2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
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)
17 #include <linux/stddef.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/kasan.h>
28 #include <linux/module.h>
29 #include <linux/suspend.h>
30 #include <linux/pagevec.h>
31 #include <linux/blkdev.h>
32 #include <linux/slab.h>
33 #include <linux/ratelimit.h>
34 #include <linux/oom.h>
35 #include <linux/notifier.h>
36 #include <linux/topology.h>
37 #include <linux/sysctl.h>
38 #include <linux/cpu.h>
39 #include <linux/cpuset.h>
40 #include <linux/memory_hotplug.h>
41 #include <linux/nodemask.h>
42 #include <linux/vmalloc.h>
43 #include <linux/vmstat.h>
44 #include <linux/mempolicy.h>
45 #include <linux/memremap.h>
46 #include <linux/stop_machine.h>
47 #include <linux/sort.h>
48 #include <linux/pfn.h>
49 #include <linux/backing-dev.h>
50 #include <linux/fault-inject.h>
51 #include <linux/page-isolation.h>
52 #include <linux/page_ext.h>
53 #include <linux/debugobjects.h>
54 #include <linux/kmemleak.h>
55 #include <linux/compaction.h>
56 #include <trace/events/kmem.h>
57 #include <trace/events/oom.h>
58 #include <linux/prefetch.h>
59 #include <linux/mm_inline.h>
60 #include <linux/migrate.h>
61 #include <linux/hugetlb.h>
62 #include <linux/sched/rt.h>
63 #include <linux/sched/mm.h>
64 #include <linux/page_owner.h>
65 #include <linux/kthread.h>
66 #include <linux/memcontrol.h>
67 #include <linux/ftrace.h>
68 #include <linux/lockdep.h>
69 #include <linux/nmi.h>
71 #include <asm/sections.h>
72 #include <asm/tlbflush.h>
73 #include <asm/div64.h>
76 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
77 static DEFINE_MUTEX(pcp_batch_high_lock);
78 #define MIN_PERCPU_PAGELIST_FRACTION (8)
80 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
81 DEFINE_PER_CPU(int, numa_node);
82 EXPORT_PER_CPU_SYMBOL(numa_node);
85 DEFINE_STATIC_KEY_TRUE(vm_numa_stat_key);
87 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
89 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
90 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
91 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
92 * defined in <linux/topology.h>.
94 DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
95 EXPORT_PER_CPU_SYMBOL(_numa_mem_);
96 int _node_numa_mem_[MAX_NUMNODES];
99 /* work_structs for global per-cpu drains */
100 DEFINE_MUTEX(pcpu_drain_mutex);
101 DEFINE_PER_CPU(struct work_struct, pcpu_drain);
103 #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY
104 volatile unsigned long latent_entropy __latent_entropy;
105 EXPORT_SYMBOL(latent_entropy);
109 * Array of node states.
111 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
112 [N_POSSIBLE] = NODE_MASK_ALL,
113 [N_ONLINE] = { { [0] = 1UL } },
115 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
116 #ifdef CONFIG_HIGHMEM
117 [N_HIGH_MEMORY] = { { [0] = 1UL } },
119 [N_MEMORY] = { { [0] = 1UL } },
120 [N_CPU] = { { [0] = 1UL } },
123 EXPORT_SYMBOL(node_states);
125 /* Protect totalram_pages and zone->managed_pages */
126 static DEFINE_SPINLOCK(managed_page_count_lock);
128 unsigned long totalram_pages __read_mostly;
129 unsigned long totalreserve_pages __read_mostly;
130 unsigned long totalcma_pages __read_mostly;
132 int percpu_pagelist_fraction;
133 gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
136 * A cached value of the page's pageblock's migratetype, used when the page is
137 * put on a pcplist. Used to avoid the pageblock migratetype lookup when
138 * freeing from pcplists in most cases, at the cost of possibly becoming stale.
139 * Also the migratetype set in the page does not necessarily match the pcplist
140 * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
141 * other index - this ensures that it will be put on the correct CMA freelist.
143 static inline int get_pcppage_migratetype(struct page *page)
148 static inline void set_pcppage_migratetype(struct page *page, int migratetype)
150 page->index = migratetype;
153 #ifdef CONFIG_PM_SLEEP
155 * The following functions are used by the suspend/hibernate code to temporarily
156 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
157 * while devices are suspended. To avoid races with the suspend/hibernate code,
158 * they should always be called with pm_mutex held (gfp_allowed_mask also should
159 * only be modified with pm_mutex held, unless the suspend/hibernate code is
160 * guaranteed not to run in parallel with that modification).
163 static gfp_t saved_gfp_mask;
165 void pm_restore_gfp_mask(void)
167 WARN_ON(!mutex_is_locked(&pm_mutex));
168 if (saved_gfp_mask) {
169 gfp_allowed_mask = saved_gfp_mask;
174 void pm_restrict_gfp_mask(void)
176 WARN_ON(!mutex_is_locked(&pm_mutex));
177 WARN_ON(saved_gfp_mask);
178 saved_gfp_mask = gfp_allowed_mask;
179 gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS);
182 bool pm_suspended_storage(void)
184 if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS))
188 #endif /* CONFIG_PM_SLEEP */
190 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
191 unsigned int pageblock_order __read_mostly;
194 static void __free_pages_ok(struct page *page, unsigned int order);
197 * results with 256, 32 in the lowmem_reserve sysctl:
198 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
199 * 1G machine -> (16M dma, 784M normal, 224M high)
200 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
201 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
202 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
204 * TBD: should special case ZONE_DMA32 machines here - in those we normally
205 * don't need any ZONE_NORMAL reservation
207 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES] = {
208 #ifdef CONFIG_ZONE_DMA
211 #ifdef CONFIG_ZONE_DMA32
215 #ifdef CONFIG_HIGHMEM
221 EXPORT_SYMBOL(totalram_pages);
223 static char * const zone_names[MAX_NR_ZONES] = {
224 #ifdef CONFIG_ZONE_DMA
227 #ifdef CONFIG_ZONE_DMA32
231 #ifdef CONFIG_HIGHMEM
235 #ifdef CONFIG_ZONE_DEVICE
240 char * const migratetype_names[MIGRATE_TYPES] = {
248 #ifdef CONFIG_MEMORY_ISOLATION
253 compound_page_dtor * const compound_page_dtors[] = {
256 #ifdef CONFIG_HUGETLB_PAGE
259 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
264 int min_free_kbytes = 1024;
265 int user_min_free_kbytes = -1;
266 int watermark_scale_factor = 10;
268 static unsigned long nr_kernel_pages __meminitdata;
269 static unsigned long nr_all_pages __meminitdata;
270 static unsigned long dma_reserve __meminitdata;
272 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
273 static unsigned long arch_zone_lowest_possible_pfn[MAX_NR_ZONES] __meminitdata;
274 static unsigned long arch_zone_highest_possible_pfn[MAX_NR_ZONES] __meminitdata;
275 static unsigned long required_kernelcore __initdata;
276 static unsigned long required_kernelcore_percent __initdata;
277 static unsigned long required_movablecore __initdata;
278 static unsigned long required_movablecore_percent __initdata;
279 static unsigned long zone_movable_pfn[MAX_NUMNODES] __meminitdata;
280 static bool mirrored_kernelcore __meminitdata;
282 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
284 EXPORT_SYMBOL(movable_zone);
285 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
288 int nr_node_ids __read_mostly = MAX_NUMNODES;
289 int nr_online_nodes __read_mostly = 1;
290 EXPORT_SYMBOL(nr_node_ids);
291 EXPORT_SYMBOL(nr_online_nodes);
294 int page_group_by_mobility_disabled __read_mostly;
296 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
297 /* Returns true if the struct page for the pfn is uninitialised */
298 static inline bool __meminit early_page_uninitialised(unsigned long pfn)
300 int nid = early_pfn_to_nid(pfn);
302 if (node_online(nid) && pfn >= NODE_DATA(nid)->first_deferred_pfn)
309 * Returns false when the remaining initialisation should be deferred until
310 * later in the boot cycle when it can be parallelised.
312 static inline bool update_defer_init(pg_data_t *pgdat,
313 unsigned long pfn, unsigned long zone_end,
314 unsigned long *nr_initialised)
316 /* Always populate low zones for address-constrained allocations */
317 if (zone_end < pgdat_end_pfn(pgdat))
320 if ((*nr_initialised > pgdat->static_init_pgcnt) &&
321 (pfn & (PAGES_PER_SECTION - 1)) == 0) {
322 pgdat->first_deferred_pfn = pfn;
329 static inline bool early_page_uninitialised(unsigned long pfn)
334 static inline bool update_defer_init(pg_data_t *pgdat,
335 unsigned long pfn, unsigned long zone_end,
336 unsigned long *nr_initialised)
342 /* Return a pointer to the bitmap storing bits affecting a block of pages */
343 static inline unsigned long *get_pageblock_bitmap(struct page *page,
346 #ifdef CONFIG_SPARSEMEM
347 return __pfn_to_section(pfn)->pageblock_flags;
349 return page_zone(page)->pageblock_flags;
350 #endif /* CONFIG_SPARSEMEM */
353 static inline int pfn_to_bitidx(struct page *page, unsigned long pfn)
355 #ifdef CONFIG_SPARSEMEM
356 pfn &= (PAGES_PER_SECTION-1);
357 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
359 pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages);
360 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
361 #endif /* CONFIG_SPARSEMEM */
365 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
366 * @page: The page within the block of interest
367 * @pfn: The target page frame number
368 * @end_bitidx: The last bit of interest to retrieve
369 * @mask: mask of bits that the caller is interested in
371 * Return: pageblock_bits flags
373 static __always_inline unsigned long __get_pfnblock_flags_mask(struct page *page,
375 unsigned long end_bitidx,
378 unsigned long *bitmap;
379 unsigned long bitidx, word_bitidx;
382 bitmap = get_pageblock_bitmap(page, pfn);
383 bitidx = pfn_to_bitidx(page, pfn);
384 word_bitidx = bitidx / BITS_PER_LONG;
385 bitidx &= (BITS_PER_LONG-1);
387 word = bitmap[word_bitidx];
388 bitidx += end_bitidx;
389 return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
392 unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
393 unsigned long end_bitidx,
396 return __get_pfnblock_flags_mask(page, pfn, end_bitidx, mask);
399 static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned long pfn)
401 return __get_pfnblock_flags_mask(page, pfn, PB_migrate_end, MIGRATETYPE_MASK);
405 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
406 * @page: The page within the block of interest
407 * @flags: The flags to set
408 * @pfn: The target page frame number
409 * @end_bitidx: The last bit of interest
410 * @mask: mask of bits that the caller is interested in
412 void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
414 unsigned long end_bitidx,
417 unsigned long *bitmap;
418 unsigned long bitidx, word_bitidx;
419 unsigned long old_word, word;
421 BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
423 bitmap = get_pageblock_bitmap(page, pfn);
424 bitidx = pfn_to_bitidx(page, pfn);
425 word_bitidx = bitidx / BITS_PER_LONG;
426 bitidx &= (BITS_PER_LONG-1);
428 VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page);
430 bitidx += end_bitidx;
431 mask <<= (BITS_PER_LONG - bitidx - 1);
432 flags <<= (BITS_PER_LONG - bitidx - 1);
434 word = READ_ONCE(bitmap[word_bitidx]);
436 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
437 if (word == old_word)
443 void set_pageblock_migratetype(struct page *page, int migratetype)
445 if (unlikely(page_group_by_mobility_disabled &&
446 migratetype < MIGRATE_PCPTYPES))
447 migratetype = MIGRATE_UNMOVABLE;
449 set_pageblock_flags_group(page, (unsigned long)migratetype,
450 PB_migrate, PB_migrate_end);
453 #ifdef CONFIG_DEBUG_VM
454 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
458 unsigned long pfn = page_to_pfn(page);
459 unsigned long sp, start_pfn;
462 seq = zone_span_seqbegin(zone);
463 start_pfn = zone->zone_start_pfn;
464 sp = zone->spanned_pages;
465 if (!zone_spans_pfn(zone, pfn))
467 } while (zone_span_seqretry(zone, seq));
470 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
471 pfn, zone_to_nid(zone), zone->name,
472 start_pfn, start_pfn + sp);
477 static int page_is_consistent(struct zone *zone, struct page *page)
479 if (!pfn_valid_within(page_to_pfn(page)))
481 if (zone != page_zone(page))
487 * Temporary debugging check for pages not lying within a given zone.
489 static int __maybe_unused bad_range(struct zone *zone, struct page *page)
491 if (page_outside_zone_boundaries(zone, page))
493 if (!page_is_consistent(zone, page))
499 static inline int __maybe_unused bad_range(struct zone *zone, struct page *page)
505 static void bad_page(struct page *page, const char *reason,
506 unsigned long bad_flags)
508 static unsigned long resume;
509 static unsigned long nr_shown;
510 static unsigned long nr_unshown;
513 * Allow a burst of 60 reports, then keep quiet for that minute;
514 * or allow a steady drip of one report per second.
516 if (nr_shown == 60) {
517 if (time_before(jiffies, resume)) {
523 "BUG: Bad page state: %lu messages suppressed\n",
530 resume = jiffies + 60 * HZ;
532 pr_alert("BUG: Bad page state in process %s pfn:%05lx\n",
533 current->comm, page_to_pfn(page));
534 __dump_page(page, reason);
535 bad_flags &= page->flags;
537 pr_alert("bad because of flags: %#lx(%pGp)\n",
538 bad_flags, &bad_flags);
539 dump_page_owner(page);
544 /* Leave bad fields for debug, except PageBuddy could make trouble */
545 page_mapcount_reset(page); /* remove PageBuddy */
546 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
550 * Higher-order pages are called "compound pages". They are structured thusly:
552 * The first PAGE_SIZE page is called the "head page" and have PG_head set.
554 * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
555 * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
557 * The first tail page's ->compound_dtor holds the offset in array of compound
558 * page destructors. See compound_page_dtors.
560 * The first tail page's ->compound_order holds the order of allocation.
561 * This usage means that zero-order pages may not be compound.
564 void free_compound_page(struct page *page)
566 __free_pages_ok(page, compound_order(page));
569 void prep_compound_page(struct page *page, unsigned int order)
572 int nr_pages = 1 << order;
574 set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
575 set_compound_order(page, order);
577 for (i = 1; i < nr_pages; i++) {
578 struct page *p = page + i;
579 set_page_count(p, 0);
580 p->mapping = TAIL_MAPPING;
581 set_compound_head(p, page);
583 atomic_set(compound_mapcount_ptr(page), -1);
586 #ifdef CONFIG_DEBUG_PAGEALLOC
587 unsigned int _debug_guardpage_minorder;
588 bool _debug_pagealloc_enabled __read_mostly
589 = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT);
590 EXPORT_SYMBOL(_debug_pagealloc_enabled);
591 bool _debug_guardpage_enabled __read_mostly;
593 static int __init early_debug_pagealloc(char *buf)
597 return kstrtobool(buf, &_debug_pagealloc_enabled);
599 early_param("debug_pagealloc", early_debug_pagealloc);
601 static bool need_debug_guardpage(void)
603 /* If we don't use debug_pagealloc, we don't need guard page */
604 if (!debug_pagealloc_enabled())
607 if (!debug_guardpage_minorder())
613 static void init_debug_guardpage(void)
615 if (!debug_pagealloc_enabled())
618 if (!debug_guardpage_minorder())
621 _debug_guardpage_enabled = true;
624 struct page_ext_operations debug_guardpage_ops = {
625 .need = need_debug_guardpage,
626 .init = init_debug_guardpage,
629 static int __init debug_guardpage_minorder_setup(char *buf)
633 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
634 pr_err("Bad debug_guardpage_minorder value\n");
637 _debug_guardpage_minorder = res;
638 pr_info("Setting debug_guardpage_minorder to %lu\n", res);
641 early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup);
643 static inline bool set_page_guard(struct zone *zone, struct page *page,
644 unsigned int order, int migratetype)
646 struct page_ext *page_ext;
648 if (!debug_guardpage_enabled())
651 if (order >= debug_guardpage_minorder())
654 page_ext = lookup_page_ext(page);
655 if (unlikely(!page_ext))
658 __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
660 INIT_LIST_HEAD(&page->lru);
661 set_page_private(page, order);
662 /* Guard pages are not available for any usage */
663 __mod_zone_freepage_state(zone, -(1 << order), migratetype);
668 static inline void clear_page_guard(struct zone *zone, struct page *page,
669 unsigned int order, int migratetype)
671 struct page_ext *page_ext;
673 if (!debug_guardpage_enabled())
676 page_ext = lookup_page_ext(page);
677 if (unlikely(!page_ext))
680 __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
682 set_page_private(page, 0);
683 if (!is_migrate_isolate(migratetype))
684 __mod_zone_freepage_state(zone, (1 << order), migratetype);
687 struct page_ext_operations debug_guardpage_ops;
688 static inline bool set_page_guard(struct zone *zone, struct page *page,
689 unsigned int order, int migratetype) { return false; }
690 static inline void clear_page_guard(struct zone *zone, struct page *page,
691 unsigned int order, int migratetype) {}
694 static inline void set_page_order(struct page *page, unsigned int order)
696 set_page_private(page, order);
697 __SetPageBuddy(page);
700 static inline void rmv_page_order(struct page *page)
702 __ClearPageBuddy(page);
703 set_page_private(page, 0);
707 * This function checks whether a page is free && is the buddy
708 * we can do coalesce a page and its buddy if
709 * (a) the buddy is not in a hole (check before calling!) &&
710 * (b) the buddy is in the buddy system &&
711 * (c) a page and its buddy have the same order &&
712 * (d) a page and its buddy are in the same zone.
714 * For recording whether a page is in the buddy system, we set ->_mapcount
715 * PAGE_BUDDY_MAPCOUNT_VALUE.
716 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
717 * serialized by zone->lock.
719 * For recording page's order, we use page_private(page).
721 static inline int page_is_buddy(struct page *page, struct page *buddy,
724 if (page_is_guard(buddy) && page_order(buddy) == order) {
725 if (page_zone_id(page) != page_zone_id(buddy))
728 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
733 if (PageBuddy(buddy) && page_order(buddy) == order) {
735 * zone check is done late to avoid uselessly
736 * calculating zone/node ids for pages that could
739 if (page_zone_id(page) != page_zone_id(buddy))
742 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
750 * Freeing function for a buddy system allocator.
752 * The concept of a buddy system is to maintain direct-mapped table
753 * (containing bit values) for memory blocks of various "orders".
754 * The bottom level table contains the map for the smallest allocatable
755 * units of memory (here, pages), and each level above it describes
756 * pairs of units from the levels below, hence, "buddies".
757 * At a high level, all that happens here is marking the table entry
758 * at the bottom level available, and propagating the changes upward
759 * as necessary, plus some accounting needed to play nicely with other
760 * parts of the VM system.
761 * At each level, we keep a list of pages, which are heads of continuous
762 * free pages of length of (1 << order) and marked with _mapcount
763 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
765 * So when we are allocating or freeing one, we can derive the state of the
766 * other. That is, if we allocate a small block, and both were
767 * free, the remainder of the region must be split into blocks.
768 * If a block is freed, and its buddy is also free, then this
769 * triggers coalescing into a block of larger size.
774 static inline void __free_one_page(struct page *page,
776 struct zone *zone, unsigned int order,
779 unsigned long combined_pfn;
780 unsigned long uninitialized_var(buddy_pfn);
782 unsigned int max_order;
784 max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1);
786 VM_BUG_ON(!zone_is_initialized(zone));
787 VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
789 VM_BUG_ON(migratetype == -1);
790 if (likely(!is_migrate_isolate(migratetype)))
791 __mod_zone_freepage_state(zone, 1 << order, migratetype);
793 VM_BUG_ON_PAGE(pfn & ((1 << order) - 1), page);
794 VM_BUG_ON_PAGE(bad_range(zone, page), page);
797 while (order < max_order - 1) {
798 buddy_pfn = __find_buddy_pfn(pfn, order);
799 buddy = page + (buddy_pfn - pfn);
801 if (!pfn_valid_within(buddy_pfn))
803 if (!page_is_buddy(page, buddy, order))
806 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
807 * merge with it and move up one order.
809 if (page_is_guard(buddy)) {
810 clear_page_guard(zone, buddy, order, migratetype);
812 list_del(&buddy->lru);
813 zone->free_area[order].nr_free--;
814 rmv_page_order(buddy);
816 combined_pfn = buddy_pfn & pfn;
817 page = page + (combined_pfn - pfn);
821 if (max_order < MAX_ORDER) {
822 /* If we are here, it means order is >= pageblock_order.
823 * We want to prevent merge between freepages on isolate
824 * pageblock and normal pageblock. Without this, pageblock
825 * isolation could cause incorrect freepage or CMA accounting.
827 * We don't want to hit this code for the more frequent
830 if (unlikely(has_isolate_pageblock(zone))) {
833 buddy_pfn = __find_buddy_pfn(pfn, order);
834 buddy = page + (buddy_pfn - pfn);
835 buddy_mt = get_pageblock_migratetype(buddy);
837 if (migratetype != buddy_mt
838 && (is_migrate_isolate(migratetype) ||
839 is_migrate_isolate(buddy_mt)))
843 goto continue_merging;
847 set_page_order(page, order);
850 * If this is not the largest possible page, check if the buddy
851 * of the next-highest order is free. If it is, it's possible
852 * that pages are being freed that will coalesce soon. In case,
853 * that is happening, add the free page to the tail of the list
854 * so it's less likely to be used soon and more likely to be merged
855 * as a higher order page
857 if ((order < MAX_ORDER-2) && pfn_valid_within(buddy_pfn)) {
858 struct page *higher_page, *higher_buddy;
859 combined_pfn = buddy_pfn & pfn;
860 higher_page = page + (combined_pfn - pfn);
861 buddy_pfn = __find_buddy_pfn(combined_pfn, order + 1);
862 higher_buddy = higher_page + (buddy_pfn - combined_pfn);
863 if (pfn_valid_within(buddy_pfn) &&
864 page_is_buddy(higher_page, higher_buddy, order + 1)) {
865 list_add_tail(&page->lru,
866 &zone->free_area[order].free_list[migratetype]);
871 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
873 zone->free_area[order].nr_free++;
877 * A bad page could be due to a number of fields. Instead of multiple branches,
878 * try and check multiple fields with one check. The caller must do a detailed
879 * check if necessary.
881 static inline bool page_expected_state(struct page *page,
882 unsigned long check_flags)
884 if (unlikely(atomic_read(&page->_mapcount) != -1))
887 if (unlikely((unsigned long)page->mapping |
888 page_ref_count(page) |
890 (unsigned long)page->mem_cgroup |
892 (page->flags & check_flags)))
898 static void free_pages_check_bad(struct page *page)
900 const char *bad_reason;
901 unsigned long bad_flags;
906 if (unlikely(atomic_read(&page->_mapcount) != -1))
907 bad_reason = "nonzero mapcount";
908 if (unlikely(page->mapping != NULL))
909 bad_reason = "non-NULL mapping";
910 if (unlikely(page_ref_count(page) != 0))
911 bad_reason = "nonzero _refcount";
912 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
913 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
914 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
917 if (unlikely(page->mem_cgroup))
918 bad_reason = "page still charged to cgroup";
920 bad_page(page, bad_reason, bad_flags);
923 static inline int free_pages_check(struct page *page)
925 if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE)))
928 /* Something has gone sideways, find it */
929 free_pages_check_bad(page);
933 static int free_tail_pages_check(struct page *head_page, struct page *page)
938 * We rely page->lru.next never has bit 0 set, unless the page
939 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
941 BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
943 if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
947 switch (page - head_page) {
949 /* the first tail page: ->mapping is compound_mapcount() */
950 if (unlikely(compound_mapcount(page))) {
951 bad_page(page, "nonzero compound_mapcount", 0);
957 * the second tail page: ->mapping is
958 * page_deferred_list().next -- ignore value.
