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/memblock.h>
  25#include <linux/compiler.h>
  26#include <linux/kernel.h>
  27#include <linux/kmemcheck.h>
  28#include <linux/kasan.h>
  29#include <linux/module.h>
  30#include <linux/suspend.h>
  31#include <linux/pagevec.h>
  32#include <linux/blkdev.h>
  33#include <linux/slab.h>
  34#include <linux/ratelimit.h>
  35#include <linux/oom.h>
  36#include <linux/notifier.h>
  37#include <linux/topology.h>
  38#include <linux/sysctl.h>
  39#include <linux/cpu.h>
  40#include <linux/cpuset.h>
  41#include <linux/memory_hotplug.h>
  42#include <linux/nodemask.h>
  43#include <linux/vmalloc.h>
  44#include <linux/vmstat.h>
  45#include <linux/mempolicy.h>
  46#include <linux/memremap.h>
  47#include <linux/stop_machine.h>
  48#include <linux/sort.h>
  49#include <linux/pfn.h>
  50#include <linux/backing-dev.h>
  51#include <linux/fault-inject.h>
  52#include <linux/page-isolation.h>
  53#include <linux/page_ext.h>
  54#include <linux/debugobjects.h>
  55#include <linux/kmemleak.h>
  56#include <linux/compaction.h>
  57#include <trace/events/kmem.h>
  58#include <linux/prefetch.h>
  59#include <linux/mm_inline.h>
  60#include <linux/migrate.h>
  61#include <linux/page_ext.h>
  62#include <linux/hugetlb.h>
  63#include <linux/sched/rt.h>
  64#include <linux/page_owner.h>
  65#include <linux/kthread.h>
  66#include <linux/memcontrol.h>
  67
  68#include <asm/sections.h>
  69#include <asm/tlbflush.h>
  70#include <asm/div64.h>
  71#include "internal.h"
  72
  73/* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
  74static DEFINE_MUTEX(pcp_batch_high_lock);
  75#define MIN_PERCPU_PAGELIST_FRACTION    (8)
  76
  77#ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
  78DEFINE_PER_CPU(int, numa_node);
  79EXPORT_PER_CPU_SYMBOL(numa_node);
  80#endif
  81
  82#ifdef CONFIG_HAVE_MEMORYLESS_NODES
  83/*
  84 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
  85 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
  86 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
  87 * defined in <linux/topology.h>.
  88 */
  89DEFINE_PER_CPU(int, _numa_mem_);                /* Kernel "local memory" node */
  90EXPORT_PER_CPU_SYMBOL(_numa_mem_);
  91int _node_numa_mem_[MAX_NUMNODES];
  92#endif
  93
  94#ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY
  95volatile unsigned long latent_entropy __latent_entropy;
  96EXPORT_SYMBOL(latent_entropy);
  97#endif
  98
  99/*
 100 * Array of node states.
 101 */
 102nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
 103        [N_POSSIBLE] = NODE_MASK_ALL,
 104        [N_ONLINE] = { { [0] = 1UL } },
 105#ifndef CONFIG_NUMA
 106        [N_NORMAL_MEMORY] = { { [0] = 1UL } },
 107#ifdef CONFIG_HIGHMEM
 108        [N_HIGH_MEMORY] = { { [0] = 1UL } },
 109#endif
 110#ifdef CONFIG_MOVABLE_NODE
 111        [N_MEMORY] = { { [0] = 1UL } },
 112#endif
 113        [N_CPU] = { { [0] = 1UL } },
 114#endif  /* NUMA */
 115};
 116EXPORT_SYMBOL(node_states);
 117
 118/* Protect totalram_pages and zone->managed_pages */
 119static DEFINE_SPINLOCK(managed_page_count_lock);
 120
 121unsigned long totalram_pages __read_mostly;
 122unsigned long totalreserve_pages __read_mostly;
 123unsigned long totalcma_pages __read_mostly;
 124
 125int percpu_pagelist_fraction;
 126gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
 127
 128/*
 129 * A cached value of the page's pageblock's migratetype, used when the page is
 130 * put on a pcplist. Used to avoid the pageblock migratetype lookup when
 131 * freeing from pcplists in most cases, at the cost of possibly becoming stale.
 132 * Also the migratetype set in the page does not necessarily match the pcplist
 133 * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
 134 * other index - this ensures that it will be put on the correct CMA freelist.
 135 */
 136static inline int get_pcppage_migratetype(struct page *page)
 137{
 138        return page->index;
 139}
 140
 141static inline void set_pcppage_migratetype(struct page *page, int migratetype)
 142{
 143        page->index = migratetype;
 144}
 145
 146#ifdef CONFIG_PM_SLEEP
 147/*
 148 * The following functions are used by the suspend/hibernate code to temporarily
 149 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
 150 * while devices are suspended.  To avoid races with the suspend/hibernate code,
 151 * they should always be called with pm_mutex held (gfp_allowed_mask also should
 152 * only be modified with pm_mutex held, unless the suspend/hibernate code is
 153 * guaranteed not to run in parallel with that modification).
 154 */
 155
 156static gfp_t saved_gfp_mask;
 157
 158void pm_restore_gfp_mask(void)
 159{
 160        WARN_ON(!mutex_is_locked(&pm_mutex));
 161        if (saved_gfp_mask) {
 162                gfp_allowed_mask = saved_gfp_mask;
 163                saved_gfp_mask = 0;
 164        }
 165}
 166
 167void pm_restrict_gfp_mask(void)
 168{
 169        WARN_ON(!mutex_is_locked(&pm_mutex));
 170        WARN_ON(saved_gfp_mask);
 171        saved_gfp_mask = gfp_allowed_mask;
 172        gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS);
 173}
 174
 175bool pm_suspended_storage(void)
 176{
 177        if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS))
 178                return false;
 179        return true;
 180}
 181#endif /* CONFIG_PM_SLEEP */
 182
 183#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
 184unsigned int pageblock_order __read_mostly;
 185#endif
 186
 187static void __free_pages_ok(struct page *page, unsigned int order);
 188
 189/*
 190 * results with 256, 32 in the lowmem_reserve sysctl:
 191 *      1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
 192 *      1G machine -> (16M dma, 784M normal, 224M high)
 193 *      NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
 194 *      HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
 195 *      HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
 196 *
 197 * TBD: should special case ZONE_DMA32 machines here - in those we normally
 198 * don't need any ZONE_NORMAL reservation
 199 */
 200int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
 201#ifdef CONFIG_ZONE_DMA
 202         256,
 203#endif
 204#ifdef CONFIG_ZONE_DMA32
 205         256,
 206#endif
 207#ifdef CONFIG_HIGHMEM
 208         32,
 209#endif
 210         32,
 211};
 212
 213EXPORT_SYMBOL(totalram_pages);
 214
 215static char * const zone_names[MAX_NR_ZONES] = {
 216#ifdef CONFIG_ZONE_DMA
 217         "DMA",
 218#endif
 219#ifdef CONFIG_ZONE_DMA32
 220         "DMA32",
 221#endif
 222         "Normal",
 223#ifdef CONFIG_HIGHMEM
 224         "HighMem",
 225#endif
 226         "Movable",
 227#ifdef CONFIG_ZONE_DEVICE
 228         "Device",
 229#endif
 230};
 231
 232char * const migratetype_names[MIGRATE_TYPES] = {
 233        "Unmovable",
 234        "Movable",
 235        "Reclaimable",
 236        "HighAtomic",
 237#ifdef CONFIG_CMA
 238        "CMA",
 239#endif
 240#ifdef CONFIG_MEMORY_ISOLATION
 241        "Isolate",
 242#endif
 243};
 244
 245compound_page_dtor * const compound_page_dtors[] = {
 246        NULL,
 247        free_compound_page,
 248#ifdef CONFIG_HUGETLB_PAGE
 249        free_huge_page,
 250#endif
 251#ifdef CONFIG_TRANSPARENT_HUGEPAGE
 252        free_transhuge_page,
 253#endif
 254};
 255
 256int min_free_kbytes = 1024;
 257int user_min_free_kbytes = -1;
 258int watermark_scale_factor = 10;
 259
 260static unsigned long __meminitdata nr_kernel_pages;
 261static unsigned long __meminitdata nr_all_pages;
 262static unsigned long __meminitdata dma_reserve;
 263
 264#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
 265static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
 266static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
 267static unsigned long __initdata required_kernelcore;
 268static unsigned long __initdata required_movablecore;
 269static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
 270static bool mirrored_kernelcore;
 271
 272/* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
 273int movable_zone;
 274EXPORT_SYMBOL(movable_zone);
 275#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
 276
 277#if MAX_NUMNODES > 1
 278int nr_node_ids __read_mostly = MAX_NUMNODES;
 279int nr_online_nodes __read_mostly = 1;
 280EXPORT_SYMBOL(nr_node_ids);
 281EXPORT_SYMBOL(nr_online_nodes);
 282#endif
 283
 284int page_group_by_mobility_disabled __read_mostly;
 285
 286#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
 287static inline void reset_deferred_meminit(pg_data_t *pgdat)
 288{
 289        pgdat->first_deferred_pfn = ULONG_MAX;
 290}
 291
 292/* Returns true if the struct page for the pfn is uninitialised */
 293static inline bool __meminit early_page_uninitialised(unsigned long pfn)
 294{
 295        int nid = early_pfn_to_nid(pfn);
 296
 297        if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn)
 298                return true;
 299
 300        return false;
 301}
 302
 303/*
 304 * Returns false when the remaining initialisation should be deferred until
 305 * later in the boot cycle when it can be parallelised.
 306 */
 307static inline bool update_defer_init(pg_data_t *pgdat,
 308                                unsigned long pfn, unsigned long zone_end,
 309                                unsigned long *nr_initialised)
 310{
 311        unsigned long max_initialise;
 312
 313        /* Always populate low zones for address-contrained allocations */
 314        if (zone_end < pgdat_end_pfn(pgdat))
 315                return true;
 316        /*
 317         * Initialise at least 2G of a node but also take into account that
 318         * two large system hashes that can take up 1GB for 0.25TB/node.
 319         */
 320        max_initialise = max(2UL << (30 - PAGE_SHIFT),
 321                (pgdat->node_spanned_pages >> 8));
 322
 323        (*nr_initialised)++;
 324        if ((*nr_initialised > max_initialise) &&
 325            (pfn & (PAGES_PER_SECTION - 1)) == 0) {
 326                pgdat->first_deferred_pfn = pfn;
 327                return false;
 328        }
 329
 330        return true;
 331}
 332#else
 333static inline void reset_deferred_meminit(pg_data_t *pgdat)
 334{
 335}
 336
 337static inline bool early_page_uninitialised(unsigned long pfn)
 338{
 339        return false;
 340}
 341
 342static inline bool update_defer_init(pg_data_t *pgdat,
 343                                unsigned long pfn, unsigned long zone_end,
 344                                unsigned long *nr_initialised)
 345{
 346        return true;
 347}
 348#endif
 349
 350/* Return a pointer to the bitmap storing bits affecting a block of pages */
 351static inline unsigned long *get_pageblock_bitmap(struct page *page,
 352                                                        unsigned long pfn)
 353{
 354#ifdef CONFIG_SPARSEMEM
 355        return __pfn_to_section(pfn)->pageblock_flags;
 356#else
 357        return page_zone(page)->pageblock_flags;
 358#endif /* CONFIG_SPARSEMEM */
 359}
 360
 361static inline int pfn_to_bitidx(struct page *page, unsigned long pfn)
 362{
 363#ifdef CONFIG_SPARSEMEM
 364        pfn &= (PAGES_PER_SECTION-1);
 365        return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
 366#else
 367        pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages);
 368        return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
 369#endif /* CONFIG_SPARSEMEM */
 370}
 371
 372/**
 373 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
 374 * @page: The page within the block of interest
 375 * @pfn: The target page frame number
 376 * @end_bitidx: The last bit of interest to retrieve
 377 * @mask: mask of bits that the caller is interested in
 378 *
 379 * Return: pageblock_bits flags
 380 */
 381static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page,
 382                                        unsigned long pfn,
 383                                        unsigned long end_bitidx,
 384                                        unsigned long mask)
 385{
 386        unsigned long *bitmap;
 387        unsigned long bitidx, word_bitidx;
 388        unsigned long word;
 389
 390        bitmap = get_pageblock_bitmap(page, pfn);
 391        bitidx = pfn_to_bitidx(page, pfn);
 392        word_bitidx = bitidx / BITS_PER_LONG;
 393        bitidx &= (BITS_PER_LONG-1);
 394
 395        word = bitmap[word_bitidx];
 396        bitidx += end_bitidx;
 397        return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
 398}
 399
 400unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
 401                                        unsigned long end_bitidx,
 402                                        unsigned long mask)
 403{
 404        return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask);
 405}
 406
 407static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn)
 408{
 409        return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK);
 410}
 411
 412/**
 413 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
 414 * @page: The page within the block of interest
 415 * @flags: The flags to set
 416 * @pfn: The target page frame number
 417 * @end_bitidx: The last bit of interest
 418 * @mask: mask of bits that the caller is interested in
 419 */
 420void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
 421                                        unsigned long pfn,
 422                                        unsigned long end_bitidx,
 423                                        unsigned long mask)
 424{
 425        unsigned long *bitmap;
 426        unsigned long bitidx, word_bitidx;
 427        unsigned long old_word, word;
 428
 429        BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
 430
 431        bitmap = get_pageblock_bitmap(page, pfn);
 432        bitidx = pfn_to_bitidx(page, pfn);
 433        word_bitidx = bitidx / BITS_PER_LONG;
 434        bitidx &= (BITS_PER_LONG-1);
 435
 436        VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page);
 437
 438        bitidx += end_bitidx;
 439        mask <<= (BITS_PER_LONG - bitidx - 1);
 440        flags <<= (BITS_PER_LONG - bitidx - 1);
 441
 442        word = READ_ONCE(bitmap[word_bitidx]);
 443        for (;;) {
 444                old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
 445                if (word == old_word)
 446                        break;
 447                word = old_word;
 448        }
 449}
 450
 451void set_pageblock_migratetype(struct page *page, int migratetype)
 452{
 453        if (unlikely(page_group_by_mobility_disabled &&
 454                     migratetype < MIGRATE_PCPTYPES))
 455                migratetype = MIGRATE_UNMOVABLE;
 456
 457        set_pageblock_flags_group(page, (unsigned long)migratetype,
 458                                        PB_migrate, PB_migrate_end);
 459}
 460
 461#ifdef CONFIG_DEBUG_VM
 462static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
 463{
 464        int ret = 0;
 465        unsigned seq;
 466        unsigned long pfn = page_to_pfn(page);
 467        unsigned long sp, start_pfn;
 468
 469        do {
 470                seq = zone_span_seqbegin(zone);
 471                start_pfn = zone->zone_start_pfn;
 472                sp = zone->spanned_pages;
 473                if (!zone_spans_pfn(zone, pfn))
 474                        ret = 1;
 475        } while (zone_span_seqretry(zone, seq));
 476
 477        if (ret)
 478                pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
 479                        pfn, zone_to_nid(zone), zone->name,
 480                        start_pfn, start_pfn + sp);
 481
 482        return ret;
 483}
 484
 485static int page_is_consistent(struct zone *zone, struct page *page)
 486{
 487        if (!pfn_valid_within(page_to_pfn(page)))
 488                return 0;
 489        if (zone != page_zone(page))
 490                return 0;
 491
 492        return 1;
 493}
 494/*
 495 * Temporary debugging check for pages not lying within a given zone.
 496 */
 497static int bad_range(struct zone *zone, struct page *page)
 498{
 499        if (page_outside_zone_boundaries(zone, page))
 500                return 1;
 501        if (!page_is_consistent(zone, page))
 502                return 1;
 503
 504        return 0;
 505}
 506#else
 507static inline int bad_range(struct zone *zone, struct page *page)
 508{
 509        return 0;
 510}
 511#endif
 512
 513static void bad_page(struct page *page, const char *reason,
 514                unsigned long bad_flags)
 515{
 516        static unsigned long resume;
 517        static unsigned long nr_shown;
 518        static unsigned long nr_unshown;
 519
 520        /*
 521         * Allow a burst of 60 reports, then keep quiet for that minute;
 522         * or allow a steady drip of one report per second.
 523         */
 524        if (nr_shown == 60) {
 525                if (time_before(jiffies, resume)) {
 526                        nr_unshown++;
 527                        goto out;
 528                }
 529                if (nr_unshown) {
 530                        pr_alert(
 531                              "BUG: Bad page state: %lu messages suppressed\n",
 532                                nr_unshown);
 533                        nr_unshown = 0;
 534                }
 535                nr_shown = 0;
 536        }
 537        if (nr_shown++ == 0)
 538                resume = jiffies + 60 * HZ;
 539
 540        pr_alert("BUG: Bad page state in process %s  pfn:%05lx\n",
 541                current->comm, page_to_pfn(page));
 542        __dump_page(page, reason);
 543        bad_flags &= page->flags;
 544        if (bad_flags)
 545                pr_alert("bad because of flags: %#lx(%pGp)\n",
 546                                                bad_flags, &bad_flags);
 547        dump_page_owner(page);
 548
 549        print_modules();
 550        dump_stack();
 551out:
 552        /* Leave bad fields for debug, except PageBuddy could make trouble */
 553        page_mapcount_reset(page); /* remove PageBuddy */
 554        add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
 555}
 556
 557/*
 558 * Higher-order pages are called "compound pages".  They are structured thusly:
 559 *
 560 * The first PAGE_SIZE page is called the "head page" and have PG_head set.
 561 *
 562 * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
 563 * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
 564 *
 565 * The first tail page's ->compound_dtor holds the offset in array of compound
 566 * page destructors. See compound_page_dtors.
 567 *
 568 * The first tail page's ->compound_order holds the order of allocation.
 569 * This usage means that zero-order pages may not be compound.
 570 */
 571
 572void free_compound_page(struct page *page)
 573{
 574        __free_pages_ok(page, compound_order(page));
 575}
 576
 577void prep_compound_page(struct page *page, unsigned int order)
 578{
 579        int i;
 580        int nr_pages = 1 << order;
 581
 582        set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
 583        set_compound_order(page, order);
 584        __SetPageHead(page);
 585        for (i = 1; i < nr_pages; i++) {
 586                struct page *p = page + i;
 587                set_page_count(p, 0);
 588                p->mapping = TAIL_MAPPING;
 589                set_compound_head(p, page);
 590        }
 591        atomic_set(compound_mapcount_ptr(page), -1);
 592}
 593
 594#ifdef CONFIG_DEBUG_PAGEALLOC
 595unsigned int _debug_guardpage_minorder;
 596bool _debug_pagealloc_enabled __read_mostly
 597                        = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT);
 598EXPORT_SYMBOL(_debug_pagealloc_enabled);
 599bool _debug_guardpage_enabled __read_mostly;
 600
 601static int __init early_debug_pagealloc(char *buf)
 602{
 603        if (!buf)
 604                return -EINVAL;
 605        return kstrtobool(buf, &_debug_pagealloc_enabled);
 606}
 607early_param("debug_pagealloc", early_debug_pagealloc);
 608
 609static bool need_debug_guardpage(void)
 610{
 611        /* If we don't use debug_pagealloc, we don't need guard page */
 612        if (!debug_pagealloc_enabled())
 613                return false;
 614
 615        if (!debug_guardpage_minorder())
 616                return false;
 617
 618        return true;
 619}
 620
 621static void init_debug_guardpage(void)
 622{
 623        if (!debug_pagealloc_enabled())
 624                return;
 625
 626        if (!debug_guardpage_minorder())
 627                return;
 628
 629        _debug_guardpage_enabled = true;
 630}
 631
 632struct page_ext_operations debug_guardpage_ops = {
 633        .need = need_debug_guardpage,
 634        .init = init_debug_guardpage,
 635};
 636
 637static int __init debug_guardpage_minorder_setup(char *buf)
 638{
 639        unsigned long res;
 640
 641        if (kstrtoul(buf, 10, &res) < 0 ||  res > MAX_ORDER / 2) {
 642                pr_err("Bad debug_guardpage_minorder value\n");
 643                return 0;
 644        }
 645        _debug_guardpage_minorder = res;
 646        pr_info("Setting debug_guardpage_minorder to %lu\n", res);
 647        return 0;
 648}
 649early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup);
 650
 651static inline bool set_page_guard(struct zone *zone, struct page *page,
 652                                unsigned int order, int migratetype)
 653{
 654        struct page_ext *page_ext;
 655
 656        if (!debug_guardpage_enabled())
 657                return false;
 658
 659        if (order >= debug_guardpage_minorder())
 660                return false;
 661
 662        page_ext = lookup_page_ext(page);
 663        if (unlikely(!page_ext))
 664                return false;
 665
 666        __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
 667
 668        INIT_LIST_HEAD(&page->lru);
 669        set_page_private(page, order);
 670        /* Guard pages are not available for any usage */
 671        __mod_zone_freepage_state(zone, -(1 << order), migratetype);
 672
 673        return true;
 674}
 675
 676static inline void clear_page_guard(struct zone *zone, struct page *page,
 677                                unsigned int order, int migratetype)
 678{
 679        struct page_ext *page_ext;
 680
 681        if (!debug_guardpage_enabled())
 682                return;
 683
 684        page_ext = lookup_page_ext(page);
 685        if (unlikely(!page_ext))
 686                return;
 687
 688        __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
 689
 690        set_page_private(page, 0);
 691        if (!is_migrate_isolate(migratetype))
 692                __mod_zone_freepage_state(zone, (1 << order), migratetype);
 693}
 694#else
 695struct page_ext_operations debug_guardpage_ops;
 696static inline bool set_page_guard(struct zone *zone, struct page *page,
 697                        unsigned int order, int migratetype) { return false; }
 698static inline void clear_page_guard(struct zone *zone, struct page *page,
 699                                unsigned int order, int migratetype) {}
 700#endif
 701
 702static inline void set_page_order(struct page *page, unsigned int order)
 703{
 704        set_page_private(page, order);
 705        __SetPageBuddy(page);
 706}
 707
 708static inline void rmv_page_order(struct page *page)
 709{
 710        __ClearPageBuddy(page);
 711        set_page_private(page, 0);
 712}
 713
 714/*
 715 * This function checks whether a page is free && is the buddy
 716 * we can do coalesce a page and its buddy if
 717 * (a) the buddy is not in a hole &&
 718 * (b) the buddy is in the buddy system &&
 719 * (c) a page and its buddy have the same order &&
 720 * (d) a page and its buddy are in the same zone.
 721 *
 722 * For recording whether a page is in the buddy system, we set ->_mapcount
 723 * PAGE_BUDDY_MAPCOUNT_VALUE.
 724 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
 725 * serialized by zone->lock.
 726 *
 727 * For recording page's order, we use page_private(page).
 728 */
 729static inline int page_is_buddy(struct page *page, struct page *buddy,
 730                                                        unsigned int order)
 731{
 732        if (!pfn_valid_within(page_to_pfn(buddy)))
 733                return 0;
 734
 735        if (page_is_guard(buddy) && page_order(buddy) == order) {
 736                if (page_zone_id(page) != page_zone_id(buddy))
 737                        return 0;
 738
 739                VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
 740
 741                return 1;
 742        }
 743
 744        if (PageBuddy(buddy) && page_order(buddy) == order) {
 745                /*
 746                 * zone check is done late to avoid uselessly
 747                 * calculating zone/node ids for pages that could
 748                 * never merge.
 749                 */
 750                if (page_zone_id(page) != page_zone_id(buddy))
 751                        return 0;
 752
 753                VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
 754
 755                return 1;
 756        }
 757        return 0;
 758}
 759
 760/*
 761 * Freeing function for a buddy system allocator.
 762 *
 763 * The concept of a buddy system is to maintain direct-mapped table
 764 * (containing bit values) for memory blocks of various "orders".
 765 * The bottom level table contains the map for the smallest allocatable
 766 * units of memory (here, pages), and each level above it describes
 767 * pairs of units from the levels below, hence, "buddies".
 768 * At a high level, all that happens here is marking the table entry
 769 * at the bottom level available, and propagating the changes upward
 770 * as necessary, plus some accounting needed to play nicely with other
 771 * parts of the VM system.
 772 * At each level, we keep a list of pages, which are heads of continuous
 773 * free pages of length of (1 << order) and marked with _mapcount
 774 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
 775 * field.
 776 * So when we are allocating or freeing one, we can derive the state of the
 777 * other.  That is, if we allocate a small block, and both were
 778 * free, the remainder of the region must be split into blocks.
 779 * If a block is freed, and its buddy is also free, then this
 780 * triggers coalescing into a block of larger size.
