem_kbytes;
4400
4401        lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
4402
4403        min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
4404        if (min_free_kbytes < 128)
4405                min_free_kbytes = 128;
4406        if (min_free_kbytes > 65536)
4407                min_free_kbytes = 65536;
4408        setup_per_zone_pages_min();
4409        setup_per_zone_lowmem_reserve();
4410        setup_per_zone_inactive_ratio();
4411        return 0;
4412}
4413module_init(init_per_zone_pages_min)
4414
4415/*
4416 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so 
4417 *      that we can call two helper functions whenever min_free_kbytes
4418 *      changes.
4419 */
4420int min_free_kbytes_sysctl_handler(ctl_table *table, int write, 
4421        struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4422{
4423        proc_dointvec(table, write, file, buffer, length, ppos);
4424        if (write)
4425                setup_per_zone_pages_min();
4426        return 0;
4427}
4428
4429#ifdef CONFIG_NUMA
4430int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table *table, int write,
4431        struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4432{
4433        struct zone *zone;
4434        int rc;
4435
4436        rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4437        if (rc)
4438                return rc;
4439
4440        for_each_zone(zone)
4441                zone->min_unmapped_pages = (zone->present_pages *
4442                                sysctl_min_unmapped_ratio) / 100;
4443        return 0;
4444}
4445
4446int sysctl_min_slab_ratio_sysctl_handler(ctl_table *table, int write,
4447        struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4448{
4449        struct zone *zone;
4450        int rc;
4451
4452        rc = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4453        if (rc)
4454                return rc;
4455
4456        for_each_zone(zone)
4457                zone->min_slab_pages = (zone->present_pages *
4458                                sysctl_min_slab_ratio) / 100;
4459        return 0;
4460}
4461#endif
4462
4463/*
4464 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4465 *      proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4466 *      whenever sysctl_lowmem_reserve_ratio changes.
4467 *
4468 * The reserve ratio obviously has absolutely no relation with the
4469 * pages_min watermarks. The lowmem reserve ratio can only make sense
4470 * if in function of the boot time zone sizes.
4471 */
4472int lowmem_reserve_ratio_sysctl_handler(ctl_table *table, int write,
4473        struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4474{
4475        proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4476        setup_per_zone_lowmem_reserve();
4477        return 0;
4478}
4479
4480/*
4481 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4482 * cpu.  It is the fraction of total pages in each zone that a hot per cpu pagelist
4483 * can have before it gets flushed back to buddy allocator.
4484 */
4485
4486int percpu_pagelist_fraction_sysctl_handler(ctl_table *table, int write,
4487        struct file *file, void __user *buffer, size_t *length, loff_t *ppos)
4488{
4489        struct zone *zone;
4490        unsigned int cpu;
4491        int ret;
4492
4493        ret = proc_dointvec_minmax(table, write, file, buffer, length, ppos);
4494        if (!write || (ret == -EINVAL))
4495                return ret;
4496        for_each_zone(zone) {
4497                for_each_online_cpu(cpu) {
4498                        unsigned long  high;
4499                        high = zone->present_pages / percpu_pagelist_fraction;
4500                        setup_pagelist_highmark(zone_pcp(zone, cpu), high);
4501                }
4502        }
4503        return 0;
4504}
4505
4506int hashdist = HASHDIST_DEFAULT;
4507
4508#ifdef CONFIG_NUMA
4509static int __init set_hashdist(char *str)
4510{
4511        if (!str)
4512                return 0;
4513        hashdist = simple_strtoul(str, &str, 0);
4514        return 1;
4515}
4516__setup("hashdist=", set_hashdist);
4517#endif
4518
4519/*
4520 * allocate a large system hash table from bootmem
4521 * - it is assumed that the hash table must contain an exact power-of-2
4522 *   quantity of entries
4523 * - limit is the number of hash buckets, not the total allocation size
4524 */
4525void *__init alloc_large_system_hash(const char *tablename,
4526                                     unsigned long bucketsize,
4527                                     unsigned long numentries,
4528                                     int scale,
4529                                     int flags,
4530                                     unsigned int *_hash_shift,
4531                                     unsigned int *_hash_mask,
4532                                     unsigned long limit)
4533{
4534        unsigned long long max = limit;
4535        unsigned long log2qty, size;
4536        void *table = NULL;
4537
4538        /* allow the kernel cmdline to have a say */
4539        if (!numentries) {
4540                /* round applicable memory size up to nearest megabyte */
4541                numentries = nr_kernel_pages;
4542                numentries += (1UL << (20 - PAGE_SHIFT)) - 1;
4543                numentries >>= 20 - PAGE_SHIFT;
4544                numentries <<= 20 - PAGE_SHIFT;
4545
4546                /* limit to 1 bucket per 2^scale bytes of low memory */
4547                if (scale > PAGE_SHIFT)
4548                        numentries >>= (scale - PAGE_SHIFT);
4549                else
4550                        numentries <<= (PAGE_SHIFT - scale);
4551
4552                /* Make sure we've got at least a 0-order allocation.. */
4553                if (unlikely((numentries * bucketsize) < PAGE_SIZE))
4554                        numentries = PAGE_SIZE / bucketsize;
4555        }
4556        numentries = roundup_pow_of_two(numentries);
4557
4558        /* limit allocation size to 1/16 total memory by default */
4559        if (max == 0) {
4560                max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
4561                do_div(max, bucketsize);
4562        }
4563
4564        if (numentries > max)
4565                numentries = max;
4566
4567        log2qty = ilog2(numentries);
4568
4569        do {
4570                size = bucketsize << log2qty;
4571                if (flags & HASH_EARLY)
4572                        table = alloc_bootmem_nopanic(size);
4573                else if (hashdist)
4574                        table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
4575                else {
4576                        unsigned long order = get_order(size);
4577                        table = (void*) __get_free_pages(GFP_ATOMIC, order);
4578                        /*
4579                         * If bucketsize is not a power-of-two, we may free
4580                         * some pages at the end of hash table.