962 if (page->mapping != TAIL_MAPPING) {
963 bad_page(page, "corrupted mapping in tail page", 0);
968 if (unlikely(!PageTail(page))) {
969 bad_page(page, "PageTail not set", 0);
972 if (unlikely(compound_head(page) != head_page)) {
973 bad_page(page, "compound_head not consistent", 0);
978 page->mapping = NULL;
979 clear_compound_head(page);
983 static __always_inline bool free_pages_prepare(struct page *page,
984 unsigned int order, bool check_free)
988 VM_BUG_ON_PAGE(PageTail(page), page);
990 trace_mm_page_free(page, order);
993 * Check tail pages before head page information is cleared to
994 * avoid checking PageCompound for order-0 pages.
996 if (unlikely(order)) {
997 bool compound = PageCompound(page);
1000 VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
1003 ClearPageDoubleMap(page);
1004 for (i = 1; i < (1 << order); i++) {
1006 bad += free_tail_pages_check(page, page + i);
1007 if (unlikely(free_pages_check(page + i))) {
1011 (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1014 if (PageMappingFlags(page))
1015 page->mapping = NULL;
1016 if (memcg_kmem_enabled() && PageKmemcg(page))
1017 memcg_kmem_uncharge(page, order);
1019 bad += free_pages_check(page);
1023 page_cpupid_reset_last(page);
1024 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1025 reset_page_owner(page, order);
1027 if (!PageHighMem(page)) {
1028 debug_check_no_locks_freed(page_address(page),
1029 PAGE_SIZE << order);
1030 debug_check_no_obj_freed(page_address(page),
1031 PAGE_SIZE << order);
1033 arch_free_page(page, order);
1034 kernel_poison_pages(page, 1 << order, 0);
1035 kernel_map_pages(page, 1 << order, 0);
1036 kasan_free_pages(page, order);
1041 #ifdef CONFIG_DEBUG_VM
1042 static inline bool free_pcp_prepare(struct page *page)
1044 return free_pages_prepare(page, 0, true);
1047 static inline bool bulkfree_pcp_prepare(struct page *page)
1052 static bool free_pcp_prepare(struct page *page)
1054 return free_pages_prepare(page, 0, false);
1057 static bool bulkfree_pcp_prepare(struct page *page)
1059 return free_pages_check(page);
1061 #endif /* CONFIG_DEBUG_VM */
1063 static inline void prefetch_buddy(struct page *page)
1065 unsigned long pfn = page_to_pfn(page);
1066 unsigned long buddy_pfn = __find_buddy_pfn(pfn, 0);
1067 struct page *buddy = page + (buddy_pfn - pfn);
1073 * Frees a number of pages from the PCP lists
1074 * Assumes all pages on list are in same zone, and of same order.
1075 * count is the number of pages to free.
1077 * If the zone was previously in an "all pages pinned" state then look to
1078 * see if this freeing clears that state.
1080 * And clear the zone's pages_scanned counter, to hold off the "all pages are
1081 * pinned" detection logic.
1083 static void free_pcppages_bulk(struct zone *zone, int count,
1084 struct per_cpu_pages *pcp)
1086 int migratetype = 0;
1088 int prefetch_nr = 0;
1089 bool isolated_pageblocks;
1090 struct page *page, *tmp;
1094 struct list_head *list;
1097 * Remove pages from lists in a round-robin fashion. A
1098 * batch_free count is maintained that is incremented when an
1099 * empty list is encountered. This is so more pages are freed
1100 * off fuller lists instead of spinning excessively around empty
1105 if (++migratetype == MIGRATE_PCPTYPES)
1107 list = &pcp->lists[migratetype];
1108 } while (list_empty(list));
1110 /* This is the only non-empty list. Free them all. */
1111 if (batch_free == MIGRATE_PCPTYPES)
1115 page = list_last_entry(list, struct page, lru);
1116 /* must delete to avoid corrupting pcp list */
1117 list_del(&page->lru);
1120 if (bulkfree_pcp_prepare(page))
1123 list_add_tail(&page->lru, &head);
1126 * We are going to put the page back to the global
1127 * pool, prefetch its buddy to speed up later access
1128 * under zone->lock. It is believed the overhead of
1129 * an additional test and calculating buddy_pfn here
1130 * can be offset by reduced memory latency later. To
1131 * avoid excessive prefetching due to large count, only
1132 * prefetch buddy for the first pcp->batch nr of pages.
1134 if (prefetch_nr++ < pcp->batch)
1135 prefetch_buddy(page);
1136 } while (--count && --batch_free && !list_empty(list));
1139 spin_lock(&zone->lock);
1140 isolated_pageblocks = has_isolate_pageblock(zone);
1143 * Use safe version since after __free_one_page(),
1144 * page->lru.next will not point to original list.
1146 list_for_each_entry_safe(page, tmp, &head, lru) {
1147 int mt = get_pcppage_migratetype(page);
1148 /* MIGRATE_ISOLATE page should not go to pcplists */
1149 VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
1150 /* Pageblock could have been isolated meanwhile */
1151 if (unlikely(isolated_pageblocks))
1152 mt = get_pageblock_migratetype(page);
1154 __free_one_page(page, page_to_pfn(page), zone, 0, mt);
1155 trace_mm_page_pcpu_drain(page, 0, mt);
1157 spin_unlock(&zone->lock);
1160 static void free_one_page(struct zone *zone,
1161 struct page *page, unsigned long pfn,
1165 spin_lock(&zone->lock);
1166 if (unlikely(has_isolate_pageblock(zone) ||
1167 is_migrate_isolate(migratetype))) {
1168 migratetype = get_pfnblock_migratetype(page, pfn);
1170 __free_one_page(page, pfn, zone, order, migratetype);
1171 spin_unlock(&zone->lock);
1174 static void __meminit __init_single_page(struct page *page, unsigned long pfn,
1175 unsigned long zone, int nid)
1177 mm_zero_struct_page(page);
1178 set_page_links(page, zone, nid, pfn);
1179 init_page_count(page);
1180 page_mapcount_reset(page);
1181 page_cpupid_reset_last(page);
1183 INIT_LIST_HEAD(&page->lru);
1184 #ifdef WANT_PAGE_VIRTUAL
1185 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1186 if (!is_highmem_idx(zone))
1187 set_page_address(page, __va(pfn << PAGE_SHIFT));
1191 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1192 static void __meminit init_reserved_page(unsigned long pfn)
1197 if (!early_page_uninitialised(pfn))
1200 nid = early_pfn_to_nid(pfn);
1201 pgdat = NODE_DATA(nid);
1203 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1204 struct zone *zone = &pgdat->node_zones[zid];
1206 if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
1209 __init_single_page(pfn_to_page(pfn), pfn, zid, nid);
1212 static inline void init_reserved_page(unsigned long pfn)
1215 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1218 * Initialised pages do not have PageReserved set. This function is
1219 * called for each range allocated by the bootmem allocator and
1220 * marks the pages PageReserved. The remaining valid pages are later
1221 * sent to the buddy page allocator.
1223 void __meminit reserve_bootmem_region(phys_addr_t start, phys_addr_t end)
1225 unsigned long start_pfn = PFN_DOWN(start);
1226 unsigned long end_pfn = PFN_UP(end);
1228 for (; start_pfn < end_pfn; start_pfn++) {
1229 if (pfn_valid(start_pfn)) {
1230 struct page *page = pfn_to_page(start_pfn);
1232 init_reserved_page(start_pfn);
1234 /* Avoid false-positive PageTail() */
1235 INIT_LIST_HEAD(&page->lru);
1237 SetPageReserved(page);
1242 static void __free_pages_ok(struct page *page, unsigned int order)
1244 unsigned long flags;
1246 unsigned long pfn = page_to_pfn(page);
1248 if (!free_pages_prepare(page, order, true))
1251 migratetype = get_pfnblock_migratetype(page, pfn);
1252 local_irq_save(flags);
1253 __count_vm_events(PGFREE, 1 << order);
1254 free_one_page(page_zone(page), page, pfn, order, migratetype);
1255 local_irq_restore(flags);
1258 static void __init __free_pages_boot_core(struct page *page, unsigned int order)
1260 unsigned int nr_pages = 1 << order;
1261 struct page *p = page;
1265 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1267 __ClearPageReserved(p);
1268 set_page_count(p, 0);
1270 __ClearPageReserved(p);
1271 set_page_count(p, 0);
1273 page_zone(page)->managed_pages += nr_pages;
1274 set_page_refcounted(page);
1275 __free_pages(page, order);
1278 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1279 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1281 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1283 int __meminit early_pfn_to_nid(unsigned long pfn)
1285 static DEFINE_SPINLOCK(early_pfn_lock);
1288 spin_lock(&early_pfn_lock);
1289 nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
1291 nid = first_online_node;
1292 spin_unlock(&early_pfn_lock);
1298 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1299 static inline bool __meminit __maybe_unused
1300 meminit_pfn_in_nid(unsigned long pfn, int node,
1301 struct mminit_pfnnid_cache *state)
1305 nid = __early_pfn_to_nid(pfn, state);
1306 if (nid >= 0 && nid != node)
1311 /* Only safe to use early in boot when initialisation is single-threaded */
1312 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1314 return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1319 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1323 static inline bool __meminit __maybe_unused
1324 meminit_pfn_in_nid(unsigned long pfn, int node,
1325 struct mminit_pfnnid_cache *state)
1332 void __init __free_pages_bootmem(struct page *page, unsigned long pfn,
1335 if (early_page_uninitialised(pfn))
1337 return __free_pages_boot_core(page, order);
1341 * Check that the whole (or subset of) a pageblock given by the interval of
1342 * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1343 * with the migration of free compaction scanner. The scanners then need to
1344 * use only pfn_valid_within() check for arches that allow holes within
1347 * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1349 * It's possible on some configurations to have a setup like node0 node1 node0
1350 * i.e. it's possible that all pages within a zones range of pages do not
1351 * belong to a single zone. We assume that a border between node0 and node1
1352 * can occur within a single pageblock, but not a node0 node1 node0
1353 * interleaving within a single pageblock. It is therefore sufficient to check
1354 * the first and last page of a pageblock and avoid checking each individual
1355 * page in a pageblock.
1357 struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
1358 unsigned long end_pfn, struct zone *zone)
1360 struct page *start_page;
1361 struct page *end_page;
1363 /* end_pfn is one past the range we are checking */
1366 if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
1369 start_page = pfn_to_online_page(start_pfn);
1373 if (page_zone(start_page) != zone)
1376 end_page = pfn_to_page(end_pfn);
1378 /* This gives a shorter code than deriving page_zone(end_page) */
1379 if (page_zone_id(start_page) != page_zone_id(end_page))
1385 void set_zone_contiguous(struct zone *zone)
1387 unsigned long block_start_pfn = zone->zone_start_pfn;
1388 unsigned long block_end_pfn;
1390 block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages);
1391 for (; block_start_pfn < zone_end_pfn(zone);
1392 block_start_pfn = block_end_pfn,
1393 block_end_pfn += pageblock_nr_pages) {
1395 block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
1397 if (!__pageblock_pfn_to_page(block_start_pfn,
1398 block_end_pfn, zone))
1402 /* We confirm that there is no hole */
1403 zone->contiguous = true;
1406 void clear_zone_contiguous(struct zone *zone)
1408 zone->contiguous = false;
1411 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1412 static void __init deferred_free_range(unsigned long pfn,
1413 unsigned long nr_pages)
1421 page = pfn_to_page(pfn);
1423 /* Free a large naturally-aligned chunk if possible */
1424 if (nr_pages == pageblock_nr_pages &&
1425 (pfn & (pageblock_nr_pages - 1)) == 0) {
1426 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1427 __free_pages_boot_core(page, pageblock_order);
1431 for (i = 0; i < nr_pages; i++, page++, pfn++) {
1432 if ((pfn & (pageblock_nr_pages - 1)) == 0)
1433 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1434 __free_pages_boot_core(page, 0);
1438 /* Completion tracking for deferred_init_memmap() threads */
1439 static atomic_t pgdat_init_n_undone __initdata;
1440 static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1442 static inline void __init pgdat_init_report_one_done(void)
1444 if (atomic_dec_and_test(&pgdat_init_n_undone))
1445 complete(&pgdat_init_all_done_comp);
1449 * Returns true if page needs to be initialized or freed to buddy allocator.
1451 * First we check if pfn is valid on architectures where it is possible to have
1452 * holes within pageblock_nr_pages. On systems where it is not possible, this
1453 * function is optimized out.
1455 * Then, we check if a current large page is valid by only checking the validity
1458 * Finally, meminit_pfn_in_nid is checked on systems where pfns can interleave
1459 * within a node: a pfn is between start and end of a node, but does not belong
1460 * to this memory node.
1462 static inline bool __init
1463 deferred_pfn_valid(int nid, unsigned long pfn,
1464 struct mminit_pfnnid_cache *nid_init_state)
1466 if (!pfn_valid_within(pfn))
1468 if (!(pfn & (pageblock_nr_pages - 1)) && !pfn_valid(pfn))
1470 if (!meminit_pfn_in_nid(pfn, nid, nid_init_state))
1476 * Free pages to buddy allocator. Try to free aligned pages in
1477 * pageblock_nr_pages sizes.
1479 static void __init deferred_free_pages(int nid, int zid, unsigned long pfn,
1480 unsigned long end_pfn)
1482 struct mminit_pfnnid_cache nid_init_state = { };
1483 unsigned long nr_pgmask = pageblock_nr_pages - 1;
1484 unsigned long nr_free = 0;
1486 for (; pfn < end_pfn; pfn++) {
1487 if (!deferred_pfn_valid(nid, pfn, &nid_init_state)) {
1488 deferred_free_range(pfn - nr_free, nr_free);
1490 } else if (!(pfn & nr_pgmask)) {
1491 deferred_free_range(pfn - nr_free, nr_free);
1493 touch_nmi_watchdog();
1498 /* Free the last block of pages to allocator */
1499 deferred_free_range(pfn - nr_free, nr_free);
1503 * Initialize struct pages. We minimize pfn page lookups and scheduler checks
1504 * by performing it only once every pageblock_nr_pages.
1505 * Return number of pages initialized.
1507 static unsigned long __init deferred_init_pages(int nid, int zid,
1509 unsigned long end_pfn)
1511 struct mminit_pfnnid_cache nid_init_state = { };
1512 unsigned long nr_pgmask = pageblock_nr_pages - 1;
1513 unsigned long nr_pages = 0;
1514 struct page *page = NULL;
1516 for (; pfn < end_pfn; pfn++) {
1517 if (!deferred_pfn_valid(nid, pfn, &nid_init_state)) {
1520 } else if (!page || !(pfn & nr_pgmask)) {
1521 page = pfn_to_page(pfn);
1522 touch_nmi_watchdog();
1526 __init_single_page(page, pfn, zid, nid);
1532 /* Initialise remaining memory on a node */
1533 static int __init deferred_init_memmap(void *data)
1535 pg_data_t *pgdat = data;
1536 int nid = pgdat->node_id;
1537 unsigned long start = jiffies;
1538 unsigned long nr_pages = 0;
1539 unsigned long spfn, epfn, first_init_pfn, flags;
1540 phys_addr_t spa, epa;
1543 const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
1546 /* Bind memory initialisation thread to a local node if possible */
1547 if (!cpumask_empty(cpumask))
1548 set_cpus_allowed_ptr(current, cpumask);
1550 pgdat_resize_lock(pgdat, &flags);
1551 first_init_pfn = pgdat->first_deferred_pfn;
1552 if (first_init_pfn == ULONG_MAX) {
1553 pgdat_resize_unlock(pgdat, &flags);
1554 pgdat_init_report_one_done();
1558 /* Sanity check boundaries */
1559 BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1560 BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1561 pgdat->first_deferred_pfn = ULONG_MAX;
1563 /* Only the highest zone is deferred so find it */
1564 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1565 zone = pgdat->node_zones + zid;
1566 if (first_init_pfn < zone_end_pfn(zone))
1569 first_init_pfn = max(zone->zone_start_pfn, first_init_pfn);
1572 * Initialize and free pages. We do it in two loops: first we initialize
1573 * struct page, than free to buddy allocator, because while we are
1574 * freeing pages we can access pages that are ahead (computing buddy
1575 * page in __free_one_page()).
1577 for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) {
1578 spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa));
1579 epfn = min_t(unsigned long, zone_end_pfn(zone), PFN_DOWN(epa));
1580 nr_pages += deferred_init_pages(nid, zid, spfn, epfn);
1582 for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) {
1583 spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa));
1584 epfn = min_t(unsigned long, zone_end_pfn(zone), PFN_DOWN(epa));
1585 deferred_free_pages(nid, zid, spfn, epfn);
1587 pgdat_resize_unlock(pgdat, &flags);
1589 /* Sanity check that the next zone really is unpopulated */
1590 WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1592 pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
1593 jiffies_to_msecs(jiffies - start));
1595 pgdat_init_report_one_done();
1600 * During boot we initialize deferred pages on-demand, as needed, but once
1601 * page_alloc_init_late() has finished, the deferred pages are all initialized,
1602 * and we can permanently disable that path.
1604 static DEFINE_STATIC_KEY_TRUE(deferred_pages);
1607 * If this zone has deferred pages, try to grow it by initializing enough
1608 * deferred pages to satisfy the allocation specified by order, rounded up to
1609 * the nearest PAGES_PER_SECTION boundary. So we're adding memory in increments
1610 * of SECTION_SIZE bytes by initializing struct pages in increments of
1611 * PAGES_PER_SECTION * sizeof(struct page) bytes.
1613 * Return true when zone was grown, otherwise return false. We return true even
1614 * when we grow less than requested, to let the caller decide if there are
1615 * enough pages to satisfy the allocation.
1617 * Note: We use noinline because this function is needed only during boot, and
1618 * it is called from a __ref function _deferred_grow_zone. This way we are
1619 * making sure that it is not inlined into permanent text section.
1621 static noinline bool __init
1622 deferred_grow_zone(struct zone *zone, unsigned int order)
1624 int zid = zone_idx(zone);
1625 int nid = zone_to_nid(zone);
1626 pg_data_t *pgdat = NODE_DATA(nid);
1627 unsigned long nr_pages_needed = ALIGN(1 << order, PAGES_PER_SECTION);
1628 unsigned long nr_pages = 0;
1629 unsigned long first_init_pfn, spfn, epfn, t, flags;
1630 unsigned long first_deferred_pfn = pgdat->first_deferred_pfn;
1631 phys_addr_t spa, epa;
1634 /* Only the last zone may have deferred pages */
1635 if (zone_end_pfn(zone) != pgdat_end_pfn(pgdat))
1638 pgdat_resize_lock(pgdat, &flags);
1641 * If deferred pages have been initialized while we were waiting for
1642 * the lock, return true, as the zone was grown. The caller will retry
1643 * this zone. We won't return to this function since the caller also
1644 * has this static branch.
1646 if (!static_branch_unlikely(&deferred_pages)) {
1647 pgdat_resize_unlock(pgdat, &flags);
1652 * If someone grew this zone while we were waiting for spinlock, return
1653 * true, as there might be enough pages already.
1655 if (first_deferred_pfn != pgdat->first_deferred_pfn) {
1656 pgdat_resize_unlock(pgdat, &flags);
1660 first_init_pfn = max(zone->zone_start_pfn, first_deferred_pfn);
1662 if (first_init_pfn >= pgdat_end_pfn(pgdat)) {
1663 pgdat_resize_unlock(pgdat, &flags);
1667 for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) {
1668 spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa));
1669 epfn = min_t(unsigned long, zone_end_pfn(zone), PFN_DOWN(epa));
1671 while (spfn < epfn && nr_pages < nr_pages_needed) {
1672 t = ALIGN(spfn + PAGES_PER_SECTION, PAGES_PER_SECTION);
1673 first_deferred_pfn = min(t, epfn);
1674 nr_pages += deferred_init_pages(nid, zid, spfn,
1675 first_deferred_pfn);
1676 spfn = first_deferred_pfn;
1679 if (nr_pages >= nr_pages_needed)
1683 for_each_free_mem_range(i, nid, MEMBLOCK_NONE, &spa, &epa, NULL) {
1684 spfn = max_t(unsigned long, first_init_pfn, PFN_UP(spa));
1685 epfn = min_t(unsigned long, first_deferred_pfn, PFN_DOWN(epa));
1686 deferred_free_pages(nid, zid, spfn, epfn);
1688 if (first_deferred_pfn == epfn)
1691 pgdat->first_deferred_pfn = first_deferred_pfn;
1692 pgdat_resize_unlock(pgdat, &flags);
1694 return nr_pages > 0;
1698 * deferred_grow_zone() is __init, but it is called from
1699 * get_page_from_freelist() during early boot until deferred_pages permanently
1700 * disables this call. This is why we have refdata wrapper to avoid warning,
1701 * and to ensure that the function body gets unloaded.
1704 _deferred_grow_zone(struct zone *zone, unsigned int order)
1706 return deferred_grow_zone(zone, order);
1709 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1711 void __init page_alloc_init_late(void)
1715 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1718 /* There will be num_node_state(N_MEMORY) threads */
1719 atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
1720 for_each_node_state(nid, N_MEMORY) {
1721 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1724 /* Block until all are initialised */
1725 wait_for_completion(&pgdat_init_all_done_comp);
1728 * We initialized the rest of the deferred pages. Permanently disable
1729 * on-demand struct page initialization.
1731 static_branch_disable(&deferred_pages);
1733 /* Reinit limits that are based on free pages after the kernel is up */
1734 files_maxfiles_init();
1736 #ifdef CONFIG_ARCH_DISCARD_MEMBLOCK
1737 /* Discard memblock private memory */
1741 for_each_populated_zone(zone)
1742 set_zone_contiguous(zone);
1746 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1747 void __init init_cma_reserved_pageblock(struct page *page)
1749 unsigned i = pageblock_nr_pages;
1750 struct page *p = page;
1753 __ClearPageReserved(p);
1754 set_page_count(p, 0);
1757 set_pageblock_migratetype(page, MIGRATE_CMA);
1759 if (pageblock_order >= MAX_ORDER) {
1760 i = pageblock_nr_pages;
1763 set_page_refcounted(p);
1764 __free_pages(p, MAX_ORDER - 1);
1765 p += MAX_ORDER_NR_PAGES;
1766 } while (i -= MAX_ORDER_NR_PAGES);
1768 set_page_refcounted(page);
1769 __free_pages(page, pageblock_order);
1772 adjust_managed_page_count(page, pageblock_nr_pages);
1777 * The order of subdivision here is critical for the IO subsystem.
1778 * Please do not alter this order without good reasons and regression
1779 * testing. Specifically, as large blocks of memory are subdivided,
1780 * the order in which smaller blocks are delivered depends on the order
1781 * they're subdivided in this function. This is the primary factor
1782 * influencing the order in which pages are delivered to the IO
1783 * subsystem according to empirical testing, and this is also justified
1784 * by considering the behavior of a buddy system containing a single
1785 * large block of memory acted on by a series of small allocations.
1786 * This behavior is a critical factor in sglist merging's success.
1790 static inline void expand(struct zone *zone, struct page *page,
1791 int low, int high, struct free_area *area,
1794 unsigned long size = 1 << high;
1796 while (high > low) {
1800 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
1803 * Mark as guard pages (or page), that will allow to
1804 * merge back to allocator when buddy will be freed.