 781 *
 782 * -- nyc
 783 */
 784
 785static inline void __free_one_page(struct page *page,
 786                unsigned long pfn,
 787                struct zone *zone, unsigned int order,
 788                int migratetype)
 789{
 790        unsigned long page_idx;
 791        unsigned long combined_idx;
 792        unsigned long uninitialized_var(buddy_idx);
 793        struct page *buddy;
 794        unsigned int max_order;
 795
 796        max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1);
 797
 798        VM_BUG_ON(!zone_is_initialized(zone));
 799        VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
 800
 801        VM_BUG_ON(migratetype == -1);
 802        if (likely(!is_migrate_isolate(migratetype)))
 803                __mod_zone_freepage_state(zone, 1 << order, migratetype);
 804
 805        page_idx = pfn & ((1 << MAX_ORDER) - 1);
 806
 807        VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
 808        VM_BUG_ON_PAGE(bad_range(zone, page), page);
 809
 810continue_merging:
 811        while (order < max_order - 1) {
 812                buddy_idx = __find_buddy_index(page_idx, order);
 813                buddy = page + (buddy_idx - page_idx);
 814                if (!page_is_buddy(page, buddy, order))
 815                        goto done_merging;
 816                /*
 817                 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
 818                 * merge with it and move up one order.
 819                 */
 820                if (page_is_guard(buddy)) {
 821                        clear_page_guard(zone, buddy, order, migratetype);
 822                } else {
 823                        list_del(&buddy->lru);
 824                        zone->free_area[order].nr_free--;
 825                        rmv_page_order(buddy);
 826                }
 827                combined_idx = buddy_idx & page_idx;
 828                page = page + (combined_idx - page_idx);
 829                page_idx = combined_idx;
 830                order++;
 831        }
 832        if (max_order < MAX_ORDER) {
 833                /* If we are here, it means order is >= pageblock_order.
 834                 * We want to prevent merge between freepages on isolate
 835                 * pageblock and normal pageblock. Without this, pageblock
 836                 * isolation could cause incorrect freepage or CMA accounting.
 837                 *
 838                 * We don't want to hit this code for the more frequent
 839                 * low-order merging.
 840                 */
 841                if (unlikely(has_isolate_pageblock(zone))) {
 842                        int buddy_mt;
 843
 844                        buddy_idx = __find_buddy_index(page_idx, order);
 845                        buddy = page + (buddy_idx - page_idx);
 846                        buddy_mt = get_pageblock_migratetype(buddy);
 847
 848                        if (migratetype != buddy_mt
 849                                        && (is_migrate_isolate(migratetype) ||
 850                                                is_migrate_isolate(buddy_mt)))
 851                                goto done_merging;
 852                }
 853                max_order++;
 854                goto continue_merging;
 855        }
 856
 857done_merging:
 858        set_page_order(page, order);
 859
 860        /*
 861         * If this is not the largest possible page, check if the buddy
 862         * of the next-highest order is free. If it is, it's possible
 863         * that pages are being freed that will coalesce soon. In case,
 864         * that is happening, add the free page to the tail of the list
 865         * so it's less likely to be used soon and more likely to be merged
 866         * as a higher order page
 867         */
 868        if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
 869                struct page *higher_page, *higher_buddy;
 870                combined_idx = buddy_idx & page_idx;
 871                higher_page = page + (combined_idx - page_idx);
 872                buddy_idx = __find_buddy_index(combined_idx, order + 1);
 873                higher_buddy = higher_page + (buddy_idx - combined_idx);
 874                if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
 875                        list_add_tail(&page->lru,
 876                                &zone->free_area[order].free_list[migratetype]);
 877                        goto out;
 878                }
 879        }
 880
 881        list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
 882out:
 883        zone->free_area[order].nr_free++;
 884}
 885
 886/*
 887 * A bad page could be due to a number of fields. Instead of multiple branches,
 888 * try and check multiple fields with one check. The caller must do a detailed
 889 * check if necessary.
 890 */
 891static inline bool page_expected_state(struct page *page,
 892                                        unsigned long check_flags)
 893{
 894        if (unlikely(atomic_read(&page->_mapcount) != -1))
 895                return false;
 896
 897        if (unlikely((unsigned long)page->mapping |
 898                        page_ref_count(page) |
 899#ifdef CONFIG_MEMCG
 900                        (unsigned long)page->mem_cgroup |
 901#endif
 902                        (page->flags & check_flags)))
 903                return false;
 904
 905        return true;
 906}
 907
 908static void free_pages_check_bad(struct page *page)
 909{
 910        const char *bad_reason;
 911        unsigned long bad_flags;
 912
 913        bad_reason = NULL;
 914        bad_flags = 0;
 915
 916        if (unlikely(atomic_read(&page->_mapcount) != -1))
 917                bad_reason = "nonzero mapcount";
 918        if (unlikely(page->mapping != NULL))
 919                bad_reason = "non-NULL mapping";
 920        if (unlikely(page_ref_count(page) != 0))
 921                bad_reason = "nonzero _refcount";
 922        if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
 923                bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
 924                bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
 925        }
 926#ifdef CONFIG_MEMCG
 927        if (unlikely(page->mem_cgroup))
 928                bad_reason = "page still charged to cgroup";
 929#endif
 930        bad_page(page, bad_reason, bad_flags);
 931}
 932
 933static inline int free_pages_check(struct page *page)
 934{
 935        if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE)))
 936                return 0;
 937
 938        /* Something has gone sideways, find it */
 939        free_pages_check_bad(page);
 940        return 1;
 941}
 942
 943static int free_tail_pages_check(struct page *head_page, struct page *page)
 944{
 945        int ret = 1;
 946
 947        /*
 948         * We rely page->lru.next never has bit 0 set, unless the page
 949         * is PageTail(). Let's make sure that's true even for poisoned ->lru.
 950         */
 951        BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
 952
 953        if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
 954                ret = 0;
 955                goto out;
 956        }
 957        switch (page - head_page) {
 958        case 1:
 959                /* the first tail page: ->mapping is compound_mapcount() */
 960                if (unlikely(compound_mapcount(page))) {
 961                        bad_page(page, "nonzero compound_mapcount", 0);
 962                        goto out;
 963                }
 964                break;
 965        case 2:
 966                /*
 967                 * the second tail page: ->mapping is
 968                 * page_deferred_list().next -- ignore value.
 969                 */
 970                break;
 971        default:
 972                if (page->mapping != TAIL_MAPPING) {
 973                        bad_page(page, "corrupted mapping in tail page", 0);
 974                        goto out;
 975                }
 976                break;
 977        }
 978        if (unlikely(!PageTail(page))) {
 979                bad_page(page, "PageTail not set", 0);
 980                goto out;
 981        }
 982        if (unlikely(compound_head(page) != head_page)) {
 983                bad_page(page, "compound_head not consistent", 0);
 984                goto out;
 985        }
 986        ret = 0;
 987out:
 988        page->mapping = NULL;
 989        clear_compound_head(page);
 990        return ret;
 991}
 992
 993static __always_inline bool free_pages_prepare(struct page *page,
 994                                        unsigned int order, bool check_free)
 995{
 996        int bad = 0;
 997
 998        VM_BUG_ON_PAGE(PageTail(page), page);
 999
1000        trace_mm_page_free(page, order);
1001        kmemcheck_free_shadow(page, order);
1002
1003        /*
1004         * Check tail pages before head page information is cleared to
1005         * avoid checking PageCompound for order-0 pages.
1006         */
1007        if (unlikely(order)) {
1008                bool compound = PageCompound(page);
1009                int i;
1010
1011                VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
1012
1013                if (compound)
1014                        ClearPageDoubleMap(page);
1015                for (i = 1; i < (1 << order); i++) {
1016                        if (compound)
1017                                bad += free_tail_pages_check(page, page + i);
1018                        if (unlikely(free_pages_check(page + i))) {
1019                                bad++;
1020                                continue;
1021                        }
1022                        (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1023                }
1024        }
1025        if (PageMappingFlags(page))
1026                page->mapping = NULL;
1027        if (memcg_kmem_enabled() && PageKmemcg(page))
1028                memcg_kmem_uncharge(page, order);
1029        if (check_free)
1030                bad += free_pages_check(page);
1031        if (bad)
1032                return false;
1033
1034        page_cpupid_reset_last(page);
1035        page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1036        reset_page_owner(page, order);
1037
1038        if (!PageHighMem(page)) {
1039                debug_check_no_locks_freed(page_address(page),
1040                                           PAGE_SIZE << order);
1041                debug_check_no_obj_freed(page_address(page),
1042                                           PAGE_SIZE << order);
1043        }
1044        arch_free_page(page, order);
1045        kernel_poison_pages(page, 1 << order, 0);
1046        kernel_map_pages(page, 1 << order, 0);
1047        kasan_free_pages(page, order);
1048
1049        return true;
1050}
1051
1052#ifdef CONFIG_DEBUG_VM
1053static inline bool free_pcp_prepare(struct page *page)
1054{
1055        return free_pages_prepare(page, 0, true);
1056}
1057
1058static inline bool bulkfree_pcp_prepare(struct page *page)
1059{
1060        return false;
1061}
1062#else
1063static bool free_pcp_prepare(struct page *page)
1064{
1065        return free_pages_prepare(page, 0, false);
1066}
1067
1068static bool bulkfree_pcp_prepare(struct page *page)
1069{
1070        return free_pages_check(page);
1071}
1072#endif /* CONFIG_DEBUG_VM */
1073
1074/*
1075 * Frees a number of pages from the PCP lists
1076 * Assumes all pages on list are in same zone, and of same order.
1077 * count is the number of pages to free.
1078 *
1079 * If the zone was previously in an "all pages pinned" state then look to
1080 * see if this freeing clears that state.
1081 *
1082 * And clear the zone's pages_scanned counter, to hold off the "all pages are
1083 * pinned" detection logic.
1084 */
1085static void free_pcppages_bulk(struct zone *zone, int count,
1086                                        struct per_cpu_pages *pcp)
1087{
1088        int migratetype = 0;
1089        int batch_free = 0;
1090        unsigned long nr_scanned;
1091        bool isolated_pageblocks;
1092
1093        spin_lock(&zone->lock);
1094        isolated_pageblocks = has_isolate_pageblock(zone);
1095        nr_scanned = node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED);
1096        if (nr_scanned)
1097                __mod_node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED, -nr_scanned);
1098
1099        while (count) {
1100                struct page *page;
1101                struct list_head *list;
1102
1103                /*
1104                 * Remove pages from lists in a round-robin fashion. A
1105                 * batch_free count is maintained that is incremented when an
1106                 * empty list is encountered.  This is so more pages are freed
1107                 * off fuller lists instead of spinning excessively around empty
1108                 * lists
1109                 */
1110                do {
1111                        batch_free++;
1112                        if (++migratetype == MIGRATE_PCPTYPES)
1113                                migratetype = 0;
1114                        list = &pcp->lists[migratetype];
1115                } while (list_empty(list));
1116
1117                /* This is the only non-empty list. Free them all. */
1118                if (batch_free == MIGRATE_PCPTYPES)
1119                        batch_free = count;
1120
1121                do {
1122                        int mt; /* migratetype of the to-be-freed page */
1123
1124                        page = list_last_entry(list, struct page, lru);
1125                        /* must delete as __free_one_page list manipulates */
1126                        list_del(&page->lru);
1127
1128                        mt = get_pcppage_migratetype(page);
1129                        /* MIGRATE_ISOLATE page should not go to pcplists */
1130                        VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
1131                        /* Pageblock could have been isolated meanwhile */
1132                        if (unlikely(isolated_pageblocks))
1133                                mt = get_pageblock_migratetype(page);
1134
1135                        if (bulkfree_pcp_prepare(page))
1136                                continue;
1137
1138                        __free_one_page(page, page_to_pfn(page), zone, 0, mt);
1139                        trace_mm_page_pcpu_drain(page, 0, mt);
1140                } while (--count && --batch_free && !list_empty(list));
1141        }
1142        spin_unlock(&zone->lock);
1143}
1144
1145static void free_one_page(struct zone *zone,
1146                                struct page *page, unsigned long pfn,
1147                                unsigned int order,
1148                                int migratetype)
1149{
1150        unsigned long nr_scanned;
1151        spin_lock(&zone->lock);
1152        nr_scanned = node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED);
1153        if (nr_scanned)
1154                __mod_node_page_state(zone->zone_pgdat, NR_PAGES_SCANNED, -nr_scanned);
1155
1156        if (unlikely(has_isolate_pageblock(zone) ||
1157                is_migrate_isolate(migratetype))) {
1158                migratetype = get_pfnblock_migratetype(page, pfn);
1159        }
1160        __free_one_page(page, pfn, zone, order, migratetype);
1161        spin_unlock(&zone->lock);
1162}
1163
1164static void __meminit __init_single_page(struct page *page, unsigned long pfn,
1165                                unsigned long zone, int nid)
1166{
1167        set_page_links(page, zone, nid, pfn);
1168        init_page_count(page);
1169        page_mapcount_reset(page);
1170        page_cpupid_reset_last(page);
1171
1172        INIT_LIST_HEAD(&page->lru);
1173#ifdef WANT_PAGE_VIRTUAL
1174        /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1175        if (!is_highmem_idx(zone))
1176                set_page_address(page, __va(pfn << PAGE_SHIFT));
1177#endif
1178}
1179
1180static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone,
1181                                        int nid)
1182{
1183        return __init_single_page(pfn_to_page(pfn), pfn, zone, nid);
1184}
1185
1186#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1187static void init_reserved_page(unsigned long pfn)
1188{
1189        pg_data_t *pgdat;
1190        int nid, zid;
1191
1192        if (!early_page_uninitialised(pfn))
1193                return;
1194
1195        nid = early_pfn_to_nid(pfn);
1196        pgdat = NODE_DATA(nid);
1197
1198        for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1199                struct zone *zone = &pgdat->node_zones[zid];
1200
1201                if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
1202                        break;
1203        }
1204        __init_single_pfn(pfn, zid, nid);
1205}
1206#else
1207static inline void init_reserved_page(unsigned long pfn)
1208{
1209}
1210#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1211
1212/*
1213 * Initialised pages do not have PageReserved set. This function is
1214 * called for each range allocated by the bootmem allocator and
1215 * marks the pages PageReserved. The remaining valid pages are later
1216 * sent to the buddy page allocator.
1217 */
1218void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
1219{
1220        unsigned long start_pfn = PFN_DOWN(start);
1221        unsigned long end_pfn = PFN_UP(end);
1222
1223        for (; start_pfn < end_pfn; start_pfn++) {
1224                if (pfn_valid(start_pfn)) {
1225                        struct page *page = pfn_to_page(start_pfn);
1226
1227                        init_reserved_page(start_pfn);
1228
1229                        /* Avoid false-positive PageTail() */
1230                        INIT_LIST_HEAD(&page->lru);
1231
1232                        SetPageReserved(page);
1233                }
1234        }
1235}
1236
1237static void __free_pages_ok(struct page *page, unsigned int order)
1238{
1239        unsigned long flags;
1240        int migratetype;
1241        unsigned long pfn = page_to_pfn(page);
1242
1243        if (!free_pages_prepare(page, order, true))
1244                return;
1245
1246        migratetype = get_pfnblock_migratetype(page, pfn);
1247        local_irq_save(flags);
1248        __count_vm_events(PGFREE, 1 << order);
1249        free_one_page(page_zone(page), page, pfn, order, migratetype);
1250        local_irq_restore(flags);
1251}
1252
1253static void __init __free_pages_boot_core(struct page *page, unsigned int order)
1254{
1255        unsigned int nr_pages = 1 << order;
1256        struct page *p = page;
1257        unsigned int loop;
1258
1259        prefetchw(p);
1260        for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1261                prefetchw(p + 1);
1262                __ClearPageReserved(p);
1263                set_page_count(p, 0);
1264        }
1265        __ClearPageReserved(p);
1266        set_page_count(p, 0);
1267
1268        page_zone(page)->managed_pages += nr_pages;
1269        set_page_refcounted(page);
1270        __free_pages(page, order);
1271}
1272
1273#if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1274        defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1275
1276static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1277
1278int __meminit early_pfn_to_nid(unsigned long pfn)
1279{
1280        static DEFINE_SPINLOCK(early_pfn_lock);
1281        int nid;
1282
1283        spin_lock(&early_pfn_lock);
1284        nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
1285        if (nid < 0)
1286                nid = first_online_node;
1287        spin_unlock(&early_pfn_lock);
1288
1289        return nid;
1290}
1291#endif
1292
1293#ifdef CONFIG_NODES_SPAN_OTHER_NODES
1294static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1295                                        struct mminit_pfnnid_cache *state)
1296{
1297        int nid;
1298
1299        nid = __early_pfn_to_nid(pfn, state);
1300        if (nid >= 0 && nid != node)
1301                return false;
1302        return true;
1303}
1304
1305/* Only safe to use early in boot when initialisation is single-threaded */
1306static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1307{
1308        return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1309}
1310
1311#else
1312
1313static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1314{
1315        return true;
1316}
1317static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1318                                        struct mminit_pfnnid_cache *state)
1319{
1320        return true;
1321}
1322#endif
1323
1324
1325void __init __free_pages_bootmem(struct page *page, unsigned long pfn,
1326                                                        unsigned int order)
1327{
1328        if (early_page_uninitialised(pfn))
1329                return;
1330        return __free_pages_boot_core(page, order);
1331}
1332
1333/*
1334 * Check that the whole (or subset of) a pageblock given by the interval of
1335 * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1336 * with the migration of free compaction scanner. The scanners then need to
1337 * use only pfn_valid_within() check for arches that allow holes within
1338 * pageblocks.
1339 *
1340 * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1341 *
1342 * It's possible on some configurations to have a setup like node0 node1 node0
1343 * i.e. it's possible that all pages within a zones range of pages do not
1344 * belong to a single zone. We assume that a border between node0 and node1
1345 * can occur within a single pageblock, but not a node0 node1 node0
1346 * interleaving within a single pageblock. It is therefore sufficient to check
1347 * the first and last page of a pageblock and avoid checking each individual
1348 * page in a pageblock.
1349 */
1350struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
1351                                     unsigned long end_pfn, struct zone *zone)
1352{
1353        struct page *start_page;
1354        struct page *end_page;
1355
1356        /* end_pfn is one past the range we are checking */
1357        end_pfn--;
1358
1359        if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
1360                return NULL;
1361
1362        start_page = pfn_to_page(start_pfn);
1363
1364        if (page_zone(start_page) != zone)
1365                return NULL;
1366
1367        end_page = pfn_to_page(end_pfn);
1368
1369        /* This gives a shorter code than deriving page_zone(end_page) */
1370        if (page_zone_id(start_page) != page_zone_id(end_page))
1371                return NULL;
1372
1373        return start_page;
1374}
1375
1376void set_zone_contiguous(struct zone *zone)
1377{
1378        unsigned long block_start_pfn = zone->zone_start_pfn;
1379        unsigned long block_end_pfn;
1380
1381        block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages);
1382        for (; block_start_pfn < zone_end_pfn(zone);
1383                        block_start_pfn = block_end_pfn,
1384                         block_end_pfn += pageblock_nr_pages) {
1385
1386                block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
1387
1388                if (!__pageblock_pfn_to_page(block_start_pfn,
1389                                             block_end_pfn, zone))
1390                        return;
1391        }
1392
1393        /* We confirm that there is no hole */
1394        zone->contiguous = true;
1395}
1396
1397void clear_zone_contiguous(struct zone *zone)
1398{
1399        zone->contiguous = false;
1400}
1401
1402#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1403static void __init deferred_free_range(struct page *page,
1404                                        unsigned long pfn, int nr_pages)
1405{
1406        int i;
1407
1408        if (!page)
1409                return;
1410
1411        /* Free a large naturally-aligned chunk if possible */
1412        if (nr_pages == pageblock_nr_pages &&
1413            (pfn & (pageblock_nr_pages - 1)) == 0) {
1414                set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1415                __free_pages_boot_core(page, pageblock_order);
1416                return;
1417        }
1418
1419        for (i = 0; i < nr_pages; i++, page++, pfn++) {
1420                if ((pfn & (pageblock_nr_pages - 1)) == 0)
1421                        set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1422                __free_pages_boot_core(page, 0);
1423        }
1424}
1425
1426/* Completion tracking for deferred_init_memmap() threads */
1427static atomic_t pgdat_init_n_undone __initdata;
1428static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1429
1430static inline void __init pgdat_init_report_one_done(void)
1431{
1432        if (atomic_dec_and_test(&pgdat_init_n_undone))
1433                complete(&pgdat_init_all_done_comp);
1434}
1435
1436/* Initialise remaining memory on a node */
1437static int __init deferred_init_memmap(void *data)
1438{
1439        pg_data_t *pgdat = data;
1440        int nid = pgdat->node_id;
1441        struct mminit_pfnnid_cache nid_init_state = { };
1442        unsigned long start = jiffies;
1443        unsigned long nr_pages = 0;
1444        unsigned long walk_start, walk_end;
1445        int i, zid;
1446        struct zone *zone;
1447        unsigned long first_init_pfn = pgdat->first_deferred_pfn;
1448        const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
1449
1450        if (first_init_pfn == ULONG_MAX) {
1451                pgdat_init_report_one_done();
1452                return 0;
1453        }
1454
1455        /* Bind memory initialisation thread to a local node if possible */
1456        if (!cpumask_empty(cpumask))
1457                set_cpus_allowed_ptr(current, cpumask);
1458
1459        /* Sanity check boundaries */
1460        BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1461        BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1462        pgdat->first_deferred_pfn = ULONG_MAX;
1463
1464        /* Only the highest zone is deferred so find it */
1465        for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1466                zone = pgdat->node_zones + zid;
1467                if (first_init_pfn < zone_end_pfn(zone))
1468                        break;
1469        }
1470
1471        for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) {
1472                unsigned long pfn, end_pfn;
1473                struct page *page = NULL;
1474                struct page *free_base_page = NULL;
1475                unsigned long free_base_pfn = 0;
1476                int nr_to_free = 0;
1477
1478                end_pfn = min(walk_end, zone_end_pfn(zone));
1479                pfn = first_init_pfn;
1480                if (pfn < walk_start)
1481                        pfn = walk_start;
1482                if (pfn < zone->zone_start_pfn)
1483                        pfn = zone->zone_start_pfn;
1484
1485                for (; pfn < end_pfn; pfn++) {
1486                        if (!pfn_valid_within(pfn))
1487                                goto free_range;
1488
1489                        /*
1490                         * Ensure pfn_valid is checked every
1491                         * pageblock_nr_pages for memory holes
1492                         */
1493                        if ((pfn & (pageblock_nr_pages - 1)) == 0) {
1494                                if (!pfn_valid(pfn)) {
1495                                        page = NULL;
1496                                        goto free_range;
1497                                }
1498                        }
1499
1500                        if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) {
1501                                page = NULL;
1502                                goto free_range;
1503                        }
1504
1505                        /* Minimise pfn page lookups and scheduler checks */
1506                        if (page && (pfn & (pageblock_nr_pages - 1)) != 0) {
1507                                page++;
1508                        } else {
1509                                nr_pages += nr_to_free;
1510                                deferred_free_range(free_base_page,
1511                                                free_base_pfn, nr_to_free);
1512                                free_base_page = NULL;
1513                                free_base_pfn = nr_to_free = 0;
1514
1515                                page = pfn_to_page(pfn);
1516                                cond_resched();
1517                        }
1518
1519                        if (page->flags) {
1520                                VM_BUG_ON(page_zone(page) != zone);
1521                                goto free_range;
1522                        }
1523
1524                        __init_single_page(page, pfn, zid, nid);
1525                        if (!free_base_page) {
1526                                free_base_page = page;
1527                                free_base_pfn = pfn;
1528                                nr_to_free = 0;
1529                        }
1530                        nr_to_free++;
1531
1532                        /* Where possible, batch up pages for a single free */
1533                        continue;
1534free_range:
1535                        /* Free the current block of pages to allocator */
1536                        nr_pages += nr_to_free;
1537                        deferred_free_range(free_base_page, free_base_pfn,
1538                                                                nr_to_free);
1539                        free_base_page = NULL;
1540                        free_base_pfn = nr_to_free = 0;
1541                }
1542                /* Free the last block of pages to allocator */
1543                nr_pages += nr_to_free;
1544                deferred_free_range(free_base_page, free_base_pfn, nr_to_free);
1545
1546                first_init_pfn = max(end_pfn, first_init_pfn);
1547        }
1548
1549        /* Sanity check that the next zone really is unpopulated */
1550        WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1551
1552        pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
1553                                        jiffies_to_msecs(jiffies - start));
1554
1555        pgdat_init_report_one_done();
1556        return 0;
1557}
1558#endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1559
1560void __init page_alloc_init_late(void)
1561{
1562        struct zone *zone;
1563
1564#ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1565        int nid;
1566
1567        /* There will be num_node_state(N_MEMORY) threads */
1568        atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
1569        for_each_node_state(nid, N_MEMORY) {
1570                kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1571        }
1572
1573        /* Block until all are initialised */
1574        wait_for_completion(&pgdat_init_all_done_comp);
1575
1576        /* Reinit limits that are based on free pages after the kernel is up */
1577        files_maxfiles_init();
1578#endif
1579
1580        for_each_populated_zone(zone)
1581                set_zone_contiguous(zone);
1582}
1583
1584#ifdef CONFIG_CMA
1585/* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1586void __init init_cma_reserved_pageblock(struct page *page)
1587{
1588        unsigned i = pageblock_nr_pages;
1589        struct page *p = page;
1590
1591        do {
1592                __ClearPageReserved(p);
1593                set_page_count(p, 0);
1594        } while (++p, --i);
1595
1596        set_pageblock_migratetype(page, MIGRATE_CMA);
1597
1598        if (pageblock_order >= MAX_ORDER) {
1599                i = pageblock_nr_pages;
1600                p = page;
1601                do {
1602                        set_page_refcounted(p);
1603                        __free_pages(p, MAX_ORDER - 1);
1604                        p += MAX_ORDER_NR_PAGES;
1605                } while (i -= MAX_ORDER_NR_PAGES);
1606        } else {
1607                set_page_refcounted(page);
1608                __free_pages(page, pageblock_order);
1609        }
1610
1611        adjust_managed_page_count(page, pageblock_nr_pages);
1612}
1613#endif
1614
1615/*
1616 * The order of subdivision here is critical for the IO subsystem.