4581                         */
4582                        if (table) {
4583                                unsigned long alloc_end = (unsigned long)table +
4584                                                (PAGE_SIZE << order);
4585                                unsigned long used = (unsigned long)table +
4586                                                PAGE_ALIGN(size);
4587                                split_page(virt_to_page(table), order);
4588                                while (used < alloc_end) {
4589                                        free_page(used);
4590                                        used += PAGE_SIZE;
4591                                }
4592                        }
4593                }
4594        } while (!table && size > PAGE_SIZE && --log2qty);
4595
4596        if (!table)
4597                panic("Failed to allocate %s hash table\n", tablename);
4598
4599        printk(KERN_INFO "%s hash table entries: %d (order: %d, %lu bytes)\n",
4600               tablename,
4601               (1U << log2qty),
4602               ilog2(size) - PAGE_SHIFT,
4603               size);
4604
4605        if (_hash_shift)
4606                *_hash_shift = log2qty;
4607        if (_hash_mask)
4608                *_hash_mask = (1 << log2qty) - 1;
4609
4610        return table;
4611}
4612
4613/* Return a pointer to the bitmap storing bits affecting a block of pages */
4614static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
4615                                                        unsigned long pfn)
4616{
4617#ifdef CONFIG_SPARSEMEM
4618        return __pfn_to_section(pfn)->pageblock_flags;
4619#else
4620        return zone->pageblock_flags;
4621#endif /* CONFIG_SPARSEMEM */
4622}
4623
4624static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
4625{
4626#ifdef CONFIG_SPARSEMEM
4627        pfn &= (PAGES_PER_SECTION-1);
4628        return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4629#else
4630        pfn = pfn - zone->zone_start_pfn;
4631        return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
4632#endif /* CONFIG_SPARSEMEM */
4633}
4634
4635/**
4636 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4637 * @page: The page within the block of interest
4638 * @start_bitidx: The first bit of interest to retrieve
4639 * @end_bitidx: The last bit of interest
4640 * returns pageblock_bits flags
4641 */
4642unsigned long get_pageblock_flags_group(struct page *page,
4643                                        int start_bitidx, int end_bitidx)
4644{
4645        struct zone *zone;
4646        unsigned long *bitmap;
4647        unsigned long pfn, bitidx;
4648        unsigned long flags = 0;
4649        unsigned long value = 1;
4650
4651        zone = page_zone(page);
4652        pfn = page_to_pfn(page);
4653        bitmap = get_pageblock_bitmap(zone, pfn);
4654        bitidx = pfn_to_bitidx(zone, pfn);
4655
4656        for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4657                if (test_bit(bitidx + start_bitidx, bitmap))
4658                        flags |= value;
4659
4660        return flags;
4661}
4662
4663/**
4664 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4665 * @page: The page within the block of interest
4666 * @start_bitidx: The first bit of interest
4667 * @end_bitidx: The last bit of interest
4668 * @flags: The flags to set
4669 */
4670void set_pageblock_flags_group(struct page *page, unsigned long flags,
4671                                        int start_bitidx, int end_bitidx)
4672{
4673        struct zone *zone;
4674        unsigned long *bitmap;
4675        unsigned long pfn, bitidx;
4676        unsigned long value = 1;
4677
4678        zone = page_zone(page);
4679        pfn = page_to_pfn(page);
4680        bitmap = get_pageblock_bitmap(zone, pfn);
4681        bitidx = pfn_to_bitidx(zone, pfn);
4682        VM_BUG_ON(pfn < zone->zone_start_pfn);
4683        VM_BUG_ON(pfn >= zone->zone_start_pfn + zone->spanned_pages);
4684
4685        for (; start_bitidx <= end_bitidx; start_bitidx++, value <<= 1)
4686                if (flags & value)
4687                        __set_bit(bitidx + start_bitidx, bitmap);
4688                else
4689                        __clear_bit(bitidx + start_bitidx, bitmap);
4690}
4691
4692/*
4693 * This is designed as sub function...plz see page_isolation.c also.
4694 * set/clear page block's type to be ISOLATE.
4695 * page allocater never alloc memory from ISOLATE block.
4696 */
4697
4698int set_migratetype_isolate(struct page *page)
4699{
4700        struct zone *zone;
4701        unsigned long flags;
4702        int ret = -EBUSY;
4703
4704        zone = page_zone(page);
4705        spin_lock_irqsave(&zone->lock, flags);
4706        /*
4707         * In future, more migrate types will be able to be isolation target.
4708         */
4709        if (get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
4710                goto out;
4711        set_pageblock_migratetype(page, MIGRATE_ISOLATE);
4712        move_freepages_block(zone, page, MIGRATE_ISOLATE);
4713        ret = 0;
4714out:
4715        spin_unlock_irqrestore(&zone->lock, flags);
4716        if (!ret)
4717                drain_all_pages();
4718        return ret;
4719}
4720
4721void unset_migratetype_isolate(struct page *page)
4722{
4723        struct zone *zone;
4724        unsigned long flags;
4725        zone = page_zone(page);
4726        spin_lock_irqsave(&zone->lock, flags);
4727        if (get_pageblock_migratetype(page) != MIGRATE_ISOLATE)
4728                goto out;
4729        set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4730        move_freepages_block(