1805 * Corresponding page table entries will not be touched,
1806 * pages will stay not present in virtual address space
1808 if (set_page_guard(zone, &page[size], high, migratetype))
1811 list_add(&page[size].lru, &area->free_list[migratetype]);
1813 set_page_order(&page[size], high);
1817 static void check_new_page_bad(struct page *page)
1819 const char *bad_reason = NULL;
1820 unsigned long bad_flags = 0;
1822 if (unlikely(atomic_read(&page->_mapcount) != -1))
1823 bad_reason = "nonzero mapcount";
1824 if (unlikely(page->mapping != NULL))
1825 bad_reason = "non-NULL mapping";
1826 if (unlikely(page_ref_count(page) != 0))
1827 bad_reason = "nonzero _count";
1828 if (unlikely(page->flags & __PG_HWPOISON)) {
1829 bad_reason = "HWPoisoned (hardware-corrupted)";
1830 bad_flags = __PG_HWPOISON;
1831 /* Don't complain about hwpoisoned pages */
1832 page_mapcount_reset(page); /* remove PageBuddy */
1835 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1836 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1837 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1840 if (unlikely(page->mem_cgroup))
1841 bad_reason = "page still charged to cgroup";
1843 bad_page(page, bad_reason, bad_flags);
1847 * This page is about to be returned from the page allocator
1849 static inline int check_new_page(struct page *page)
1851 if (likely(page_expected_state(page,
1852 PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON)))
1855 check_new_page_bad(page);
1859 static inline bool free_pages_prezeroed(void)
1861 return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) &&
1862 page_poisoning_enabled();
1865 #ifdef CONFIG_DEBUG_VM
1866 static bool check_pcp_refill(struct page *page)
1871 static bool check_new_pcp(struct page *page)
1873 return check_new_page(page);
1876 static bool check_pcp_refill(struct page *page)
1878 return check_new_page(page);
1880 static bool check_new_pcp(struct page *page)
1884 #endif /* CONFIG_DEBUG_VM */
1886 static bool check_new_pages(struct page *page, unsigned int order)
1889 for (i = 0; i < (1 << order); i++) {
1890 struct page *p = page + i;
1892 if (unlikely(check_new_page(p)))
1899 inline void post_alloc_hook(struct page *page, unsigned int order,
1902 set_page_private(page, 0);
1903 set_page_refcounted(page);
1905 arch_alloc_page(page, order);
1906 kernel_map_pages(page, 1 << order, 1);
1907 kernel_poison_pages(page, 1 << order, 1);
1908 kasan_alloc_pages(page, order);
1909 set_page_owner(page, order, gfp_flags);
1912 static void prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
1913 unsigned int alloc_flags)
1917 post_alloc_hook(page, order, gfp_flags);
1919 if (!free_pages_prezeroed() && (gfp_flags & __GFP_ZERO))
1920 for (i = 0; i < (1 << order); i++)
1921 clear_highpage(page + i);
1923 if (order && (gfp_flags & __GFP_COMP))
1924 prep_compound_page(page, order);
1927 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1928 * allocate the page. The expectation is that the caller is taking
1929 * steps that will free more memory. The caller should avoid the page
1930 * being used for !PFMEMALLOC purposes.
1932 if (alloc_flags & ALLOC_NO_WATERMARKS)
1933 set_page_pfmemalloc(page);
1935 clear_page_pfmemalloc(page);
1939 * Go through the free lists for the given migratetype and remove
1940 * the smallest available page from the freelists
1942 static __always_inline
1943 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
1946 unsigned int current_order;
1947 struct free_area *area;
1950 /* Find a page of the appropriate size in the preferred list */
1951 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1952 area = &(zone->free_area[current_order]);
1953 page = list_first_entry_or_null(&area->free_list[migratetype],
1957 list_del(&page->lru);
1958 rmv_page_order(page);
1960 expand(zone, page, order, current_order, area, migratetype);
1961 set_pcppage_migratetype(page, migratetype);
1970 * This array describes the order lists are fallen back to when
1971 * the free lists for the desirable migrate type are depleted
1973 static int fallbacks[MIGRATE_TYPES][4] = {
1974 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
1975 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
1976 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
1978 [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */
1980 #ifdef CONFIG_MEMORY_ISOLATION
1981 [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */
1986 static __always_inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1989 return __rmqueue_smallest(zone, order, MIGRATE_CMA);
1992 static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1993 unsigned int order) { return NULL; }
1997 * Move the free pages in a range to the free lists of the requested type.
1998 * Note that start_page and end_pages are not aligned on a pageblock
1999 * boundary. If alignment is required, use move_freepages_block()
2001 static int move_freepages(struct zone *zone,
2002 struct page *start_page, struct page *end_page,
2003 int migratetype, int *num_movable)
2007 int pages_moved = 0;
2009 #ifndef CONFIG_HOLES_IN_ZONE
2011 * page_zone is not safe to call in this context when
2012 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
2013 * anyway as we check zone boundaries in move_freepages_block().
2014 * Remove at a later date when no bug reports exist related to
2015 * grouping pages by mobility
2017 VM_BUG_ON(pfn_valid(page_to_pfn(start_page)) &&
2018 pfn_valid(page_to_pfn(end_page)) &&
2019 page_zone(start_page) != page_zone(end_page));
2025 for (page = start_page; page <= end_page;) {
2026 if (!pfn_valid_within(page_to_pfn(page))) {
2031 /* Make sure we are not inadvertently changing nodes */
2032 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
2034 if (!PageBuddy(page)) {
2036 * We assume that pages that could be isolated for
2037 * migration are movable. But we don't actually try
2038 * isolating, as that would be expensive.
2041 (PageLRU(page) || __PageMovable(page)))
2048 order = page_order(page);
2049 list_move(&page->lru,
2050 &zone->free_area[order].free_list[migratetype]);
2052 pages_moved += 1 << order;
2058 int move_freepages_block(struct zone *zone, struct page *page,
2059 int migratetype, int *num_movable)
2061 unsigned long start_pfn, end_pfn;
2062 struct page *start_page, *end_page;
2064 start_pfn = page_to_pfn(page);
2065 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
2066 start_page = pfn_to_page(start_pfn);
2067 end_page = start_page + pageblock_nr_pages - 1;
2068 end_pfn = start_pfn + pageblock_nr_pages - 1;
2070 /* Do not cross zone boundaries */
2071 if (!zone_spans_pfn(zone, start_pfn))
2073 if (!zone_spans_pfn(zone, end_pfn))
2076 return move_freepages(zone, start_page, end_page, migratetype,
2080 static void change_pageblock_range(struct page *pageblock_page,
2081 int start_order, int migratetype)
2083 int nr_pageblocks = 1 << (start_order - pageblock_order);
2085 while (nr_pageblocks--) {
2086 set_pageblock_migratetype(pageblock_page, migratetype);
2087 pageblock_page += pageblock_nr_pages;
2092 * When we are falling back to another migratetype during allocation, try to
2093 * steal extra free pages from the same pageblocks to satisfy further
2094 * allocations, instead of polluting multiple pageblocks.
2096 * If we are stealing a relatively large buddy page, it is likely there will
2097 * be more free pages in the pageblock, so try to steal them all. For
2098 * reclaimable and unmovable allocations, we steal regardless of page size,
2099 * as fragmentation caused by those allocations polluting movable pageblocks
2100 * is worse than movable allocations stealing from unmovable and reclaimable
2103 static bool can_steal_fallback(unsigned int order, int start_mt)
2106 * Leaving this order check is intended, although there is
2107 * relaxed order check in next check. The reason is that
2108 * we can actually steal whole pageblock if this condition met,
2109 * but, below check doesn't guarantee it and that is just heuristic
2110 * so could be changed anytime.
2112 if (order >= pageblock_order)
2115 if (order >= pageblock_order / 2 ||
2116 start_mt == MIGRATE_RECLAIMABLE ||
2117 start_mt == MIGRATE_UNMOVABLE ||
2118 page_group_by_mobility_disabled)
2125 * This function implements actual steal behaviour. If order is large enough,
2126 * we can steal whole pageblock. If not, we first move freepages in this
2127 * pageblock to our migratetype and determine how many already-allocated pages
2128 * are there in the pageblock with a compatible migratetype. If at least half
2129 * of pages are free or compatible, we can change migratetype of the pageblock
2130 * itself, so pages freed in the future will be put on the correct free list.
2132 static void steal_suitable_fallback(struct zone *zone, struct page *page,
2133 int start_type, bool whole_block)
2135 unsigned int current_order = page_order(page);
2136 struct free_area *area;
2137 int free_pages, movable_pages, alike_pages;
2140 old_block_type = get_pageblock_migratetype(page);
2143 * This can happen due to races and we want to prevent broken
2144 * highatomic accounting.
2146 if (is_migrate_highatomic(old_block_type))
2149 /* Take ownership for orders >= pageblock_order */
2150 if (current_order >= pageblock_order) {
2151 change_pageblock_range(page, current_order, start_type);
2155 /* We are not allowed to try stealing from the whole block */
2159 free_pages = move_freepages_block(zone, page, start_type,
2162 * Determine how many pages are compatible with our allocation.
2163 * For movable allocation, it's the number of movable pages which
2164 * we just obtained. For other types it's a bit more tricky.
2166 if (start_type == MIGRATE_MOVABLE) {
2167 alike_pages = movable_pages;
2170 * If we are falling back a RECLAIMABLE or UNMOVABLE allocation
2171 * to MOVABLE pageblock, consider all non-movable pages as
2172 * compatible. If it's UNMOVABLE falling back to RECLAIMABLE or
2173 * vice versa, be conservative since we can't distinguish the
2174 * exact migratetype of non-movable pages.
2176 if (old_block_type == MIGRATE_MOVABLE)
2177 alike_pages = pageblock_nr_pages
2178 - (free_pages + movable_pages);
2183 /* moving whole block can fail due to zone boundary conditions */
2188 * If a sufficient number of pages in the block are either free or of
2189 * comparable migratability as our allocation, claim the whole block.
2191 if (free_pages + alike_pages >= (1 << (pageblock_order-1)) ||
2192 page_group_by_mobility_disabled)
2193 set_pageblock_migratetype(page, start_type);
2198 area = &zone->free_area[current_order];
2199 list_move(&page->lru, &area->free_list[start_type]);
2203 * Check whether there is a suitable fallback freepage with requested order.
2204 * If only_stealable is true, this function returns fallback_mt only if
2205 * we can steal other freepages all together. This would help to reduce
2206 * fragmentation due to mixed migratetype pages in one pageblock.
2208 int find_suitable_fallback(struct free_area *area, unsigned int order,
2209 int migratetype, bool only_stealable, bool *can_steal)
2214 if (area->nr_free == 0)
2219 fallback_mt = fallbacks[migratetype][i];
2220 if (fallback_mt == MIGRATE_TYPES)
2223 if (list_empty(&area->free_list[fallback_mt]))
2226 if (can_steal_fallback(order, migratetype))
2229 if (!only_stealable)
2240 * Reserve a pageblock for exclusive use of high-order atomic allocations if
2241 * there are no empty page blocks that contain a page with a suitable order
2243 static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
2244 unsigned int alloc_order)
2247 unsigned long max_managed, flags;
2250 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
2251 * Check is race-prone but harmless.
2253 max_managed = (zone->managed_pages / 100) + pageblock_nr_pages;
2254 if (zone->nr_reserved_highatomic >= max_managed)
2257 spin_lock_irqsave(&zone->lock, flags);
2259 /* Recheck the nr_reserved_highatomic limit under the lock */
2260 if (zone->nr_reserved_highatomic >= max_managed)
2264 mt = get_pageblock_migratetype(page);
2265 if (!is_migrate_highatomic(mt) && !is_migrate_isolate(mt)
2266 && !is_migrate_cma(mt)) {
2267 zone->nr_reserved_highatomic += pageblock_nr_pages;
2268 set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
2269 move_freepages_block(zone, page, MIGRATE_HIGHATOMIC, NULL);
2273 spin_unlock_irqrestore(&zone->lock, flags);
2277 * Used when an allocation is about to fail under memory pressure. This
2278 * potentially hurts the reliability of high-order allocations when under
2279 * intense memory pressure but failed atomic allocations should be easier
2280 * to recover from than an OOM.
2282 * If @force is true, try to unreserve a pageblock even though highatomic
2283 * pageblock is exhausted.
2285 static bool unreserve_highatomic_pageblock(const struct alloc_context *ac,
2288 struct zonelist *zonelist = ac->zonelist;
2289 unsigned long flags;
2296 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2299 * Preserve at least one pageblock unless memory pressure
2302 if (!force && zone->nr_reserved_highatomic <=
2306 spin_lock_irqsave(&zone->lock, flags);
2307 for (order = 0; order < MAX_ORDER; order++) {
2308 struct free_area *area = &(zone->free_area[order]);
2310 page = list_first_entry_or_null(
2311 &area->free_list[MIGRATE_HIGHATOMIC],
2317 * In page freeing path, migratetype change is racy so
2318 * we can counter several free pages in a pageblock
2319 * in this loop althoug we changed the pageblock type
2320 * from highatomic to ac->migratetype. So we should
2321 * adjust the count once.
2323 if (is_migrate_highatomic_page(page)) {
2325 * It should never happen but changes to
2326 * locking could inadvertently allow a per-cpu
2327 * drain to add pages to MIGRATE_HIGHATOMIC
2328 * while unreserving so be safe and watch for
2331 zone->nr_reserved_highatomic -= min(
2333 zone->nr_reserved_highatomic);
2337 * Convert to ac->migratetype and avoid the normal
2338 * pageblock stealing heuristics. Minimally, the caller
2339 * is doing the work and needs the pages. More
2340 * importantly, if the block was always converted to
2341 * MIGRATE_UNMOVABLE or another type then the number
2342 * of pageblocks that cannot be completely freed
2345 set_pageblock_migratetype(page, ac->migratetype);
2346 ret = move_freepages_block(zone, page, ac->migratetype,
2349 spin_unlock_irqrestore(&zone->lock, flags);
2353 spin_unlock_irqrestore(&zone->lock, flags);
2360 * Try finding a free buddy page on the fallback list and put it on the free
2361 * list of requested migratetype, possibly along with other pages from the same
2362 * block, depending on fragmentation avoidance heuristics. Returns true if
2363 * fallback was found so that __rmqueue_smallest() can grab it.
2365 * The use of signed ints for order and current_order is a deliberate
2366 * deviation from the rest of this file, to make the for loop
2367 * condition simpler.
2369 static __always_inline bool
2370 __rmqueue_fallback(struct zone *zone, int order, int start_migratetype)
2372 struct free_area *area;
2379 * Find the largest available free page in the other list. This roughly
2380 * approximates finding the pageblock with the most free pages, which
2381 * would be too costly to do exactly.
2383 for (current_order = MAX_ORDER - 1; current_order >= order;
2385 area = &(zone->free_area[current_order]);
2386 fallback_mt = find_suitable_fallback(area, current_order,
2387 start_migratetype, false, &can_steal);
2388 if (fallback_mt == -1)
2392 * We cannot steal all free pages from the pageblock and the
2393 * requested migratetype is movable. In that case it's better to
2394 * steal and split the smallest available page instead of the
2395 * largest available page, because even if the next movable
2396 * allocation falls back into a different pageblock than this
2397 * one, it won't cause permanent fragmentation.
2399 if (!can_steal && start_migratetype == MIGRATE_MOVABLE
2400 && current_order > order)
2409 for (current_order = order; current_order < MAX_ORDER;
2411 area = &(zone->free_area[current_order]);
2412 fallback_mt = find_suitable_fallback(area, current_order,
2413 start_migratetype, false, &can_steal);
2414 if (fallback_mt != -1)
2419 * This should not happen - we already found a suitable fallback
2420 * when looking for the largest page.
2422 VM_BUG_ON(current_order == MAX_ORDER);
2425 page = list_first_entry(&area->free_list[fallback_mt],
2428 steal_suitable_fallback(zone, page, start_migratetype, can_steal);
2430 trace_mm_page_alloc_extfrag(page, order, current_order,
2431 start_migratetype, fallback_mt);
2438 * Do the hard work of removing an element from the buddy allocator.
2439 * Call me with the zone->lock already held.
2441 static __always_inline struct page *
2442 __rmqueue(struct zone *zone, unsigned int order, int migratetype)
2447 page = __rmqueue_smallest(zone, order, migratetype);
2448 if (unlikely(!page)) {
2449 if (migratetype == MIGRATE_MOVABLE)
2450 page = __rmqueue_cma_fallback(zone, order);
2452 if (!page && __rmqueue_fallback(zone, order, migratetype))
2456 trace_mm_page_alloc_zone_locked(page, order, migratetype);
2461 * Obtain a specified number of elements from the buddy allocator, all under
2462 * a single hold of the lock, for efficiency. Add them to the supplied list.
2463 * Returns the number of new pages which were placed at *list.
2465 static int rmqueue_bulk(struct zone *zone, unsigned int order,
2466 unsigned long count, struct list_head *list,
2471 spin_lock(&zone->lock);
2472 for (i = 0; i < count; ++i) {
2473 struct page *page = __rmqueue(zone, order, migratetype);
2474 if (unlikely(page == NULL))
2477 if (unlikely(check_pcp_refill(page)))
2481 * Split buddy pages returned by expand() are received here in
2482 * physical page order. The page is added to the tail of
2483 * caller's list. From the callers perspective, the linked list
2484 * is ordered by page number under some conditions. This is
2485 * useful for IO devices that can forward direction from the
2486 * head, thus also in the physical page order. This is useful
2487 * for IO devices that can merge IO requests if the physical
2488 * pages are ordered properly.
2490 list_add_tail(&page->lru, list);
2492 if (is_migrate_cma(get_pcppage_migratetype(page)))
2493 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
2498 * i pages were removed from the buddy list even if some leak due
2499 * to check_pcp_refill failing so adjust NR_FREE_PAGES based
2500 * on i. Do not confuse with 'alloced' which is the number of
2501 * pages added to the pcp list.
2503 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
2504 spin_unlock(&zone->lock);
2510 * Called from the vmstat counter updater to drain pagesets of this
2511 * currently executing processor on remote nodes after they have
2514 * Note that this function must be called with the thread pinned to
2515 * a single processor.
2517 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
2519 unsigned long flags;
2520 int to_drain, batch;
2522 local_irq_save(flags);
2523 batch = READ_ONCE(pcp->batch);
2524 to_drain = min(pcp->count, batch);
2526 free_pcppages_bulk(zone, to_drain, pcp);
2527 local_irq_restore(flags);
2532 * Drain pcplists of the indicated processor and zone.
2534 * The processor must either be the current processor and the
2535 * thread pinned to the current processor or a processor that
2538 static void drain_pages_zone(unsigned int cpu, struct zone *zone)
2540 unsigned long flags;
2541 struct per_cpu_pageset *pset;
2542 struct per_cpu_pages *pcp;
2544 local_irq_save(flags);
2545 pset = per_cpu_ptr(zone->pageset, cpu);
2549 free_pcppages_bulk(zone, pcp->count, pcp);
2550 local_irq_restore(flags);
2554 * Drain pcplists of all zones on the indicated processor.
2556 * The processor must either be the current processor and the
2557 * thread pinned to the current processor or a processor that
2560 static void drain_pages(unsigned int cpu)
2564 for_each_populated_zone(zone) {
2565 drain_pages_zone(cpu, zone);
2570 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
2572 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2573 * the single zone's pages.
2575 void drain_local_pages(struct zone *zone)
2577 int cpu = smp_processor_id();
2580 drain_pages_zone(cpu, zone);
2585 static void drain_local_pages_wq(struct work_struct *work)
2588 * drain_all_pages doesn't use proper cpu hotplug protection so
2589 * we can race with cpu offline when the WQ can move this from
2590 * a cpu pinned worker to an unbound one. We can operate on a different
2591 * cpu which is allright but we also have to make sure to not move to
2595 drain_local_pages(NULL);
2600 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2602 * When zone parameter is non-NULL, spill just the single zone's pages.
2604 * Note that this can be extremely slow as the draining happens in a workqueue.
2606 void drain_all_pages(struct zone *zone)
2611 * Allocate in the BSS so we wont require allocation in
2612 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2614 static cpumask_t cpus_with_pcps;
2617 * Make sure nobody triggers this path before mm_percpu_wq is fully
2620 if (WARN_ON_ONCE(!mm_percpu_wq))
2624 * Do not drain if one is already in progress unless it's specific to
2625 * a zone. Such callers are primarily CMA and memory hotplug and need
2626 * the drain to be complete when the call returns.
2628 if (unlikely(!mutex_trylock(&pcpu_drain_mutex))) {
2631 mutex_lock(&pcpu_drain_mutex);
2635 * We don't care about racing with CPU hotplug event
2636 * as offline notification will cause the notified
2637 * cpu to drain that CPU pcps and on_each_cpu_mask
2638 * disables preemption as part of its processing
2640 for_each_online_cpu(cpu) {
2641 struct per_cpu_pageset *pcp;
2643 bool has_pcps = false;
2646 pcp = per_cpu_ptr(zone->pageset, cpu);
2650 for_each_populated_zone(z) {
2651 pcp = per_cpu_ptr(z->pageset, cpu);
2652 if (pcp->pcp.count) {
2660 cpumask_set_cpu(cpu, &cpus_with_pcps);
2662 cpumask_clear_cpu(cpu, &cpus_with_pcps);
2665 for_each_cpu(cpu, &cpus_with_pcps) {
2666 struct work_struct *work = per_cpu_ptr(&pcpu_drain, cpu);
2667 INIT_WORK(work, drain_local_pages_wq);
2668 queue_work_on(cpu, mm_percpu_wq, work);
2670 for_each_cpu(cpu, &cpus_with_pcps)
2671 flush_work(per_cpu_ptr(&pcpu_drain, cpu));
2673 mutex_unlock(&pcpu_drain_mutex);
2676 #ifdef CONFIG_HIBERNATION
2679 * Touch the watchdog for every WD_PAGE_COUNT pages.
2681 #define WD_PAGE_COUNT (128*1024)
2683 void mark_free_pages(struct zone *zone)
2685 unsigned long pfn, max_zone_pfn, page_count = WD_PAGE_COUNT;
2686 unsigned long flags;
2687 unsigned int order, t;
2690 if (zone_is_empty(zone))
2693 spin_lock_irqsave(&zone->lock, flags);
2695 max_zone_pfn = zone_end_pfn(zone);
2696 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
2697 if (pfn_valid(pfn)) {
2698 page = pfn_to_page(pfn);
2700 if (!--page_count) {
2701 touch_nmi_watchdog();
2702 page_count = WD_PAGE_COUNT;
2705 if (page_zone(page) != zone)
2708 if (!swsusp_page_is_forbidden(page))
2709 swsusp_unset_page_free(page);
2712 for_each_migratetype_order(order, t) {
2713 list_for_each_entry(page,
2714 &zone->free_area[order].free_list[t], lru) {
2717 pfn = page_to_pfn(page);
2718 for (i = 0; i < (1UL << order); i++) {
2719 if (!--page_count) {
2720 touch_nmi_watchdog();
2721 page_count = WD_PAGE_COUNT;
2723 swsusp_set_page_free(pfn_to_page(pfn + i));
2727 spin_unlock_irqrestore(&zone->lock, flags);
2729 #endif /* CONFIG_PM */
2731 static bool free_unref_page_prepare(struct page *page, unsigned long pfn)
2735 if (!free_pcp_prepare(page))
2738 migratetype = get_pfnblock_migratetype(page, pfn);
2739 set_pcppage_migratetype(page, migratetype);
2743 static void free_unref_page_commit(struct page *page, unsigned long pfn)
2745 struct zone *zone = page_zone(page);
2746 struct per_cpu_pages *pcp;
2749 migratetype = get_pcppage_migratetype(page);
2750 __count_vm_event(PGFREE);
2753 * We only track unmovable, reclaimable and movable on pcp lists.
2754 * Free ISOLATE pages back to the allocator because they are being
2755 * offlined but treat HIGHATOMIC as movable pages so we can get those
2756 * areas back if necessary. Otherwise, we may have to free
2757 * excessively into the page allocator
2759 if (migratetype >= MIGRATE_PCPTYPES) {
2760 if (unlikely(is_migrate_isolate(migratetype))) {
2761 free_one_page(zone, page, pfn, 0, migratetype);
2764 migratetype = MIGRATE_MOVABLE;
2767 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2768 list_add(&page->lru, &pcp->lists[migratetype]);
2770 if (pcp->count >= pcp->high) {
2771 unsigned long batch = READ_ONCE(pcp->batch);
2772 free_pcppages_bulk(zone, batch, pcp);
2777 * Free a 0-order page
2779 void free_unref_page(struct page *page)
2781 unsigned long flags;
2782 unsigned long pfn = page_to_pfn(page);
2784 if (!free_unref_page_prepare(page, pfn))
2787 local_irq_save(flags);
2788 free_unref_page_commit(page, pfn);
2789 local_irq_restore(flags);
2793 * Free a list of 0-order pages
2795 void free_unref_page_list(struct list_head *list)
2797 struct page *page, *next;
2798 unsigned long flags, pfn;
2799 int batch_count = 0;
2801 /* Prepare pages for freeing */
2802 list_for_each_entry_safe(page, next, list, lru) {
2803 pfn = page_to_pfn(page);
2804 if (!free_unref_page_prepare(page, pfn))
2805 list_del(&page->lru);
2806 set_page_private(page, pfn);
2809 local_irq_save(flags);
2810 list_for_each_entry_safe(page, next, list, lru) {
2811 unsigned long pfn = page_private(page);
2813 set_page_private(page, 0);
2814 trace_mm_page_free_batched(page);
2815 free_unref_page_commit(page, pfn);
2818 * Guard against excessive IRQ disabled times when we get
2819 * a large list of pages to free.