1617 * Please do not alter this order without good reasons and regression
1618 * testing. Specifically, as large blocks of memory are subdivided,
1619 * the order in which smaller blocks are delivered depends on the order
1620 * they're subdivided in this function. This is the primary factor
1621 * influencing the order in which pages are delivered to the IO
1622 * subsystem according to empirical testing, and this is also justified
1623 * by considering the behavior of a buddy system containing a single
1624 * large block of memory acted on by a series of small allocations.
1625 * This behavior is a critical factor in sglist merging's success.
1626 *
1627 * -- nyc
1628 */
1629static inline void expand(struct zone *zone, struct page *page,
1630        int low, int high, struct free_area *area,
1631        int migratetype)
1632{
1633        unsigned long size = 1 << high;
1634
1635        while (high > low) {
1636                area--;
1637                high--;
1638                size >>= 1;
1639                VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
1640
1641                /*
1642                 * Mark as guard pages (or page), that will allow to
1643                 * merge back to allocator when buddy will be freed.
1644                 * Corresponding page table entries will not be touched,
1645                 * pages will stay not present in virtual address space
1646                 */
1647                if (set_page_guard(zone, &page[size], high, migratetype))
1648                        continue;
1649
1650                list_add(&page[size].lru, &area->free_list[migratetype]);
1651                area->nr_free++;
1652                set_page_order(&page[size], high);
1653        }
1654}
1655
1656static void check_new_page_bad(struct page *page)
1657{
1658        const char *bad_reason = NULL;
1659        unsigned long bad_flags = 0;
1660
1661        if (unlikely(atomic_read(&page->_mapcount) != -1))
1662                bad_reason = "nonzero mapcount";
1663        if (unlikely(page->mapping != NULL))
1664                bad_reason = "non-NULL mapping";
1665        if (unlikely(page_ref_count(page) != 0))
1666                bad_reason = "nonzero _count";
1667        if (unlikely(page->flags & __PG_HWPOISON)) {
1668                bad_reason = "HWPoisoned (hardware-corrupted)";
1669                bad_flags = __PG_HWPOISON;
1670                /* Don't complain about hwpoisoned pages */
1671                page_mapcount_reset(page); /* remove PageBuddy */
1672                return;
1673        }
1674        if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1675                bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1676                bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1677        }
1678#ifdef CONFIG_MEMCG
1679        if (unlikely(page->mem_cgroup))
1680                bad_reason = "page still charged to cgroup";
1681#endif
1682        bad_page(page, bad_reason, bad_flags);
1683}
1684
1685/*
1686 * This page is about to be returned from the page allocator
1687 */
1688static inline int check_new_page(struct page *page)
1689{
1690        if (likely(page_expected_state(page,
1691                                PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON)))
1692                return 0;
1693
1694        check_new_page_bad(page);
1695        return 1;
1696}
1697
1698static inline bool free_pages_prezeroed(bool poisoned)
1699{
1700        return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) &&
1701                page_poisoning_enabled() && poisoned;
1702}
1703
1704#ifdef CONFIG_DEBUG_VM
1705static bool check_pcp_refill(struct page *page)
1706{
1707        return false;
1708}
1709
1710static bool check_new_pcp(struct page *page)
1711{
1712        return check_new_page(page);
1713}
1714#else
1715static bool check_pcp_refill(struct page *page)
1716{
1717        return check_new_page(page);
1718}
1719static bool check_new_pcp(struct page *page)
1720{
1721        return false;
1722}
1723#endif /* CONFIG_DEBUG_VM */
1724
1725static bool check_new_pages(struct page *page, unsigned int order)
1726{
1727        int i;
1728        for (i = 0; i < (1 << order); i++) {
1729                struct page *p = page + i;
1730
1731                if (unlikely(check_new_page(p)))
1732                        return true;
1733        }
1734
1735        return false;
1736}
1737
1738inline void post_alloc_hook(struct page *page, unsigned int order,
1739                                gfp_t gfp_flags)
1740{
1741        set_page_private(page, 0);
1742        set_page_refcounted(page);
1743
1744        arch_alloc_page(page, order);
1745        kernel_map_pages(page, 1 << order, 1);
1746        kernel_poison_pages(page, 1 << order, 1);
1747        kasan_alloc_pages(page, order);
1748        set_page_owner(page, order, gfp_flags);
1749}
1750
1751static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
1752                                                        unsigned int alloc_flags)
1753{
1754        int i;
1755        bool poisoned = true;
1756
1757        for (i = 0; i < (1 << order); i++) {
1758                struct page *p = page + i;
1759                if (poisoned)
1760                        poisoned &= page_is_poisoned(p);
1761        }
1762
1763        post_alloc_hook(page, order, gfp_flags);
1764
1765        if (!free_pages_prezeroed(poisoned) && (gfp_flags & __GFP_ZERO))
1766                for (i = 0; i < (1 << order); i++)
1767                        clear_highpage(page + i);
1768
1769        if (order && (gfp_flags & __GFP_COMP))
1770                prep_compound_page(page, order);
1771
1772        /*
1773         * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1774         * allocate the page. The expectation is that the caller is taking
1775         * steps that will free more memory. The caller should avoid the page
1776         * being used for !PFMEMALLOC purposes.
1777         */
1778        if (alloc_flags & ALLOC_NO_WATERMARKS)
1779                set_page_pfmemalloc(page);
1780        else
1781                clear_page_pfmemalloc(page);
1782}
1783
1784/*
1785 * Go through the free lists for the given migratetype and remove
1786 * the smallest available page from the freelists
1787 */
1788static inline
1789struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
1790                                                int migratetype)
1791{
1792        unsigned int current_order;
1793        struct free_area *area;
1794        struct page *page;
1795
1796        /* Find a page of the appropriate size in the preferred list */
1797        for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1798                area = &(zone->free_area[current_order]);
1799                page = list_first_entry_or_null(&area->free_list[migratetype],
1800                                                        struct page, lru);
1801                if (!page)
1802                        continue;
1803                list_del(&page->lru);
1804                rmv_page_order(page);
1805                area->nr_free--;
1806                expand(zone, page, order, current_order, area, migratetype);
1807                set_pcppage_migratetype(page, migratetype);
1808                return page;
1809        }
1810
1811        return NULL;
1812}
1813
1814
1815/*
1816 * This array describes the order lists are fallen back to when
1817 * the free lists for the desirable migrate type are depleted
1818 */
1819static int fallbacks[MIGRATE_TYPES][4] = {
1820        [MIGRATE_UNMOVABLE]   = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE,   MIGRATE_TYPES },
1821        [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE,   MIGRATE_MOVABLE,   MIGRATE_TYPES },
1822        [MIGRATE_MOVABLE]     = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
1823#ifdef CONFIG_CMA
1824        [MIGRATE_CMA]         = { MIGRATE_TYPES }, /* Never used */
1825#endif
1826#ifdef CONFIG_MEMORY_ISOLATION
1827        [MIGRATE_ISOLATE]     = { MIGRATE_TYPES }, /* Never used */
1828#endif
1829};
1830
1831#ifdef CONFIG_CMA
1832static struct page *__rmqueue_cma_fallback(struct zone *zone,
1833                                        unsigned int order)
1834{
1835        return __rmqueue_smallest(zone, order, MIGRATE_CMA);
1836}
1837#else
1838static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1839                                        unsigned int order) { return NULL; }
1840#endif
1841
1842/*
1843 * Move the free pages in a range to the free lists of the requested type.
1844 * Note that start_page and end_pages are not aligned on a pageblock
1845 * boundary. If alignment is required, use move_freepages_block()
1846 */
1847int move_freepages(struct zone *zone,
1848                          struct page *start_page, struct page *end_page,
1849                          int migratetype)
1850{
1851        struct page *page;
1852        unsigned int order;
1853        int pages_moved = 0;
1854
1855#ifndef CONFIG_HOLES_IN_ZONE
1856        /*
1857         * page_zone is not safe to call in this context when
1858         * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1859         * anyway as we check zone boundaries in move_freepages_block().
1860         * Remove at a later date when no bug reports exist related to
1861         * grouping pages by mobility
1862         */
1863        VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
1864#endif
1865
1866        for (page = start_page; page <= end_page;) {
1867                /* Make sure we are not inadvertently changing nodes */
1868                VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
1869
1870                if (!pfn_valid_within(page_to_pfn(page))) {
1871                        page++;
1872                        continue;
1873                }
1874
1875                if (!PageBuddy(page)) {
1876                        page++;
1877                        continue;
1878                }
1879
1880                order = page_order(page);
1881                list_move(&page->lru,
1882                          &zone->free_area[order].free_list[migratetype]);
1883                page += 1 << order;
1884                pages_moved += 1 << order;
1885        }
1886
1887        return pages_moved;
1888}
1889
1890int move_freepages_block(struct zone *zone, struct page *page,
1891                                int migratetype)
1892{
1893        unsigned long start_pfn, end_pfn;
1894        struct page *start_page, *end_page;
1895
1896        start_pfn = page_to_pfn(page);
1897        start_pfn = start_pfn & ~(pageblock_nr_pages-1);
1898        start_page = pfn_to_page(start_pfn);
1899        end_page = start_page + pageblock_nr_pages - 1;
1900        end_pfn = start_pfn + pageblock_nr_pages - 1;
1901
1902        /* Do not cross zone boundaries */
1903        if (!zone_spans_pfn(zone, start_pfn))
1904                start_page = page;
1905        if (!zone_spans_pfn(zone, end_pfn))
1906                return 0;
1907
1908        return move_freepages(zone, start_page, end_page, migratetype);
1909}
1910
1911static void change_pageblock_range(struct page *pageblock_page,
1912                                        int start_order, int migratetype)
1913{
1914        int nr_pageblocks = 1 << (start_order - pageblock_order);
1915
1916        while (nr_pageblocks--) {
1917                set_pageblock_migratetype(pageblock_page, migratetype);
1918                pageblock_page += pageblock_nr_pages;
1919        }
1920}
1921
1922/*
1923 * When we are falling back to another migratetype during allocation, try to
1924 * steal extra free pages from the same pageblocks to satisfy further
1925 * allocations, instead of polluting multiple pageblocks.
1926 *
1927 * If we are stealing a relatively large buddy page, it is likely there will
1928 * be more free pages in the pageblock, so try to steal them all. For
1929 * reclaimable and unmovable allocations, we steal regardless of page size,
1930 * as fragmentation caused by those allocations polluting movable pageblocks
1931 * is worse than movable allocations stealing from unmovable and reclaimable
1932 * pageblocks.
1933 */
1934static bool can_steal_fallback(unsigned int order, int start_mt)
1935{
1936        /*
1937         * Leaving this order check is intended, although there is
1938         * relaxed order check in next check. The reason is that
1939         * we can actually steal whole pageblock if this condition met,
1940         * but, below check doesn't guarantee it and that is just heuristic
1941         * so could be changed anytime.
1942         */
1943        if (order >= pageblock_order)
1944                return true;
1945
1946        if (order >= pageblock_order / 2 ||
1947                start_mt == MIGRATE_RECLAIMABLE ||
1948                start_mt == MIGRATE_UNMOVABLE ||
1949                page_group_by_mobility_disabled)
1950                return true;
1951
1952        return false;
1953}
1954
1955/*
1956 * This function implements actual steal behaviour. If order is large enough,
1957 * we can steal whole pageblock. If not, we first move freepages in this
1958 * pageblock and check whether half of pages are moved or not. If half of
1959 * pages are moved, we can change migratetype of pageblock and permanently
1960 * use it's pages as requested migratetype in the future.
1961 */
1962static void steal_suitable_fallback(struct zone *zone, struct page *page,
1963                                                          int start_type)
1964{
1965        unsigned int current_order = page_order(page);
1966        int pages;
1967
1968        /* Take ownership for orders >= pageblock_order */
1969        if (current_order >= pageblock_order) {
1970                change_pageblock_range(page, current_order, start_type);
1971                return;
1972        }
1973
1974        pages = move_freepages_block(zone, page, start_type);
1975
1976        /* Claim the whole block if over half of it is free */
1977        if (pages >= (1 << (pageblock_order-1)) ||
1978                        page_group_by_mobility_disabled)
1979                set_pageblock_migratetype(page, start_type);
1980}
1981
1982/*
1983 * Check whether there is a suitable fallback freepage with requested order.
1984 * If only_stealable is true, this function returns fallback_mt only if
1985 * we can steal other freepages all together. This would help to reduce
1986 * fragmentation due to mixed migratetype pages in one pageblock.
1987 */
1988int find_suitable_fallback(struct free_area *area, unsigned int order,
1989                        int migratetype, bool only_stealable, bool *can_steal)
1990{
1991        int i;
1992        int fallback_mt;
1993
1994        if (area->nr_free == 0)
1995                return -1;
1996
1997        *can_steal = false;
1998        for (i = 0;; i++) {
1999                fallback_mt = fallbacks[migratetype][i];
2000                if (fallback_mt == MIGRATE_TYPES)
2001                        break;
2002
2003                if (list_empty(&area->free_list[fallback_mt]))
2004                        continue;
2005
2006                if (can_steal_fallback(order, migratetype))
2007                        *can_steal = true;
2008
2009                if (!only_stealable)
2010                        return fallback_mt;
2011
2012                if (*can_steal)
2013                        return fallback_mt;
2014        }
2015
2016        return -1;
2017}
2018
2019/*
2020 * Reserve a pageblock for exclusive use of high-order atomic allocations if
2021 * there are no empty page blocks that contain a page with a suitable order
2022 */
2023static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
2024                                unsigned int alloc_order)
2025{
2026        int mt;
2027        unsigned long max_managed, flags;
2028
2029        /*
2030         * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
2031         * Check is race-prone but harmless.
2032         */
2033        max_managed = (zone->managed_pages / 100) + pageblock_nr_pages;
2034        if (zone->nr_reserved_highatomic >= max_managed)
2035                return;
2036
2037        spin_lock_irqsave(&zone->lock, flags);
2038
2039        /* Recheck the nr_reserved_highatomic limit under the lock */
2040        if (zone->nr_reserved_highatomic >= max_managed)
2041                goto out_unlock;
2042
2043        /* Yoink! */
2044        mt = get_pageblock_migratetype(page);
2045        if (mt != MIGRATE_HIGHATOMIC &&
2046                        !is_migrate_isolate(mt) && !is_migrate_cma(mt)) {
2047                zone->nr_reserved_highatomic += pageblock_nr_pages;
2048                set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
2049                move_freepages_block(zone, page, MIGRATE_HIGHATOMIC);
2050        }
2051
2052out_unlock:
2053        spin_unlock_irqrestore(&zone->lock, flags);
2054}
2055
2056/*
2057 * Used when an allocation is about to fail under memory pressure. This
2058 * potentially hurts the reliability of high-order allocations when under
2059 * intense memory pressure but failed atomic allocations should be easier
2060 * to recover from than an OOM.
2061 */
2062static void unreserve_highatomic_pageblock(const struct alloc_context *ac)
2063{
2064        struct zonelist *zonelist = ac->zonelist;
2065        unsigned long flags;
2066        struct zoneref *z;
2067        struct zone *zone;
2068        struct page *page;
2069        int order;
2070
2071        for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2072                                                                ac->nodemask) {
2073                /* Preserve at least one pageblock */
2074                if (zone->nr_reserved_highatomic <= pageblock_nr_pages)
2075                        continue;
2076
2077                spin_lock_irqsave(&zone->lock, flags);
2078                for (order = 0; order < MAX_ORDER; order++) {
2079                        struct free_area *area = &(zone->free_area[order]);
2080
2081                        page = list_first_entry_or_null(
2082                                        &area->free_list[MIGRATE_HIGHATOMIC],
2083                                        struct page, lru);
2084                        if (!page)
2085                                continue;
2086
2087                        /*
2088                         * It should never happen but changes to locking could
2089                         * inadvertently allow a per-cpu drain to add pages
2090                         * to MIGRATE_HIGHATOMIC while unreserving so be safe
2091                         * and watch for underflows.
2092                         */
2093                        zone->nr_reserved_highatomic -= min(pageblock_nr_pages,
2094                                zone->nr_reserved_highatomic);
2095
2096                        /*
2097                         * Convert to ac->migratetype and avoid the normal
2098                         * pageblock stealing heuristics. Minimally, the caller
2099                         * is doing the work and needs the pages. More
2100                         * importantly, if the block was always converted to
2101                         * MIGRATE_UNMOVABLE or another type then the number
2102                         * of pageblocks that cannot be completely freed
2103                         * may increase.
2104                         */
2105                        set_pageblock_migratetype(page, ac->migratetype);
2106                        move_freepages_block(zone, page, ac->migratetype);
2107                        spin_unlock_irqrestore(&zone->lock, flags);
2108                        return;
2109                }
2110                spin_unlock_irqrestore(&zone->lock, flags);
2111        }
2112}
2113
2114/* Remove an element from the buddy allocator from the fallback list */
2115static inline struct page *
2116__rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
2117{
2118        struct free_area *area;
2119        unsigned int current_order;
2120        struct page *page;
2121        int fallback_mt;
2122        bool can_steal;
2123
2124        /* Find the largest possible block of pages in the other list */
2125        for (current_order = MAX_ORDER-1;
2126                                current_order >= order && current_order <= MAX_ORDER-1;
2127                                --current_order) {
2128                area = &(zone->free_area[current_order]);
2129                fallback_mt = find_suitable_fallback(area, current_order,
2130                                start_migratetype, false, &can_steal);
2131                if (fallback_mt == -1)
2132                        continue;
2133
2134                page = list_first_entry(&area->free_list[fallback_mt],
2135                                                struct page, lru);
2136                if (can_steal)
2137                        steal_suitable_fallback(zone, page, start_migratetype);
2138
2139                /* Remove the page from the freelists */
2140                area->nr_free--;
2141                list_del(&page->lru);
2142                rmv_page_order(page);
2143
2144                expand(zone, page, order, current_order, area,
2145                                        start_migratetype);
2146                /*
2147                 * The pcppage_migratetype may differ from pageblock's
2148                 * migratetype depending on the decisions in
2149                 * find_suitable_fallback(). This is OK as long as it does not
2150                 * differ for MIGRATE_CMA pageblocks. Those can be used as
2151                 * fallback only via special __rmqueue_cma_fallback() function
2152                 */
2153                set_pcppage_migratetype(page, start_migratetype);
2154
2155                trace_mm_page_alloc_extfrag(page, order, current_order,
2156                        start_migratetype, fallback_mt);
2157
2158                return page;
2159        }
2160
2161        return NULL;
2162}
2163
2164/*
2165 * Do the hard work of removing an element from the buddy allocator.
2166 * Call me with the zone->lock already held.
2167 */
2168static struct page *__rmqueue(struct zone *zone, unsigned int order,
2169                                int migratetype)
2170{
2171        struct page *page;
2172
2173        page = __rmqueue_smallest(zone, order, migratetype);
2174        if (unlikely(!page)) {
2175                if (migratetype == MIGRATE_MOVABLE)
2176                        page = __rmqueue_cma_fallback(zone, order);
2177
2178                if (!page)
2179                        page = __rmqueue_fallback(zone, order, migratetype);
2180        }
2181
2182        trace_mm_page_alloc_zone_locked(page, order, migratetype);
2183        return page;
2184}
2185
2186/*
2187 * Obtain a specified number of elements from the buddy allocator, all under
2188 * a single hold of the lock, for efficiency.  Add them to the supplied list.
2189 * Returns the number of new pages which were placed at *list.
2190 */
2191static int rmqueue_bulk(struct zone *zone, unsigned int order,
2192                        unsigned long count, struct list_head *list,
2193                        int migratetype, bool cold)
2194{
2195        int i;
2196
2197        spin_lock(&zone->lock);
2198        for (i = 0; i < count; ++i) {
2199                struct page *page = __rmqueue(zone, order, migratetype);
2200                if (unlikely(page == NULL))
2201                        break;
2202
2203                if (unlikely(check_pcp_refill(page)))
2204                        continue;
2205
2206                /*
2207                 * Split buddy pages returned by expand() are received here
2208                 * in physical page order. The page is added to the callers and
2209                 * list and the list head then moves forward. From the callers
2210                 * perspective, the linked list is ordered by page number in
2211                 * some conditions. This is useful for IO devices that can
2212                 * merge IO requests if the physical pages are ordered
2213                 * properly.
2214                 */
2215                if (likely(!cold))
2216                        list_add(&page->lru, list);
2217                else
2218                        list_add_tail(&page->lru, list);
2219                list = &page->lru;
2220                if (is_migrate_cma(get_pcppage_migratetype(page)))
2221                        __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
2222                                              -(1 << order));
2223        }
2224        __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
2225        spin_unlock(&zone->lock);
2226        return i;
2227}
2228
2229#ifdef CONFIG_NUMA
2230/*
2231 * Called from the vmstat counter updater to drain pagesets of this
2232 * currently executing processor on remote nodes after they have
2233 * expired.
2234 *
2235 * Note that this function must be called with the thread pinned to
2236 * a single processor.
2237 */
2238void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
2239{
2240        unsigned long flags;
2241        int to_drain, batch;
2242
2243        local_irq_save(flags);
2244        batch = READ_ONCE(pcp->batch);
2245        to_drain = min(pcp->count, batch);
2246        if (to_drain > 0) {
2247                free_pcppages_bulk(zone, to_drain, pcp);
2248                pcp->count -= to_drain;
2249        }
2250        local_irq_restore(flags);
2251}
2252#endif
2253
2254/*
2255 * Drain pcplists of the indicated processor and zone.
2256 *
2257 * The processor must either be the current processor and the
2258 * thread pinned to the current processor or a processor that
2259 * is not online.
2260 */
2261static void drain_pages_zone(unsigned int cpu, struct zone *zone)
2262{
2263        unsigned long flags;
2264        struct per_cpu_pageset *pset;
2265        struct per_cpu_pages *pcp;
2266
2267        local_irq_save(flags);
2268        pset = per_cpu_ptr(zone->pageset, cpu);
2269
2270        pcp = &pset->pcp;
2271        if (pcp->count) {
2272                free_pcppages_bulk(zone, pcp->count, pcp);
2273                pcp->count = 0;
2274        }
2275        local_irq_restore(flags);
2276}
2277
2278/*
2279 * Drain pcplists of all zones on the indicated processor.
2280 *
2281 * The processor must either be the current processor and the
2282 * thread pinned to the current processor or a processor that
2283 * is not online.
2284 */
2285static void drain_pages(unsigned int cpu)
2286{
2287        struct zone *zone;
2288
2289        for_each_populated_zone(zone) {
2290                drain_pages_zone(cpu, zone);
2291        }
2292}
2293
2294/*
2295 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
2296 *
2297 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2298 * the single zone's pages.
2299 */
2300void drain_local_pages(struct zone *zone)
2301{
2302        int cpu = smp_processor_id();
2303
2304        if (zone)
2305                drain_pages_zone(cpu, zone);
2306        else
2307                drain_pages(cpu);
2308}
2309
2310/*
2311 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2312 *
2313 * When zone parameter is non-NULL, spill just the single zone's pages.