2821 if (++batch_count == SWAP_CLUSTER_MAX) {
2822 local_irq_restore(flags);
2824 local_irq_save(flags);
2827 local_irq_restore(flags);
2831 * split_page takes a non-compound higher-order page, and splits it into
2832 * n (1<<order) sub-pages: page[0..n]
2833 * Each sub-page must be freed individually.
2835 * Note: this is probably too low level an operation for use in drivers.
2836 * Please consult with lkml before using this in your driver.
2838 void split_page(struct page *page, unsigned int order)
2842 VM_BUG_ON_PAGE(PageCompound(page), page);
2843 VM_BUG_ON_PAGE(!page_count(page), page);
2845 for (i = 1; i < (1 << order); i++)
2846 set_page_refcounted(page + i);
2847 split_page_owner(page, order);
2849 EXPORT_SYMBOL_GPL(split_page);
2851 int __isolate_free_page(struct page *page, unsigned int order)
2853 unsigned long watermark;
2857 BUG_ON(!PageBuddy(page));
2859 zone = page_zone(page);
2860 mt = get_pageblock_migratetype(page);
2862 if (!is_migrate_isolate(mt)) {
2864 * Obey watermarks as if the page was being allocated. We can
2865 * emulate a high-order watermark check with a raised order-0
2866 * watermark, because we already know our high-order page
2869 watermark = min_wmark_pages(zone) + (1UL << order);
2870 if (!zone_watermark_ok(zone, 0, watermark, 0, ALLOC_CMA))
2873 __mod_zone_freepage_state(zone, -(1UL << order), mt);
2876 /* Remove page from free list */
2877 list_del(&page->lru);
2878 zone->free_area[order].nr_free--;
2879 rmv_page_order(page);
2882 * Set the pageblock if the isolated page is at least half of a
2885 if (order >= pageblock_order - 1) {
2886 struct page *endpage = page + (1 << order) - 1;
2887 for (; page < endpage; page += pageblock_nr_pages) {
2888 int mt = get_pageblock_migratetype(page);
2889 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt)
2890 && !is_migrate_highatomic(mt))
2891 set_pageblock_migratetype(page,
2897 return 1UL << order;
2901 * Update NUMA hit/miss statistics
2903 * Must be called with interrupts disabled.
2905 static inline void zone_statistics(struct zone *preferred_zone, struct zone *z)
2908 enum numa_stat_item local_stat = NUMA_LOCAL;
2910 /* skip numa counters update if numa stats is disabled */
2911 if (!static_branch_likely(&vm_numa_stat_key))
2914 if (z->node != numa_node_id())
2915 local_stat = NUMA_OTHER;
2917 if (z->node == preferred_zone->node)
2918 __inc_numa_state(z, NUMA_HIT);
2920 __inc_numa_state(z, NUMA_MISS);
2921 __inc_numa_state(preferred_zone, NUMA_FOREIGN);
2923 __inc_numa_state(z, local_stat);
2927 /* Remove page from the per-cpu list, caller must protect the list */
2928 static struct page *__rmqueue_pcplist(struct zone *zone, int migratetype,
2929 struct per_cpu_pages *pcp,
2930 struct list_head *list)
2935 if (list_empty(list)) {
2936 pcp->count += rmqueue_bulk(zone, 0,
2939 if (unlikely(list_empty(list)))
2943 page = list_first_entry(list, struct page, lru);
2944 list_del(&page->lru);
2946 } while (check_new_pcp(page));
2951 /* Lock and remove page from the per-cpu list */
2952 static struct page *rmqueue_pcplist(struct zone *preferred_zone,
2953 struct zone *zone, unsigned int order,
2954 gfp_t gfp_flags, int migratetype)
2956 struct per_cpu_pages *pcp;
2957 struct list_head *list;
2959 unsigned long flags;
2961 local_irq_save(flags);
2962 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2963 list = &pcp->lists[migratetype];
2964 page = __rmqueue_pcplist(zone, migratetype, pcp, list);
2966 __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
2967 zone_statistics(preferred_zone, zone);
2969 local_irq_restore(flags);
2974 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2977 struct page *rmqueue(struct zone *preferred_zone,
2978 struct zone *zone, unsigned int order,
2979 gfp_t gfp_flags, unsigned int alloc_flags,
2982 unsigned long flags;
2985 if (likely(order == 0)) {
2986 page = rmqueue_pcplist(preferred_zone, zone, order,
2987 gfp_flags, migratetype);
2992 * We most definitely don't want callers attempting to
2993 * allocate greater than order-1 page units with __GFP_NOFAIL.
2995 WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
2996 spin_lock_irqsave(&zone->lock, flags);
3000 if (alloc_flags & ALLOC_HARDER) {
3001 page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
3003 trace_mm_page_alloc_zone_locked(page, order, migratetype);
3006 page = __rmqueue(zone, order, migratetype);
3007 } while (page && check_new_pages(page, order));
3008 spin_unlock(&zone->lock);
3011 __mod_zone_freepage_state(zone, -(1 << order),
3012 get_pcppage_migratetype(page));
3014 __count_zid_vm_events(PGALLOC, page_zonenum(page), 1 << order);
3015 zone_statistics(preferred_zone, zone);
3016 local_irq_restore(flags);
3019 VM_BUG_ON_PAGE(page && bad_range(zone, page), page);
3023 local_irq_restore(flags);
3027 #ifdef CONFIG_FAIL_PAGE_ALLOC
3030 struct fault_attr attr;
3032 bool ignore_gfp_highmem;
3033 bool ignore_gfp_reclaim;
3035 } fail_page_alloc = {
3036 .attr = FAULT_ATTR_INITIALIZER,
3037 .ignore_gfp_reclaim = true,
3038 .ignore_gfp_highmem = true,
3042 static int __init setup_fail_page_alloc(char *str)
3044 return setup_fault_attr(&fail_page_alloc.attr, str);
3046 __setup("fail_page_alloc=", setup_fail_page_alloc);
3048 static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3050 if (order < fail_page_alloc.min_order)
3052 if (gfp_mask & __GFP_NOFAIL)
3054 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
3056 if (fail_page_alloc.ignore_gfp_reclaim &&
3057 (gfp_mask & __GFP_DIRECT_RECLAIM))
3060 return should_fail(&fail_page_alloc.attr, 1 << order);
3063 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
3065 static int __init fail_page_alloc_debugfs(void)
3067 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
3070 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
3071 &fail_page_alloc.attr);
3073 return PTR_ERR(dir);
3075 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
3076 &fail_page_alloc.ignore_gfp_reclaim))
3078 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
3079 &fail_page_alloc.ignore_gfp_highmem))
3081 if (!debugfs_create_u32("min-order", mode, dir,
3082 &fail_page_alloc.min_order))
3087 debugfs_remove_recursive(dir);
3092 late_initcall(fail_page_alloc_debugfs);
3094 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
3096 #else /* CONFIG_FAIL_PAGE_ALLOC */
3098 static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
3103 #endif /* CONFIG_FAIL_PAGE_ALLOC */
3106 * Return true if free base pages are above 'mark'. For high-order checks it
3107 * will return true of the order-0 watermark is reached and there is at least
3108 * one free page of a suitable size. Checking now avoids taking the zone lock
3109 * to check in the allocation paths if no pages are free.
3111 bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
3112 int classzone_idx, unsigned int alloc_flags,
3117 const bool alloc_harder = (alloc_flags & (ALLOC_HARDER|ALLOC_OOM));
3119 /* free_pages may go negative - that's OK */
3120 free_pages -= (1 << order) - 1;
3122 if (alloc_flags & ALLOC_HIGH)
3126 * If the caller does not have rights to ALLOC_HARDER then subtract
3127 * the high-atomic reserves. This will over-estimate the size of the
3128 * atomic reserve but it avoids a search.
3130 if (likely(!alloc_harder)) {
3131 free_pages -= z->nr_reserved_highatomic;
3134 * OOM victims can try even harder than normal ALLOC_HARDER
3135 * users on the grounds that it's definitely going to be in
3136 * the exit path shortly and free memory. Any allocation it
3137 * makes during the free path will be small and short-lived.
3139 if (alloc_flags & ALLOC_OOM)
3147 /* If allocation can't use CMA areas don't use free CMA pages */
3148 if (!(alloc_flags & ALLOC_CMA))
3149 free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES);
3153 * Check watermarks for an order-0 allocation request. If these
3154 * are not met, then a high-order request also cannot go ahead
3155 * even if a suitable page happened to be free.
3157 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
3160 /* If this is an order-0 request then the watermark is fine */
3164 /* For a high-order request, check at least one suitable page is free */
3165 for (o = order; o < MAX_ORDER; o++) {
3166 struct free_area *area = &z->free_area[o];
3172 for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
3173 if (!list_empty(&area->free_list[mt]))
3178 if ((alloc_flags & ALLOC_CMA) &&
3179 !list_empty(&area->free_list[MIGRATE_CMA])) {
3184 !list_empty(&area->free_list[MIGRATE_HIGHATOMIC]))
3190 bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
3191 int classzone_idx, unsigned int alloc_flags)
3193 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
3194 zone_page_state(z, NR_FREE_PAGES));
3197 static inline bool zone_watermark_fast(struct zone *z, unsigned int order,
3198 unsigned long mark, int classzone_idx, unsigned int alloc_flags)
3200 long free_pages = zone_page_state(z, NR_FREE_PAGES);
3204 /* If allocation can't use CMA areas don't use free CMA pages */
3205 if (!(alloc_flags & ALLOC_CMA))
3206 cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES);
3210 * Fast check for order-0 only. If this fails then the reserves
3211 * need to be calculated. There is a corner case where the check
3212 * passes but only the high-order atomic reserve are free. If
3213 * the caller is !atomic then it'll uselessly search the free
3214 * list. That corner case is then slower but it is harmless.
3216 if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx])
3219 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
3223 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
3224 unsigned long mark, int classzone_idx)
3226 long free_pages = zone_page_state(z, NR_FREE_PAGES);
3228 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
3229 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
3231 return __zone_watermark_ok(z, order, mark, classzone_idx, 0,
3236 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
3238 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <=
3241 #else /* CONFIG_NUMA */
3242 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
3246 #endif /* CONFIG_NUMA */
3249 * get_page_from_freelist goes through the zonelist trying to allocate
3252 static struct page *
3253 get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
3254 const struct alloc_context *ac)
3256 struct zoneref *z = ac->preferred_zoneref;
3258 struct pglist_data *last_pgdat_dirty_limit = NULL;
3261 * Scan zonelist, looking for a zone with enough free.
3262 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
3264 for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3269 if (cpusets_enabled() &&
3270 (alloc_flags & ALLOC_CPUSET) &&
3271 !__cpuset_zone_allowed(zone, gfp_mask))
3274 * When allocating a page cache page for writing, we
3275 * want to get it from a node that is within its dirty
3276 * limit, such that no single node holds more than its
3277 * proportional share of globally allowed dirty pages.
3278 * The dirty limits take into account the node's
3279 * lowmem reserves and high watermark so that kswapd
3280 * should be able to balance it without having to
3281 * write pages from its LRU list.
3283 * XXX: For now, allow allocations to potentially
3284 * exceed the per-node dirty limit in the slowpath
3285 * (spread_dirty_pages unset) before going into reclaim,
3286 * which is important when on a NUMA setup the allowed
3287 * nodes are together not big enough to reach the
3288 * global limit. The proper fix for these situations
3289 * will require awareness of nodes in the
3290 * dirty-throttling and the flusher threads.
3292 if (ac->spread_dirty_pages) {
3293 if (last_pgdat_dirty_limit == zone->zone_pgdat)
3296 if (!node_dirty_ok(zone->zone_pgdat)) {
3297 last_pgdat_dirty_limit = zone->zone_pgdat;
3302 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
3303 if (!zone_watermark_fast(zone, order, mark,
3304 ac_classzone_idx(ac), alloc_flags)) {
3307 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
3309 * Watermark failed for this zone, but see if we can
3310 * grow this zone if it contains deferred pages.
3312 if (static_branch_unlikely(&deferred_pages)) {
3313 if (_deferred_grow_zone(zone, order))
3317 /* Checked here to keep the fast path fast */
3318 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
3319 if (alloc_flags & ALLOC_NO_WATERMARKS)
3322 if (node_reclaim_mode == 0 ||
3323 !zone_allows_reclaim(ac->preferred_zoneref->zone, zone))
3326 ret = node_reclaim(zone->zone_pgdat, gfp_mask, order);
3328 case NODE_RECLAIM_NOSCAN:
3331 case NODE_RECLAIM_FULL:
3332 /* scanned but unreclaimable */
3335 /* did we reclaim enough */
3336 if (zone_watermark_ok(zone, order, mark,
3337 ac_classzone_idx(ac), alloc_flags))
3345 page = rmqueue(ac->preferred_zoneref->zone, zone, order,
3346 gfp_mask, alloc_flags, ac->migratetype);
3348 prep_new_page(page, order, gfp_mask, alloc_flags);
3351 * If this is a high-order atomic allocation then check
3352 * if the pageblock should be reserved for the future
3354 if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
3355 reserve_highatomic_pageblock(page, zone, order);
3359 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
3360 /* Try again if zone has deferred pages */
3361 if (static_branch_unlikely(&deferred_pages)) {
3362 if (_deferred_grow_zone(zone, order))
3373 * Large machines with many possible nodes should not always dump per-node
3374 * meminfo in irq context.
3376 static inline bool should_suppress_show_mem(void)
3381 ret = in_interrupt();
3386 static void warn_alloc_show_mem(gfp_t gfp_mask, nodemask_t *nodemask)
3388 unsigned int filter = SHOW_MEM_FILTER_NODES;
3389 static DEFINE_RATELIMIT_STATE(show_mem_rs, HZ, 1);
3391 if (should_suppress_show_mem() || !__ratelimit(&show_mem_rs))
3395 * This documents exceptions given to allocations in certain
3396 * contexts that are allowed to allocate outside current's set
3399 if (!(gfp_mask & __GFP_NOMEMALLOC))
3400 if (tsk_is_oom_victim(current) ||
3401 (current->flags & (PF_MEMALLOC | PF_EXITING)))
3402 filter &= ~SHOW_MEM_FILTER_NODES;
3403 if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
3404 filter &= ~SHOW_MEM_FILTER_NODES;
3406 show_mem(filter, nodemask);
3409 void warn_alloc(gfp_t gfp_mask, nodemask_t *nodemask, const char *fmt, ...)
3411 struct va_format vaf;
3413 static DEFINE_RATELIMIT_STATE(nopage_rs, DEFAULT_RATELIMIT_INTERVAL,
3414 DEFAULT_RATELIMIT_BURST);
3416 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs))
3419 va_start(args, fmt);
3422 pr_warn("%s: %pV, mode:%#x(%pGg), nodemask=%*pbl\n",
3423 current->comm, &vaf, gfp_mask, &gfp_mask,
3424 nodemask_pr_args(nodemask));
3427 cpuset_print_current_mems_allowed();
3430 warn_alloc_show_mem(gfp_mask, nodemask);
3433 static inline struct page *
3434 __alloc_pages_cpuset_fallback(gfp_t gfp_mask, unsigned int order,
3435 unsigned int alloc_flags,
3436 const struct alloc_context *ac)
3440 page = get_page_from_freelist(gfp_mask, order,
3441 alloc_flags|ALLOC_CPUSET, ac);
3443 * fallback to ignore cpuset restriction if our nodes
3447 page = get_page_from_freelist(gfp_mask, order,
3453 static inline struct page *
3454 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
3455 const struct alloc_context *ac, unsigned long *did_some_progress)
3457 struct oom_control oc = {
3458 .zonelist = ac->zonelist,
3459 .nodemask = ac->nodemask,
3461 .gfp_mask = gfp_mask,
3466 *did_some_progress = 0;
3469 * Acquire the oom lock. If that fails, somebody else is
3470 * making progress for us.
3472 if (!mutex_trylock(&oom_lock)) {
3473 *did_some_progress = 1;
3474 schedule_timeout_uninterruptible(1);
3479 * Go through the zonelist yet one more time, keep very high watermark
3480 * here, this is only to catch a parallel oom killing, we must fail if
3481 * we're still under heavy pressure. But make sure that this reclaim
3482 * attempt shall not depend on __GFP_DIRECT_RECLAIM && !__GFP_NORETRY
3483 * allocation which will never fail due to oom_lock already held.
3485 page = get_page_from_freelist((gfp_mask | __GFP_HARDWALL) &
3486 ~__GFP_DIRECT_RECLAIM, order,
3487 ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
3491 /* Coredumps can quickly deplete all memory reserves */
3492 if (current->flags & PF_DUMPCORE)
3494 /* The OOM killer will not help higher order allocs */
3495 if (order > PAGE_ALLOC_COSTLY_ORDER)
3498 * We have already exhausted all our reclaim opportunities without any
3499 * success so it is time to admit defeat. We will skip the OOM killer
3500 * because it is very likely that the caller has a more reasonable
3501 * fallback than shooting a random task.
3503 if (gfp_mask & __GFP_RETRY_MAYFAIL)
3505 /* The OOM killer does not needlessly kill tasks for lowmem */
3506 if (ac->high_zoneidx < ZONE_NORMAL)
3508 if (pm_suspended_storage())
3511 * XXX: GFP_NOFS allocations should rather fail than rely on
3512 * other request to make a forward progress.
3513 * We are in an unfortunate situation where out_of_memory cannot
3514 * do much for this context but let's try it to at least get
3515 * access to memory reserved if the current task is killed (see
3516 * out_of_memory). Once filesystems are ready to handle allocation
3517 * failures more gracefully we should just bail out here.
3520 /* The OOM killer may not free memory on a specific node */
3521 if (gfp_mask & __GFP_THISNODE)
3524 /* Exhausted what can be done so it's blame time */
3525 if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
3526 *did_some_progress = 1;
3529 * Help non-failing allocations by giving them access to memory
3532 if (gfp_mask & __GFP_NOFAIL)
3533 page = __alloc_pages_cpuset_fallback(gfp_mask, order,
3534 ALLOC_NO_WATERMARKS, ac);
3537 mutex_unlock(&oom_lock);
3542 * Maximum number of compaction retries wit a progress before OOM
3543 * killer is consider as the only way to move forward.
3545 #define MAX_COMPACT_RETRIES 16
3547 #ifdef CONFIG_COMPACTION
3548 /* Try memory compaction for high-order allocations before reclaim */
3549 static struct page *
3550 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
3551 unsigned int alloc_flags, const struct alloc_context *ac,
3552 enum compact_priority prio, enum compact_result *compact_result)
3555 unsigned int noreclaim_flag;
3560 noreclaim_flag = memalloc_noreclaim_save();
3561 *compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
3563 memalloc_noreclaim_restore(noreclaim_flag);
3565 if (*compact_result <= COMPACT_INACTIVE)
3569 * At least in one zone compaction wasn't deferred or skipped, so let's
3570 * count a compaction stall
3572 count_vm_event(COMPACTSTALL);
3574 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
3577 struct zone *zone = page_zone(page);
3579 zone->compact_blockskip_flush = false;
3580 compaction_defer_reset(zone, order, true);
3581 count_vm_event(COMPACTSUCCESS);
3586 * It's bad if compaction run occurs and fails. The most likely reason
3587 * is that pages exist, but not enough to satisfy watermarks.
3589 count_vm_event(COMPACTFAIL);
3597 should_compact_retry(struct alloc_context *ac, int order, int alloc_flags,
3598 enum compact_result compact_result,
3599 enum compact_priority *compact_priority,
3600 int *compaction_retries)
3602 int max_retries = MAX_COMPACT_RETRIES;
3605 int retries = *compaction_retries;
3606 enum compact_priority priority = *compact_priority;
3611 if (compaction_made_progress(compact_result))
3612 (*compaction_retries)++;
3615 * compaction considers all the zone as desperately out of memory
3616 * so it doesn't really make much sense to retry except when the
3617 * failure could be caused by insufficient priority
3619 if (compaction_failed(compact_result))
3620 goto check_priority;
3623 * make sure the compaction wasn't deferred or didn't bail out early
3624 * due to locks contention before we declare that we should give up.
3625 * But do not retry if the given zonelist is not suitable for
3628 if (compaction_withdrawn(compact_result)) {
3629 ret = compaction_zonelist_suitable(ac, order, alloc_flags);
3634 * !costly requests are much more important than __GFP_RETRY_MAYFAIL
3635 * costly ones because they are de facto nofail and invoke OOM
3636 * killer to move on while costly can fail and users are ready
3637 * to cope with that. 1/4 retries is rather arbitrary but we
3638 * would need much more detailed feedback from compaction to
3639 * make a better decision.
3641 if (order > PAGE_ALLOC_COSTLY_ORDER)
3643 if (*compaction_retries <= max_retries) {
3649 * Make sure there are attempts at the highest priority if we exhausted
3650 * all retries or failed at the lower priorities.
3653 min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ?
3654 MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY;
3656 if (*compact_priority > min_priority) {
3657 (*compact_priority)--;
3658 *compaction_retries = 0;
3662 trace_compact_retry(order, priority, compact_result, retries, max_retries, ret);
3666 static inline struct page *
3667 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
3668 unsigned int alloc_flags, const struct alloc_context *ac,
3669 enum compact_priority prio, enum compact_result *compact_result)
3671 *compact_result = COMPACT_SKIPPED;
3676 should_compact_retry(struct alloc_context *ac, unsigned int order, int alloc_flags,
3677 enum compact_result compact_result,
3678 enum compact_priority *compact_priority,
3679 int *compaction_retries)
3684 if (!order || order > PAGE_ALLOC_COSTLY_ORDER)
3688 * There are setups with compaction disabled which would prefer to loop
3689 * inside the allocator rather than hit the oom killer prematurely.
3690 * Let's give them a good hope and keep retrying while the order-0
3691 * watermarks are OK.
3693 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3695 if (zone_watermark_ok(zone, 0, min_wmark_pages(zone),
3696 ac_classzone_idx(ac), alloc_flags))
3701 #endif /* CONFIG_COMPACTION */
3703 #ifdef CONFIG_LOCKDEP
3704 struct lockdep_map __fs_reclaim_map =
3705 STATIC_LOCKDEP_MAP_INIT("fs_reclaim", &__fs_reclaim_map);
3707 static bool __need_fs_reclaim(gfp_t gfp_mask)
3709 gfp_mask = current_gfp_context(gfp_mask);
3711 /* no reclaim without waiting on it */
3712 if (!(gfp_mask & __GFP_DIRECT_RECLAIM))
3715 /* this guy won't enter reclaim */
3716 if (current->flags & PF_MEMALLOC)
3719 /* We're only interested __GFP_FS allocations for now */
3720 if (!(gfp_mask & __GFP_FS))
3723 if (gfp_mask & __GFP_NOLOCKDEP)
3729 void fs_reclaim_acquire(gfp_t gfp_mask)
3731 if (__need_fs_reclaim(gfp_mask))
3732 lock_map_acquire(&__fs_reclaim_map);
3734 EXPORT_SYMBOL_GPL(fs_reclaim_acquire);
3736 void fs_reclaim_release(gfp_t gfp_mask)
3738 if (__need_fs_reclaim(gfp_mask))
3739 lock_map_release(&__fs_reclaim_map);
3741 EXPORT_SYMBOL_GPL(fs_reclaim_release);
3744 /* Perform direct synchronous page reclaim */
3746 __perform_reclaim(gfp_t gfp_mask, unsigned int order,
3747 const struct alloc_context *ac)
3749 struct reclaim_state reclaim_state;
3751 unsigned int noreclaim_flag;
3755 /* We now go into synchronous reclaim */
3756 cpuset_memory_pressure_bump();
3757 noreclaim_flag = memalloc_noreclaim_save();
3758 fs_reclaim_acquire(gfp_mask);
3759 reclaim_state.reclaimed_slab = 0;
3760 current->reclaim_state = &reclaim_state;
3762 progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
3765 current->reclaim_state = NULL;
3766 fs_reclaim_release(gfp_mask);
3767 memalloc_noreclaim_restore(noreclaim_flag);
3774 /* The really slow allocator path where we enter direct reclaim */
3775 static inline struct page *
3776 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
3777 unsigned int alloc_flags, const struct alloc_context *ac,
3778 unsigned long *did_some_progress)
3780 struct page *page = NULL;
3781 bool drained = false;
3783 *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
3784 if (unlikely(!(*did_some_progress)))
3788 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
3791 * If an allocation failed after direct reclaim, it could be because
3792 * pages are pinned on the per-cpu lists or in high alloc reserves.