2314 *
2315 * Note that this code is protected against sending an IPI to an offline
2316 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
2317 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
2318 * nothing keeps CPUs from showing up after we populated the cpumask and
2319 * before the call to on_each_cpu_mask().
2320 */
2321void drain_all_pages(struct zone *zone)
2322{
2323        int cpu;
2324
2325        /*
2326         * Allocate in the BSS so we wont require allocation in
2327         * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2328         */
2329        static cpumask_t cpus_with_pcps;
2330
2331        /*
2332         * We don't care about racing with CPU hotplug event
2333         * as offline notification will cause the notified
2334         * cpu to drain that CPU pcps and on_each_cpu_mask
2335         * disables preemption as part of its processing
2336         */
2337        for_each_online_cpu(cpu) {
2338                struct per_cpu_pageset *pcp;
2339                struct zone *z;
2340                bool has_pcps = false;
2341
2342                if (zone) {
2343                        pcp = per_cpu_ptr(zone->pageset, cpu);
2344                        if (pcp->pcp.count)
2345                                has_pcps = true;
2346                } else {
2347                        for_each_populated_zone(z) {
2348                                pcp = per_cpu_ptr(z->pageset, cpu);
2349                                if (pcp->pcp.count) {
2350                                        has_pcps = true;
2351                                        break;
2352                                }
2353                        }
2354                }
2355
2356                if (has_pcps)
2357                        cpumask_set_cpu(cpu, &cpus_with_pcps);
2358                else
2359                        cpumask_clear_cpu(cpu, &cpus_with_pcps);
2360        }
2361        on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages,
2362                                                                zone, 1);
2363}
2364
2365#ifdef CONFIG_HIBERNATION
2366
2367void mark_free_pages(struct zone *zone)
2368{
2369        unsigned long pfn, max_zone_pfn;
2370        unsigned long flags;
2371        unsigned int order, t;
2372        struct page *page;
2373
2374        if (zone_is_empty(zone))
2375                return;
2376
2377        spin_lock_irqsave(&zone->lock, flags);
2378
2379        max_zone_pfn = zone_end_pfn(zone);
2380        for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
2381                if (pfn_valid(pfn)) {
2382                        page = pfn_to_page(pfn);
2383
2384                        if (page_zone(page) != zone)
2385                                continue;
2386
2387                        if (!swsusp_page_is_forbidden(page))
2388                                swsusp_unset_page_free(page);
2389                }
2390
2391        for_each_migratetype_order(order, t) {
2392                list_for_each_entry(page,
2393                                &zone->free_area[order].free_list[t], lru) {
2394                        unsigned long i;
2395
2396                        pfn = page_to_pfn(page);
2397                        for (i = 0; i < (1UL << order); i++)
2398                                swsusp_set_page_free(pfn_to_page(pfn + i));
2399                }
2400        }
2401        spin_unlock_irqrestore(&zone->lock, flags);
2402}
2403#endif /* CONFIG_PM */
2404
2405/*
2406 * Free a 0-order page
2407 * cold == true ? free a cold page : free a hot page
2408 */
2409void free_hot_cold_page(struct page *page, bool cold)
2410{
2411        struct zone *zone = page_zone(page);
2412        struct per_cpu_pages *pcp;
2413        unsigned long flags;
2414        unsigned long pfn = page_to_pfn(page);
2415        int migratetype;
2416
2417        if (!free_pcp_prepare(page))
2418                return;
2419
2420        migratetype = get_pfnblock_migratetype(page, pfn);
2421        set_pcppage_migratetype(page, migratetype);
2422        local_irq_save(flags);
2423        __count_vm_event(PGFREE);
2424
2425        /*
2426         * We only track unmovable, reclaimable and movable on pcp lists.
2427         * Free ISOLATE pages back to the allocator because they are being
2428         * offlined but treat RESERVE as movable pages so we can get those
2429         * areas back if necessary. Otherwise, we may have to free
2430         * excessively into the page allocator
2431         */
2432        if (migratetype >= MIGRATE_PCPTYPES) {
2433                if (unlikely(is_migrate_isolate(migratetype))) {
2434                        free_one_page(zone, page, pfn, 0, migratetype);
2435                        goto out;
2436                }
2437                migratetype = MIGRATE_MOVABLE;
2438        }
2439
2440        pcp = &this_cpu_ptr(zone->pageset)->pcp;
2441        if (!cold)
2442                list_add(&page->lru, &pcp->lists[migratetype]);
2443        else
2444                list_add_tail(&page->lru, &pcp->lists[migratetype]);
2445        pcp->count++;
2446        if (pcp->count >= pcp->high) {
2447                unsigned long batch = READ_ONCE(pcp->batch);
2448                free_pcppages_bulk(zone, batch, pcp);
2449                pcp->count -= batch;
2450        }
2451
2452out:
2453        local_irq_restore(flags);
2454}
2455
2456/*
2457 * Free a list of 0-order pages
2458 */
2459void free_hot_cold_page_list(struct list_head *list, bool cold)
2460{
2461        struct page *page, *next;
2462
2463        list_for_each_entry_safe(page, next, list, lru) {
2464                trace_mm_page_free_batched(page, cold);
2465                free_hot_cold_page(page, cold);
2466        }
2467}
2468
2469/*
2470 * split_page takes a non-compound higher-order page, and splits it into
2471 * n (1<<order) sub-pages: page[0..n]
2472 * Each sub-page must be freed individually.
2473 *
2474 * Note: this is probably too low level an operation for use in drivers.
2475 * Please consult with lkml before using this in your driver.
2476 */
2477void split_page(struct page *page, unsigned int order)
2478{
2479        int i;
2480
2481        VM_BUG_ON_PAGE(PageCompound(page), page);
2482        VM_BUG_ON_PAGE(!page_count(page), page);
2483
2484#ifdef CONFIG_KMEMCHECK
2485        /*
2486         * Split shadow pages too, because free(page[0]) would
2487         * otherwise free the whole shadow.
2488         */
2489        if (kmemcheck_page_is_tracked(page))
2490                split_page(virt_to_page(page[0].shadow), order);
2491#endif
2492
2493        for (i = 1; i < (1 << order); i++)
2494                set_page_refcounted(page + i);
2495        split_page_owner(page, order);
2496}
2497EXPORT_SYMBOL_GPL(split_page);
2498
2499int __isolate_free_page(struct page *page, unsigned int order)
2500{
2501        unsigned long watermark;
2502        struct zone *zone;
2503        int mt;
2504
2505        BUG_ON(!PageBuddy(page));
2506
2507        zone = page_zone(page);
2508        mt = get_pageblock_migratetype(page);
2509
2510        if (!is_migrate_isolate(mt)) {
2511                /*
2512                 * Obey watermarks as if the page was being allocated. We can
2513                 * emulate a high-order watermark check with a raised order-0
2514                 * watermark, because we already know our high-order page
2515                 * exists.
2516                 */
2517                watermark = min_wmark_pages(zone) + (1UL << order);
2518                if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA))
2519                        return 0;
2520
2521                __mod_zone_freepage_state(zone, -(1UL << order), mt);
2522        }
2523
2524        /* Remove page from free list */
2525        list_del(&page->lru);
2526        zone->free_area[order].nr_free--;
2527        rmv_page_order(page);
2528
2529        /*
2530         * Set the pageblock if the isolated page is at least half of a
2531         * pageblock
2532         */
2533        if (order >= pageblock_order - 1) {
2534                struct page *endpage = page + (1 << order) - 1;
2535                for (; page < endpage; page += pageblock_nr_pages) {
2536                        int mt = get_pageblock_migratetype(page);
2537                        if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
2538                                set_pageblock_migratetype(page,
2539                                                          MIGRATE_MOVABLE);
2540                }
2541        }
2542
2543
2544        return 1UL << order;
2545}
2546
2547/*
2548 * Update NUMA hit/miss statistics
2549 *
2550 * Must be called with interrupts disabled.
2551 *
2552 * When __GFP_OTHER_NODE is set assume the node of the preferred
2553 * zone is the local node. This is useful for daemons who allocate
2554 * memory on behalf of other processes.
2555 */
2556static inline void zone_statistics(struct zone *preferred_zone, struct zone *z,
2557                                                                gfp_t flags)
2558{
2559#ifdef CONFIG_NUMA
2560        int local_nid = numa_node_id();
2561        enum zone_stat_item local_stat = NUMA_LOCAL;
2562
2563        if (unlikely(flags & __GFP_OTHER_NODE)) {
2564                local_stat = NUMA_OTHER;
2565                local_nid = preferred_zone->node;
2566        }
2567
2568        if (z->node == local_nid) {
2569                __inc_zone_state(z, NUMA_HIT);
2570                __inc_zone_state(z, local_stat);
2571        } else {
2572                __inc_zone_state(z, NUMA_MISS);
2573                __inc_zone_state(preferred_zone, NUMA_FOREIGN);
2574        }
2575#endif
2576}
2577
2578/*
2579 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2580 */
2581static inline
2582struct page *buffered_rmqueue(struct zone *preferred_zone,
2583                        struct zone *zone, unsigned int order,
2584                        gfp_t gfp_flags, unsigned int alloc_flags,
2585                        int migratetype)
2586{
2587        unsigned long flags;
2588        struct page *page;
2589        bool cold = ((gfp_flags & __GFP_COLD) != 0);
2590
2591        if (likely(order == 0)) {
2592                struct per_cpu_pages *pcp;
2593                struct list_head *list;
2594
2595                local_irq_save(flags);
2596                do {
2597                        pcp = &this_cpu_ptr(zone->pageset)->pcp;
2598                        list = &pcp->lists[migratetype];
2599                        if (list_empty(list)) {
2600                                pcp->count += rmqueue_bulk(zone, 0,
2601                                                pcp->batch, list,
2602                                                migratetype, cold);
2603                                if (unlikely(list_empty(list)))
2604                                        goto failed;
2605                        }
2606
2607                        if (cold)
2608                                page = list_last_entry(list, struct page, lru);
2609                        else
2610                                page = list_first_entry(list, struct page, lru);
2611
2612                        list_del(&page->lru);
2613                        pcp->count--;
2614
2615                } while (check_new_pcp(page));
2616        } else {
2617                /*
2618                 * We most definitely don't want callers attempting to
2619                 * allocate greater than order-1 page units with __GFP_NOFAIL.
2620                 */
2621                WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
2622                spin_lock_irqsave(&zone->lock, flags);
2623
2624                do {
2625                        page = NULL;
2626                        if (alloc_flags & ALLOC_HARDER) {
2627                                page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
2628                                if (page)
2629                                        trace_mm_page_alloc_zone_locked(page, order, migratetype);
2630                        }
2631                        if (!page)
2632                                page = __rmqueue(zone, order, migratetype);
2633                } while (page && check_new_pages(page, order));
2634                spin_unlock(&zone->lock);
2635                if (!page)
2636                        goto failed;
2637                __mod_zone_freepage_state(zone, -(1 << order),
2638                                          get_pcppage_migratetype(page));
2639        }
2640
2641        __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
2642        zone_statistics(preferred_zone, zone, gfp_flags);
2643        local_irq_restore(flags);
2644
2645        VM_BUG_ON_PAGE(bad_range(zone, page), page);
2646        return page;
2647
2648failed:
2649        local_irq_restore(flags);
2650        return NULL;
2651}
2652
2653#ifdef CONFIG_FAIL_PAGE_ALLOC
2654
2655static struct {
2656        struct fault_attr attr;
2657
2658        bool ignore_gfp_highmem;
2659        bool ignore_gfp_reclaim;
2660        u32 min_order;
2661} fail_page_alloc = {
2662        .attr = FAULT_ATTR_INITIALIZER,
2663        .ignore_gfp_reclaim = true,
2664        .ignore_gfp_highmem = true,
2665        .min_order = 1,
2666};
2667
2668static int __init setup_fail_page_alloc(char *str)
2669{
2670        return setup_fault_attr(&fail_page_alloc.attr, str);
2671}
2672__setup("fail_page_alloc=", setup_fail_page_alloc);
2673
2674static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
2675{
2676        if (order < fail_page_alloc.min_order)
2677                return false;
2678        if (gfp_mask & __GFP_NOFAIL)
2679                return false;
2680        if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
2681                return false;
2682        if (fail_page_alloc.ignore_gfp_reclaim &&
2683                        (gfp_mask & __GFP_DIRECT_RECLAIM))
2684                return false;
2685
2686        return should_fail(&fail_page_alloc.attr, 1 << order);
2687}
2688
2689#ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2690
2691static int __init fail_page_alloc_debugfs(void)
2692{
2693        umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
2694        struct dentry *dir;
2695
2696        dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
2697                                        &fail_page_alloc.attr);
2698        if (IS_ERR(dir))
2699                return PTR_ERR(dir);
2700
2701        if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
2702                                &fail_page_alloc.ignore_gfp_reclaim))
2703                goto fail;
2704        if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
2705                                &fail_page_alloc.ignore_gfp_highmem))
2706                goto fail;
2707        if (!debugfs_create_u32("min-order", mode, dir,
2708                                &fail_page_alloc.min_order))
2709                goto fail;
2710
2711        return 0;
2712fail:
2713        debugfs_remove_recursive(dir);
2714
2715        return -ENOMEM;
2716}
2717
2718late_initcall(fail_page_alloc_debugfs);
2719
2720#endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2721
2722#else /* CONFIG_FAIL_PAGE_ALLOC */
2723
2724static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
2725{
2726        return false;
2727}
2728
2729#endif /* CONFIG_FAIL_PAGE_ALLOC */
2730
2731/*
2732 * Return true if free base pages are above 'mark'. For high-order checks it
2733 * will return true of the order-0 watermark is reached and there is at least
2734 * one free page of a suitable size. Checking now avoids taking the zone lock
2735 * to check in the allocation paths if no pages are free.
2736 */
2737bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
2738                         int classzone_idx, unsigned int alloc_flags,
2739                         long free_pages)
2740{
2741        long min = mark;
2742        int o;
2743        const bool alloc_harder = (alloc_flags & ALLOC_HARDER);
2744
2745        /* free_pages may go negative - that's OK */
2746        free_pages -= (1 << order) - 1;
2747
2748        if (alloc_flags & ALLOC_HIGH)
2749                min -= min / 2;
2750
2751        /*
2752         * If the caller does not have rights to ALLOC_HARDER then subtract
2753         * the high-atomic reserves. This will over-estimate the size of the
2754         * atomic reserve but it avoids a search.
2755         */
2756        if (likely(!alloc_harder))
2757                free_pages -= z->nr_reserved_highatomic;
2758        else
2759                min -= min / 4;
2760
2761#ifdef CONFIG_CMA
2762        /* If allocation can't use CMA areas don't use free CMA pages */
2763        if (!(alloc_flags & ALLOC_CMA))
2764                free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES);
2765#endif
2766
2767        /*
2768         * Check watermarks for an order-0 allocation request. If these
2769         * are not met, then a high-order request also cannot go ahead
2770         * even if a suitable page happened to be free.
2771         */
2772        if (free_pages <= min + z->lowmem_reserve[classzone_idx])
2773                return false;
2774
2775        /* If this is an order-0 request then the watermark is fine */
2776        if (!order)
2777                return true;
2778
2779        /* For a high-order request, check at least one suitable page is free */
2780        for (o = order; o < MAX_ORDER; o++) {
2781                struct free_area *area = &z->free_area[o];
2782                int mt;
2783
2784                if (!area->nr_free)
2785                        continue;
2786
2787                if (alloc_harder)
2788                        return true;
2789
2790                for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
2791                        if (!list_empty(&area->free_list[mt]))
2792                                return true;
2793                }
2794
2795#ifdef CONFIG_CMA
2796                if ((alloc_flags & ALLOC_CMA) &&
2797                    !list_empty(&area->free_list[MIGRATE_CMA])) {
2798                        return true;
2799                }
2800#endif
2801        }
2802        return false;
2803}
2804
2805bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
2806                      int classzone_idx, unsigned int alloc_flags)
2807{
2808        return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2809                                        zone_page_state(z, NR_FREE_PAGES));
2810}
2811
2812static inline bool zone_watermark_fast(struct zone *z, unsigned int order,
2813                unsigned long mark, int classzone_idx, unsigned int alloc_flags)
2814{
2815        long free_pages = zone_page_state(z, NR_FREE_PAGES);
2816        long cma_pages = 0;
2817
2818#ifdef CONFIG_CMA
2819        /* If allocation can't use CMA areas don't use free CMA pages */
2820        if (!(alloc_flags & ALLOC_CMA))
2821                cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES);
2822#endif
2823
2824        /*
2825         * Fast check for order-0 only. If this fails then the reserves
2826         * need to be calculated. There is a corner case where the check
2827         * passes but only the high-order atomic reserve are free. If
2828         * the caller is !atomic then it'll uselessly search the free
2829         * list. That corner case is then slower but it is harmless.
2830         */
2831        if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx])
2832                return true;
2833
2834        return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2835                                        free_pages);
2836}
2837
2838bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
2839                        unsigned long mark, int classzone_idx)
2840{
2841        long free_pages = zone_page_state(z, NR_FREE_PAGES);
2842
2843        if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
2844                free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
2845
2846        return __zone_watermark_ok(z, order, mark, classzone_idx, 0,
2847                                                                free_pages);
2848}
2849
2850#ifdef CONFIG_NUMA
2851static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2852{
2853        return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
2854                                RECLAIM_DISTANCE;
2855}
2856#else   /* CONFIG_NUMA */
2857static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2858{
2859        return true;
2860}
2861#endif  /* CONFIG_NUMA */
2862
2863/*
2864 * get_page_from_freelist goes through the zonelist trying to allocate
2865 * a page.
2866 */
2867static struct page *
2868get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
2869                                                const struct alloc_context *ac)
2870{
2871        struct zoneref *z = ac->preferred_zoneref;
2872        struct zone *zone;
2873        struct pglist_data *last_pgdat_dirty_limit = NULL;
2874
2875        /*
2876         * Scan zonelist, looking for a zone with enough free.
2877         * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2878         */
2879        for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
2880                                                                ac->nodemask) {
2881                struct page *page;
2882                unsigned long mark;
2883
2884                if (cpusets_enabled() &&
2885                        (alloc_flags & ALLOC_CPUSET) &&
2886                        !__cpuset_zone_allowed(zone, gfp_mask))
2887                                continue;
2888                /*
2889                 * When allocating a page cache page for writing, we
2890                 * want to get it from a node that is within its dirty
2891                 * limit, such that no single node holds more than its
2892                 * proportional share of globally allowed dirty pages.
2893                 * The dirty limits take into account the node's
2894                 * lowmem reserves and high watermark so that kswapd
2895                 * should be able to balance it without having to
2896                 * write pages from its LRU list.
2897                 *
2898                 * XXX: For now, allow allocations to potentially
2899                 * exceed the per-node dirty limit in the slowpath
2900                 * (spread_dirty_pages unset) before going into reclaim,
2901                 * which is important when on a NUMA setup the allowed
2902                 * nodes are together not big enough to reach the
2903                 * global limit.  The proper fix for these situations
2904                 * will require awareness of nodes in the
2905                 * dirty-throttling and the flusher threads.
2906                 */
2907                if (ac->spread_dirty_pages) {
2908                        if (last_pgdat_dirty_limit == zone->zone_pgdat)
2909                                continue;
2910
2911                        if (!node_dirty_ok(zone->zone_pgdat)) {
2912                                last_pgdat_dirty_limit = zone->zone_pgdat;
2913                                continue;
2914                        }
2915                }
2916
2917                mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2918                if (!zone_watermark_fast(zone, order, mark,
2919                                       ac_classzone_idx(ac), alloc_flags)) {
2920                        int ret;
2921
2922                        /* Checked here to keep the fast path fast */
2923                        BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2924                        if (alloc_flags & ALLOC_NO_WATERMARKS)
2925                                goto try_this_zone;
2926
2927                        if (node_reclaim_mode == 0 ||
2928                            !zone_allows_reclaim(ac->preferred_zoneref->zone, zone))
2929                                continue;
2930
2931                        ret = node_reclaim(zone->zone_pgdat, gfp_mask, order);
2932                        switch (ret) {
2933                        case NODE_RECLAIM_NOSCAN:
2934                                /* did not scan */
2935                                continue;
2936                        case NODE_RECLAIM_FULL:
2937                                /* scanned but unreclaimable */
2938                                continue;
2939                        default:
2940                                /* did we reclaim enough */
2941                                if (zone_watermark_ok(zone, order, mark,
2942                                                ac_classzone_idx(ac), alloc_flags))
2943                                        goto try_this_zone;
2944
2945                                continue;
2946                        }
2947                }
2948
2949try_this_zone:
2950                page = buffered_rmqueue(ac->preferred_zoneref->zone, zone, order,
2951                                gfp_mask, alloc_flags, ac->migratetype);
2952                if (page) {
2953                        prep_new_page(page, order, gfp_mask, alloc_flags);
2954
2955                        /*
2956                         * If this is a high-order atomic allocation then check
2957                         * if the pageblock should be reserved for the future
2958                         */
2959                        if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
2960                                reserve_highatomic_pageblock(page, zone, order);
2961
2962                        return page;
2963                }
2964        }
2965
2966        return NULL;
2967}
2968
2969/*
2970 * Large machines with many possible nodes should not always dump per-node
2971 * meminfo in irq context.
2972 */
2973static inline bool should_suppress_show_mem(void)
2974{
2975        bool ret = false;
2976
2977#if NODES_SHIFT > 8
2978        ret = in_interrupt();
2979#endif
2980        return ret;
2981}
2982
2983static DEFINE_RATELIMIT_STATE(nopage_rs,
2984                DEFAULT_RATELIMIT_INTERVAL,
2985                DEFAULT_RATELIMIT_BURST);
2986
2987void warn_alloc(gfp_t gfp_mask, const char *fmt, ...)
2988{
2989        unsigned int filter = SHOW_MEM_FILTER_NODES;
2990        struct va_format vaf;
2991        va_list args;
2992
2993        if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2994            debug_guardpage_minorder() > 0)
2995                return;
2996
2997        /*
2998         * This documents exceptions given to allocations in certain
2999         * contexts that are allowed to allocate outside current's set
3000         * of allowed nodes.
3001         */
3002        if (!(gfp_mask & __GFP_NOMEMALLOC))
3003                if (test_thread_flag(TIF_MEMDIE) ||
3004                    (current->flags & (PF_MEMALLOC | PF_EXITING)))
3005                        filter &= ~SHOW_MEM_FILTER_NODES;
3006        if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
3007                filter &= ~SHOW_MEM_FILTER_NODES;
3008
3009        pr_warn("%s: ", current->comm);
3010
3011        va_start(args, fmt);
3012        vaf.fmt = fmt;
3013        vaf.va = &args;
3014        pr_cont("%pV", &vaf);
3015        va_end(args);
3016
3017        pr_cont(", mode:%#x(%pGg)\n", gfp_mask, &gfp_mask);
3018
3019        dump_stack();
3020        if (!should_suppress_show_mem())
3021                show_mem(filter);
3022}
3023
3024static inline struct page *
3025__alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
3026        const struct alloc_context *ac, unsigned long *did_some_progress)
3027{
3028        struct oom_control oc = {
3029                .zonelist = ac->zonelist,
3030                .nodemask = ac->nodemask,
3031                .memcg = NULL,
3032                .gfp_mask = gfp_mask,
3033                .order = order,
3034        };
3035        struct page *page;
3036
3037        *did_some_progress = 0;
3038
3039        /*
3040         * Acquire the oom lock.  If that fails, somebody else is
3041         * making progress for us.
3042         */
3043        if (!mutex_trylock(&oom_lock)) {
3044                *did_some_progress = 1;
3045                schedule_timeout_uninterruptible(1);
3046                return NULL;
3047        }
3048
3049        /*
3050         * Go through the zonelist yet one more time, keep very high watermark
3051         * here, this is only to catch a parallel oom killing, we must fail if
3052         * we're still under heavy pressure.
3053         */
3054        page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
3055                                        ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
3056        if (page)
3057                goto out;
3058
3059        if (!(gfp_mask & __GFP_NOFAIL)) {
3060                /* Coredumps can quickly deplete all memory reserves */
3061                if (current->flags & PF_DUMPCORE)
3062                        goto out;
3063                /* The OOM killer will not help higher order allocs */
3064                if (order > PAGE_ALLOC_COSTLY_ORDER)
3065                        goto out;
3066                /* The OOM killer does not needlessly kill tasks for lowmem */
3067                if (ac->high_zoneidx < ZONE_NORMAL)
3068                        goto out;
3069                if (pm_suspended_storage())
3070                        goto out;
3071                /*
3072                 * XXX: GFP_NOFS allocations should rather fail than rely on
3073                 * other request to make a forward progress.
3074                 * We are in an unfortunate situation where out_of_memory cannot
3075                 * do much for this context but let's try it to at least get
3076                 * access to memory reserved if the current task is killed (see
3077                 * out_of_memory). Once filesystems are ready to handle allocation
3078                 * failures more gracefully we should just bail out here.
3079                 */
3080
3081                /* The OOM killer may not free memory on a specific node */
3082                if (gfp_mask & __GFP_THISNODE)
3083                        goto out;
3084        }
3085        /* Exhausted what can be done so it's blamo time */
3086        if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
3087                *did_some_progress = 1;
3088
3089                if (gfp_mask & __GFP_NOFAIL) {
3090                        page = get_page_from_freelist(gfp_mask, order,
3091                                        ALLOC_NO_WATERMARKS|ALLOC_CPUSET, ac);
3092                        /*
3093                         * fallback to ignore cpuset restriction if our nodes
3094                         * are depleted
3095                         */
3096                        if (!page)
3097                                page = get_page_from_freelist(gfp_mask, order,
3098                                        ALLOC_NO_WATERMARKS, ac);
3099                }
3100        }
3101out:
3102        mutex_unlock(&oom_lock);
3103        return page;
3104}
3105
3106/*
3107 * Maximum number of compaction retries wit a progress before OOM
3108 * killer is consider as the only way to move forward.
3109 */
3110#define MAX_COMPACT_RETRIES 16
3111
3112#ifdef CONFIG_COMPACTION
3113/* Try memory compaction for high-order allocations before reclaim */
3114static struct page *
3115__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
3116                unsigned int alloc_flags, const struct alloc_context *ac,
3117                enum compact_priority prio, enum compact_result *compact_result)
3118{
3119        struct page *page;
3120
3121        if (!order)
3122                return NULL;
3123
3124        current->flags |= PF_MEMALLOC;
3125        *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
3126                                                                        prio);
3127        current->flags &= ~PF_MEMALLOC;
3128
3129        if (*compact_result <= COMPACT_INACTIVE)
3130                return NULL;
3131
3132        /*
3133         * At least in one zone compaction wasn't deferred or skipped, so let's
3134         * count a compaction stall
3135         */
3136        count_vm_event(COMPACTSTALL);
3137
3138        page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
3139
3140        if (page) {
3141                struct zone *zone = page_zone(page);
3142
3143                zone->compact_blockskip_flush = false;
3144                compaction_defer_reset(zone, order, true);
3145                count_vm_event(COMPACTSUCCESS);
3146                return page;
3147        }
3148
3149        /*
3150         * It's bad if compaction run occurs and fails. The most likely reason
3151         * is that pages exist, but not enough to satisfy watermarks.
3152         */
3153        count_vm_event(COMPACTFAIL);
3154
3155        cond_resched();
3156
3157        return NULL;
3158}
3159
3160static inline bool
3161should_compact_retry(struct alloc_context *ac, int order, int alloc_flags,
3162                     enum compact_result compact_result,
3163                     enum compact_priority *compact_priority,
3164                     int *compaction_retries)
3165{
3166        int max_retries = MAX_COMPACT_RETRIES;
3167        int min_priority;
3168
3169        if (!order)
3170                return false;
3171
3172        if (compaction_made_progress(compact_result))
3173                (*compaction_retries)++;
3174
3175        /*
3176         * compaction considers all the zone as desperately out of memory
3177         * so it doesn't really make much sense to retry except when the
3178         * failure could be caused by insufficient priority
3179         */
3180        if (compaction_failed(compact_result))
3181                goto check_priority;
3182
3183        /*
3184         * make sure the compaction wasn't deferred or didn't bail out early
3185         * due to locks contention before we declare that we should give up.
3186         * But do not retry if the given zonelist is not suitable for
3187         * compaction.
3188         */
3189        if (compaction_withdrawn(compact_result))
3190                return compaction_zonelist_suitable(ac, order, alloc_flags);
3191
3192        /*
3193         * !costly requests are much more important than __GFP_REPEAT
3194         * costly ones because they are de facto nofail and invoke OOM
3195         * killer to move on while costly can fail and users are ready
3196         * to cope with that. 1/4 retries is rather arbitrary but we
3197         * would need much more detailed feedback from compaction to
3198         * make a better decision.
3199         */
3200        if (order > PAGE_ALLOC_COSTLY_ORDER)
3201                max_retries /= 4;
3202        if (*compaction_retries <= max_retries)
3203                return true;
3204
3205        /*
3206         * Make sure there are attempts at the highest priority if we exhausted
3207         * all retries or failed at the lower priorities.
3208         */
3209check_priority:
3210        min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ?
3211                        MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY;
3212        if (*compact_priority > min_priority) {
3213                (*compact_priority)--;
3214                *compaction_retries = 0;
3215                return true;
3216        }
3217        return false;
3218}
3219#else
3220static inline struct page *
3221__alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
3222                unsigned int alloc_flags, const struct alloc_context *ac,
3223                enum compact_priority prio, enum compact_result *compact_result)
3224{
3225        *compact_result = COMPACT_SKIPPED;
3226        return NULL;
3227}
3228
3229static inline bool
3230should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags,
3231                     enum compact_result compact_result,
3232                     enum compact_priority *compact_priority,
3233                     int *compaction_retries)
3234{
3235        struct zone *zone;
3236        struct zoneref *z;
3237
3238        if (!order || order > PAGE_ALLOC_COSTLY_ORDER)
3239                return false;
3240
3241        /*
3242         * There are setups with compaction disabled which would prefer to loop
3243         * inside the allocator rather than hit the oom killer prematurely.
3244         * Let's give them a good hope and keep retrying while the order-0
3245         * watermarks are OK.
3246         */
3247        for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3248                                        ac->nodemask) {
3249                if (zone_watermark_ok(zone, 0, min_wmark_pages(zone),
3250                                        ac_classzone_idx(ac), alloc_flags))
3251                        return true;
3252        }
3253        return false;
3254}
3255#endif /* CONFIG_COMPACTION */
3256
3257/* Perform direct synchronous page reclaim */
3258static int
3259__perform_reclaim(gfp_t gfp_mask, unsigned int order,
3260                                        const struct alloc_context *ac)
3261{
3262        struct reclaim_state reclaim_state;
3263        int progress;
3264
3265        cond_resched();
3266
3267        /* We now go into synchronous reclaim */
3268        cpuset_memory_pressure_bump();
3269        current->flags |= PF_MEMALLOC;
3270        lockdep_set_current_reclaim_state(gfp_mask);
3271        reclaim_state.reclaimed_slab = 0;
3272        current->reclaim_state = &reclaim_state;
3273
3274        progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
3275                                                                ac->nodemask);
3276
3277        current->reclaim_state = NULL;
3278        lockdep_clear_current_reclaim_state();
3279        current->flags &= ~PF_MEMALLOC;
3280
3281        cond_resched();
3282
3283        return progress;
3284}
3285
3286/* The really slow allocator path where we enter direct reclaim */
3287static inline struct page *
3288__alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
3289                unsigned int alloc_flags, const struct alloc_context *ac,
3290                unsigned long *did_some_progress)
3291{
3292        struct page *page = NULL;
3293        bool drained = false;
3294
3295        *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
3296        if (unlikely(!(*did_some_progress)))
3297                return NULL;
3298
3299retry:
3300        page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
3301
3302        /*
3303         * If an allocation failed after direct reclaim, it could be because
3304         * pages are pinned on the per-cpu lists or in high alloc reserves.
3305         * Shrink them them and try again
3306         */
3307        if (!page && !drained) {
3308                unreserve_highatomic_pageblock(ac);
3309                drain_all_pages(NULL);
3310                drained = true;
3311                goto retry;
3312        }
3313
3314        return page;
3315}
3316
3317static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
3318{
3319        struct zoneref *z;
3320        struct zone *zone;
3321        pg_data_t *last_pgdat = NULL;
3322
3323        for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
3324                                        ac->high_zoneidx, ac->nodemask) {
3325                if (last_pgdat != zone->zone_pgdat)
3326                        wakeup_kswapd(zone, order, ac->high_zoneidx);
3327                last_pgdat = zone->zone_pgdat;
3328        }
3329}
3330
3331static inline unsigned int
3332gfp_to_alloc_flags(gfp_t gfp_mask)
3333{
3334        unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
3335
3336        /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
3337        BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
3338
3339        /*
3340         * The caller may dip into page reserves a bit more if the caller
3341         * cannot run direct reclaim, or if the caller has realtime scheduling
3342         * policy or is asking for __GFP_HIGH memory.  GFP_ATOMIC requests will
3343         * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
3344         */
3345        alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
3346
3347        if (gfp_mask & __GFP_ATOMIC) {
3348                /*
3349                 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
3350                 * if it can't schedule.
3351                 */
3352                if (!(gfp_mask & __GFP_NOMEMALLOC))
3353                        alloc_flags |= ALLOC_HARDER;
3354                /*
3355                 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
3356                 * comment for __cpuset_node_allowed().
3357                 */
3358                alloc_flags &= ~ALLOC_CPUSET;
3359        } else if (unlikely(rt_task(current)) && !in_interrupt())
3360                alloc_flags |= ALLOC_HARDER;
3361
3362#ifdef CONFIG_CMA
3363        if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
3364                alloc_flags |= ALLOC_CMA;
3365#endif
3366        return alloc_flags;
3367}
3368
3369bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
3370{
3371        if (unlikely(gfp_mask & __GFP_NOMEMALLOC))
3372                return false;
3373
3374        if (gfp_mask & __GFP_MEMALLOC)
3375                return true;
3376        if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
3377                return true;
3378        if (!in_interrupt() &&
3379                        ((current->flags & PF_MEMALLOC) ||
3380                         unlikely(test_thread_flag(TIF_MEMDIE))))
3381                return true;
3382
3383        return false;
3384}
3385
3386/*
3387 * Maximum number of reclaim retries without any progress before OOM killer
3388 * is consider as the only way to move forward.
3389 */
3390#define MAX_RECLAIM_RETRIES 16
3391
3392/*
3393 * Checks whether it makes sense to retry the reclaim to make a forward progress
3394 * for the given allocation request.
3395 * The reclaim feedback represented by did_some_progress (any progress during
3396 * the last reclaim round) and no_progress_loops (number of reclaim rounds without
3397 * any progress in a row) is considered as well as the reclaimable pages on the
3398 * applicable zone list (with a backoff mechanism which is a function of
3399 * no_progress_loops).
3400 *
3401 * Returns true if a retry is viable or false to enter the oom path.
3402 */
3403static inline bool
3404should_reclaim_retry(gfp_t gfp_mask, unsigned order,
3405                     struct alloc_context *ac, int alloc_flags,
3406                     bool did_some_progress, int *no_progress_loops)
3407{
3408        struct zone *zone;
3409        struct zoneref *z;
3410
3411        /*
3412         * Costly allocations might have made a progress but this doesn't mean
3413         * their order will become available due to high fragmentation so
3414         * always increment the no progress counter for them
3415         */
3416        if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER)
3417                *no_progress_loops = 0;
3418        else
3419                (*no_progress_loops)++;
3420
3421        /*
3422         * Make sure we converge to OOM if we cannot make any progress
3423         * several times in the row.
3424         */
3425        if (*no_progress_loops > MAX_RECLAIM_RETRIES)
3426                return false;
3427
3428        /*
3429         * Keep reclaiming pages while there is a chance this will lead
3430         * somewhere.  If none of the target zones can satisfy our allocation
3431         * request even if all reclaimable pages are considered then we are
3432         * screwed and have to go OOM.
3433         */
3434        for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3435                                        ac->nodemask) {
3436                unsigned long available;
3437                unsigned long reclaimable;
3438
3439                available = reclaimable = zone_reclaimable_pages(zone);
3440                available -= DIV_ROUND_UP((*no_progress_loops) * available,
3441                                          MAX_RECLAIM_RETRIES);
3442                available += zone_page_state_snapshot(zone, NR_FREE_PAGES);
3443
3444                /*
3445                 * Would the allocation succeed if we reclaimed the whole
3446                 * available?
3447                 */
3448                if (__zone_watermark_ok(zone, order, min_wmark_pages(zone),
3449                                ac_classzone_idx(ac), alloc_flags, available)) {
3450                        /*
3451                         * If we didn't make any progress and have a lot of
3452                         * dirty + writeback pages then we should wait for
3453                         * an IO to complete to slow down the reclaim and
3454                         * prevent from pre mature OOM
3455                         */
3456                        if (!did_some_progress) {
3457                                unsigned long write_pending;
3458
3459                                write_pending = zone_page_state_snapshot(zone,
3460                                                        NR_ZONE_WRITE_PENDING);
3461
3462                                if (2 * write_pending > reclaimable) {
3463                                        congestion_wait(BLK_RW_ASYNC, HZ/10);
3464                                        return true;
3465                                }
3466                        }
3467
3468                        /*
3469                         * Memory allocation/reclaim might be called from a WQ
3470                         * context and the current implementation of the WQ
3471                         * concurrency control doesn't recognize that
3472                         * a particular WQ is congested if the worker thread is
3473                         * looping without ever sleeping. Therefore we have to
3474                         * do a short sleep here rather than calling
3475                         * cond_resched().
3476                         */
3477                        if (current->flags & PF_WQ_WORKER)
3478                                schedule_timeout_uninterruptible(1);
3479                        else
3480                                cond_resched();
3481
3482                        return true;
3483                }
3484        }
3485
3486        return false;
3487}
3488
3489static inline struct page *
3490__alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
3491                                                struct alloc_context *ac)
3492{
3493        bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
3494        struct page *page = NULL;
3495        unsigned int alloc_flags;
3496        unsigned long did_some_progress;
3497        enum compact_priority compact_priority = DEF_COMPACT_PRIORITY;
3498        enum compact_result compact_result;
3499        int compaction_retries = 0;
3500        int no_progress_loops = 0;
3501        unsigned long alloc_start = jiffies;
3502        unsigned int stall_timeout = 10 * HZ;
3503
3504        /*
3505         * In the slowpath, we sanity check order to avoid ever trying to
3506         * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3507         * be using allocators in order of preference for an area that is
3508         * too large.
3509         */
3510        if (order >= MAX_ORDER) {
3511                WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
3512                return NULL;
3513        }
3514
3515        /*
3516         * We also sanity check to catch abuse of atomic reserves being used by
3517         * callers that are not in atomic context.
3518         */
3519        if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
3520                                (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
3521                gfp_mask &= ~__GFP_ATOMIC;
3522
3523        /*
3524         * The fast path uses conservative alloc_flags to succeed only until
3525         * kswapd needs to be woken up, and to avoid the cost of setting up
3526         * alloc_flags precisely. So we do that now.
3527         */
3528        alloc_flags = gfp_to_alloc_flags(gfp_mask);
3529
3530        if (gfp_mask & __GFP_KSWAPD_RECLAIM)
3531                wake_all_kswapds(order, ac);
3532
3533        /*
3534         * The adjusted alloc_flags might result in immediate success, so try
3535         * that first
3536         */
3537        page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
3538        if (page)
3539                goto got_pg;
3540
3541        /*
3542         * For costly allocations, try direct compaction first, as it's likely
3543         * that we have enough base pages and don't need to reclaim. Don't try
3544         * that for allocations that are allowed to ignore watermarks, as the
3545         * ALLOC_NO_WATERMARKS attempt didn't yet happen.
3546         */
3547        if (can_direct_reclaim && order > PAGE_ALLOC_COSTLY_ORDER &&
3548                !gfp_pfmemalloc_allowed(gfp_mask)) {
3549                page = __alloc_pages_direct_compact(gfp_mask, order,
3550                                                alloc_flags, ac,
3551                                                INIT_COMPACT_PRIORITY,
3552                                                &compact_result);
3553                if (page)
3554                        goto got_pg;
3555
3556                /*
3557                 * Checks for costly allocations with __GFP_NORETRY, which
3558                 * includes THP page fault allocations
3559                 */
3560                if (gfp_mask & __GFP_NORETRY) {
3561                        /*
3562                         * If compaction is deferred for high-order allocations,
3563                         * it is because sync compaction recently failed. If
3564                         * this is the case and the caller requested a THP
3565                         * allocation, we do not want to heavily disrupt the
3566                         * system, so we fail the allocation instead of entering
3567                         * direct reclaim.
3568                         */
3569                        if (compact_result == COMPACT_DEFERRED)
3570                                goto nopage;
3571
3572                        /*
3573                         * Looks like reclaim/compaction is worth trying, but
3574                         * sync compaction could be very expensive, so keep
3575                         * using async compaction.
3576                         */
3577                        compact_priority = INIT_COMPACT_PRIORITY;
3578                }
3579        }
3580
3581retry:
3582        /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */
3583        if (gfp_mask & __GFP_KSWAPD_RECLAIM)
3584                wake_all_kswapds(order, ac);
3585
3586        if (gfp_pfmemalloc_allowed(gfp_mask))
3587                alloc_flags = ALLOC_NO_WATERMARKS;
3588
3589        /*
3590         * Reset the zonelist iterators if memory policies can be ignored.
3591         * These allocations are high priority and system rather than user
3592         * orientated.
3593         */
3594        if (!(alloc_flags & ALLOC_CPUSET) || (alloc_flags & ALLOC_NO_WATERMARKS)) {
3595                ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
3596                ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
3597                                        ac->high_zoneidx, ac->nodemask);
3598        }
3599
3600        /* Attempt with potentially adjusted zonelist and alloc_flags */
3601        page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
3602        if (page)
3603                goto got_pg;
3604
3605        /* Caller is not willing to reclaim, we can't balance anything */
3606        if (!can_direct_reclaim) {
3607                /*
3608                 * All existing users of the __GFP_NOFAIL are blockable, so warn
3609                 * of any new users that actually allow this type of allocation
3610                 * to fail.
3611                 */
3612                WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
3613                goto nopage;
3614        }
3615
3616        /* Avoid recursion of direct reclaim */
3617        if (current->flags & PF_MEMALLOC) {
3618                /*
3619                 * __GFP_NOFAIL request from this context is rather bizarre
3620                 * because we cannot reclaim anything and only can loop waiting
3621                 * for somebody to do a work for us.
3622                 */
3623                if (WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
3624                        cond_resched();
3625                        goto retry;
3626                }
3627                goto nopage;
3628        }
3629
3630        /* Avoid allocations with no watermarks from looping endlessly */
3631        if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
3632                goto nopage;
3633
3634
3635        /* Try direct reclaim and then allocating */
3636        page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
3637                                                        &did_some_progress);
3638        if (page)
3639                goto got_pg;
3640
3641        /* Try direct compaction and then allocating */
3642        page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
3643                                        compact_priority, &compact_result);
3644        if (page)
3645                goto got_pg;
3646
3647        /* Do not loop if specifically requested */
3648        if (gfp_mask & __GFP_NORETRY)
3649                goto nopage;
3650
3651        /*
3652         * Do not retry costly high order allocations unless they are
3653         * __GFP_REPEAT
3654         */
3655        if (order > PAGE_ALLOC_COSTLY_ORDER && !(gfp_mask & __GFP_REPEAT))
3656                goto nopage;
3657
3658        /* Make sure we know about allocations which stall for too long */
3659        if (time_after(jiffies, alloc_start + stall_timeout)) {
3660                warn_alloc(gfp_mask,
3661                        "page allocation stalls for %ums, order:%u",
3662                        jiffies_to_msecs(jiffies-alloc_start), order);
3663                stall_timeout += 10 * HZ;
3664        }
3665
3666        if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags,
3667                                 did_some_progress > 0, &no_progress_loops))
3668                goto retry;
3669
3670        /*
3671         * It doesn't make any sense to retry for the compaction if the order-0
3672         * reclaim is not able to make any progress because the current
3673         * implementation of the compaction depends on the sufficient amount
3674         * of free memory (see __compaction_suitable)
3675         */
3676        if (did_some_progress > 0 &&
3677                        should_compact_retry(ac, order, alloc_flags,
3678                                compact_result, &compact_priority,
3679                                &compaction_retries))
3680                goto retry;
3681
3682        /* Reclaim has failed us, start killing things */
3683        page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
3684        if (page)
3685                goto got_pg;
3686
3687        /* Retry as long as the OOM killer is making progress */
3688        if (did_some_progress) {
3689                no_progress_loops = 0;
3690                goto retry;
3691        }
3692
3693nopage:
3694        warn_alloc(gfp_mask,
3695                        "page allocation failure: order:%u", order);
3696got_pg:
3697        return page;
3698}
3699
3700/*
3701 * This is the 'heart' of the zoned buddy allocator.
3702 */
3703struct page *
3704__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
3705                        struct zonelist *zonelist, nodemask_t *nodemask)
3706{
3707        struct page *page;
3708        unsigned int cpuset_mems_cookie;
3709        unsigned int alloc_flags = ALLOC_WMARK_LOW;
3710        gfp_t alloc_mask = gfp_mask; /* The gfp_t that was actually used for allocation */
3711        struct alloc_context ac = {
3712                .high_zoneidx = gfp_zone(gfp_mask),
3713                .zonelist = zonelist,
3714                .nodemask = nodemask,
3715                .migratetype = gfpflags_to_migratetype(gfp_mask),
3716        };
3717
3718        if (cpusets_enabled()) {
3719                alloc_mask |= __GFP_HARDWALL;
3720                alloc_flags |= ALLOC_CPUSET;
3721                if (!ac.nodemask)
3722                        ac.nodemask = &cpuset_current_mems_allowed;
3723        }
3724
3725        gfp_mask &= gfp_allowed_mask;
3726
3727        lockdep_trace_alloc(gfp_mask);
3728
3729        might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
3730
3731        if (should_fail_alloc_page(gfp_mask, order))
3732                return NULL;
3733
3734        /*
3735         * Check the zones suitable for the gfp_mask contain at least one
3736         * valid zone. It's possible to have an empty zonelist as a result
3737         * of __GFP_THISNODE and a memoryless node
3738         */
3739        if (unlikely(!zonelist->_zonerefs->zone))
3740                return NULL;
3741
3742        if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE)
3743                alloc_flags |= ALLOC_CMA;
3744
3745retry_cpuset:
3746        cpuset_mems_cookie = read_mems_allowed_begin();
3747
3748        /* Dirty zone balancing only done in the fast path */
3749        ac.spread_dirty_pages = (gfp_mask & __GFP_WRITE);
3750
3751        /*
3752         * The preferred zone is used for statistics but crucially it is
3753         * also used as the starting point for the zonelist iterator. It
3754         * may get reset for allocations that ignore memory policies.
3755         */
3756        ac.preferred_zoneref = first_zones_zonelist(ac.zonelist,
3757                                        ac.high_zoneidx, ac.nodemask);
3758        if (!ac.preferred_zoneref) {
3759                page = NULL;
3760                goto no_zone;
3761        }
3762
3763        /* First allocation attempt */
3764        page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
3765        if (likely(page))
3766                goto out;
3767
3768        /*
3769         * Runtime PM, block IO and its error handling path can deadlock
3770         * because I/O on the device might not complete.
3771         */
3772        alloc_mask = memalloc_noio_flags(gfp_mask);
3773        ac.spread_dirty_pages = false;
3774
3775        /*
3776         * Restore the original nodemask if it was potentially replaced with
3777         * &cpuset_current_mems_allowed to optimize the fast-path attempt.
3778         */
3779        if (cpusets_enabled())
3780                ac.nodemask = nodemask;
3781        page = __alloc_pages_slowpath(alloc_mask, order, &ac);
3782
3783no_zone:
3784        /*
3785         * When updating a task's mems_allowed, it is possible to race with
3786         * parallel threads in such a way that an allocation can fail while
3787         * the mask is being updated. If a page allocation is about to fail,
3788         * check if the cpuset changed during allocation and if so, retry.
3789         */
3790        if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) {
3791                alloc_mask = gfp_mask;
3792                goto retry_cpuset;
3793        }
3794
3795out:
3796        if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page &&
3797            unlikely(memcg_kmem_charge(page, gfp_mask, order) != 0)) {
3798                __free_pages(page, order);
3799                page = NULL;
3800        }
3801
3802        if (kmemcheck_enabled && page)
3803                kmemcheck_pagealloc_alloc(page, order, gfp_mask);
3804
3805        trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
3806
3807        return page;
3808}
3809EXPORT_SYMBOL(__alloc_pages_nodemask);
3810
3811/*
3812 * Common helper functions.