3793 * Shrink them them and try again
3795 if (!page && !drained) {
3796 unreserve_highatomic_pageblock(ac, false);
3797 drain_all_pages(NULL);
3805 static void wake_all_kswapds(unsigned int order, gfp_t gfp_mask,
3806 const struct alloc_context *ac)
3810 pg_data_t *last_pgdat = NULL;
3811 enum zone_type high_zoneidx = ac->high_zoneidx;
3813 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, high_zoneidx,
3815 if (last_pgdat != zone->zone_pgdat)
3816 wakeup_kswapd(zone, gfp_mask, order, high_zoneidx);
3817 last_pgdat = zone->zone_pgdat;
3821 static inline unsigned int
3822 gfp_to_alloc_flags(gfp_t gfp_mask)
3824 unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
3826 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
3827 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
3830 * The caller may dip into page reserves a bit more if the caller
3831 * cannot run direct reclaim, or if the caller has realtime scheduling
3832 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
3833 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
3835 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
3837 if (gfp_mask & __GFP_ATOMIC) {
3839 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
3840 * if it can't schedule.
3842 if (!(gfp_mask & __GFP_NOMEMALLOC))
3843 alloc_flags |= ALLOC_HARDER;
3845 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
3846 * comment for __cpuset_node_allowed().
3848 alloc_flags &= ~ALLOC_CPUSET;
3849 } else if (unlikely(rt_task(current)) && !in_interrupt())
3850 alloc_flags |= ALLOC_HARDER;
3853 if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
3854 alloc_flags |= ALLOC_CMA;
3859 static bool oom_reserves_allowed(struct task_struct *tsk)
3861 if (!tsk_is_oom_victim(tsk))
3865 * !MMU doesn't have oom reaper so give access to memory reserves
3866 * only to the thread with TIF_MEMDIE set
3868 if (!IS_ENABLED(CONFIG_MMU) && !test_thread_flag(TIF_MEMDIE))
3875 * Distinguish requests which really need access to full memory
3876 * reserves from oom victims which can live with a portion of it
3878 static inline int __gfp_pfmemalloc_flags(gfp_t gfp_mask)
3880 if (unlikely(gfp_mask & __GFP_NOMEMALLOC))
3882 if (gfp_mask & __GFP_MEMALLOC)
3883 return ALLOC_NO_WATERMARKS;
3884 if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
3885 return ALLOC_NO_WATERMARKS;
3886 if (!in_interrupt()) {
3887 if (current->flags & PF_MEMALLOC)
3888 return ALLOC_NO_WATERMARKS;
3889 else if (oom_reserves_allowed(current))
3896 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
3898 return !!__gfp_pfmemalloc_flags(gfp_mask);
3902 * Checks whether it makes sense to retry the reclaim to make a forward progress
3903 * for the given allocation request.
3905 * We give up when we either have tried MAX_RECLAIM_RETRIES in a row
3906 * without success, or when we couldn't even meet the watermark if we
3907 * reclaimed all remaining pages on the LRU lists.
3909 * Returns true if a retry is viable or false to enter the oom path.
3912 should_reclaim_retry(gfp_t gfp_mask, unsigned order,
3913 struct alloc_context *ac, int alloc_flags,
3914 bool did_some_progress, int *no_progress_loops)
3920 * Costly allocations might have made a progress but this doesn't mean
3921 * their order will become available due to high fragmentation so
3922 * always increment the no progress counter for them
3924 if (did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER)
3925 *no_progress_loops = 0;
3927 (*no_progress_loops)++;
3930 * Make sure we converge to OOM if we cannot make any progress
3931 * several times in the row.
3933 if (*no_progress_loops > MAX_RECLAIM_RETRIES) {
3934 /* Before OOM, exhaust highatomic_reserve */
3935 return unreserve_highatomic_pageblock(ac, true);
3939 * Keep reclaiming pages while there is a chance this will lead
3940 * somewhere. If none of the target zones can satisfy our allocation
3941 * request even if all reclaimable pages are considered then we are
3942 * screwed and have to go OOM.
3944 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
3946 unsigned long available;
3947 unsigned long reclaimable;
3948 unsigned long min_wmark = min_wmark_pages(zone);
3951 available = reclaimable = zone_reclaimable_pages(zone);
3952 available += zone_page_state_snapshot(zone, NR_FREE_PAGES);
3955 * Would the allocation succeed if we reclaimed all
3956 * reclaimable pages?
3958 wmark = __zone_watermark_ok(zone, order, min_wmark,
3959 ac_classzone_idx(ac), alloc_flags, available);
3960 trace_reclaim_retry_zone(z, order, reclaimable,
3961 available, min_wmark, *no_progress_loops, wmark);
3964 * If we didn't make any progress and have a lot of
3965 * dirty + writeback pages then we should wait for
3966 * an IO to complete to slow down the reclaim and
3967 * prevent from pre mature OOM
3969 if (!did_some_progress) {
3970 unsigned long write_pending;
3972 write_pending = zone_page_state_snapshot(zone,
3973 NR_ZONE_WRITE_PENDING);
3975 if (2 * write_pending > reclaimable) {
3976 congestion_wait(BLK_RW_ASYNC, HZ/10);
3982 * Memory allocation/reclaim might be called from a WQ
3983 * context and the current implementation of the WQ
3984 * concurrency control doesn't recognize that
3985 * a particular WQ is congested if the worker thread is
3986 * looping without ever sleeping. Therefore we have to
3987 * do a short sleep here rather than calling
3990 if (current->flags & PF_WQ_WORKER)
3991 schedule_timeout_uninterruptible(1);
4003 check_retry_cpuset(int cpuset_mems_cookie, struct alloc_context *ac)
4006 * It's possible that cpuset's mems_allowed and the nodemask from
4007 * mempolicy don't intersect. This should be normally dealt with by
4008 * policy_nodemask(), but it's possible to race with cpuset update in
4009 * such a way the check therein was true, and then it became false
4010 * before we got our cpuset_mems_cookie here.
4011 * This assumes that for all allocations, ac->nodemask can come only
4012 * from MPOL_BIND mempolicy (whose documented semantics is to be ignored
4013 * when it does not intersect with the cpuset restrictions) or the
4014 * caller can deal with a violated nodemask.
4016 if (cpusets_enabled() && ac->nodemask &&
4017 !cpuset_nodemask_valid_mems_allowed(ac->nodemask)) {
4018 ac->nodemask = NULL;
4023 * When updating a task's mems_allowed or mempolicy nodemask, it is
4024 * possible to race with parallel threads in such a way that our
4025 * allocation can fail while the mask is being updated. If we are about
4026 * to fail, check if the cpuset changed during allocation and if so,
4029 if (read_mems_allowed_retry(cpuset_mems_cookie))
4035 static inline struct page *
4036 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
4037 struct alloc_context *ac)
4039 bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
4040 const bool costly_order = order > PAGE_ALLOC_COSTLY_ORDER;
4041 struct page *page = NULL;
4042 unsigned int alloc_flags;
4043 unsigned long did_some_progress;
4044 enum compact_priority compact_priority;
4045 enum compact_result compact_result;
4046 int compaction_retries;
4047 int no_progress_loops;
4048 unsigned int cpuset_mems_cookie;
4052 * In the slowpath, we sanity check order to avoid ever trying to
4053 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
4054 * be using allocators in order of preference for an area that is
4057 if (order >= MAX_ORDER) {
4058 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
4063 * We also sanity check to catch abuse of atomic reserves being used by
4064 * callers that are not in atomic context.
4066 if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
4067 (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
4068 gfp_mask &= ~__GFP_ATOMIC;
4071 compaction_retries = 0;
4072 no_progress_loops = 0;
4073 compact_priority = DEF_COMPACT_PRIORITY;
4074 cpuset_mems_cookie = read_mems_allowed_begin();
4077 * The fast path uses conservative alloc_flags to succeed only until
4078 * kswapd needs to be woken up, and to avoid the cost of setting up
4079 * alloc_flags precisely. So we do that now.
4081 alloc_flags = gfp_to_alloc_flags(gfp_mask);
4084 * We need to recalculate the starting point for the zonelist iterator
4085 * because we might have used different nodemask in the fast path, or
4086 * there was a cpuset modification and we are retrying - otherwise we
4087 * could end up iterating over non-eligible zones endlessly.
4089 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4090 ac->high_zoneidx, ac->nodemask);
4091 if (!ac->preferred_zoneref->zone)
4094 if (gfp_mask & __GFP_KSWAPD_RECLAIM)
4095 wake_all_kswapds(order, gfp_mask, ac);
4098 * The adjusted alloc_flags might result in immediate success, so try
4101 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4106 * For costly allocations, try direct compaction first, as it's likely
4107 * that we have enough base pages and don't need to reclaim. For non-
4108 * movable high-order allocations, do that as well, as compaction will
4109 * try prevent permanent fragmentation by migrating from blocks of the
4111 * Don't try this for allocations that are allowed to ignore
4112 * watermarks, as the ALLOC_NO_WATERMARKS attempt didn't yet happen.
4114 if (can_direct_reclaim &&
4116 (order > 0 && ac->migratetype != MIGRATE_MOVABLE))
4117 && !gfp_pfmemalloc_allowed(gfp_mask)) {
4118 page = __alloc_pages_direct_compact(gfp_mask, order,
4120 INIT_COMPACT_PRIORITY,
4126 * Checks for costly allocations with __GFP_NORETRY, which
4127 * includes THP page fault allocations
4129 if (costly_order && (gfp_mask & __GFP_NORETRY)) {
4131 * If compaction is deferred for high-order allocations,
4132 * it is because sync compaction recently failed. If
4133 * this is the case and the caller requested a THP
4134 * allocation, we do not want to heavily disrupt the
4135 * system, so we fail the allocation instead of entering
4138 if (compact_result == COMPACT_DEFERRED)
4142 * Looks like reclaim/compaction is worth trying, but
4143 * sync compaction could be very expensive, so keep
4144 * using async compaction.
4146 compact_priority = INIT_COMPACT_PRIORITY;
4151 /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */
4152 if (gfp_mask & __GFP_KSWAPD_RECLAIM)
4153 wake_all_kswapds(order, gfp_mask, ac);
4155 reserve_flags = __gfp_pfmemalloc_flags(gfp_mask);
4157 alloc_flags = reserve_flags;
4160 * Reset the zonelist iterators if memory policies can be ignored.
4161 * These allocations are high priority and system rather than user
4164 if (!(alloc_flags & ALLOC_CPUSET) || reserve_flags) {
4165 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
4166 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4167 ac->high_zoneidx, ac->nodemask);
4170 /* Attempt with potentially adjusted zonelist and alloc_flags */
4171 page = get_page_from_freelist(gfp_mask, order, alloc_flags, ac);
4175 /* Caller is not willing to reclaim, we can't balance anything */
4176 if (!can_direct_reclaim)
4179 /* Avoid recursion of direct reclaim */
4180 if (current->flags & PF_MEMALLOC)
4183 /* Try direct reclaim and then allocating */
4184 page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
4185 &did_some_progress);
4189 /* Try direct compaction and then allocating */
4190 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
4191 compact_priority, &compact_result);
4195 /* Do not loop if specifically requested */
4196 if (gfp_mask & __GFP_NORETRY)
4200 * Do not retry costly high order allocations unless they are
4201 * __GFP_RETRY_MAYFAIL
4203 if (costly_order && !(gfp_mask & __GFP_RETRY_MAYFAIL))
4206 if (should_reclaim_retry(gfp_mask, order, ac, alloc_flags,
4207 did_some_progress > 0, &no_progress_loops))
4211 * It doesn't make any sense to retry for the compaction if the order-0
4212 * reclaim is not able to make any progress because the current
4213 * implementation of the compaction depends on the sufficient amount
4214 * of free memory (see __compaction_suitable)
4216 if (did_some_progress > 0 &&
4217 should_compact_retry(ac, order, alloc_flags,
4218 compact_result, &compact_priority,
4219 &compaction_retries))
4223 /* Deal with possible cpuset update races before we start OOM killing */
4224 if (check_retry_cpuset(cpuset_mems_cookie, ac))
4227 /* Reclaim has failed us, start killing things */
4228 page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
4232 /* Avoid allocations with no watermarks from looping endlessly */
4233 if (tsk_is_oom_victim(current) &&
4234 (alloc_flags == ALLOC_OOM ||
4235 (gfp_mask & __GFP_NOMEMALLOC)))
4238 /* Retry as long as the OOM killer is making progress */
4239 if (did_some_progress) {
4240 no_progress_loops = 0;
4245 /* Deal with possible cpuset update races before we fail */
4246 if (check_retry_cpuset(cpuset_mems_cookie, ac))
4250 * Make sure that __GFP_NOFAIL request doesn't leak out and make sure
4253 if (gfp_mask & __GFP_NOFAIL) {
4255 * All existing users of the __GFP_NOFAIL are blockable, so warn
4256 * of any new users that actually require GFP_NOWAIT
4258 if (WARN_ON_ONCE(!can_direct_reclaim))
4262 * PF_MEMALLOC request from this context is rather bizarre
4263 * because we cannot reclaim anything and only can loop waiting
4264 * for somebody to do a work for us
4266 WARN_ON_ONCE(current->flags & PF_MEMALLOC);
4269 * non failing costly orders are a hard requirement which we
4270 * are not prepared for much so let's warn about these users
4271 * so that we can identify them and convert them to something
4274 WARN_ON_ONCE(order > PAGE_ALLOC_COSTLY_ORDER);
4277 * Help non-failing allocations by giving them access to memory
4278 * reserves but do not use ALLOC_NO_WATERMARKS because this
4279 * could deplete whole memory reserves which would just make
4280 * the situation worse
4282 page = __alloc_pages_cpuset_fallback(gfp_mask, order, ALLOC_HARDER, ac);
4290 warn_alloc(gfp_mask, ac->nodemask,
4291 "page allocation failure: order:%u", order);
4296 static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order,
4297 int preferred_nid, nodemask_t *nodemask,
4298 struct alloc_context *ac, gfp_t *alloc_mask,
4299 unsigned int *alloc_flags)
4301 ac->high_zoneidx = gfp_zone(gfp_mask);
4302 ac->zonelist = node_zonelist(preferred_nid, gfp_mask);
4303 ac->nodemask = nodemask;
4304 ac->migratetype = gfpflags_to_migratetype(gfp_mask);
4306 if (cpusets_enabled()) {
4307 *alloc_mask |= __GFP_HARDWALL;
4309 ac->nodemask = &cpuset_current_mems_allowed;
4311 *alloc_flags |= ALLOC_CPUSET;
4314 fs_reclaim_acquire(gfp_mask);
4315 fs_reclaim_release(gfp_mask);
4317 might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
4319 if (should_fail_alloc_page(gfp_mask, order))
4322 if (IS_ENABLED(CONFIG_CMA) && ac->migratetype == MIGRATE_MOVABLE)
4323 *alloc_flags |= ALLOC_CMA;
4328 /* Determine whether to spread dirty pages and what the first usable zone */
4329 static inline void finalise_ac(gfp_t gfp_mask,
4330 unsigned int order, struct alloc_context *ac)
4332 /* Dirty zone balancing only done in the fast path */
4333 ac->spread_dirty_pages = (gfp_mask & __GFP_WRITE);
4336 * The preferred zone is used for statistics but crucially it is
4337 * also used as the starting point for the zonelist iterator. It
4338 * may get reset for allocations that ignore memory policies.
4340 ac->preferred_zoneref = first_zones_zonelist(ac->zonelist,
4341 ac->high_zoneidx, ac->nodemask);
4345 * This is the 'heart' of the zoned buddy allocator.
4348 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid,
4349 nodemask_t *nodemask)
4352 unsigned int alloc_flags = ALLOC_WMARK_LOW;
4353 gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
4354 struct alloc_context ac = { };
4356 gfp_mask &= gfp_allowed_mask;
4357 alloc_mask = gfp_mask;
4358 if (!prepare_alloc_pages(gfp_mask, order, preferred_nid, nodemask, &ac, &alloc_mask, &alloc_flags))
4361 finalise_ac(gfp_mask, order, &ac);
4363 /* First allocation attempt */
4364 page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
4369 * Apply scoped allocation constraints. This is mainly about GFP_NOFS
4370 * resp. GFP_NOIO which has to be inherited for all allocation requests
4371 * from a particular context which has been marked by
4372 * memalloc_no{fs,io}_{save,restore}.
4374 alloc_mask = current_gfp_context(gfp_mask);
4375 ac.spread_dirty_pages = false;
4378 * Restore the original nodemask if it was potentially replaced with
4379 * &cpuset_current_mems_allowed to optimize the fast-path attempt.
4381 if (unlikely(ac.nodemask != nodemask))
4382 ac.nodemask = nodemask;
4384 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
4387 if (memcg_kmem_enabled() && (gfp_mask & __GFP_ACCOUNT) && page &&
4388 unlikely(memcg_kmem_charge(page, gfp_mask, order) != 0)) {
4389 __free_pages(page, order);
4393 trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
4397 EXPORT_SYMBOL(__alloc_pages_nodemask);
4400 * Common helper functions.
4402 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
4407 * __get_free_pages() returns a virtual address, which cannot represent
4410 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
4412 page = alloc_pages(gfp_mask, order);
4415 return (unsigned long) page_address(page);
4417 EXPORT_SYMBOL(__get_free_pages);
4419 unsigned long get_zeroed_page(gfp_t gfp_mask)
4421 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
4423 EXPORT_SYMBOL(get_zeroed_page);
4425 void __free_pages(struct page *page, unsigned int order)
4427 if (put_page_testzero(page)) {
4429 free_unref_page(page);
4431 __free_pages_ok(page, order);
4435 EXPORT_SYMBOL(__free_pages);
4437 void free_pages(unsigned long addr, unsigned int order)
4440 VM_BUG_ON(!virt_addr_valid((void *)addr));
4441 __free_pages(virt_to_page((void *)addr), order);
4445 EXPORT_SYMBOL(free_pages);
4449 * An arbitrary-length arbitrary-offset area of memory which resides
4450 * within a 0 or higher order page. Multiple fragments within that page
4451 * are individually refcounted, in the page's reference counter.
4453 * The page_frag functions below provide a simple allocation framework for
4454 * page fragments. This is used by the network stack and network device
4455 * drivers to provide a backing region of memory for use as either an
4456 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
4458 static struct page *__page_frag_cache_refill(struct page_frag_cache *nc,
4461 struct page *page = NULL;
4462 gfp_t gfp = gfp_mask;
4464 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4465 gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
4467 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
4468 PAGE_FRAG_CACHE_MAX_ORDER);
4469 nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
4471 if (unlikely(!page))
4472 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
4474 nc->va = page ? page_address(page) : NULL;
4479 void __page_frag_cache_drain(struct page *page, unsigned int count)
4481 VM_BUG_ON_PAGE(page_ref_count(page) == 0, page);
4483 if (page_ref_sub_and_test(page, count)) {
4484 unsigned int order = compound_order(page);
4487 free_unref_page(page);
4489 __free_pages_ok(page, order);
4492 EXPORT_SYMBOL(__page_frag_cache_drain);
4494 void *page_frag_alloc(struct page_frag_cache *nc,
4495 unsigned int fragsz, gfp_t gfp_mask)
4497 unsigned int size = PAGE_SIZE;
4501 if (unlikely(!nc->va)) {
4503 page = __page_frag_cache_refill(nc, gfp_mask);
4507 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4508 /* if size can vary use size else just use PAGE_SIZE */
4511 /* Even if we own the page, we do not use atomic_set().
4512 * This would break get_page_unless_zero() users.
4514 page_ref_add(page, size - 1);
4516 /* reset page count bias and offset to start of new frag */
4517 nc->pfmemalloc = page_is_pfmemalloc(page);
4518 nc->pagecnt_bias = size;
4522 offset = nc->offset - fragsz;
4523 if (unlikely(offset < 0)) {
4524 page = virt_to_page(nc->va);
4526 if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
4529 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4530 /* if size can vary use size else just use PAGE_SIZE */
4533 /* OK, page count is 0, we can safely set it */
4534 set_page_count(page, size);
4536 /* reset page count bias and offset to start of new frag */
4537 nc->pagecnt_bias = size;
4538 offset = size - fragsz;
4542 nc->offset = offset;
4544 return nc->va + offset;
4546 EXPORT_SYMBOL(page_frag_alloc);
4549 * Frees a page fragment allocated out of either a compound or order 0 page.
4551 void page_frag_free(void *addr)
4553 struct page *page = virt_to_head_page(addr);
4555 if (unlikely(put_page_testzero(page)))
4556 __free_pages_ok(page, compound_order(page));
4558 EXPORT_SYMBOL(page_frag_free);
4560 static void *make_alloc_exact(unsigned long addr, unsigned int order,
4564 unsigned long alloc_end = addr + (PAGE_SIZE << order);
4565 unsigned long used = addr + PAGE_ALIGN(size);
4567 split_page(virt_to_page((void *)addr), order);
4568 while (used < alloc_end) {
4573 return (void *)addr;
4577 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
4578 * @size: the number of bytes to allocate
4579 * @gfp_mask: GFP flags for the allocation
4581 * This function is similar to alloc_pages(), except that it allocates the
4582 * minimum number of pages to satisfy the request. alloc_pages() can only
4583 * allocate memory in power-of-two pages.
4585 * This function is also limited by MAX_ORDER.
4587 * Memory allocated by this function must be released by free_pages_exact().
4589 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
4591 unsigned int order = get_order(size);
4594 addr = __get_free_pages(gfp_mask, order);
4595 return make_alloc_exact(addr, order, size);
4597 EXPORT_SYMBOL(alloc_pages_exact);
4600 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
4602 * @nid: the preferred node ID where memory should be allocated
4603 * @size: the number of bytes to allocate
4604 * @gfp_mask: GFP flags for the allocation
4606 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
4609 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
4611 unsigned int order = get_order(size);
4612 struct page *p = alloc_pages_node(nid, gfp_mask, order);
4615 return make_alloc_exact((unsigned long)page_address(p), order, size);
4619 * free_pages_exact - release memory allocated via alloc_pages_exact()
4620 * @virt: the value returned by alloc_pages_exact.
4621 * @size: size of allocation, same value as passed to alloc_pages_exact().
4623 * Release the memory allocated by a previous call to alloc_pages_exact.
4625 void free_pages_exact(void *virt, size_t size)
4627 unsigned long addr = (unsigned long)virt;
4628 unsigned long end = addr + PAGE_ALIGN(size);
4630 while (addr < end) {
4635 EXPORT_SYMBOL(free_pages_exact);
4638 * nr_free_zone_pages - count number of pages beyond high watermark
4639 * @offset: The zone index of the highest zone
4641 * nr_free_zone_pages() counts the number of counts pages which are beyond the
4642 * high watermark within all zones at or below a given zone index. For each
4643 * zone, the number of pages is calculated as:
4645 * nr_free_zone_pages = managed_pages - high_pages
4647 static unsigned long nr_free_zone_pages(int offset)
4652 /* Just pick one node, since fallback list is circular */
4653 unsigned long sum = 0;
4655 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
4657 for_each_zone_zonelist(zone, z, zonelist, offset) {
4658 unsigned long size = zone->managed_pages;
4659 unsigned long high = high_wmark_pages(zone);
4668 * nr_free_buffer_pages - count number of pages beyond high watermark
4670 * nr_free_buffer_pages() counts the number of pages which are beyond the high
4671 * watermark within ZONE_DMA and ZONE_NORMAL.
4673 unsigned long nr_free_buffer_pages(void)
4675 return nr_free_zone_pages(gfp_zone(GFP_USER));
4677 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
4680 * nr_free_pagecache_pages - count number of pages beyond high watermark
4682 * nr_free_pagecache_pages() counts the number of pages which are beyond the
4683 * high watermark within all zones.
4685 unsigned long nr_free_pagecache_pages(void)
4687 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
4690 static inline void show_node(struct zone *zone)
4692 if (IS_ENABLED(CONFIG_NUMA))
4693 printk("Node %d ", zone_to_nid(zone));
4696 long si_mem_available(void)
4699 unsigned long pagecache;
4700 unsigned long wmark_low = 0;
4701 unsigned long pages[NR_LRU_LISTS];
4705 for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++)
4706 pages[lru] = global_node_page_state(NR_LRU_BASE + lru);
4709 wmark_low += zone->watermark[WMARK_LOW];
4712 * Estimate the amount of memory available for userspace allocations,
4713 * without causing swapping.
4715 available = global_zone_page_state(NR_FREE_PAGES) - totalreserve_pages;
4718 * Not all the page cache can be freed, otherwise the system will
4719 * start swapping. Assume at least half of the page cache, or the
4720 * low watermark worth of cache, needs to stay.
4722 pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE];
4723 pagecache -= min(pagecache / 2, wmark_low);
4724 available += pagecache;
4727 * Part of the reclaimable slab consists of items that are in use,
4728 * and cannot be freed. Cap this estimate at the low watermark.
4730 available += global_node_page_state(NR_SLAB_RECLAIMABLE) -
4731 min(global_node_page_state(NR_SLAB_RECLAIMABLE) / 2,
4735 * Part of the kernel memory, which can be released under memory
4738 available += global_node_page_state(NR_INDIRECTLY_RECLAIMABLE_BYTES) >>
4745 EXPORT_SYMBOL_GPL(si_mem_available);
4747 void si_meminfo(struct sysinfo *val)
4749 val->totalram = totalram_pages;
4750 val->sharedram = global_node_page_state(NR_SHMEM);
4751 val->freeram = global_zone_page_state(NR_FREE_PAGES);
4752 val->bufferram = nr_blockdev_pages();
4753 val->totalhigh = totalhigh_pages;
4754 val->freehigh = nr_free_highpages();
4755 val->mem_unit = PAGE_SIZE;
4758 EXPORT_SYMBOL(si_meminfo);
4761 void si_meminfo_node(struct sysinfo *val, int nid)
4763 int zone_type; /* needs to be signed */
4764 unsigned long managed_pages = 0;
4765 unsigned long managed_highpages = 0;
4766 unsigned long free_highpages = 0;
4767 pg_data_t *pgdat = NODE_DATA(nid);
4769 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
4770 managed_pages += pgdat->node_zones[zone_type].managed_pages;
4771 val->totalram = managed_pages;
4772 val->sharedram = node_page_state(pgdat, NR_SHMEM);
4773 val->freeram = sum_zone_node_page_state(nid, NR_FREE_PAGES);
4774 #ifdef CONFIG_HIGHMEM
4775 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
4776 struct zone *zone = &pgdat->node_zones[zone_type];
4778 if (is_highmem(zone)) {
4779 managed_highpages += zone->managed_pages;
4780 free_highpages += zone_page_state(zone, NR_FREE_PAGES);
4783 val->totalhigh = managed_highpages;
4784 val->freehigh = free_highpages;
4786 val->totalhigh = managed_highpages;
4787 val->freehigh = free_highpages;
4789 val->mem_unit = PAGE_SIZE;
4794 * Determine whether the node should be displayed or not, depending on whether
4795 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
4797 static bool show_mem_node_skip(unsigned int flags, int nid, nodemask_t *nodemask)
4799 if (!(flags & SHOW_MEM_FILTER_NODES))
4803 * no node mask - aka implicit memory numa policy. Do not bother with
4804 * the synchronization - read_mems_allowed_begin - because we do not
4805 * have to be precise here.
4808 nodemask = &cpuset_current_mems_allowed;
4810 return !node_isset(nid, *nodemask);
4813 #define K(x) ((x) << (PAGE_SHIFT-10))
4815 static void show_migration_types(unsigned char type)
4817 static const char types[MIGRATE_TYPES] = {
4818 [MIGRATE_UNMOVABLE] = 'U',
4819 [MIGRATE_MOVABLE] = 'M',
4820 [MIGRATE_RECLAIMABLE] = 'E',
4821 [MIGRATE_HIGHATOMIC] = 'H',
4823 [MIGRATE_CMA] = 'C',
4825 #ifdef CONFIG_MEMORY_ISOLATION
4826 [MIGRATE_ISOLATE] = 'I',
4829 char tmp[MIGRATE_TYPES + 1];
4833 for (i = 0; i < MIGRATE_TYPES; i++) {
4834 if (type & (1 << i))
4839 printk(KERN_CONT "(%s) ", tmp);
4843 * Show free area list (used inside shift_scroll-lock stuff)
4844 * We also calculate the percentage fragmentation. We do this by counting the
4845 * memory on each free list with the exception of the first item on the list.
4848 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
4851 void show_free_areas(unsigned int filter, nodemask_t *nodemask)
4853 unsigned long free_pcp = 0;
4858 for_each_populated_zone(zone) {
4859 if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
4862 for_each_online_cpu(cpu)
4863 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
4866 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
4867 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
4868 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
4869 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
4870 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
4871 " free:%lu free_pcp:%lu free_cma:%lu\n",
4872 global_node_page_state(NR_ACTIVE_ANON),
4873 global_node_page_state(NR_INACTIVE_ANON),
4874 global_node_page_state(NR_ISOLATED_ANON),
4875 global_node_page_state(NR_ACTIVE_FILE),
4876 global_node_page_state(NR_INACTIVE_FILE),
4877 global_node_page_state(NR_ISOLATED_FILE),
4878 global_node_page_state(NR_UNEVICTABLE),
4879 global_node_page_state(NR_FILE_DIRTY),
4880 global_node_page_state(NR_WRITEBACK),
4881 global_node_page_state(NR_UNSTABLE_NFS),
4882 global_node_page_state(NR_SLAB_RECLAIMABLE),
4883 global_node_page_state(NR_SLAB_UNRECLAIMABLE),
4884 global_node_page_state(NR_FILE_MAPPED),
4885 global_node_page_state(NR_SHMEM),
4886 global_zone_page_state(NR_PAGETABLE),
4887 global_zone_page_state(NR_BOUNCE),
4888 global_zone_page_state(NR_FREE_PAGES),
4890 global_zone_page_state(NR_FREE_CMA_PAGES));
4892 for_each_online_pgdat(pgdat) {
4893 if (show_mem_node_skip(filter, pgdat->node_id, nodemask))
4897 " active_anon:%lukB"
4898 " inactive_anon:%lukB"
4899 " active_file:%lukB"
4900 " inactive_file:%lukB"
4901 " unevictable:%lukB"
4902 " isolated(anon):%lukB"
4903 " isolated(file):%lukB"
4908 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4910 " shmem_pmdmapped: %lukB"
4913 " writeback_tmp:%lukB"
4915 " all_unreclaimable? %s"
4918 K(node_page_state(pgdat, NR_ACTIVE_ANON)),
4919 K(node_page_state(pgdat, NR_INACTIVE_ANON)),
4920 K(node_page_state(pgdat, NR_ACTIVE_FILE)),
4921 K(node_page_state(pgdat, NR_INACTIVE_FILE)),
4922 K(node_page_state(pgdat, NR_UNEVICTABLE)),
4923 K(node_page_state(pgdat, NR_ISOLATED_ANON)),
4924 K(node_page_state(pgdat, NR_ISOLATED_FILE)),
4925 K(node_page_state(pgdat, NR_FILE_MAPPED)),
4926 K(node_page_state(pgdat, NR_FILE_DIRTY)),
4927 K(node_page_state(pgdat, NR_WRITEBACK)),
4928 K(node_page_state(pgdat, NR_SHMEM)),
4929 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4930 K(node_page_state(pgdat, NR_SHMEM_THPS) * HPAGE_PMD_NR),
4931 K(node_page_state(pgdat, NR_SHMEM_PMDMAPPED)
4933 K(node_page_state(pgdat, NR_ANON_THPS) * HPAGE_PMD_NR),
4935 K(node_page_state(pgdat, NR_WRITEBACK_TEMP)),
4936 K(node_page_state(pgdat, NR_UNSTABLE_NFS)),
4937 pgdat->kswapd_failures >= MAX_RECLAIM_RETRIES ?
4941 for_each_populated_zone(zone) {
4944 if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
4948 for_each_online_cpu(cpu)
4949 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
4958 " active_anon:%lukB"
4959 " inactive_anon:%lukB"
4960 " active_file:%lukB"
4961 " inactive_file:%lukB"
4962 " unevictable:%lukB"
4963 " writepending:%lukB"
4967 " kernel_stack:%lukB"
4975 K(zone_page_state(zone, NR_FREE_PAGES)),
4976 K(min_wmark_pages(zone)),
4977 K(low_wmark_pages(zone)),
4978 K(high_wmark_pages(zone)),
4979 K(zone_page_state(zone, NR_ZONE_ACTIVE_ANON)),
4980 K(zone_page_state(zone, NR_ZONE_INACTIVE_ANON)),
4981 K(zone_page_state(zone, NR_ZONE_ACTIVE_FILE)),
4982 K(zone_page_state(zone, NR_ZONE_INACTIVE_FILE)),
4983 K(zone_page_state(zone, NR_ZONE_UNEVICTABLE)),
4984 K(zone_page_state(zone, NR_ZONE_WRITE_PENDING)),
4985 K(zone->present_pages),
4986 K(zone->managed_pages),
4987 K(zone_page_state(zone, NR_MLOCK)),
4988 zone_page_state(zone, NR_KERNEL_STACK_KB),
4989 K(zone_page_state(zone, NR_PAGETABLE)),
4990 K(zone_page_state(zone, NR_BOUNCE)),
4992 K(this_cpu_read(zone->pageset->pcp.count)),
4993 K(zone_page_state(zone, NR_FREE_CMA_PAGES)));
4994 printk("lowmem_reserve[]:");
4995 for (i = 0; i < MAX_NR_ZONES; i++)
4996 printk(KERN_CONT " %ld", zone->lowmem_reserve[i]);
4997 printk(KERN_CONT "\n");
5000 for_each_populated_zone(zone) {
5002 unsigned long nr[MAX_ORDER], flags, total = 0;
5003 unsigned char types[MAX_ORDER];
5005 if (show_mem_node_skip(filter, zone_to_nid(zone), nodemask))
5008 printk(KERN_CONT "%s: ", zone->name);
5010 spin_lock_irqsave(&zone->lock, flags);
5011 for (order = 0; order < MAX_ORDER; order++) {
5012 struct free_area *area = &zone->free_area[order];
5015 nr[order] = area->nr_free;
5016 total += nr[order] << order;
5019 for (type = 0; type < MIGRATE_TYPES; type++) {
5020 if (!list_empty(&area->free_list[type]))
5021 types[order] |= 1 << type;
5024 spin_unlock_irqrestore(&zone->lock, flags);
5025 for (order = 0; order < MAX_ORDER; order++) {
5026 printk(KERN_CONT "%lu*%lukB ",
5027 nr[order], K(1UL) << order);
5029 show_migration_types(types[order]);
5031 printk(KERN_CONT "= %lukB\n", K(total));
5034 hugetlb_show_meminfo();
5036 printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES));
5038 show_swap_cache_info();
5041 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
5043 zoneref->zone = zone;
5044 zoneref->zone_idx = zone_idx(zone);
5048 * Builds allocation fallback zone lists.
5050 * Add all populated zones of a node to the zonelist.
5052 static int build_zonerefs_node(pg_data_t *pgdat, struct zoneref *zonerefs)
5055 enum zone_type zone_type = MAX_NR_ZONES;
5060 zone = pgdat->node_zones + zone_type;
5061 if (managed_zone(zone)) {
5062 zoneref_set_zone(zone, &zonerefs[nr_zones++]);
5063 check_highest_zone(zone_type);
5065 } while (zone_type);
5072 static int __parse_numa_zonelist_order(char *s)
5075 * We used to support different zonlists modes but they turned
5076 * out to be just not useful. Let's keep the warning in place
5077 * if somebody still use the cmd line parameter so that we do
5078 * not fail it silently
5080 if (!(*s == 'd' || *s == 'D' || *s == 'n' || *s == 'N')) {
5081 pr_warn("Ignoring unsupported numa_zonelist_order value: %s\n", s);
5087 static __init int setup_numa_zonelist_order(char *s)
5092 return __parse_numa_zonelist_order(s);
5094 early_param("numa_zonelist_order", setup_numa_zonelist_order);
5096 char numa_zonelist_order[] = "Node";
5099 * sysctl handler for numa_zonelist_order
5101 int numa_zonelist_order_handler(struct ctl_table *table, int write,
5102 void __user *buffer, size_t *length,
5109 return proc_dostring(table, write, buffer, length, ppos);
5110 str = memdup_user_nul(buffer, 16);
5112 return PTR_ERR(str);
5114 ret = __parse_numa_zonelist_order(str);
5120 #define MAX_NODE_LOAD (nr_online_nodes)
5121 static int node_load[MAX_NUMNODES];
5124 * find_next_best_node - find the next node that should appear in a given node's fallback list
5125 * @node: node whose fallback list we're appending
5126 * @used_node_mask: nodemask_t of already used nodes
5128 * We use a number of factors to determine which is the next node that should
5129 * appear on a given node's fallback list. The node should not have appeared
5130 * already in @node's fallback list, and it should be the next closest node
5131 * according to the distance array (which contains arbitrary distance values
5132 * from each node to each node in the system), and should also prefer nodes
5133 * with no CPUs, since presumably they'll have very little allocation pressure
5134 * on them otherwise.
5135 * It returns -1 if no node is found.
5137 static int find_next_best_node(int node, nodemask_t *used_node_mask)
5140 int min_val = INT_MAX;
5141 int best_node = NUMA_NO_NODE;
5142 const struct cpumask *tmp = cpumask_of_node(0);
5144 /* Use the local node if we haven't already */
5145 if (!node_isset(node, *used_node_mask)) {
5146 node_set(node, *used_node_mask);
5150 for_each_node_state(n, N_MEMORY) {
5152 /* Don't want a node to appear more than once */
5153 if (node_isset(n, *used_node_mask))
5156 /* Use the distance array to find the distance */
5157 val = node_distance(node, n);
5159 /* Penalize nodes under us ("prefer the next node") */
5162 /* Give preference to headless and unused nodes */
5163 tmp = cpumask_of_node(n);
5164 if (!cpumask_empty(tmp))
5165 val += PENALTY_FOR_NODE_WITH_CPUS;
5167 /* Slight preference for less loaded node */
5168 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
5169 val += node_load[n];
5171 if (val < min_val) {
5178 node_set(best_node, *used_node_mask);
5185 * Build zonelists ordered by node and zones within node.
5186 * This results in maximum locality--normal zone overflows into local
5187 * DMA zone, if any--but risks exhausting DMA zone.
5189 static void build_zonelists_in_node_order(pg_data_t *pgdat, int *node_order,
5192 struct zoneref *zonerefs;
5195 zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs;
5197 for (i = 0; i < nr_nodes; i++) {
5200 pg_data_t *node = NODE_DATA(node_order[i]);
5202 nr_zones = build_zonerefs_node(node, zonerefs);
5203 zonerefs += nr_zones;
5205 zonerefs->zone = NULL;
5206 zonerefs->zone_idx = 0;
5210 * Build gfp_thisnode zonelists
5212 static void build_thisnode_zonelists(pg_data_t *pgdat)
5214 struct zoneref *zonerefs;
5217 zonerefs = pgdat->node_zonelists[ZONELIST_NOFALLBACK]._zonerefs;
5218 nr_zones = build_zonerefs_node(pgdat, zonerefs);
5219 zonerefs += nr_zones;
5220 zonerefs->zone = NULL;
5221 zonerefs->zone_idx = 0;
5225 * Build zonelists ordered by zone and nodes within zones.
5226 * This results in conserving DMA zone[s] until all Normal memory is
5227 * exhausted, but results in overflowing to remote node while memory
5228 * may still exist in local DMA zone.
5231 static void build_zonelists(pg_data_t *pgdat)
5233 static int node_order[MAX_NUMNODES];
5234 int node, load, nr_nodes = 0;
5235 nodemask_t used_mask;
5236 int local_node, prev_node;
5238 /* NUMA-aware ordering of nodes */
5239 local_node = pgdat->node_id;
5240 load = nr_online_nodes;
5241 prev_node = local_node;
5242 nodes_clear(used_mask);
5244 memset(node_order, 0, sizeof(node_order));
5245 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
5247 * We don't want to pressure a particular node.
5248 * So adding penalty to the first node in same
5249 * distance group to make it round-robin.
5251 if (node_distance(local_node, node) !=
5252 node_distance(local_node, prev_node))
5253 node_load[node] = load;
5255 node_order[nr_nodes++] = node;
5260 build_zonelists_in_node_order(pgdat, node_order, nr_nodes);
5261 build_thisnode_zonelists(pgdat);
5264 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
5266 * Return node id of node used for "local" allocations.
5267 * I.e., first node id of first zone in arg node's generic zonelist.
5268 * Used for initializing percpu 'numa_mem', which is used primarily
5269 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
5271 int local_memory_node(int node)
5275 z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
5276 gfp_zone(GFP_KERNEL),
5278 return z->zone->node;
5282 static void setup_min_unmapped_ratio(void);
5283 static void setup_min_slab_ratio(void);
5284 #else /* CONFIG_NUMA */
5286 static void build_zonelists(pg_data_t *pgdat)
5288 int node, local_node;
5289 struct zoneref *zonerefs;
5292 local_node = pgdat->node_id;
5294 zonerefs = pgdat->node_zonelists[ZONELIST_FALLBACK]._zonerefs;
5295 nr_zones = build_zonerefs_node(pgdat, zonerefs);
5296 zonerefs += nr_zones;
5299 * Now we build the zonelist so that it contains the zones
5300 * of all the other nodes.
5301 * We don't want to pressure a particular node, so when
5302 * building the zones for node N, we make sure that the
5303 * zones coming right after the local ones are those from
5304 * node N+1 (modulo N)
5306 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
5307 if (!node_online(node))
5309 nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs);
5310 zonerefs += nr_zones;
5312 for (node = 0; node < local_node; node++) {
5313 if (!node_online(node))
5315 nr_zones = build_zonerefs_node(NODE_DATA(node), zonerefs);
5316 zonerefs += nr_zones;
5319 zonerefs->zone = NULL;
5320 zonerefs->zone_idx = 0;
5323 #endif /* CONFIG_NUMA */
5326 * Boot pageset table. One per cpu which is going to be used for all
5327 * zones and all nodes. The parameters will be set in such a way
5328 * that an item put on a list will immediately be handed over to
5329 * the buddy list. This is safe since pageset manipulation is done
5330 * with interrupts disabled.
5332 * The boot_pagesets must be kept even after bootup is complete for
5333 * unused processors and/or zones. They do play a role for bootstrapping
5334 * hotplugged processors.
5336 * zoneinfo_show() and maybe other functions do
5337 * not check if the processor is online before following the pageset pointer.
5338 * Other parts of the kernel may not check if the zone is available.
5340 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
5341 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
5342 static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats);
5344 static void __build_all_zonelists(void *data)
5347 int __maybe_unused cpu;
5348 pg_data_t *self = data;
5349 static DEFINE_SPINLOCK(lock);
5354 memset(node_load, 0, sizeof(node_load));
5358 * This node is hotadded and no memory is yet present. So just
5359 * building zonelists is fine - no need to touch other nodes.
5361 if (self && !node_online(self->node_id)) {
5362 build_zonelists(self);
5364 for_each_online_node(nid) {
5365 pg_data_t *pgdat = NODE_DATA(nid);
5367 build_zonelists(pgdat);
5370 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
5372 * We now know the "local memory node" for each node--
5373 * i.e., the node of the first zone in the generic zonelist.
5374 * Set up numa_mem percpu variable for on-line cpus. During
5375 * boot, only the boot cpu should be on-line; we'll init the
5376 * secondary cpus' numa_mem as they come on-line. During
5377 * node/memory hotplug, we'll fixup all on-line cpus.
5379 for_each_online_cpu(cpu)
5380 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
5387 static noinline void __init
5388 build_all_zonelists_init(void)
5392 __build_all_zonelists(NULL);
5395 * Initialize the boot_pagesets that are going to be used
5396 * for bootstrapping processors. The real pagesets for
5397 * each zone will be allocated later when the per cpu
5398 * allocator is available.
5400 * boot_pagesets are used also for bootstrapping offline
5401 * cpus if the system is already booted because the pagesets
5402 * are needed to initialize allocators on a specific cpu too.
5403 * F.e. the percpu allocator needs the page allocator which
5404 * needs the percpu allocator in order to allocate its pagesets
5405 * (a chicken-egg dilemma).
5407 for_each_possible_cpu(cpu)
5408 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
5410 mminit_verify_zonelist();
5411 cpuset_init_current_mems_allowed();
5415 * unless system_state == SYSTEM_BOOTING.
5417 * __ref due to call of __init annotated helper build_all_zonelists_init
5418 * [protected by SYSTEM_BOOTING].
5420 void __ref build_all_zonelists(pg_data_t *pgdat)
5422 if (system_state == SYSTEM_BOOTING) {
5423 build_all_zonelists_init();
5425 __build_all_zonelists(pgdat);
5426 /* cpuset refresh routine should be here */
5428 vm_total_pages = nr_free_pagecache_pages();
5430 * Disable grouping by mobility if the number of pages in the
5431 * system is too low to allow the mechanism to work. It would be
5432 * more accurate, but expensive to check per-zone. This check is
5433 * made on memory-hotadd so a system can start with mobility
5434 * disabled and enable it later
5436 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
5437 page_group_by_mobility_disabled = 1;
5439 page_group_by_mobility_disabled = 0;
5441 pr_info("Built %i zonelists, mobility grouping %s. Total pages: %ld\n",
5443 page_group_by_mobility_disabled ? "off" : "on",
5446 pr_info("Policy zone: %s\n", zone_names[policy_zone]);
5451 * Initially all pages are reserved - free ones are freed
5452 * up by free_all_bootmem() once the early boot process is
5453 * done. Non-atomic initialization, single-pass.
5455 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
5456 unsigned long start_pfn, enum memmap_context context,
5457 struct vmem_altmap *altmap)
5459 unsigned long end_pfn = start_pfn + size;
5460 pg_data_t *pgdat = NODE_DATA(nid);
5462 unsigned long nr_initialised = 0;
5464 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5465 struct memblock_region *r = NULL, *tmp;
5468 if (highest_memmap_pfn < end_pfn - 1)
5469 highest_memmap_pfn = end_pfn - 1;
5472 * Honor reservation requested by the driver for this ZONE_DEVICE
5475 if (altmap && start_pfn == altmap->base_pfn)
5476 start_pfn += altmap->reserve;
5478 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
5480 * There can be holes in boot-time mem_map[]s handed to this
5481 * function. They do not exist on hotplugged memory.
5483 if (context != MEMMAP_EARLY)
5486 if (!early_pfn_valid(pfn))
5488 if (!early_pfn_in_nid(pfn, nid))
5490 if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised))
5493 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5495 * Check given memblock attribute by firmware which can affect
5496 * kernel memory layout. If zone==ZONE_MOVABLE but memory is
5497 * mirrored, it's an overlapped memmap init. skip it.
5499 if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
5500 if (!r || pfn >= memblock_region_memory_end_pfn(r)) {
5501 for_each_memblock(memory, tmp)
5502 if (pfn < memblock_region_memory_end_pfn(tmp))
5506 if (pfn >= memblock_region_memory_base_pfn(r) &&
5507 memblock_is_mirror(r)) {
5508 /* already initialized as NORMAL */
5509 pfn = memblock_region_memory_end_pfn(r);
5516 page = pfn_to_page(pfn);
5517 __init_single_page(page, pfn, zone, nid);
5518 if (context == MEMMAP_HOTPLUG)
5519 SetPageReserved(page);
5522 * Mark the block movable so that blocks are reserved for
5523 * movable at startup. This will force kernel allocations
5524 * to reserve their blocks rather than leaking throughout
5525 * the address space during boot when many long-lived
5526 * kernel allocations are made.