3813 */
3814unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
3815{
3816        struct page *page;
3817
3818        /*
3819         * __get_free_pages() returns a 32-bit address, which cannot represent
3820         * a highmem page
3821         */
3822        VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
3823
3824        page = alloc_pages(gfp_mask, order);
3825        if (!page)
3826                return 0;
3827        return (unsigned long) page_address(page);
3828}
3829EXPORT_SYMBOL(__get_free_pages);
3830
3831unsigned long get_zeroed_page(gfp_t gfp_mask)
3832{
3833        return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
3834}
3835EXPORT_SYMBOL(get_zeroed_page);
3836
3837void __free_pages(struct page *page, unsigned int order)
3838{
3839        if (put_page_testzero(page)) {
3840                if (order == 0)
3841                        free_hot_cold_page(page, false);
3842                else
3843                        __free_pages_ok(page, order);
3844        }
3845}
3846
3847EXPORT_SYMBOL(__free_pages);
3848
3849void free_pages(unsigned long addr, unsigned int order)
3850{
3851        if (addr != 0) {
3852                VM_BUG_ON(!virt_addr_valid((void *)addr));
3853                __free_pages(virt_to_page((void *)addr), order);
3854        }
3855}
3856
3857EXPORT_SYMBOL(free_pages);
3858
3859/*
3860 * Page Fragment:
3861 *  An arbitrary-length arbitrary-offset area of memory which resides
3862 *  within a 0 or higher order page.  Multiple fragments within that page
3863 *  are individually refcounted, in the page's reference counter.
3864 *
3865 * The page_frag functions below provide a simple allocation framework for
3866 * page fragments.  This is used by the network stack and network device
3867 * drivers to provide a backing region of memory for use as either an
3868 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3869 */
3870static struct page *__page_frag_refill(struct page_frag_cache *nc,
3871                                       gfp_t gfp_mask)
3872{
3873        struct page *page = NULL;
3874        gfp_t gfp = gfp_mask;
3875
3876#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3877        gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
3878                    __GFP_NOMEMALLOC;
3879        page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
3880                                PAGE_FRAG_CACHE_MAX_ORDER);
3881        nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
3882#endif
3883        if (unlikely(!page))
3884                page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
3885
3886        nc->va = page ? page_address(page) : NULL;
3887
3888        return page;
3889}
3890
3891void *__alloc_page_frag(struct page_frag_cache *nc,
3892                        unsigned int fragsz, gfp_t gfp_mask)
3893{
3894        unsigned int size = PAGE_SIZE;
3895        struct page *page;
3896        int offset;
3897
3898        if (unlikely(!nc->va)) {
3899refill:
3900                page = __page_frag_refill(nc, gfp_mask);
3901                if (!page)
3902                        return NULL;
3903
3904#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3905                /* if size can vary use size else just use PAGE_SIZE */
3906                size = nc->size;
3907#endif
3908                /* Even if we own the page, we do not use atomic_set().
3909                 * This would break get_page_unless_zero() users.
3910                 */
3911                page_ref_add(page, size - 1);
3912
3913                /* reset page count bias and offset to start of new frag */
3914                nc->pfmemalloc = page_is_pfmemalloc(page);
3915                nc->pagecnt_bias = size;
3916                nc->offset = size;
3917        }
3918
3919        offset = nc->offset - fragsz;
3920        if (unlikely(offset < 0)) {
3921                page = virt_to_page(nc->va);
3922
3923                if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
3924                        goto refill;
3925
3926#if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3927                /* if size can vary use size else just use PAGE_SIZE */
3928                size = nc->size;
3929#endif
3930                /* OK, page count is 0, we can safely set it */
3931                set_page_count(page, size);
3932
3933                /* reset page count bias and offset to start of new frag */
3934                nc->pagecnt_bias = size;
3935                offset = size - fragsz;
3936        }
3937
3938        nc->pagecnt_bias--;
3939        nc->offset = offset;
3940
3941        return nc->va + offset;
3942}
3943EXPORT_SYMBOL(__alloc_page_frag);
3944
3945/*
3946 * Frees a page fragment allocated out of either a compound or order 0 page.
3947 */
3948void __free_page_frag(void *addr)
3949{
3950        struct page *page = virt_to_head_page(addr);
3951
3952        if (unlikely(put_page_testzero(page)))
3953                __free_pages_ok(page, compound_order(page));
3954}
3955EXPORT_SYMBOL(__free_page_frag);
3956
3957static void *make_alloc_exact(unsigned long addr, unsigned int order,
3958                size_t size)
3959{
3960        if (addr) {
3961                unsigned long alloc_end = addr + (PAGE_SIZE << order);
3962                unsigned long used = addr + PAGE_ALIGN(size);
3963
3964                split_page(virt_to_page((void *)addr), order);
3965                while (used < alloc_end) {
3966                        free_page(used);
3967                        used += PAGE_SIZE;
3968                }
3969        }
3970        return (void *)addr;
3971}
3972
3973/**
3974 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3975 * @size: the number of bytes to allocate
3976 * @gfp_mask: GFP flags for the allocation
3977 *
3978 * This function is similar to alloc_pages(), except that it allocates the
3979 * minimum number of pages to satisfy the request.  alloc_pages() can only
3980 * allocate memory in power-of-two pages.
3981 *
3982 * This function is also limited by MAX_ORDER.
3983 *
3984 * Memory allocated by this function must be released by free_pages_exact().
3985 */
3986void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
3987{
3988        unsigned int order = get_order(size);
3989        unsigned long addr;
3990
3991        addr = __get_free_pages(gfp_mask, order);
3992        return make_alloc_exact(addr, order, size);
3993}
3994EXPORT_SYMBOL(alloc_pages_exact);
3995
3996/**
3997 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3998 *                         pages on a node.
3999 * @nid: the preferred node ID where memory should be allocated
4000 * @size: the number of bytes to allocate
4001 * @gfp_mask: GFP flags for the allocation
4002 *
4003 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
4004 * back.
4005 */
4006void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
4007{
4008        unsigned int order = get_order(size);
4009        struct page *p = alloc_pages_node(nid, gfp_mask, order);
4010        if (!p)
4011                return NULL;
4012        return make_alloc_exact((unsigned long)page_address(p), order, size);
4013}
4014
4015/**
4016 * free_pages_exact - release memory allocated via alloc_pages_exact()
4017 * @virt: the value returned by alloc_pages_exact.
4018 * @size: size of allocation, same value as passed to alloc_pages_exact().
4019 *
4020 * Release the memory allocated by a previous call to alloc_pages_exact.
4021 */
4022void free_pages_exact(void *virt, size_t size)
4023{
4024        unsigned long addr = (unsigned long)virt;
4025        unsigned long end = addr + PAGE_ALIGN(size);
4026
4027        while (addr < end) {
4028                free_page(addr);
4029                addr += PAGE_SIZE;
4030        }
4031}
4032EXPORT_SYMBOL(free_pages_exact);
4033
4034/**
4035 * nr_free_zone_pages - count number of pages beyond high watermark
4036 * @offset: The zone index of the highest zone
4037 *
4038 * nr_free_zone_pages() counts the number of counts pages which are beyond the
4039 * high watermark within all zones at or below a given zone index.  For each
4040 * zone, the number of pages is calculated as:
4041 *     managed_pages - high_pages
4042 */
4043static unsigned long nr_free_zone_pages(int offset)
4044{
4045        struct zoneref *z;
4046        struct zone *zone;
4047
4048        /* Just pick one node, since fallback list is circular */
4049        unsigned long sum = 0;
4050
4051        struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
4052
4053        for_each_zone_zonelist(zone, z, zonelist, offset) {
4054                unsigned long size = zone->managed_pages;
4055                unsigned long high = high_wmark_pages(zone);
4056                if (size > high)
4057                        sum += size - high;
4058        }
4059
4060        return sum;
4061}
4062
4063/**
4064 * nr_free_buffer_pages - count number of pages beyond high watermark
4065 *
4066 * nr_free_buffer_pages() counts the number of pages which are beyond the high
4067 * watermark within ZONE_DMA and ZONE_NORMAL.
4068 */
4069unsigned long nr_free_buffer_pages(void)
4070{
4071        return nr_free_zone_pages(gfp_zone(GFP_USER));
4072}
4073EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
4074
4075/**
4076 * nr_free_pagecache_pages - count number of pages beyond high watermark
4077 *
4078 * nr_free_pagecache_pages() counts the number of pages which are beyond the
4079 * high watermark within all zones.
4080 */
4081unsigned long nr_free_pagecache_pages(void)
4082{
4083        return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
4084}
4085
4086static inline void show_node(struct zone *zone)
4087{
4088        if (IS_ENABLED(CONFIG_NUMA))
4089                printk("Node %d ", zone_to_nid(zone));
4090}
4091
4092long si_mem_available(void)
4093{
4094        long available;
4095        unsigned long pagecache;
4096        unsigned long wmark_low = 0;
4097        unsigned long pages[NR_LRU_LISTS];
4098        struct zone *zone;
4099        int lru;
4100
4101        for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++)
4102                pages[lru] = global_node_page_state(NR_LRU_BASE + lru);
4103
4104        for_each_zone(zone)
4105                wmark_low += zone->watermark[WMARK_LOW];
4106
4107        /*
4108         * Estimate the amount of memory available for userspace allocations,
4109         * without causing swapping.
4110         */
4111        available = global_page_state(NR_FREE_PAGES) - totalreserve_pages;
4112
4113        /*
4114         * Not all the page cache can be freed, otherwise the system will
4115         * start swapping. Assume at least half of the page cache, or the
4116         * low watermark worth of cache, needs to stay.
4117         */
4118        pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE];
4119        pagecache -= min(pagecache / 2, wmark_low);
4120        available += pagecache;
4121
4122        /*
4123         * Part of the reclaimable slab consists of items that are in use,
4124         * and cannot be freed. Cap this estimate at the low watermark.
4125         */
4126        available += global_page_state(NR_SLAB_RECLAIMABLE) -
4127                     min(global_page_state(NR_SLAB_RECLAIMABLE) / 2, wmark_low);
4128
4129        if (available < 0)
4130                available = 0;
4131        return available;
4132}
4133EXPORT_SYMBOL_GPL(si_mem_available);
4134
4135void si_meminfo(struct sysinfo *val)
4136{
4137        val->totalram = totalram_pages;
4138        val->sharedram = global_node_page_state(NR_SHMEM);
4139        val->freeram = global_page_state(NR_FREE_PAGES);
4140        val->bufferram = nr_blockdev_pages();
4141        val->totalhigh = totalhigh_pages;
4142        val->freehigh = nr_free_highpages();
4143        val->mem_unit = PAGE_SIZE;
4144}
4145
4146EXPORT_SYMBOL(si_meminfo);
4147
4148#ifdef CONFIG_NUMA
4149void si_meminfo_node(struct sysinfo *val, int nid)
4150{
4151        int zone_type;          /* needs to be signed */
4152        unsigned long managed_pages = 0;
4153        unsigned long managed_highpages = 0;
4154        unsigned long free_highpages = 0;
4155        pg_data_t *pgdat = NODE_DATA(nid);
4156
4157        for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
4158                managed_pages += pgdat->node_zones[zone_type].managed_pages;
4159        val->totalram = managed_pages;
4160        val->sharedram = node_page_state(pgdat, NR_SHMEM);
4161        val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES);
4162#ifdef CONFIG_HIGHMEM
4163        for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
4164                struct zone *zone = &pgdat->node_zones[zone_type];
4165
4166                if (is_highmem(zone)) {
4167                        managed_highpages += zone->managed_pages;
4168                        free_highpages += zone_page_state(zone, NR_FREE_PAGES);
4169                }
4170        }
4171        val->totalhigh = managed_highpages;
4172        val->freehigh = free_highpages;
4173#else
4174        val->totalhigh = managed_highpages;
4175        val->freehigh = free_highpages;
4176#endif
4177        val->mem_unit = PAGE_SIZE;
4178}
4179#endif
4180
4181/*
4182 * Determine whether the node should be displayed or not, depending on whether
4183 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
4184 */
4185bool skip_free_areas_node(unsigned int flags, int nid)
4186{
4187        bool ret = false;
4188        unsigned int cpuset_mems_cookie;
4189
4190        if (!(flags & SHOW_MEM_FILTER_NODES))
4191                goto out;
4192
4193        do {
4194                cpuset_mems_cookie = read_mems_allowed_begin();
4195                ret = !node_isset(nid, cpuset_current_mems_allowed);
4196        } while (read_mems_allowed_retry(cpuset_mems_cookie));
4197out:
4198        return ret;
4199}
4200
4201#define K(x) ((x) << (PAGE_SHIFT-10))
4202
4203static void show_migration_types(unsigned char type)
4204{
4205        static const char types[MIGRATE_TYPES] = {
4206                [MIGRATE_UNMOVABLE]     = 'U',
4207                [MIGRATE_MOVABLE]       = 'M',
4208                [MIGRATE_RECLAIMABLE]   = 'E',
4209                [MIGRATE_HIGHATOMIC]    = 'H',
4210#ifdef CONFIG_CMA
4211                [MIGRATE_CMA]           = 'C',
4212#endif
4213#ifdef CONFIG_MEMORY_ISOLATION
4214                [MIGRATE_ISOLATE]       = 'I',
4215#endif
4216        };
4217        char tmp[MIGRATE_TYPES + 1];
4218        char *p = tmp;
4219        int i;
4220
4221        for (i = 0; i < MIGRATE_TYPES; i++) {
4222                if (type & (1 << i))
4223                        *p++ = types[i];
4224        }
4225
4226        *p = '\0';
4227        printk(KERN_CONT "(%s) ", tmp);
4228}
4229
4230/*
4231 * Show free area list (used inside shift_scroll-lock stuff)
4232 * We also calculate the percentage fragmentation. We do this by counting the
4233 * memory on each free list with the exception of the first item on the list.
4234 *
4235 * Bits in @filter:
4236 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
4237 *   cpuset.
4238 */
4239void show_free_areas(unsigned int filter)
4240{
4241        unsigned long free_pcp = 0;
4242        int cpu;
4243        struct zone *zone;
4244        pg_data_t *pgdat;
4245
4246        for_each_populated_zone(zone) {
4247                if (skip_free_areas_node(filter, zone_to_nid(zone)))
4248                        continue;
4249
4250                for_each_online_cpu(cpu)
4251                        free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
4252        }
4253
4254        printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
4255                " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
4256                " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
4257                " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
4258                " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
4259                " free:%lu free_pcp:%lu free_cma:%lu\n",
4260                global_node_page_state(NR_ACTIVE_ANON),
4261                global_node_page_state(NR_INACTIVE_ANON),
4262                global_node_page_state(NR_ISOLATED_ANON),
4263                global_node_page_state(NR_ACTIVE_FILE),
4264                global_node_page_state(NR_INACTIVE_FILE),
4265                global_node_page_state(NR_ISOLATED_FILE),
4266                global_node_page_state(NR_UNEVICTABLE),
4267                global_node_page_state(NR_FILE_DIRTY),
4268                global_node_page_state(NR_WRITEBACK),
4269                global_node_page_state(NR_UNSTABLE_NFS),
4270                global_page_state(NR_SLAB_RECLAIMABLE),
4271                global_page_state(NR_SLAB_UNRECLAIMABLE),
4272                global_node_page_state(NR_FILE_MAPPED),
4273                global_node_page_state(NR_SHMEM),
4274                global_page_state(NR_PAGETABLE),
4275                global_page_state(NR_BOUNCE),
4276                global_page_state(NR_FREE_PAGES),
4277                free_pcp,
4278                global_page_state(NR_FREE_CMA_PAGES));
4279
4280        for_each_online_pgdat(pgdat) {
4281                printk("Node %d"
4282                        " active_anon:%lukB"
4283                        " inactive_anon:%lukB"
4284                        " active_file:%lukB"
4285                        " inactive_file:%lukB"
4286                        " unevictable:%lukB"
4287                        " isolated(anon):%lukB"
4288                        " isolated(file):%lukB"
4289                        " mapped:%lukB"
4290                        " dirty:%lukB"
4291                        " writeback:%lukB"
4292                        " shmem:%lukB"
4293#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4294                        " shmem_thp: %lukB"
4295                        " shmem_pmdmapped: %lukB"
4296                        " anon_thp: %lukB"
4297#endif
4298                        " writeback_tmp:%lukB"
4299                        " unstable:%lukB"
4300                        " pages_scanned:%lu"
4301                        " all_unreclaimable? %s"
4302                        "\n",
4303                        pgdat->node_id,
4304                        K(node_page_state(pgdat, NR_ACTIVE_ANON)),
4305                        K(node_page_state(pgdat, NR_INACTIVE_ANON)),
4306                        K(node_page_state(pgdat, NR_ACTIVE_FILE)),
4307                        K(node_page_state(pgdat, NR_INACTIVE_FILE)),
4308                        K(node_page_state(pgdat, NR_UNEVICTABLE)),
4309                        K(node_page_state(pgdat, NR_ISOLATED_ANON)),
4310                        K(node_page_state(pgdat, NR_ISOLATED_FILE)),
4311                        K(node_page_state(pgdat, NR_FILE_MAPPED)),
4312                        K(node_page_state(pgdat, NR_FILE_DIRTY)),
4313                        K(node_page_state(pgdat, NR_WRITEBACK)),
4314#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4315                        K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR),
4316                        K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)
4317                                        * HPAGE_PMD_NR),
4318                        K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR),
4319#endif
4320                        K(node_page_state(pgdat, NR_SHMEM)),
4321                        K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
4322                        K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
4323                        node_page_state(pgdat, NR_PAGES_SCANNED),
4324                        !pgdat_reclaimable(pgdat) ? "yes" : "no");
4325        }
4326
4327        for_each_populated_zone(zone) {
4328                int i;
4329
4330                if (skip_free_areas_node(filter, zone_to_nid(zone)))
4331                        continue;
4332
4333                free_pcp = 0;
4334                for_each_online_cpu(cpu)
4335                        free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
4336
4337                show_node(zone);
4338                printk(KERN_CONT
4339                        "%s"
4340                        " free:%lukB"
4341                        " min:%lukB"
4342                        " low:%lukB"
4343                        " high:%lukB"
4344                        " active_anon:%lukB"
4345                        " inactive_anon:%lukB"
4346                        " active_file:%lukB"
4347                        " inactive_file:%lukB"
4348                        " unevictable:%lukB"
4349                        " writepending:%lukB"
4350                        " present:%lukB"
4351                        " managed:%lukB"
4352                        " mlocked:%lukB"
4353                        " slab_reclaimable:%lukB"
4354                        " slab_unreclaimable:%lukB"
4355                        " kernel_stack:%lukB"
4356                        " pagetables:%lukB"
4357                        " bounce:%lukB"
4358                        " free_pcp:%lukB"
4359                        " local_pcp:%ukB"
4360                        " free_cma:%lukB"
4361                        "\n",
4362                        zone->name,
4363                        K(zone_page_state(zone, NR_FREE_PAGES)),
4364                        K(min_wmark_pages(zone)),
4365                        K(low_wmark_pages(zone)),
4366                        K(high_wmark_pages(zone)),
4367                        K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)),
4368                        K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)),
4369                        K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)),
4370                        K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)),
4371                        K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)),
4372                        K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)),
4373                        K(zone->present_pages),
4374                        K(zone->managed_pages),
4375                        K(zone_page_state(zone, NR_MLOCK)),
4376                        K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
4377                        K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
4378                        zone_page_state(zone, NR_KERNEL_STACK_KB),
4379                        K(zone_page_state(zone, NR_PAGETABLE)),
4380                        K(zone_page_state(zone, NR_BOUNCE)),
4381                        K(free_pcp),
4382                        K(this_cpu_read(zone->pageset->pcp.count)),
4383                        K(zone_page_state(zone, NR_FREE_CMA_PAGES)));
4384                printk("lowmem_reserve[]:");
4385                for (i = 0; i < MAX_NR_ZONES; i++)
4386                        printk(KERN_CONT " %ld", zone->lowmem_reserve[i]);
4387                printk(KERN_CONT "\n");
4388        }
4389
4390        for_each_populated_zone(zone) {
4391                unsigned int order;
4392                unsigned long nr[MAX_ORDER], flags, total = 0;
4393                unsigned char types[MAX_ORDER];
4394
4395                if (skip_free_areas_node(filter, zone_to_nid(zone)))
4396                        continue;
4397                show_node(zone);
4398                printk(KERN_CONT "%s: ", zone->name);
4399
4400                spin_lock_irqsave(&zone->lock, flags);
4401                for (order = 0; order < MAX_ORDER; order++) {
4402                        struct free_area *area = &zone->free_area[order];
4403                        int type;
4404
4405                        nr[order] = area->nr_free;
4406                        total += nr[order] << order;
4407
4408                        types[order] = 0;
4409                        for (type = 0; type < MIGRATE_TYPES; type++) {
4410                                if (!list_empty(&area->free_list[type]))
4411                                        types[order] |= 1 << type;
4412                        }
4413                }
4414                spin_unlock_irqrestore(&zone->lock, flags);
4415                for (order = 0; order < MAX_ORDER; order++) {
4416                        printk(KERN_CONT "%lu*%lukB ",
4417                               nr[order], K(1UL) << order);
4418                        if (nr[order])
4419                                show_migration_types(types[order]);
4420                }
4421                printk(KERN_CONT "= %lukB\n", K(total));
4422        }
4423
4424        hugetlb_show_meminfo();
4425
4426        printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES));
4427
4428        show_swap_cache_info();
4429}
4430
4431static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
4432{
4433        zoneref->zone = zone;
4434        zoneref->zone_idx = zone_idx(zone);
4435}
4436
4437/*
4438 * Builds allocation fallback zone lists.
4439 *
4440 * Add all populated zones of a node to the zonelist.
4441 */
4442static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
4443                                int nr_zones)
4444{
4445        struct zone *zone;
4446        enum zone_type zone_type = MAX_NR_ZONES;
4447
4448        do {
4449                zone_type--;
4450                zone = pgdat->node_zones + zone_type;
4451                if (managed_zone(zone)) {
4452                        zoneref_set_zone(zone,
4453                                &zonelist->_zonerefs[nr_zones++]);
4454                        check_highest_zone(zone_type);
4455                }
4456        } while (zone_type);
4457
4458        return nr_zones;
4459}
4460
4461
4462/*
4463 *  zonelist_order:
4464 *  0 = automatic detection of better ordering.
4465 *  1 = order by ([node] distance, -zonetype)
4466 *  2 = order by (-zonetype, [node] distance)
4467 *
4468 *  If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
4469 *  the same zonelist. So only NUMA can configure this param.
4470 */
4471#define ZONELIST_ORDER_DEFAULT  0
4472#define ZONELIST_ORDER_NODE     1
4473#define ZONELIST_ORDER_ZONE     2
4474
4475/* zonelist order in the kernel.
4476 * set_zonelist_order() will set this to NODE or ZONE.
4477 */
4478static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
4479static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
4480
4481
4482#ifdef CONFIG_NUMA
4483/* The value user specified ....changed by config */
4484static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
4485/* string for sysctl */
4486#define NUMA_ZONELIST_ORDER_LEN 16
4487char numa_zonelist_order[16] = "default";
4488
4489/*
4490 * interface for configure zonelist ordering.
4491 * command line option "numa_zonelist_order"
4492 *      = "[dD]efault   - default, automatic configuration.
4493 *      = "[nN]ode      - order by node locality, then by zone within node
4494 *      = "[zZ]one      - order by zone, then by locality within zone
4495 */
4496
4497static int __parse_numa_zonelist_order(char *s)
4498{
4499        if (*s == 'd' || *s == 'D') {
4500                user_zonelist_order = ZONELIST_ORDER_DEFAULT;
4501        } else if (*s == 'n' || *s == 'N') {
4502                user_zonelist_order = ZONELIST_ORDER_NODE;
4503        } else if (*s == 'z' || *s == 'Z') {
4504                user_zonelist_order = ZONELIST_ORDER_ZONE;
4505        } else {
4506                pr_warn("Ignoring invalid numa_zonelist_order value:  %s\n", s);
4507                return -EINVAL;
4508        }
4509        return 0;
4510}
4511
4512static __init int setup_numa_zonelist_order(char *s)
4513{
4514        int ret;
4515
4516        if (!s)
4517                return 0;
4518
4519        ret = __parse_numa_zonelist_order(s);
4520        if (ret == 0)
4521                strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
4522
4523        return ret;
4524}
4525early_param("numa_zonelist_order", setup_numa_zonelist_order);
4526
4527/*
4528 * sysctl handler for numa_zonelist_order
4529 */
4530int numa_zonelist_order_handler(struct ctl_table *table, int write,
4531                void __user *buffer, size_t *length,
4532                loff_t *ppos)
4533{
4534        char saved_string[NUMA_ZONELIST_ORDER_LEN];
4535        int ret;
4536        static DEFINE_MUTEX(zl_order_mutex);
4537
4538        mutex_lock(&zl_order_mutex);
4539        if (write) {
4540                if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
4541                        ret = -EINVAL;
4542                        goto out;
4543                }
4544                strcpy(saved_string, (char *)table->data);
4545        }
4546        ret = proc_dostring(table, write, buffer, length, ppos);
4547        if (ret)
4548                goto out;
4549        if (write) {
4550                int oldval = user_zonelist_order;
4551
4552                ret = __parse_numa_zonelist_order((char *)table->data);
4553                if (ret) {
4554                        /*
4555                         * bogus value.  restore saved string
4556                         */
4557                        strncpy((char *)table->data, saved_string,
4558                                NUMA_ZONELIST_ORDER_LEN);
4559                        user_zonelist_order = oldval;
4560                } else if (oldval != user_zonelist_order) {
4561                        mutex_lock(&zonelists_mutex);
4562                        build_all_zonelists(NULL, NULL);
4563                        mutex_unlock(&zonelists_mutex);
4564                }
4565        }
4566out:
4567        mutex_unlock(&zl_order_mutex);
4568        return ret;
4569}
4570
4571
4572#define MAX_NODE_LOAD (nr_online_nodes)
4573static int node_load[MAX_NUMNODES];
4574
4575/**
4576 * find_next_best_node - find the next node that should appear in a given node's fallback list
4577 * @node: node whose fallback list we're appending
4578 * @used_node_mask: nodemask_t of already used nodes
4579 *
4580 * We use a number of factors to determine which is the next node that should
4581 * appear on a given node's fallback list.  The node should not have appeared
4582 * already in @node's fallback list, and it should be the next closest node
4583 * according to the distance array (which contains arbitrary distance values
4584 * from each node to each node in the system), and should also prefer nodes
4585 * with no CPUs, since presumably they'll have very little allocation pressure
4586 * on them otherwise.
4587 * It returns -1 if no node is found.
4588 */
4589static int find_next_best_node(int node, nodemask_t *used_node_mask)
4590{
4591        int n, val;
4592        int min_val = INT_MAX;
4593        int best_node = NUMA_NO_NODE;
4594        const struct cpumask *tmp = cpumask_of_node(0);
4595
4596        /* Use the local node if we haven't already */
4597        if (!node_isset(node, *used_node_mask)) {
4598                node_set(node, *used_node_mask);
4599                return node;
4600        }
4601
4602        for_each_node_state(n, N_MEMORY) {
4603
4604                /* Don't want a node to appear more than once */
4605                if (node_isset(n, *used_node_mask))
4606                        continue;
4607
4608                /* Use the distance array to find the distance */
4609                val = node_distance(node, n);
4610
4611                /* Penalize nodes under us ("prefer the next node") */
4612                val += (n < node);
4613
4614                /* Give preference to headless and unused nodes */
4615                tmp = cpumask_of_node(n);
4616                if (!cpumask_empty(tmp))
4617                        val += PENALTY_FOR_NODE_WITH_CPUS;
4618
4619                /* Slight preference for less loaded node */
4620                val *= (MAX_NODE_LOAD*MAX_NUMNODES);
4621                val += node_load[n];
4622
4623                if (val < min_val) {
4624                        min_val = val;
4625                        best_node = n;
4626                }
4627        }
4628
4629        if (best_node >= 0)
4630                node_set(best_node, *used_node_mask);
4631
4632        return best_node;
4633}
4634
4635
4636/*
4637 * Build zonelists ordered by node and zones within node.
4638 * This results in maximum locality--normal zone overflows into local
4639 * DMA zone, if any--but risks exhausting DMA zone.
4640 */
4641static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
4642{
4643        int j;
4644        struct zonelist *zonelist;
4645
4646        zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK];
4647        for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
4648                ;
4649        j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4650        zonelist->_zonerefs[j].zone = NULL;
4651        zonelist->_zonerefs[j].zone_idx = 0;
4652}
4653
4654/*
4655 * Build gfp_thisnode zonelists
4656 */
4657static void build_thisnode_zonelists(pg_data_t *pgdat)
4658{
4659        int j;
4660        struct zonelist *zonelist;
4661
4662        zonelist = &pgdat->node_zonelists[ZONELIST_NOFALLBACK];
4663        j = build_zonelists_node(pgdat, zonelist, 0);
4664        zonelist->_zonerefs[j].zone = NULL;
4665        zonelist->_zonerefs[j].zone_idx = 0;
4666}
4667
4668/*
4669 * Build zonelists ordered by zone and nodes within zones.
4670 * This results in conserving DMA zone[s] until all Normal memory is
4671 * exhausted, but results in overflowing to remote node while memory
4672 * may still exist in local DMA zone.
4673 */
4674static int node_order[MAX_NUMNODES];
4675
4676static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
4677{
4678        int pos, j, node;
4679        int zone_type;          /* needs to be signed */
4680        struct zone *z;
4681        struct zonelist *zonelist;
4682
4683        zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK];
4684        pos = 0;
4685        for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
4686                for (j = 0; j < nr_nodes; j++) {
4687                        node = node_order[j];
4688                        z = &NODE_DATA(node)->node_zones[zone_type];
4689                        if (managed_zone(z)) {
4690                                zoneref_set_zone(z,
4691                                        &zonelist->_zonerefs[pos++]);
4692                                check_highest_zone(zone_type);
4693                        }
4694                }
4695        }
4696        zonelist->_zonerefs[pos].zone = NULL;
4697        zonelist->_zonerefs[pos].zone_idx = 0;
4698}
4699
4700#if defined(CONFIG_64BIT)
4701/*
4702 * Devices that require DMA32/DMA are relatively rare and do not justify a
4703 * penalty to every machine in case the specialised case applies. Default
4704 * to Node-ordering on 64-bit NUMA machines
4705 */
4706static int default_zonelist_order(void)
4707{
4708        return ZONELIST_ORDER_NODE;
4709}
4710#else
4711/*
4712 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4713 * by the kernel. If processes running on node 0 deplete the low memory zone
4714 * then reclaim will occur more frequency increasing stalls and potentially
4715 * be easier to OOM if a large percentage of the zone is under writeback or
4716 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4717 * Hence, default to zone ordering on 32-bit.
4718 */
4719static int default_zonelist_order(void)
4720{
4721        return ZONELIST_ORDER_ZONE;
4722}
4723#endif /* CONFIG_64BIT */
4724
4725static void set_zonelist_order(void)
4726{
4727        if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
4728                current_zonelist_order = default_zonelist_order();
4729        else
4730                current_zonelist_order = user_zonelist_order;
4731}
4732
4733static void build_zonelists(pg_data_t *pgdat)
4734{
4735        int i, node, load;
4736        nodemask_t used_mask;
4737        int local_node, prev_node;
4738        struct zonelist *zonelist;
4739        unsigned int order = current_zonelist_order;
4740
4741        /* initialize zonelists */
4742        for (i = 0; i < MAX_ZONELISTS; i++) {
4743                zonelist = pgdat->node_zonelists + i;
4744                zonelist->_zonerefs[0].zone = NULL;
4745                zonelist->_zonerefs[0].zone_idx = 0;
4746        }
4747
4748        /* NUMA-aware ordering of nodes */
4749        local_node = pgdat->node_id;
4750        load = nr_online_nodes;
4751        prev_node = local_node;
4752        nodes_clear(used_mask);
4753
4754        memset(node_order, 0, sizeof(node_order));
4755        i = 0;
4756
4757        while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
4758                /*
4759                 * We don't want to pressure a particular node.
4760                 * So adding penalty to the first node in same
4761                 * distance group to make it round-robin.
4762                 */
4763                if (node_distance(local_node, node) !=
4764                    node_distance(local_node, prev_node))
4765                        node_load[node] = load;
4766
4767                prev_node = node;
4768                load--;
4769                if (order == ZONELIST_ORDER_NODE)
4770                        build_zonelists_in_node_order(pgdat, node);
4771                else
4772                        node_order[i++] = node; /* remember order */
4773        }
4774
4775        if (order == ZONELIST_ORDER_ZONE) {
4776                /* calculate node order -- i.e., DMA last! */
4777                build_zonelists_in_zone_order(pgdat, i);
4778        }
4779
4780        build_thisnode_zonelists(pgdat);
4781}
4782
4783#ifdef CONFIG_HAVE_MEMORYLESS_NODES
4784/*
4785 * Return node id of node used for "local" allocations.
4786 * I.e., first node id of first zone in arg node's generic zonelist.
4787 * Used for initializing percpu 'numa_mem', which is used primarily
4788 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4789 */
4790int local_memory_node(int node)
4791{
4792        struct zoneref *z;
4793
4794        z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
4795                                   gfp_zone(GFP_KERNEL),
4796                                   NULL);
4797        return z->zone->node;
4798}
4799#endif
4800
4801static void setup_min_unmapped_ratio(void);
4802static void setup_min_slab_ratio(void);
4803#else   /* CONFIG_NUMA */
4804
4805static void set_zonelist_order(void)
4806{
4807        current_zonelist_order = ZONELIST_ORDER_ZONE;
4808}
4809
4810static void build_zonelists(pg_data_t *pgdat)
4811{
4812        int node, local_node;
4813        enum zone_type j;
4814        struct zonelist *zonelist;
4815
4816        local_node = pgdat->node_id;
4817
4818        zonelist = &pgdat->node_zonelists[ZONELIST_FALLBACK];
4819        j = build_zonelists_node(pgdat, zonelist, 0);
4820
4821        /*
4822         * Now we build the zonelist so that it contains the zones
4823         * of all the other nodes.
4824         * We don't want to pressure a particular node, so when
4825         * building the zones for node N, we make sure that the
4826         * zones coming right after the local ones are those from
4827         * node N+1 (modulo N)
4828         */
4829        for (node = local_node + 1; node < MAX_NUMNODES; node++) {
4830                if (!node_online(node))
4831                        continue;
4832                j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4833        }
4834        for (node = 0; node < local_node; node++) {
4835                if (!node_online(node))
4836                        continue;
4837                j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4838        }
4839
4840        zonelist->_zonerefs[j].zone = NULL;
4841        zonelist->_zonerefs[j].zone_idx = 0;
4842}
4843
4844#endif  /* CONFIG_NUMA */
4845
4846/*
4847 * Boot pageset table. One per cpu which is going to be used for all
4848 * zones and all nodes. The parameters will be set in such a way
4849 * that an item put on a list will immediately be handed over to
4850 * the buddy list. This is safe since pageset manipulation is done
4851 * with interrupts disabled.
4852 *
4853 * The boot_pagesets must be kept even after bootup is complete for
4854 * unused processors and/or zones. They do play a role for bootstrapping
4855 * hotplugged processors.
4856 *
4857 * zoneinfo_show() and maybe other functions do
4858 * not check if the processor is online before following the pageset pointer.
4859 * Other parts of the kernel may not check if the zone is available.
4860 */
4861static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
4862static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
4863static void setup_zone_pageset(struct zone *zone);
4864
4865/*
4866 * Global mutex to protect against size modification of zonelists
4867 * as well as to serialize pageset setup for the new populated zone.
4868 */
4869DEFINE_MUTEX(zonelists_mutex);
4870
4871/* return values int ....just for stop_machine() */
4872static int __build_all_zonelists(void *data)
4873{
4874        int nid;
4875        int cpu;
4876        pg_data_t *self = data;
4877
4878#ifdef CONFIG_NUMA
4879        memset(node_load, 0, sizeof(node_load));
4880#endif
4881
4882        if (self && !node_online(self->node_id)) {
4883                build_zonelists(self);
4884        }
4885
4886        for_each_online_node(nid) {
4887                pg_data_t *pgdat = NODE_DATA(nid);
4888
4889                build_zonelists(pgdat);
4890        }
4891
4892        /*
4893         * Initialize the boot_pagesets that are going to be used
4894         * for bootstrapping processors. The real pagesets for
4895         * each zone will be allocated later when the per cpu
4896         * allocator is available.
4897         *
4898         * boot_pagesets are used also for bootstrapping offline
4899         * cpus if the system is already booted because the pagesets
4900         * are needed to initialize allocators on a specific cpu too.
4901         * F.e. the percpu allocator needs the page allocator which
4902         * needs the percpu allocator in order to allocate its pagesets
4903         * (a chicken-egg dilemma).
4904         */
4905        for_each_possible_cpu(cpu) {
4906                setup_pageset(&per_cpu(boot_pageset, cpu), 0);
4907
4908#ifdef CONFIG_HAVE_MEMORYLESS_NODES
4909                /*
4910                 * We now know the "local memory node" for each node--
4911                 * i.e., the node of the first zone in the generic zonelist.
4912                 * Set up numa_mem percpu variable for on-line cpus.  During
4913                 * boot, only the boot cpu should be on-line;  we'll init the
4914                 * secondary cpus' numa_mem as they come on-line.  During
4915                 * node/memory hotplug, we'll fixup all on-line cpus.
4916                 */
4917                if (cpu_online(cpu))
4918                        set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
4919#endif
4920        }
4921
4922        return 0;
4923}
4924
4925static noinline void __init
4926build_all_zonelists_init(void)
4927{
4928        __build_all_zonelists(NULL);
4929        mminit_verify_zonelist();
4930        cpuset_init_current_mems_allowed();
4931}
4932
4933/*
4934 * Called with zonelists_mutex held always
4935 * unless system_state == SYSTEM_BOOTING.
4936 *
4937 * __ref due to (1) call of __meminit annotated setup_zone_pageset
4938 * [we're only called with non-NULL zone through __meminit paths] and
4939 * (2) call of __init annotated helper build_all_zonelists_init
4940 * [protected by SYSTEM_BOOTING].
4941 */
4942void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
4943{
4944        set_zonelist_order();
4945
4946        if (system_state == SYSTEM_BOOTING) {
4947                build_all_zonelists_init();
4948        } else {
4949#ifdef CONFIG_MEMORY_HOTPLUG
4950                if (zone)
4951                        setup_zone_pageset(zone);
4952#endif
4953                /* we have to stop all cpus to guarantee there is no user
4954                   of zonelist */
4955                stop_machine(__build_all_zonelists, pgdat, NULL);
4956                /* cpuset refresh routine should be here */
4957        }
4958        vm_total_pages = nr_free_pagecache_pages();
4959        /*
4960         * Disable grouping by mobility if the number of pages in the
4961         * system is too low to allow the mechanism to work. It would be
4962         * more accurate, but expensive to check per-zone. This check is
4963         * made on memory-hotadd so a system can start with mobility
4964         * disabled and enable it later
4965         */
4966        if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
4967                page_group_by_mobility_disabled = 1;
4968        else
4969                page_group_by_mobility_disabled = 0;
4970
4971        pr_info("Built %i zonelists in %s order, mobility grouping %s.  Total pages: %ld\n",
4972                nr_online_nodes,
4973                zonelist_order_name[current_zonelist_order],
4974                page_group_by_mobility_disabled ? "off" : "on",
4975                vm_total_pages);
4976#ifdef CONFIG_NUMA
4977        pr_info("Policy zone: %s\n", zone_names[policy_zone]);
4978#endif
4979}
4980
4981/*
4982 * Initially all pages are reserved - free ones are freed
4983 * up by free_all_bootmem() once the early boot process is
4984 * done. Non-atomic initialization, single-pass.
4985 */
4986void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
4987                unsigned long start_pfn, enum memmap_context context)
4988{
4989        struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn));
4990        unsigned long end_pfn = start_pfn + size;
4991        pg_data_t *pgdat = NODE_DATA(nid);
4992        unsigned long pfn;
4993        unsigned long nr_initialised = 0;
4994#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4995        struct memblock_region *r = NULL, *tmp;
4996#endif
4997
4998        if (highest_memmap_pfn < end_pfn - 1)
4999                highest_memmap_pfn = end_pfn - 1;
5000
5001        /*
5002         * Honor reservation requested by the driver for this ZONE_DEVICE
5003         * memory
5004         */
5005        if (altmap && start_pfn == altmap->base_pfn)
5006                start_pfn += altmap->reserve;
5007
5008        for (pfn = start_pfn; pfn < end_pfn; pfn++) {
5009                /*
5010                 * There can be holes in boot-time mem_map[]s handed to this
5011                 * function.  They do not exist on hotplugged memory.
5012                 */
5013                if (context != MEMMAP_EARLY)
5014                        goto not_early;
5015
5016                if (!early_pfn_valid(pfn))
5017                        continue;
5018                if (!early_pfn_in_nid(pfn, nid))
5019                        continue;
5020                if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised))
5021                        break;
5022
5023#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5024                /*
5025                 * Check given memblock attribute by firmware which can affect
5026                 * kernel memory layout.  If zone==ZONE_MOVABLE but memory is
5027                 * mirrored, it's an overlapped memmap init. skip it.
5028                 */
5029                if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
5030                        if (!r || pfn >= memblock_region_memory_end_pfn(r)) {
5031                                for_each_memblock(memory, tmp)
5032                                        if (pfn < memblock_region_memory_end_pfn(tmp))
5033                                                break;
5034                                r = tmp;
5035                        }
5036                        if (pfn >= memblock_region_memory_base_pfn(r) &&
5037                            memblock_is_mirror(r)) {
5038                                /* already initialized as NORMAL */
5039                                pfn = memblock_region_memory_end_pfn(r);
5040                                continue;
5041                        }
5042                }
5043#endif
5044
5045not_early:
5046                /*
5047                 * Mark the block movable so that blocks are reserved for
5048                 * movable at startup. This will force kernel allocations
5049                 * to reserve their blocks rather than leaking throughout
5050                 * the address space during boot when many long-lived
5051                 * kernel allocations are made.
5052                 *
5053                 * bitmap is created for zone's valid pfn range. but memmap
5054                 * can be created for invalid pages (for alignment)
5055                 * check here not to call set_pageblock_migratetype() against
5056                 * pfn out of zone.
5057                 */
5058                if (!(pfn & (pageblock_nr_pages - 1))) {
5059                        struct page *page = pfn_to_page(pfn);
5060
5061                        __init_single_page(page, pfn, zone, nid);
5062                        set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5063                } else {
5064                        __init_single_pfn(pfn, zone, nid);
5065                }
5066        }
5067}
5068
5069static void __meminit zone_init_free_lists(struct zone *zone)
5070{
5071        unsigned int order, t;
5072        for_each_migratetype_order(order, t) {
5073                INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
5074                zone->free_area[order].nr_free = 0;
5075        }
5076}
5077
5078#ifndef __HAVE_ARCH_MEMMAP_INIT
5079#define memmap_init(size, nid, zone, start_pfn) \
5080        memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
5081#endif
5082
5083static int zone_batchsize(struct zone *zone)
5084{
5085#ifdef CONFIG_MMU
5086        int batch;
5087
5088        /*
5089         * The per-cpu-pages pools are set to around 1000th of the
5090         * size of the zone.  But no more than 1/2 of a meg.
5091         *
5092         * OK, so we don't know how big the cache is.  So guess.
5093         */
5094        batch = zone->managed_pages / 1024;
5095        if (batch * PAGE_SIZE > 512 * 1024)
5096                batch = (512 * 1024) / PAGE_SIZE;
5097        batch /= 4;             /* We effectively *= 4 below */
5098        if (batch < 1)
5099                batch = 1;
5100
5101        /*
5102         * Clamp the batch to a 2^n - 1 value. Having a power
5103         * of 2 value was found to be more likely to have
5104         * suboptimal cache aliasing properties in some cases.
5105         *
5106         * For example if 2 tasks are alternately allocating
5107         * batches of pages, one task can end up with a lot
5108         * of pages of one half of the possible page colors
5109         * and the other with pages of the other colors.
5110         */
5111        batch = rounddown_pow_of_two(batch + batch/2) - 1;
5112
5113        return batch;
5114
5115#else
5116        /* The deferral and batching of frees should be suppressed under NOMMU
5117         * conditions.
5118         *
5119         * The problem is that NOMMU needs to be able to allocate large chunks
5120         * of contiguous memory as there's no hardware page translation to
5121         * assemble apparent contiguous memory from discontiguous pages.
5122         *
5123         * Queueing large contiguous runs of pages for batching, however,
5124         * causes the pages to actually be freed in smaller chunks.  As there
5125         * can be a significant delay between the individual batches being
5126         * recycled, this leads to the once large chunks of space being
5127         * fragmented and becoming unavailable for high-order allocations.
5128         */
5129        return 0;
5130#endif
5131}
5132
5133/*
5134 * pcp->high and pcp->batch values are related and dependent on one another:
5135 * ->batch must never be higher then ->high.
5136 * The following function updates them in a safe manner without read side
5137 * locking.
5138 *
5139 * Any new users of pcp->batch and pcp->high should ensure they can cope with
5140 * those fields changing asynchronously (acording the the above rule).
5141 *
5142 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
5143 * outside of boot time (or some other assurance that no concurrent updaters
5144 * exist).
5145 */
5146static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
5147                unsigned long batch)
5148{
5149       /* start with a fail safe value for batch */
5150        pcp->batch = 1;
5151        smp_wmb();
5152
5153       /* Update high, then batch, in order */
5154        pcp->high = high;
5155        smp_wmb();
5156
5157        pcp->batch = batch;
5158}
5159
5160/* a companion to pageset_set_high() */
5161static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
5162{
5163        pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
5164}
5165
5166static void pageset_init(struct per_cpu_pageset *p)
5167{
5168        struct per_cpu_pages *pcp;
5169        int migratetype;
5170
5171        memset(p, 0, sizeof(*p));
5172
5173        pcp = &p->pcp;
5174        pcp->count = 0;
5175        for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
5176                INIT_LIST_HEAD(&pcp->lists[migratetype]);
5177}
5178
5179static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
5180{
5181        pageset_init(p);
5182        pageset_set_batch(p, batch);
5183}
5184
5185/*
5186 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
5187 * to the value high for the pageset p.
5188 */
5189static void pageset_set_high(struct per_cpu_pageset *p,
5190                                unsigned long high)
5191{
5192        unsigned long batch = max(1UL, high / 4);
5193        if ((high / 4) > (PAGE_SHIFT * 8))
5194                batch = PAGE_SHIFT * 8;
5195
5196        pageset_update(&p->pcp, high, batch);
5197}
5198
5199static void pageset_set_high_and_batch(struct zone *zone,
5200                                       struct per_cpu_pageset *pcp)
5201{
5202        if (percpu_pagelist_fraction)
5203                pageset_set_high(pcp,
5204                        (zone->managed_pages /
5205                                percpu_pagelist_fraction));
5206        else
5207                pageset_set_batch(pcp, zone_batchsize(zone));
5208}
5209
5210static void __meminit zone_pageset_init(struct zone *zone, int cpu)
5211{
5212        struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
5213
5214        pageset_init(pcp);
5215        pageset_set_high_and_batch(zone, pcp);
5216}
5217
5218static void __meminit setup_zone_pageset(struct zone *zone)
5219{
5220        int cpu;
5221        zone->pageset = alloc_percpu(struct per_cpu_pageset);
5222        for_each_possible_cpu(cpu)
5223                zone_pageset_init(zone, cpu);
5224}
5225
5226/*
5227 * Allocate per cpu pagesets and initialize them.
5228 * Before this call only boot pagesets were available.
5229 */
5230void __init setup_per_cpu_pageset(void)
5231{
5232        struct pglist_data *pgdat;
5233        struct zone *zone;
5234
5235        for_each_populated_zone(zone)
5236                setup_zone_pageset(zone);
5237
5238        for_each_online_pgdat(pgdat)
5239                pgdat->per_cpu_nodestats =
5240                        alloc_percpu(struct per_cpu_nodestat);
5241}
5242
5243static __meminit void zone_pcp_init(struct zone *zone)
5244{
5245        /*
5246         * per cpu subsystem is not up at this point. The following code
5247         * relies on the ability of the linker to provide the
5248         * offset of a (static) per cpu variable into the per cpu area.