5528 * bitmap is created for zone's valid pfn range. but memmap
5529 * can be created for invalid pages (for alignment)
5530 * check here not to call set_pageblock_migratetype() against
5533 * Please note that MEMMAP_HOTPLUG path doesn't clear memmap
5534 * because this is done early in sparse_add_one_section
5536 if (!(pfn & (pageblock_nr_pages - 1))) {
5537 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
5543 static void __meminit zone_init_free_lists(struct zone *zone)
5545 unsigned int order, t;
5546 for_each_migratetype_order(order, t) {
5547 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
5548 zone->free_area[order].nr_free = 0;
5552 #ifndef __HAVE_ARCH_MEMMAP_INIT
5553 #define memmap_init(size, nid, zone, start_pfn) \
5554 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY, NULL)
5557 static int zone_batchsize(struct zone *zone)
5563 * The per-cpu-pages pools are set to around 1000th of the
5564 * size of the zone. But no more than 1/2 of a meg.
5566 * OK, so we don't know how big the cache is. So guess.
5568 batch = zone->managed_pages / 1024;
5569 if (batch * PAGE_SIZE > 512 * 1024)
5570 batch = (512 * 1024) / PAGE_SIZE;
5571 batch /= 4; /* We effectively *= 4 below */
5576 * Clamp the batch to a 2^n - 1 value. Having a power
5577 * of 2 value was found to be more likely to have
5578 * suboptimal cache aliasing properties in some cases.
5580 * For example if 2 tasks are alternately allocating
5581 * batches of pages, one task can end up with a lot
5582 * of pages of one half of the possible page colors
5583 * and the other with pages of the other colors.
5585 batch = rounddown_pow_of_two(batch + batch/2) - 1;
5590 /* The deferral and batching of frees should be suppressed under NOMMU
5593 * The problem is that NOMMU needs to be able to allocate large chunks
5594 * of contiguous memory as there's no hardware page translation to
5595 * assemble apparent contiguous memory from discontiguous pages.
5597 * Queueing large contiguous runs of pages for batching, however,
5598 * causes the pages to actually be freed in smaller chunks. As there
5599 * can be a significant delay between the individual batches being
5600 * recycled, this leads to the once large chunks of space being
5601 * fragmented and becoming unavailable for high-order allocations.
5608 * pcp->high and pcp->batch values are related and dependent on one another:
5609 * ->batch must never be higher then ->high.
5610 * The following function updates them in a safe manner without read side
5613 * Any new users of pcp->batch and pcp->high should ensure they can cope with
5614 * those fields changing asynchronously (acording the the above rule).
5616 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
5617 * outside of boot time (or some other assurance that no concurrent updaters
5620 static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
5621 unsigned long batch)
5623 /* start with a fail safe value for batch */
5627 /* Update high, then batch, in order */
5634 /* a companion to pageset_set_high() */
5635 static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
5637 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
5640 static void pageset_init(struct per_cpu_pageset *p)
5642 struct per_cpu_pages *pcp;
5645 memset(p, 0, sizeof(*p));
5649 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
5650 INIT_LIST_HEAD(&pcp->lists[migratetype]);
5653 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
5656 pageset_set_batch(p, batch);
5660 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
5661 * to the value high for the pageset p.
5663 static void pageset_set_high(struct per_cpu_pageset *p,
5666 unsigned long batch = max(1UL, high / 4);
5667 if ((high / 4) > (PAGE_SHIFT * 8))
5668 batch = PAGE_SHIFT * 8;
5670 pageset_update(&p->pcp, high, batch);
5673 static void pageset_set_high_and_batch(struct zone *zone,
5674 struct per_cpu_pageset *pcp)
5676 if (percpu_pagelist_fraction)
5677 pageset_set_high(pcp,
5678 (zone->managed_pages /
5679 percpu_pagelist_fraction));
5681 pageset_set_batch(pcp, zone_batchsize(zone));
5684 static void __meminit zone_pageset_init(struct zone *zone, int cpu)
5686 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
5689 pageset_set_high_and_batch(zone, pcp);
5692 void __meminit setup_zone_pageset(struct zone *zone)
5695 zone->pageset = alloc_percpu(struct per_cpu_pageset);
5696 for_each_possible_cpu(cpu)
5697 zone_pageset_init(zone, cpu);
5701 * Allocate per cpu pagesets and initialize them.
5702 * Before this call only boot pagesets were available.
5704 void __init setup_per_cpu_pageset(void)
5706 struct pglist_data *pgdat;
5709 for_each_populated_zone(zone)
5710 setup_zone_pageset(zone);
5712 for_each_online_pgdat(pgdat)
5713 pgdat->per_cpu_nodestats =
5714 alloc_percpu(struct per_cpu_nodestat);
5717 static __meminit void zone_pcp_init(struct zone *zone)
5720 * per cpu subsystem is not up at this point. The following code
5721 * relies on the ability of the linker to provide the
5722 * offset of a (static) per cpu variable into the per cpu area.
5724 zone->pageset = &boot_pageset;
5726 if (populated_zone(zone))
5727 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
5728 zone->name, zone->present_pages,
5729 zone_batchsize(zone));
5732 void __meminit init_currently_empty_zone(struct zone *zone,
5733 unsigned long zone_start_pfn,
5736 struct pglist_data *pgdat = zone->zone_pgdat;
5738 pgdat->nr_zones = zone_idx(zone) + 1;
5740 zone->zone_start_pfn = zone_start_pfn;
5742 mminit_dprintk(MMINIT_TRACE, "memmap_init",
5743 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
5745 (unsigned long)zone_idx(zone),
5746 zone_start_pfn, (zone_start_pfn + size));
5748 zone_init_free_lists(zone);
5749 zone->initialized = 1;
5752 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5753 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
5756 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
5758 int __meminit __early_pfn_to_nid(unsigned long pfn,
5759 struct mminit_pfnnid_cache *state)
5761 unsigned long start_pfn, end_pfn;
5764 if (state->last_start <= pfn && pfn < state->last_end)
5765 return state->last_nid;
5767 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
5769 state->last_start = start_pfn;
5770 state->last_end = end_pfn;
5771 state->last_nid = nid;
5776 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
5779 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
5780 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
5781 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
5783 * If an architecture guarantees that all ranges registered contain no holes
5784 * and may be freed, this this function may be used instead of calling
5785 * memblock_free_early_nid() manually.
5787 void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
5789 unsigned long start_pfn, end_pfn;
5792 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
5793 start_pfn = min(start_pfn, max_low_pfn);
5794 end_pfn = min(end_pfn, max_low_pfn);
5796 if (start_pfn < end_pfn)
5797 memblock_free_early_nid(PFN_PHYS(start_pfn),
5798 (end_pfn - start_pfn) << PAGE_SHIFT,
5804 * sparse_memory_present_with_active_regions - Call memory_present for each active range
5805 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
5807 * If an architecture guarantees that all ranges registered contain no holes and may
5808 * be freed, this function may be used instead of calling memory_present() manually.
5810 void __init sparse_memory_present_with_active_regions(int nid)
5812 unsigned long start_pfn, end_pfn;
5815 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
5816 memory_present(this_nid, start_pfn, end_pfn);
5820 * get_pfn_range_for_nid - Return the start and end page frames for a node
5821 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
5822 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
5823 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
5825 * It returns the start and end page frame of a node based on information
5826 * provided by memblock_set_node(). If called for a node
5827 * with no available memory, a warning is printed and the start and end
5830 void __meminit get_pfn_range_for_nid(unsigned int nid,
5831 unsigned long *start_pfn, unsigned long *end_pfn)
5833 unsigned long this_start_pfn, this_end_pfn;
5839 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
5840 *start_pfn = min(*start_pfn, this_start_pfn);
5841 *end_pfn = max(*end_pfn, this_end_pfn);
5844 if (*start_pfn == -1UL)
5849 * This finds a zone that can be used for ZONE_MOVABLE pages. The
5850 * assumption is made that zones within a node are ordered in monotonic
5851 * increasing memory addresses so that the "highest" populated zone is used
5853 static void __init find_usable_zone_for_movable(void)
5856 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
5857 if (zone_index == ZONE_MOVABLE)
5860 if (arch_zone_highest_possible_pfn[zone_index] >
5861 arch_zone_lowest_possible_pfn[zone_index])
5865 VM_BUG_ON(zone_index == -1);
5866 movable_zone = zone_index;
5870 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
5871 * because it is sized independent of architecture. Unlike the other zones,
5872 * the starting point for ZONE_MOVABLE is not fixed. It may be different
5873 * in each node depending on the size of each node and how evenly kernelcore
5874 * is distributed. This helper function adjusts the zone ranges
5875 * provided by the architecture for a given node by using the end of the
5876 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5877 * zones within a node are in order of monotonic increases memory addresses
5879 static void __meminit adjust_zone_range_for_zone_movable(int nid,
5880 unsigned long zone_type,
5881 unsigned long node_start_pfn,
5882 unsigned long node_end_pfn,
5883 unsigned long *zone_start_pfn,
5884 unsigned long *zone_end_pfn)
5886 /* Only adjust if ZONE_MOVABLE is on this node */
5887 if (zone_movable_pfn[nid]) {
5888 /* Size ZONE_MOVABLE */
5889 if (zone_type == ZONE_MOVABLE) {
5890 *zone_start_pfn = zone_movable_pfn[nid];
5891 *zone_end_pfn = min(node_end_pfn,
5892 arch_zone_highest_possible_pfn[movable_zone]);
5894 /* Adjust for ZONE_MOVABLE starting within this range */
5895 } else if (!mirrored_kernelcore &&
5896 *zone_start_pfn < zone_movable_pfn[nid] &&
5897 *zone_end_pfn > zone_movable_pfn[nid]) {
5898 *zone_end_pfn = zone_movable_pfn[nid];
5900 /* Check if this whole range is within ZONE_MOVABLE */
5901 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
5902 *zone_start_pfn = *zone_end_pfn;
5907 * Return the number of pages a zone spans in a node, including holes
5908 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5910 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
5911 unsigned long zone_type,
5912 unsigned long node_start_pfn,
5913 unsigned long node_end_pfn,
5914 unsigned long *zone_start_pfn,
5915 unsigned long *zone_end_pfn,
5916 unsigned long *ignored)
5918 /* When hotadd a new node from cpu_up(), the node should be empty */
5919 if (!node_start_pfn && !node_end_pfn)
5922 /* Get the start and end of the zone */
5923 *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
5924 *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
5925 adjust_zone_range_for_zone_movable(nid, zone_type,
5926 node_start_pfn, node_end_pfn,
5927 zone_start_pfn, zone_end_pfn);
5929 /* Check that this node has pages within the zone's required range */
5930 if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
5933 /* Move the zone boundaries inside the node if necessary */
5934 *zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
5935 *zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
5937 /* Return the spanned pages */
5938 return *zone_end_pfn - *zone_start_pfn;
5942 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
5943 * then all holes in the requested range will be accounted for.
5945 unsigned long __meminit __absent_pages_in_range(int nid,
5946 unsigned long range_start_pfn,
5947 unsigned long range_end_pfn)
5949 unsigned long nr_absent = range_end_pfn - range_start_pfn;
5950 unsigned long start_pfn, end_pfn;
5953 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5954 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
5955 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
5956 nr_absent -= end_pfn - start_pfn;
5962 * absent_pages_in_range - Return number of page frames in holes within a range
5963 * @start_pfn: The start PFN to start searching for holes
5964 * @end_pfn: The end PFN to stop searching for holes
5966 * It returns the number of pages frames in memory holes within a range.
5968 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
5969 unsigned long end_pfn)
5971 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
5974 /* Return the number of page frames in holes in a zone on a node */
5975 static unsigned long __meminit zone_absent_pages_in_node(int nid,
5976 unsigned long zone_type,
5977 unsigned long node_start_pfn,
5978 unsigned long node_end_pfn,
5979 unsigned long *ignored)
5981 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
5982 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
5983 unsigned long zone_start_pfn, zone_end_pfn;
5984 unsigned long nr_absent;
5986 /* When hotadd a new node from cpu_up(), the node should be empty */
5987 if (!node_start_pfn && !node_end_pfn)
5990 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
5991 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
5993 adjust_zone_range_for_zone_movable(nid, zone_type,
5994 node_start_pfn, node_end_pfn,
5995 &zone_start_pfn, &zone_end_pfn);
5996 nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
5999 * ZONE_MOVABLE handling.
6000 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
6003 if (mirrored_kernelcore && zone_movable_pfn[nid]) {
6004 unsigned long start_pfn, end_pfn;
6005 struct memblock_region *r;
6007 for_each_memblock(memory, r) {
6008 start_pfn = clamp(memblock_region_memory_base_pfn(r),
6009 zone_start_pfn, zone_end_pfn);
6010 end_pfn = clamp(memblock_region_memory_end_pfn(r),
6011 zone_start_pfn, zone_end_pfn);
6013 if (zone_type == ZONE_MOVABLE &&
6014 memblock_is_mirror(r))
6015 nr_absent += end_pfn - start_pfn;
6017 if (zone_type == ZONE_NORMAL &&
6018 !memblock_is_mirror(r))
6019 nr_absent += end_pfn - start_pfn;
6026 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6027 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
6028 unsigned long zone_type,
6029 unsigned long node_start_pfn,
6030 unsigned long node_end_pfn,
6031 unsigned long *zone_start_pfn,
6032 unsigned long *zone_end_pfn,
6033 unsigned long *zones_size)
6037 *zone_start_pfn = node_start_pfn;
6038 for (zone = 0; zone < zone_type; zone++)
6039 *zone_start_pfn += zones_size[zone];
6041 *zone_end_pfn = *zone_start_pfn + zones_size[zone_type];
6043 return zones_size[zone_type];
6046 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
6047 unsigned long zone_type,
6048 unsigned long node_start_pfn,
6049 unsigned long node_end_pfn,
6050 unsigned long *zholes_size)
6055 return zholes_size[zone_type];
6058 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6060 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
6061 unsigned long node_start_pfn,
6062 unsigned long node_end_pfn,
6063 unsigned long *zones_size,
6064 unsigned long *zholes_size)
6066 unsigned long realtotalpages = 0, totalpages = 0;
6069 for (i = 0; i < MAX_NR_ZONES; i++) {
6070 struct zone *zone = pgdat->node_zones + i;
6071 unsigned long zone_start_pfn, zone_end_pfn;
6072 unsigned long size, real_size;
6074 size = zone_spanned_pages_in_node(pgdat->node_id, i,
6080 real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
6081 node_start_pfn, node_end_pfn,
6084 zone->zone_start_pfn = zone_start_pfn;
6086 zone->zone_start_pfn = 0;
6087 zone->spanned_pages = size;
6088 zone->present_pages = real_size;
6091 realtotalpages += real_size;
6094 pgdat->node_spanned_pages = totalpages;
6095 pgdat->node_present_pages = realtotalpages;
6096 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
6100 #ifndef CONFIG_SPARSEMEM
6102 * Calculate the size of the zone->blockflags rounded to an unsigned long
6103 * Start by making sure zonesize is a multiple of pageblock_order by rounding
6104 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
6105 * round what is now in bits to nearest long in bits, then return it in
6108 static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
6110 unsigned long usemapsize;
6112 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
6113 usemapsize = roundup(zonesize, pageblock_nr_pages);
6114 usemapsize = usemapsize >> pageblock_order;
6115 usemapsize *= NR_PAGEBLOCK_BITS;
6116 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
6118 return usemapsize / 8;
6121 static void __init setup_usemap(struct pglist_data *pgdat,
6123 unsigned long zone_start_pfn,
6124 unsigned long zonesize)
6126 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
6127 zone->pageblock_flags = NULL;
6129 zone->pageblock_flags =
6130 memblock_virt_alloc_node_nopanic(usemapsize,
6134 static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
6135 unsigned long zone_start_pfn, unsigned long zonesize) {}
6136 #endif /* CONFIG_SPARSEMEM */
6138 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
6140 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
6141 void __paginginit set_pageblock_order(void)
6145 /* Check that pageblock_nr_pages has not already been setup */
6146 if (pageblock_order)
6149 if (HPAGE_SHIFT > PAGE_SHIFT)
6150 order = HUGETLB_PAGE_ORDER;
6152 order = MAX_ORDER - 1;
6155 * Assume the largest contiguous order of interest is a huge page.
6156 * This value may be variable depending on boot parameters on IA64 and
6159 pageblock_order = order;
6161 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
6164 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
6165 * is unused as pageblock_order is set at compile-time. See
6166 * include/linux/pageblock-flags.h for the values of pageblock_order based on
6169 void __paginginit set_pageblock_order(void)
6173 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
6175 static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
6176 unsigned long present_pages)
6178 unsigned long pages = spanned_pages;
6181 * Provide a more accurate estimation if there are holes within
6182 * the zone and SPARSEMEM is in use. If there are holes within the
6183 * zone, each populated memory region may cost us one or two extra
6184 * memmap pages due to alignment because memmap pages for each
6185 * populated regions may not be naturally aligned on page boundary.
6186 * So the (present_pages >> 4) heuristic is a tradeoff for that.
6188 if (spanned_pages > present_pages + (present_pages >> 4) &&
6189 IS_ENABLED(CONFIG_SPARSEMEM))
6190 pages = present_pages;
6192 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
6196 * Set up the zone data structures:
6197 * - mark all pages reserved
6198 * - mark all memory queues empty
6199 * - clear the memory bitmaps
6201 * NOTE: pgdat should get zeroed by caller.
6203 static void __paginginit free_area_init_core(struct pglist_data *pgdat)
6206 int nid = pgdat->node_id;
6208 pgdat_resize_init(pgdat);
6209 #ifdef CONFIG_NUMA_BALANCING
6210 spin_lock_init(&pgdat->numabalancing_migrate_lock);
6211 pgdat->numabalancing_migrate_nr_pages = 0;
6212 pgdat->numabalancing_migrate_next_window = jiffies;
6214 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
6215 spin_lock_init(&pgdat->split_queue_lock);
6216 INIT_LIST_HEAD(&pgdat->split_queue);
6217 pgdat->split_queue_len = 0;
6219 init_waitqueue_head(&pgdat->kswapd_wait);
6220 init_waitqueue_head(&pgdat->pfmemalloc_wait);
6221 #ifdef CONFIG_COMPACTION
6222 init_waitqueue_head(&pgdat->kcompactd_wait);
6224 pgdat_page_ext_init(pgdat);
6225 spin_lock_init(&pgdat->lru_lock);
6226 lruvec_init(node_lruvec(pgdat));
6228 pgdat->per_cpu_nodestats = &boot_nodestats;
6230 for (j = 0; j < MAX_NR_ZONES; j++) {
6231 struct zone *zone = pgdat->node_zones + j;
6232 unsigned long size, realsize, freesize, memmap_pages;
6233 unsigned long zone_start_pfn = zone->zone_start_pfn;
6235 size = zone->spanned_pages;
6236 realsize = freesize = zone->present_pages;
6239 * Adjust freesize so that it accounts for how much memory
6240 * is used by this zone for memmap. This affects the watermark
6241 * and per-cpu initialisations
6243 memmap_pages = calc_memmap_size(size, realsize);
6244 if (!is_highmem_idx(j)) {
6245 if (freesize >= memmap_pages) {
6246 freesize -= memmap_pages;
6249 " %s zone: %lu pages used for memmap\n",
6250 zone_names[j], memmap_pages);
6252 pr_warn(" %s zone: %lu pages exceeds freesize %lu\n",
6253 zone_names[j], memmap_pages, freesize);
6256 /* Account for reserved pages */
6257 if (j == 0 && freesize > dma_reserve) {
6258 freesize -= dma_reserve;
6259 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
6260 zone_names[0], dma_reserve);
6263 if (!is_highmem_idx(j))
6264 nr_kernel_pages += freesize;
6265 /* Charge for highmem memmap if there are enough kernel pages */
6266 else if (nr_kernel_pages > memmap_pages * 2)
6267 nr_kernel_pages -= memmap_pages;
6268 nr_all_pages += freesize;
6271 * Set an approximate value for lowmem here, it will be adjusted
6272 * when the bootmem allocator frees pages into the buddy system.
6273 * And all highmem pages will be managed by the buddy system.
6275 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
6279 zone->name = zone_names[j];
6280 zone->zone_pgdat = pgdat;
6281 spin_lock_init(&zone->lock);
6282 zone_seqlock_init(zone);
6283 zone_pcp_init(zone);
6288 set_pageblock_order();
6289 setup_usemap(pgdat, zone, zone_start_pfn, size);
6290 init_currently_empty_zone(zone, zone_start_pfn, size);
6291 memmap_init(size, nid, j, zone_start_pfn);
6295 #ifdef CONFIG_FLAT_NODE_MEM_MAP
6296 static void __ref alloc_node_mem_map(struct pglist_data *pgdat)
6298 unsigned long __maybe_unused start = 0;
6299 unsigned long __maybe_unused offset = 0;
6301 /* Skip empty nodes */
6302 if (!pgdat->node_spanned_pages)
6305 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
6306 offset = pgdat->node_start_pfn - start;
6307 /* ia64 gets its own node_mem_map, before this, without bootmem */
6308 if (!pgdat->node_mem_map) {
6309 unsigned long size, end;
6313 * The zone's endpoints aren't required to be MAX_ORDER
6314 * aligned but the node_mem_map endpoints must be in order
6315 * for the buddy allocator to function correctly.
6317 end = pgdat_end_pfn(pgdat);
6318 end = ALIGN(end, MAX_ORDER_NR_PAGES);
6319 size = (end - start) * sizeof(struct page);
6320 map = memblock_virt_alloc_node_nopanic(size, pgdat->node_id);
6321 pgdat->node_mem_map = map + offset;
6323 pr_debug("%s: node %d, pgdat %08lx, node_mem_map %08lx\n",
6324 __func__, pgdat->node_id, (unsigned long)pgdat,
6325 (unsigned long)pgdat->node_mem_map);
6326 #ifndef CONFIG_NEED_MULTIPLE_NODES
6328 * With no DISCONTIG, the global mem_map is just set as node 0's
6330 if (pgdat == NODE_DATA(0)) {
6331 mem_map = NODE_DATA(0)->node_mem_map;
6332 #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
6333 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
6335 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6340 static void __ref alloc_node_mem_map(struct pglist_data *pgdat) { }
6341 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
6343 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
6344 unsigned long node_start_pfn, unsigned long *zholes_size)
6346 pg_data_t *pgdat = NODE_DATA(nid);
6347 unsigned long start_pfn = 0;
6348 unsigned long end_pfn = 0;
6350 /* pg_data_t should be reset to zero when it's allocated */
6351 WARN_ON(pgdat->nr_zones || pgdat->kswapd_classzone_idx);
6353 pgdat->node_id = nid;
6354 pgdat->node_start_pfn = node_start_pfn;
6355 pgdat->per_cpu_nodestats = NULL;
6356 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6357 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
6358 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
6359 (u64)start_pfn << PAGE_SHIFT,
6360 end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
6362 start_pfn = node_start_pfn;
6364 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
6365 zones_size, zholes_size);
6367 alloc_node_mem_map(pgdat);
6369 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
6371 * We start only with one section of pages, more pages are added as
6372 * needed until the rest of deferred pages are initialized.
6374 pgdat->static_init_pgcnt = min_t(unsigned long, PAGES_PER_SECTION,
6375 pgdat->node_spanned_pages);
6376 pgdat->first_deferred_pfn = ULONG_MAX;
6378 free_area_init_core(pgdat);
6381 #ifdef CONFIG_HAVE_MEMBLOCK
6383 * Only struct pages that are backed by physical memory are zeroed and
6384 * initialized by going through __init_single_page(). But, there are some
6385 * struct pages which are reserved in memblock allocator and their fields
6386 * may be accessed (for example page_to_pfn() on some configuration accesses
6387 * flags). We must explicitly zero those struct pages.