5249         */
5250        zone->pageset = &boot_pageset;
5251
5252        if (populated_zone(zone))
5253                printk(KERN_DEBUG "  %s zone: %lu pages, LIFO batch:%u\n",
5254                        zone->name, zone->present_pages,
5255                                         zone_batchsize(zone));
5256}
5257
5258int __meminit init_currently_empty_zone(struct zone *zone,
5259                                        unsigned long zone_start_pfn,
5260                                        unsigned long size)
5261{
5262        struct pglist_data *pgdat = zone->zone_pgdat;
5263
5264        pgdat->nr_zones = zone_idx(zone) + 1;
5265
5266        zone->zone_start_pfn = zone_start_pfn;
5267
5268        mminit_dprintk(MMINIT_TRACE, "memmap_init",
5269                        "Initialising map node %d zone %lu pfns %lu -> %lu\n",
5270                        pgdat->node_id,
5271                        (unsigned long)zone_idx(zone),
5272                        zone_start_pfn, (zone_start_pfn + size));
5273
5274        zone_init_free_lists(zone);
5275        zone->initialized = 1;
5276
5277        return 0;
5278}
5279
5280#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5281#ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
5282
5283/*
5284 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
5285 */
5286int __meminit __early_pfn_to_nid(unsigned long pfn,
5287                                        struct mminit_pfnnid_cache *state)
5288{
5289        unsigned long start_pfn, end_pfn;
5290        int nid;
5291
5292        if (state->last_start <= pfn && pfn < state->last_end)
5293                return state->last_nid;
5294
5295        nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
5296        if (nid != -1) {
5297                state->last_start = start_pfn;
5298                state->last_end = end_pfn;
5299                state->last_nid = nid;
5300        }
5301
5302        return nid;
5303}
5304#endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
5305
5306/**
5307 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
5308 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
5309 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
5310 *
5311 * If an architecture guarantees that all ranges registered contain no holes
5312 * and may be freed, this this function may be used instead of calling
5313 * memblock_free_early_nid() manually.
5314 */
5315void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
5316{
5317        unsigned long start_pfn, end_pfn;
5318        int i, this_nid;
5319
5320        for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
5321                start_pfn = min(start_pfn, max_low_pfn);
5322                end_pfn = min(end_pfn, max_low_pfn);
5323
5324                if (start_pfn < end_pfn)
5325                        memblock_free_early_nid(PFN_PHYS(start_pfn),
5326                                        (end_pfn - start_pfn) << PAGE_SHIFT,
5327                                        this_nid);
5328        }
5329}
5330
5331/**
5332 * sparse_memory_present_with_active_regions - Call memory_present for each active range
5333 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
5334 *
5335 * If an architecture guarantees that all ranges registered contain no holes and may
5336 * be freed, this function may be used instead of calling memory_present() manually.
5337 */
5338void __init sparse_memory_present_with_active_regions(int nid)
5339{
5340        unsigned long start_pfn, end_pfn;
5341        int i, this_nid;
5342
5343        for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
5344                memory_present(this_nid, start_pfn, end_pfn);
5345}
5346
5347/**
5348 * get_pfn_range_for_nid - Return the start and end page frames for a node
5349 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
5350 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
5351 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
5352 *
5353 * It returns the start and end page frame of a node based on information
5354 * provided by memblock_set_node(). If called for a node
5355 * with no available memory, a warning is printed and the start and end
5356 * PFNs will be 0.
5357 */
5358void __meminit get_pfn_range_for_nid(unsigned int nid,
5359                        unsigned long *start_pfn, unsigned long *end_pfn)
5360{
5361        unsigned long this_start_pfn, this_end_pfn;
5362        int i;
5363
5364        *start_pfn = -1UL;
5365        *end_pfn = 0;
5366
5367        for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
5368                *start_pfn = min(*start_pfn, this_start_pfn);
5369                *end_pfn = max(*end_pfn, this_end_pfn);
5370        }
5371
5372        if (*start_pfn == -1UL)
5373                *start_pfn = 0;
5374}
5375
5376/*
5377 * This finds a zone that can be used for ZONE_MOVABLE pages. The
5378 * assumption is made that zones within a node are ordered in monotonic
5379 * increasing memory addresses so that the "highest" populated zone is used
5380 */
5381static void __init find_usable_zone_for_movable(void)
5382{
5383        int zone_index;
5384        for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
5385                if (zone_index == ZONE_MOVABLE)
5386                        continue;
5387
5388                if (arch_zone_highest_possible_pfn[zone_index] >
5389                                arch_zone_lowest_possible_pfn[zone_index])
5390                        break;
5391        }
5392
5393        VM_BUG_ON(zone_index == -1);
5394        movable_zone = zone_index;
5395}
5396
5397/*
5398 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
5399 * because it is sized independent of architecture. Unlike the other zones,
5400 * the starting point for ZONE_MOVABLE is not fixed. It may be different
5401 * in each node depending on the size of each node and how evenly kernelcore
5402 * is distributed. This helper function adjusts the zone ranges
5403 * provided by the architecture for a given node by using the end of the
5404 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5405 * zones within a node are in order of monotonic increases memory addresses
5406 */
5407static void __meminit adjust_zone_range_for_zone_movable(int nid,
5408                                        unsigned long zone_type,
5409                                        unsigned long node_start_pfn,
5410                                        unsigned long node_end_pfn,
5411                                        unsigned long *zone_start_pfn,
5412                                        unsigned long *zone_end_pfn)
5413{
5414        /* Only adjust if ZONE_MOVABLE is on this node */
5415        if (zone_movable_pfn[nid]) {
5416                /* Size ZONE_MOVABLE */
5417                if (zone_type == ZONE_MOVABLE) {
5418                        *zone_start_pfn = zone_movable_pfn[nid];
5419                        *zone_end_pfn = min(node_end_pfn,
5420                                arch_zone_highest_possible_pfn[movable_zone]);
5421
5422                /* Adjust for ZONE_MOVABLE starting within this range */
5423                } else if (!mirrored_kernelcore &&
5424                        *zone_start_pfn < zone_movable_pfn[nid] &&
5425                        *zone_end_pfn > zone_movable_pfn[nid]) {
5426                        *zone_end_pfn = zone_movable_pfn[nid];
5427
5428                /* Check if this whole range is within ZONE_MOVABLE */
5429                } else if (*zone_start_pfn >= zone_movable_pfn[nid])
5430                        *zone_start_pfn = *zone_end_pfn;
5431        }
5432}
5433
5434/*
5435 * Return the number of pages a zone spans in a node, including holes
5436 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5437 */
5438static unsigned long __meminit zone_spanned_pages_in_node(int nid,
5439                                        unsigned long zone_type,
5440                                        unsigned long node_start_pfn,
5441                                        unsigned long node_end_pfn,
5442                                        unsigned long *zone_start_pfn,
5443                                        unsigned long *zone_end_pfn,
5444                                        unsigned long *ignored)
5445{
5446        /* When hotadd a new node from cpu_up(), the node should be empty */
5447        if (!node_start_pfn && !node_end_pfn)
5448                return 0;
5449
5450        /* Get the start and end of the zone */
5451        *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
5452        *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
5453        adjust_zone_range_for_zone_movable(nid, zone_type,
5454                                node_start_pfn, node_end_pfn,
5455                                zone_start_pfn, zone_end_pfn);
5456
5457        /* Check that this node has pages within the zone's required range */
5458        if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
5459                return 0;
5460
5461        /* Move the zone boundaries inside the node if necessary */
5462        *zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
5463        *zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
5464
5465        /* Return the spanned pages */
5466        return *zone_end_pfn - *zone_start_pfn;
5467}
5468
5469/*
5470 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
5471 * then all holes in the requested range will be accounted for.
5472 */
5473unsigned long __meminit __absent_pages_in_range(int nid,
5474                                unsigned long range_start_pfn,
5475                                unsigned long range_end_pfn)
5476{
5477        unsigned long nr_absent = range_end_pfn - range_start_pfn;
5478        unsigned long start_pfn, end_pfn;
5479        int i;
5480
5481        for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5482                start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
5483                end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
5484                nr_absent -= end_pfn - start_pfn;
5485        }
5486        return nr_absent;
5487}
5488
5489/**
5490 * absent_pages_in_range - Return number of page frames in holes within a range
5491 * @start_pfn: The start PFN to start searching for holes
5492 * @end_pfn: The end PFN to stop searching for holes
5493 *
5494 * It returns the number of pages frames in memory holes within a range.
5495 */
5496unsigned long __init absent_pages_in_range(unsigned long start_pfn,
5497                                                        unsigned long end_pfn)
5498{
5499        return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
5500}
5501
5502/* Return the number of page frames in holes in a zone on a node */
5503static unsigned long __meminit zone_absent_pages_in_node(int nid,
5504                                        unsigned long zone_type,
5505                                        unsigned long node_start_pfn,
5506                                        unsigned long node_end_pfn,
5507                                        unsigned long *ignored)
5508{
5509        unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
5510        unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
5511        unsigned long zone_start_pfn, zone_end_pfn;
5512        unsigned long nr_absent;
5513
5514        /* When hotadd a new node from cpu_up(), the node should be empty */
5515        if (!node_start_pfn && !node_end_pfn)
5516                return 0;
5517
5518        zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
5519        zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
5520
5521        adjust_zone_range_for_zone_movable(nid, zone_type,
5522                        node_start_pfn, node_end_pfn,
5523                        &zone_start_pfn, &zone_end_pfn);
5524        nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
5525
5526        /*
5527         * ZONE_MOVABLE handling.
5528         * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
5529         * and vice versa.
5530         */
5531        if (mirrored_kernelcore && zone_movable_pfn[nid]) {
5532                unsigned long start_pfn, end_pfn;
5533                struct memblock_region *r;
5534
5535                for_each_memblock(memory, r) {
5536                        start_pfn = clamp(memblock_region_memory_base_pfn(r),
5537                                          zone_start_pfn, zone_end_pfn);
5538                        end_pfn = clamp(memblock_region_memory_end_pfn(r),
5539                                        zone_start_pfn, zone_end_pfn);
5540
5541                        if (zone_type == ZONE_MOVABLE &&
5542                            memblock_is_mirror(r))
5543                                nr_absent += end_pfn - start_pfn;
5544
5545                        if (zone_type == ZONE_NORMAL &&
5546                            !memblock_is_mirror(r))
5547                                nr_absent += end_pfn - start_pfn;
5548                }
5549        }
5550
5551        return nr_absent;
5552}
5553
5554#else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5555static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
5556                                        unsigned long zone_type,
5557                                        unsigned long node_start_pfn,
5558                                        unsigned long node_end_pfn,
5559                                        unsigned long *zone_start_pfn,
5560                                        unsigned long *zone_end_pfn,
5561                                        unsigned long *zones_size)
5562{
5563        unsigned int zone;
5564
5565        *zone_start_pfn = node_start_pfn;
5566        for (zone = 0; zone < zone_type; zone++)
5567                *zone_start_pfn += zones_size[zone];
5568
5569        *zone_end_pfn = *zone_start_pfn + zones_size[zone_type];
5570
5571        return zones_size[zone_type];
5572}
5573
5574static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
5575                                                unsigned long zone_type,
5576                                                unsigned long node_start_pfn,
5577                                                unsigned long node_end_pfn,
5578                                                unsigned long *zholes_size)
5579{
5580        if (!zholes_size)
5581                return 0;
5582
5583        return zholes_size[zone_type];
5584}
5585
5586#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5587
5588static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
5589                                                unsigned long node_start_pfn,
5590                                                unsigned long node_end_pfn,
5591                                                unsigned long *zones_size,
5592                                                unsigned long *zholes_size)
5593{
5594        unsigned long realtotalpages = 0, totalpages = 0;
5595        enum zone_type i;
5596
5597        for (i = 0; i < MAX_NR_ZONES; i++) {
5598                struct zone *zone = pgdat->node_zones + i;
5599                unsigned long zone_start_pfn, zone_end_pfn;
5600                unsigned long size, real_size;
5601
5602                size = zone_spanned_pages_in_node(pgdat->node_id, i,
5603                                                  node_start_pfn,
5604                                                  node_end_pfn,
5605                                                  &zone_start_pfn,
5606                                                  &zone_end_pfn,
5607                                                  zones_size);
5608                real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
5609                                                  node_start_pfn, node_end_pfn,
5610                                                  zholes_size);
5611                if (size)
5612                        zone->zone_start_pfn = zone_start_pfn;
5613                else
5614                        zone->zone_start_pfn = 0;
5615                zone->spanned_pages = size;
5616                zone->present_pages = real_size;
5617
5618                totalpages += size;
5619                realtotalpages += real_size;
5620        }
5621
5622        pgdat->node_spanned_pages = totalpages;
5623        pgdat->node_present_pages = realtotalpages;
5624        printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
5625                                                        realtotalpages);
5626}
5627
5628#ifndef CONFIG_SPARSEMEM
5629/*
5630 * Calculate the size of the zone->blockflags rounded to an unsigned long
5631 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5632 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
5633 * round what is now in bits to nearest long in bits, then return it in
5634 * bytes.
5635 */
5636static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
5637{
5638        unsigned long usemapsize;
5639
5640        zonesize += zone_start_pfn & (pageblock_nr_pages-1);
5641        usemapsize = roundup(zonesize, pageblock_nr_pages);
5642        usemapsize = usemapsize >> pageblock_order;
5643        usemapsize *= NR_PAGEBLOCK_BITS;
5644        usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
5645
5646        return usemapsize / 8;
5647}
5648
5649static void __init setup_usemap(struct pglist_data *pgdat,
5650                                struct zone *zone,
5651                                unsigned long zone_start_pfn,
5652                                unsigned long zonesize)
5653{
5654        unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
5655        zone->pageblock_flags = NULL;
5656        if (usemapsize)
5657                zone->pageblock_flags =
5658                        memblock_virt_alloc_node_nopanic(usemapsize,
5659                                                         pgdat->node_id);
5660}
5661#else
5662static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
5663                                unsigned long zone_start_pfn, unsigned long zonesize) {}
5664#endif /* CONFIG_SPARSEMEM */
5665
5666#ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
5667
5668/* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
5669void __paginginit set_pageblock_order(void)
5670{
5671        unsigned int order;
5672
5673        /* Check that pageblock_nr_pages has not already been setup */
5674        if (pageblock_order)
5675                return;
5676
5677        if (HPAGE_SHIFT > PAGE_SHIFT)
5678                order = HUGETLB_PAGE_ORDER;
5679        else
5680                order = MAX_ORDER - 1;
5681
5682        /*
5683         * Assume the largest contiguous order of interest is a huge page.
5684         * This value may be variable depending on boot parameters on IA64 and
5685         * powerpc.
5686         */
5687        pageblock_order = order;
5688}
5689#else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5690
5691/*
5692 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
5693 * is unused as pageblock_order is set at compile-time. See
5694 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5695 * the kernel config
5696 */
5697void __paginginit set_pageblock_order(void)
5698{
5699}
5700
5701#endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5702
5703static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
5704                                                   unsigned long present_pages)
5705{
5706        unsigned long pages = spanned_pages;
5707
5708        /*
5709         * Provide a more accurate estimation if there are holes within
5710         * the zone and SPARSEMEM is in use. If there are holes within the
5711         * zone, each populated memory region may cost us one or two extra
5712         * memmap pages due to alignment because memmap pages for each
5713         * populated regions may not naturally algined on page boundary.
5714         * So the (present_pages >> 4) heuristic is a tradeoff for that.
5715         */
5716        if (spanned_pages > present_pages + (present_pages >> 4) &&
5717            IS_ENABLED(CONFIG_SPARSEMEM))
5718                pages = present_pages;
5719
5720        return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
5721}
5722
5723/*
5724 * Set up the zone data structures:
5725 *   - mark all pages reserved
5726 *   - mark all memory queues empty
5727 *   - clear the memory bitmaps
5728 *
5729 * NOTE: pgdat should get zeroed by caller.
5730 */
5731static void __paginginit free_area_init_core(struct pglist_data *pgdat)
5732{
5733        enum zone_type j;
5734        int nid = pgdat->node_id;
5735        int ret;
5736
5737        pgdat_resize_init(pgdat);
5738#ifdef CONFIG_NUMA_BALANCING
5739        spin_lock_init(&pgdat->numabalancing_migrate_lock);
5740        pgdat->numabalancing_migrate_nr_pages = 0;
5741        pgdat->numabalancing_migrate_next_window = jiffies;
5742#endif
5743#ifdef CONFIG_TRANSPARENT_HUGEPAGE
5744        spin_lock_init(&pgdat->split_queue_lock);
5745        INIT_LIST_HEAD(&pgdat->split_queue);
5746        pgdat->split_queue_len = 0;
5747#endif
5748        init_waitqueue_head(&pgdat->kswapd_wait);
5749        init_waitqueue_head(&pgdat->pfmemalloc_wait);
5750#ifdef CONFIG_COMPACTION
5751        init_waitqueue_head(&pgdat->kcompactd_wait);
5752#endif
5753        pgdat_page_ext_init(pgdat);
5754        spin_lock_init(&pgdat->lru_lock);
5755        lruvec_init(node_lruvec(pgdat));
5756
5757        for (j = 0; j < MAX_NR_ZONES; j++) {
5758                struct zone *zone = pgdat->node_zones + j;
5759                unsigned long size, realsize, freesize, memmap_pages;
5760                unsigned long zone_start_pfn = zone->zone_start_pfn;
5761
5762                size = zone->spanned_pages;
5763                realsize = freesize = zone->present_pages;
5764
5765                /*
5766                 * Adjust freesize so that it accounts for how much memory
5767                 * is used by this zone for memmap. This affects the watermark
5768                 * and per-cpu initialisations
5769                 */
5770                memmap_pages = calc_memmap_size(size, realsize);
5771                if (!is_highmem_idx(j)) {
5772                        if (freesize >= memmap_pages) {
5773                                freesize -= memmap_pages;
5774                                if (memmap_pages)
5775                                        printk(KERN_DEBUG
5776                                               "  %s zone: %lu pages used for memmap\n",
5777                                               zone_names[j], memmap_pages);
5778                        } else
5779                                pr_warn("  %s zone: %lu pages exceeds freesize %lu\n",
5780                                        zone_names[j], memmap_pages, freesize);
5781                }
5782
5783                /* Account for reserved pages */
5784                if (j == 0 && freesize > dma_reserve) {
5785                        freesize -= dma_reserve;
5786                        printk(KERN_DEBUG "  %s zone: %lu pages reserved\n",
5787                                        zone_names[0], dma_reserve);
5788                }
5789
5790                if (!is_highmem_idx(j))
5791                        nr_kernel_pages += freesize;
5792                /* Charge for highmem memmap if there are enough kernel pages */
5793                else if (nr_kernel_pages > memmap_pages * 2)
5794                        nr_kernel_pages -= memmap_pages;
5795                nr_all_pages += freesize;
5796
5797                /*
5798                 * Set an approximate value for lowmem here, it will be adjusted
5799                 * when the bootmem allocator frees pages into the buddy system.
5800                 * And all highmem pages will be managed by the buddy system.
5801                 */
5802                zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
5803#ifdef CONFIG_NUMA
5804                zone->node = nid;
5805#endif
5806                zone->name = zone_names[j];
5807                zone->zone_pgdat = pgdat;
5808                spin_lock_init(&zone->lock);
5809                zone_seqlock_init(zone);
5810                zone_pcp_init(zone);
5811
5812                if (!size)
5813                        continue;
5814
5815                set_pageblock_order();
5816                setup_usemap(pgdat, zone, zone_start_pfn, size);
5817                ret = init_currently_empty_zone(zone, zone_start_pfn, size);
5818                BUG_ON(ret);
5819                memmap_init(size, nid, j, zone_start_pfn);
5820        }
5821}
5822
5823static void __ref alloc_node_mem_map(struct pglist_data *pgdat)
5824{
5825        unsigned long __maybe_unused start = 0;
5826        unsigned long __maybe_unused offset = 0;
5827
5828        /* Skip empty nodes */
5829        if (!pgdat->node_spanned_pages)
5830                return;
5831
5832#ifdef CONFIG_FLAT_NODE_MEM_MAP
5833        start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
5834        offset = pgdat->node_start_pfn - start;
5835        /* ia64 gets its own node_mem_map, before this, without bootmem */
5836        if (!pgdat->node_mem_map) {
5837                unsigned long size, end;
5838                struct page *map;
5839
5840                /*
5841                 * The zone's endpoints aren't required to be MAX_ORDER
5842                 * aligned but the node_mem_map endpoints must be in order
5843                 * for the buddy allocator to function correctly.
5844                 */
5845                end = pgdat_end_pfn(pgdat);
5846                end = ALIGN(end, MAX_ORDER_NR_PAGES);
5847                size =  (end - start) * sizeof(struct page);
5848                map = alloc_remap(pgdat->node_id, size);
5849                if (!map)
5850                        map = memblock_virt_alloc_node_nopanic(size,
5851                                                               pgdat->node_id);
5852                pgdat->node_mem_map = map + offset;
5853        }
5854#ifndef CONFIG_NEED_MULTIPLE_NODES
5855        /*
5856         * With no DISCONTIG, the global mem_map is just set as node 0's
5857         */
5858        if (pgdat == NODE_DATA(0)) {
5859                mem_map = NODE_DATA(0)->node_mem_map;
5860#if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
5861                if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
5862                        mem_map -= offset;
5863#endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5864        }
5865#endif
5866#endif /* CONFIG_FLAT_NODE_MEM_MAP */
5867}
5868
5869void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
5870                unsigned long node_start_pfn, unsigned long *zholes_size)
5871{
5872        pg_data_t *pgdat = NODE_DATA(nid);
5873        unsigned long start_pfn = 0;
5874        unsigned long end_pfn = 0;
5875
5876        /* pg_data_t should be reset to zero when it's allocated */
5877        WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx);
5878
5879        reset_deferred_meminit(pgdat);
5880        pgdat->node_id = nid;
5881        pgdat->node_start_pfn = node_start_pfn;
5882        pgdat->per_cpu_nodestats = NULL;
5883#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5884        get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
5885        pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
5886                (u64)start_pfn << PAGE_SHIFT,
5887                end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
5888#else
5889        start_pfn = node_start_pfn;
5890#endif
5891        calculate_node_totalpages(pgdat, start_pfn, end_pfn,
5892                                  zones_size, zholes_size);
5893
5894        alloc_node_mem_map(pgdat);
5895#ifdef CONFIG_FLAT_NODE_MEM_MAP
5896        printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5897                nid, (unsigned long)pgdat,
5898                (unsigned long)pgdat->node_mem_map);
5899#endif
5900
5901        free_area_init_core(pgdat);
5902}
5903
5904#ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5905
5906#if MAX_NUMNODES > 1
5907/*
5908 * Figure out the number of possible node ids.
5909 */
5910void __init setup_nr_node_ids(void)
5911{
5912        unsigned int highest;
5913
5914        highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
5915        nr_node_ids = highest + 1;
5916}
5917#endif
5918
5919/**
5920 * node_map_pfn_alignment - determine the maximum internode alignment
5921 *
5922 * This function should be called after node map is populated and sorted.
5923 * It calculates the maximum power of two alignment which can distinguish
5924 * all the nodes.
5925 *
5926 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5927 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)).  If the
5928 * nodes are shifted by 256MiB, 256MiB.  Note that if only the last node is
5929 * shifted, 1GiB is enough and this function will indicate so.
5930 *
5931 * This is used to test whether pfn -> nid mapping of the chosen memory
5932 * model has fine enough granularity to avoid incorrect mapping for the
5933 * populated node map.
5934 *
5935 * Returns the determined alignment in pfn's.  0 if there is no alignment
5936 * requirement (single node).
5937 */
5938unsigned long __init node_map_pfn_alignment(void)
5939{
5940        unsigned long accl_mask = 0, last_end = 0;
5941        unsigned long start, end, mask;
5942        int last_nid = -1;
5943        int i, nid;
5944
5945        for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
5946                if (!start || last_nid < 0 || last_nid == nid) {
5947                        last_nid = nid;
5948                        last_end = end;
5949                        continue;
5950                }
5951
5952                /*
5953                 * Start with a mask granular enough to pin-point to the
5954                 * start pfn and tick off bits one-by-one until it becomes
5955                 * too coarse to separate the current node from the last.
5956                 */
5957                mask = ~((1 << __ffs(start)) - 1);
5958                while (mask && last_end <= (start & (mask << 1)))
5959                        mask <<= 1;
5960
5961                /* accumulate all internode masks */
5962                accl_mask |= mask;
5963        }
5964
5965        /* convert mask to number of pages */
5966        return ~accl_mask + 1;
5967}
5968
5969/* Find the lowest pfn for a node */
5970static unsigned long __init find_min_pfn_for_node(int nid)
5971{
5972        unsigned long min_pfn = ULONG_MAX;
5973        unsigned long start_pfn;
5974        int i;
5975
5976        for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5977                min_pfn = min(min_pfn, start_pfn);
5978
5979        if (min_pfn == ULONG_MAX) {
5980                pr_warn("Could not find start_pfn for node %d\n", nid);
5981                return 0;
5982        }
5983
5984        return min_pfn;
5985}
5986
5987/**
5988 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5989 *
5990 * It returns the minimum PFN based on information provided via
5991 * memblock_set_node().
5992 */
5993unsigned long __init find_min_pfn_with_active_regions(void)
5994{
5995        return find_min_pfn_for_node(MAX_NUMNODES);
5996}
5997
5998/*
5999 * early_calculate_totalpages()
6000 * Sum pages in active regions for movable zone.