6389 void __paginginit zero_resv_unavail(void)
6391 phys_addr_t start, end;
6396 * Loop through ranges that are reserved, but do not have reported
6397 * physical memory backing.
6400 for_each_resv_unavail_range(i, &start, &end) {
6401 for (pfn = PFN_DOWN(start); pfn < PFN_UP(end); pfn++) {
6402 if (!pfn_valid(ALIGN_DOWN(pfn, pageblock_nr_pages)))
6404 mm_zero_struct_page(pfn_to_page(pfn));
6410 * Struct pages that do not have backing memory. This could be because
6411 * firmware is using some of this memory, or for some other reasons.
6412 * Once memblock is changed so such behaviour is not allowed: i.e.
6413 * list of "reserved" memory must be a subset of list of "memory", then
6414 * this code can be removed.
6417 pr_info("Reserved but unavailable: %lld pages", pgcnt);
6419 #endif /* CONFIG_HAVE_MEMBLOCK */
6421 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6423 #if MAX_NUMNODES > 1
6425 * Figure out the number of possible node ids.
6427 void __init setup_nr_node_ids(void)
6429 unsigned int highest;
6431 highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
6432 nr_node_ids = highest + 1;
6437 * node_map_pfn_alignment - determine the maximum internode alignment
6439 * This function should be called after node map is populated and sorted.
6440 * It calculates the maximum power of two alignment which can distinguish
6443 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
6444 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
6445 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
6446 * shifted, 1GiB is enough and this function will indicate so.
6448 * This is used to test whether pfn -> nid mapping of the chosen memory
6449 * model has fine enough granularity to avoid incorrect mapping for the
6450 * populated node map.
6452 * Returns the determined alignment in pfn's. 0 if there is no alignment
6453 * requirement (single node).
6455 unsigned long __init node_map_pfn_alignment(void)
6457 unsigned long accl_mask = 0, last_end = 0;
6458 unsigned long start, end, mask;
6462 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
6463 if (!start || last_nid < 0 || last_nid == nid) {
6470 * Start with a mask granular enough to pin-point to the
6471 * start pfn and tick off bits one-by-one until it becomes
6472 * too coarse to separate the current node from the last.
6474 mask = ~((1 << __ffs(start)) - 1);
6475 while (mask && last_end <= (start & (mask << 1)))
6478 /* accumulate all internode masks */
6482 /* convert mask to number of pages */
6483 return ~accl_mask + 1;
6486 /* Find the lowest pfn for a node */
6487 static unsigned long __init find_min_pfn_for_node(int nid)
6489 unsigned long min_pfn = ULONG_MAX;
6490 unsigned long start_pfn;
6493 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
6494 min_pfn = min(min_pfn, start_pfn);
6496 if (min_pfn == ULONG_MAX) {
6497 pr_warn("Could not find start_pfn for node %d\n", nid);
6505 * find_min_pfn_with_active_regions - Find the minimum PFN registered
6507 * It returns the minimum PFN based on information provided via
6508 * memblock_set_node().
6510 unsigned long __init find_min_pfn_with_active_regions(void)
6512 return find_min_pfn_for_node(MAX_NUMNODES);
6516 * early_calculate_totalpages()
6517 * Sum pages in active regions for movable zone.
6518 * Populate N_MEMORY for calculating usable_nodes.
6520 static unsigned long __init early_calculate_totalpages(void)
6522 unsigned long totalpages = 0;
6523 unsigned long start_pfn, end_pfn;
6526 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
6527 unsigned long pages = end_pfn - start_pfn;
6529 totalpages += pages;
6531 node_set_state(nid, N_MEMORY);
6537 * Find the PFN the Movable zone begins in each node. Kernel memory
6538 * is spread evenly between nodes as long as the nodes have enough
6539 * memory. When they don't, some nodes will have more kernelcore than
6542 static void __init find_zone_movable_pfns_for_nodes(void)
6545 unsigned long usable_startpfn;
6546 unsigned long kernelcore_node, kernelcore_remaining;
6547 /* save the state before borrow the nodemask */
6548 nodemask_t saved_node_state = node_states[N_MEMORY];
6549 unsigned long totalpages = early_calculate_totalpages();
6550 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
6551 struct memblock_region *r;
6553 /* Need to find movable_zone earlier when movable_node is specified. */
6554 find_usable_zone_for_movable();
6557 * If movable_node is specified, ignore kernelcore and movablecore
6560 if (movable_node_is_enabled()) {
6561 for_each_memblock(memory, r) {
6562 if (!memblock_is_hotpluggable(r))
6567 usable_startpfn = PFN_DOWN(r->base);
6568 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
6569 min(usable_startpfn, zone_movable_pfn[nid]) :
6577 * If kernelcore=mirror is specified, ignore movablecore option
6579 if (mirrored_kernelcore) {
6580 bool mem_below_4gb_not_mirrored = false;
6582 for_each_memblock(memory, r) {
6583 if (memblock_is_mirror(r))
6588 usable_startpfn = memblock_region_memory_base_pfn(r);
6590 if (usable_startpfn < 0x100000) {
6591 mem_below_4gb_not_mirrored = true;
6595 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
6596 min(usable_startpfn, zone_movable_pfn[nid]) :
6600 if (mem_below_4gb_not_mirrored)
6601 pr_warn("This configuration results in unmirrored kernel memory.");
6607 * If kernelcore=nn% or movablecore=nn% was specified, calculate the
6608 * amount of necessary memory.
6610 if (required_kernelcore_percent)
6611 required_kernelcore = (totalpages * 100 * required_kernelcore_percent) /
6613 if (required_movablecore_percent)
6614 required_movablecore = (totalpages * 100 * required_movablecore_percent) /
6618 * If movablecore= was specified, calculate what size of
6619 * kernelcore that corresponds so that memory usable for
6620 * any allocation type is evenly spread. If both kernelcore
6621 * and movablecore are specified, then the value of kernelcore
6622 * will be used for required_kernelcore if it's greater than
6623 * what movablecore would have allowed.
6625 if (required_movablecore) {
6626 unsigned long corepages;
6629 * Round-up so that ZONE_MOVABLE is at least as large as what
6630 * was requested by the user
6632 required_movablecore =
6633 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
6634 required_movablecore = min(totalpages, required_movablecore);
6635 corepages = totalpages - required_movablecore;
6637 required_kernelcore = max(required_kernelcore, corepages);
6641 * If kernelcore was not specified or kernelcore size is larger
6642 * than totalpages, there is no ZONE_MOVABLE.
6644 if (!required_kernelcore || required_kernelcore >= totalpages)
6647 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
6648 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
6651 /* Spread kernelcore memory as evenly as possible throughout nodes */
6652 kernelcore_node = required_kernelcore / usable_nodes;
6653 for_each_node_state(nid, N_MEMORY) {
6654 unsigned long start_pfn, end_pfn;
6657 * Recalculate kernelcore_node if the division per node
6658 * now exceeds what is necessary to satisfy the requested
6659 * amount of memory for the kernel
6661 if (required_kernelcore < kernelcore_node)
6662 kernelcore_node = required_kernelcore / usable_nodes;
6665 * As the map is walked, we track how much memory is usable
6666 * by the kernel using kernelcore_remaining. When it is
6667 * 0, the rest of the node is usable by ZONE_MOVABLE
6669 kernelcore_remaining = kernelcore_node;
6671 /* Go through each range of PFNs within this node */
6672 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
6673 unsigned long size_pages;
6675 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
6676 if (start_pfn >= end_pfn)
6679 /* Account for what is only usable for kernelcore */
6680 if (start_pfn < usable_startpfn) {
6681 unsigned long kernel_pages;
6682 kernel_pages = min(end_pfn, usable_startpfn)
6685 kernelcore_remaining -= min(kernel_pages,
6686 kernelcore_remaining);
6687 required_kernelcore -= min(kernel_pages,
6688 required_kernelcore);
6690 /* Continue if range is now fully accounted */
6691 if (end_pfn <= usable_startpfn) {
6694 * Push zone_movable_pfn to the end so
6695 * that if we have to rebalance
6696 * kernelcore across nodes, we will
6697 * not double account here
6699 zone_movable_pfn[nid] = end_pfn;
6702 start_pfn = usable_startpfn;
6706 * The usable PFN range for ZONE_MOVABLE is from
6707 * start_pfn->end_pfn. Calculate size_pages as the
6708 * number of pages used as kernelcore
6710 size_pages = end_pfn - start_pfn;
6711 if (size_pages > kernelcore_remaining)
6712 size_pages = kernelcore_remaining;
6713 zone_movable_pfn[nid] = start_pfn + size_pages;
6716 * Some kernelcore has been met, update counts and
6717 * break if the kernelcore for this node has been
6720 required_kernelcore -= min(required_kernelcore,
6722 kernelcore_remaining -= size_pages;
6723 if (!kernelcore_remaining)
6729 * If there is still required_kernelcore, we do another pass with one
6730 * less node in the count. This will push zone_movable_pfn[nid] further
6731 * along on the nodes that still have memory until kernelcore is
6735 if (usable_nodes && required_kernelcore > usable_nodes)
6739 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
6740 for (nid = 0; nid < MAX_NUMNODES; nid++)
6741 zone_movable_pfn[nid] =
6742 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
6745 /* restore the node_state */
6746 node_states[N_MEMORY] = saved_node_state;
6749 /* Any regular or high memory on that node ? */
6750 static void check_for_memory(pg_data_t *pgdat, int nid)
6752 enum zone_type zone_type;
6754 if (N_MEMORY == N_NORMAL_MEMORY)
6757 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
6758 struct zone *zone = &pgdat->node_zones[zone_type];
6759 if (populated_zone(zone)) {
6760 node_set_state(nid, N_HIGH_MEMORY);
6761 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
6762 zone_type <= ZONE_NORMAL)
6763 node_set_state(nid, N_NORMAL_MEMORY);
6770 * free_area_init_nodes - Initialise all pg_data_t and zone data
6771 * @max_zone_pfn: an array of max PFNs for each zone
6773 * This will call free_area_init_node() for each active node in the system.
6774 * Using the page ranges provided by memblock_set_node(), the size of each
6775 * zone in each node and their holes is calculated. If the maximum PFN
6776 * between two adjacent zones match, it is assumed that the zone is empty.
6777 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
6778 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
6779 * starts where the previous one ended. For example, ZONE_DMA32 starts
6780 * at arch_max_dma_pfn.
6782 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
6784 unsigned long start_pfn, end_pfn;
6787 /* Record where the zone boundaries are */
6788 memset(arch_zone_lowest_possible_pfn, 0,
6789 sizeof(arch_zone_lowest_possible_pfn));
6790 memset(arch_zone_highest_possible_pfn, 0,
6791 sizeof(arch_zone_highest_possible_pfn));
6793 start_pfn = find_min_pfn_with_active_regions();
6795 for (i = 0; i < MAX_NR_ZONES; i++) {
6796 if (i == ZONE_MOVABLE)
6799 end_pfn = max(max_zone_pfn[i], start_pfn);
6800 arch_zone_lowest_possible_pfn[i] = start_pfn;
6801 arch_zone_highest_possible_pfn[i] = end_pfn;
6803 start_pfn = end_pfn;
6806 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
6807 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
6808 find_zone_movable_pfns_for_nodes();
6810 /* Print out the zone ranges */
6811 pr_info("Zone ranges:\n");
6812 for (i = 0; i < MAX_NR_ZONES; i++) {
6813 if (i == ZONE_MOVABLE)
6815 pr_info(" %-8s ", zone_names[i]);
6816 if (arch_zone_lowest_possible_pfn[i] ==
6817 arch_zone_highest_possible_pfn[i])
6820 pr_cont("[mem %#018Lx-%#018Lx]\n",
6821 (u64)arch_zone_lowest_possible_pfn[i]
6823 ((u64)arch_zone_highest_possible_pfn[i]
6824 << PAGE_SHIFT) - 1);
6827 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
6828 pr_info("Movable zone start for each node\n");
6829 for (i = 0; i < MAX_NUMNODES; i++) {
6830 if (zone_movable_pfn[i])
6831 pr_info(" Node %d: %#018Lx\n", i,
6832 (u64)zone_movable_pfn[i] << PAGE_SHIFT);
6835 /* Print out the early node map */
6836 pr_info("Early memory node ranges\n");
6837 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
6838 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
6839 (u64)start_pfn << PAGE_SHIFT,
6840 ((u64)end_pfn << PAGE_SHIFT) - 1);
6842 /* Initialise every node */
6843 mminit_verify_pageflags_layout();
6844 setup_nr_node_ids();
6845 for_each_online_node(nid) {
6846 pg_data_t *pgdat = NODE_DATA(nid);
6847 free_area_init_node(nid, NULL,
6848 find_min_pfn_for_node(nid), NULL);
6850 /* Any memory on that node */
6851 if (pgdat->node_present_pages)
6852 node_set_state(nid, N_MEMORY);
6853 check_for_memory(pgdat, nid);
6855 zero_resv_unavail();
6858 static int __init cmdline_parse_core(char *p, unsigned long *core,
6859 unsigned long *percent)
6861 unsigned long long coremem;
6867 /* Value may be a percentage of total memory, otherwise bytes */
6868 coremem = simple_strtoull(p, &endptr, 0);
6869 if (*endptr == '%') {
6870 /* Paranoid check for percent values greater than 100 */
6871 WARN_ON(coremem > 100);
6875 coremem = memparse(p, &p);
6876 /* Paranoid check that UL is enough for the coremem value */
6877 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
6879 *core = coremem >> PAGE_SHIFT;
6886 * kernelcore=size sets the amount of memory for use for allocations that
6887 * cannot be reclaimed or migrated.
6889 static int __init cmdline_parse_kernelcore(char *p)
6891 /* parse kernelcore=mirror */
6892 if (parse_option_str(p, "mirror")) {
6893 mirrored_kernelcore = true;
6897 return cmdline_parse_core(p, &required_kernelcore,
6898 &required_kernelcore_percent);
6902 * movablecore=size sets the amount of memory for use for allocations that
6903 * can be reclaimed or migrated.
6905 static int __init cmdline_parse_movablecore(char *p)
6907 return cmdline_parse_core(p, &required_movablecore,
6908 &required_movablecore_percent);
6911 early_param("kernelcore", cmdline_parse_kernelcore);
6912 early_param("movablecore", cmdline_parse_movablecore);
6914 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6916 void adjust_managed_page_count(struct page *page, long count)
6918 spin_lock(&managed_page_count_lock);
6919 page_zone(page)->managed_pages += count;
6920 totalram_pages += count;
6921 #ifdef CONFIG_HIGHMEM
6922 if (PageHighMem(page))
6923 totalhigh_pages += count;
6925 spin_unlock(&managed_page_count_lock);
6927 EXPORT_SYMBOL(adjust_managed_page_count);
6929 unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
6932 unsigned long pages = 0;
6934 start = (void *)PAGE_ALIGN((unsigned long)start);
6935 end = (void *)((unsigned long)end & PAGE_MASK);
6936 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
6937 if ((unsigned int)poison <= 0xFF)
6938 memset(pos, poison, PAGE_SIZE);
6939 free_reserved_page(virt_to_page(pos));
6943 pr_info("Freeing %s memory: %ldK\n",
6944 s, pages << (PAGE_SHIFT - 10));
6948 EXPORT_SYMBOL(free_reserved_area);
6950 #ifdef CONFIG_HIGHMEM
6951 void free_highmem_page(struct page *page)
6953 __free_reserved_page(page);
6955 page_zone(page)->managed_pages++;
6961 void __init mem_init_print_info(const char *str)
6963 unsigned long physpages, codesize, datasize, rosize, bss_size;
6964 unsigned long init_code_size, init_data_size;
6966 physpages = get_num_physpages();
6967 codesize = _etext - _stext;
6968 datasize = _edata - _sdata;
6969 rosize = __end_rodata - __start_rodata;
6970 bss_size = __bss_stop - __bss_start;
6971 init_data_size = __init_end - __init_begin;
6972 init_code_size = _einittext - _sinittext;
6975 * Detect special cases and adjust section sizes accordingly:
6976 * 1) .init.* may be embedded into .data sections
6977 * 2) .init.text.* may be out of [__init_begin, __init_end],
6978 * please refer to arch/tile/kernel/vmlinux.lds.S.
6979 * 3) .rodata.* may be embedded into .text or .data sections.
6981 #define adj_init_size(start, end, size, pos, adj) \
6983 if (start <= pos && pos < end && size > adj) \
6987 adj_init_size(__init_begin, __init_end, init_data_size,
6988 _sinittext, init_code_size);
6989 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
6990 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
6991 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
6992 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
6994 #undef adj_init_size
6996 pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
6997 #ifdef CONFIG_HIGHMEM
7001 nr_free_pages() << (PAGE_SHIFT - 10),
7002 physpages << (PAGE_SHIFT - 10),
7003 codesize >> 10, datasize >> 10, rosize >> 10,
7004 (init_data_size + init_code_size) >> 10, bss_size >> 10,
7005 (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10),
7006 totalcma_pages << (PAGE_SHIFT - 10),
7007 #ifdef CONFIG_HIGHMEM
7008 totalhigh_pages << (PAGE_SHIFT - 10),
7010 str ? ", " : "", str ? str : "");
7014 * set_dma_reserve - set the specified number of pages reserved in the first zone
7015 * @new_dma_reserve: The number of pages to mark reserved
7017 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
7018 * In the DMA zone, a significant percentage may be consumed by kernel image
7019 * and other unfreeable allocations which can skew the watermarks badly. This
7020 * function may optionally be used to account for unfreeable pages in the
7021 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
7022 * smaller per-cpu batchsize.
7024 void __init set_dma_reserve(unsigned long new_dma_reserve)
7026 dma_reserve = new_dma_reserve;
7029 void __init free_area_init(unsigned long *zones_size)
7031 free_area_init_node(0, zones_size,
7032 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
7033 zero_resv_unavail();
7036 static int page_alloc_cpu_dead(unsigned int cpu)
7039 lru_add_drain_cpu(cpu);
7043 * Spill the event counters of the dead processor
7044 * into the current processors event counters.
7045 * This artificially elevates the count of the current
7048 vm_events_fold_cpu(cpu);
7051 * Zero the differential counters of the dead processor
7052 * so that the vm statistics are consistent.
7054 * This is only okay since the processor is dead and cannot
7055 * race with what we are doing.
7057 cpu_vm_stats_fold(cpu);
7061 void __init page_alloc_init(void)
7065 ret = cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD,
7066 "mm/page_alloc:dead", NULL,
7067 page_alloc_cpu_dead);
7072 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
7073 * or min_free_kbytes changes.
7075 static void calculate_totalreserve_pages(void)
7077 struct pglist_data *pgdat;
7078 unsigned long reserve_pages = 0;
7079 enum zone_type i, j;
7081 for_each_online_pgdat(pgdat) {
7083 pgdat->totalreserve_pages = 0;
7085 for (i = 0; i < MAX_NR_ZONES; i++) {
7086 struct zone *zone = pgdat->node_zones + i;
7089 /* Find valid and maximum lowmem_reserve in the zone */
7090 for (j = i; j < MAX_NR_ZONES; j++) {
7091 if (zone->lowmem_reserve[j] > max)
7092 max = zone->lowmem_reserve[j];
7095 /* we treat the high watermark as reserved pages. */
7096 max += high_wmark_pages(zone);
7098 if (max > zone->managed_pages)
7099 max = zone->managed_pages;
7101 pgdat->totalreserve_pages += max;
7103 reserve_pages += max;
7106 totalreserve_pages = reserve_pages;
7110 * setup_per_zone_lowmem_reserve - called whenever
7111 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
7112 * has a correct pages reserved value, so an adequate number of
7113 * pages are left in the zone after a successful __alloc_pages().
7115 static void setup_per_zone_lowmem_reserve(void)
7117 struct pglist_data *pgdat;
7118 enum zone_type j, idx;
7120 for_each_online_pgdat(pgdat) {
7121 for (j = 0; j < MAX_NR_ZONES; j++) {
7122 struct zone *zone = pgdat->node_zones + j;
7123 unsigned long managed_pages = zone->managed_pages;
7125 zone->lowmem_reserve[j] = 0;
7129 struct zone *lower_zone;
7132 lower_zone = pgdat->node_zones + idx;
7134 if (sysctl_lowmem_reserve_ratio[idx] < 1) {
7135 sysctl_lowmem_reserve_ratio[idx] = 0;
7136 lower_zone->lowmem_reserve[j] = 0;
7138 lower_zone->lowmem_reserve[j] =
7139 managed_pages / sysctl_lowmem_reserve_ratio[idx];
7141 managed_pages += lower_zone->managed_pages;
7146 /* update totalreserve_pages */
7147 calculate_totalreserve_pages();
7150 static void __setup_per_zone_wmarks(void)
7152 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
7153 unsigned long lowmem_pages = 0;
7155 unsigned long flags;
7157 /* Calculate total number of !ZONE_HIGHMEM pages */
7158 for_each_zone(zone) {
7159 if (!is_highmem(zone))
7160 lowmem_pages += zone->managed_pages;
7163 for_each_zone(zone) {
7166 spin_lock_irqsave(&zone->lock, flags);
7167 tmp = (u64)pages_min * zone->managed_pages;
7168 do_div(tmp, lowmem_pages);
7169 if (is_highmem(zone)) {
7171 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
7172 * need highmem pages, so cap pages_min to a small
7175 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
7176 * deltas control asynch page reclaim, and so should
7177 * not be capped for highmem.
7179 unsigned long min_pages;
7181 min_pages = zone->managed_pages / 1024;
7182 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
7183 zone->watermark[WMARK_MIN] = min_pages;
7186 * If it's a lowmem zone, reserve a number of pages
7187 * proportionate to the zone's size.
7189 zone->watermark[WMARK_MIN] = tmp;
7193 * Set the kswapd watermarks distance according to the
7194 * scale factor in proportion to available memory, but
7195 * ensure a minimum size on small systems.
7197 tmp = max_t(u64, tmp >> 2,
7198 mult_frac(zone->managed_pages,
7199 watermark_scale_factor, 10000));
7201 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp;
7202 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2;
7204 spin_unlock_irqrestore(&zone->lock, flags);
7207 /* update totalreserve_pages */
7208 calculate_totalreserve_pages();
7212 * setup_per_zone_wmarks - called when min_free_kbytes changes
7213 * or when memory is hot-{added|removed}
7215 * Ensures that the watermark[min,low,high] values for each zone are set
7216 * correctly with respect to min_free_kbytes.
7218 void setup_per_zone_wmarks(void)
7220 static DEFINE_SPINLOCK(lock);
7223 __setup_per_zone_wmarks();
7228 * Initialise min_free_kbytes.
7230 * For small machines we want it small (128k min). For large machines
7231 * we want it large (64MB max). But it is not linear, because network
7232 * bandwidth does not increase linearly with machine size. We use
7234 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
7235 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
7251 int __meminit init_per_zone_wmark_min(void)
7253 unsigned long lowmem_kbytes;
7254 int new_min_free_kbytes;
7256 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
7257 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
7259 if (new_min_free_kbytes > user_min_free_kbytes) {
7260 min_free_kbytes = new_min_free_kbytes;
7261 if (min_free_kbytes < 128)
7262 min_free_kbytes = 128;
7263 if (min_free_kbytes > 65536)
7264 min_free_kbytes = 65536;
7266 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
7267 new_min_free_kbytes, user_min_free_kbytes);
7269 setup_per_zone_wmarks();
7270 refresh_zone_stat_thresholds();
7271 setup_per_zone_lowmem_reserve();
7274 setup_min_unmapped_ratio();
7275 setup_min_slab_ratio();
7280 core_initcall(init_per_zone_wmark_min)
7283 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
7284 * that we can call two helper functions whenever min_free_kbytes
7287 int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
7288 void __user *buffer, size_t *length, loff_t *ppos)
7292 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7297 user_min_free_kbytes = min_free_kbytes;
7298 setup_per_zone_wmarks();
7303 int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write,
7304 void __user *buffer, size_t *length, loff_t *ppos)
7308 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7313 setup_per_zone_wmarks();
7319 static void setup_min_unmapped_ratio(void)
7324 for_each_online_pgdat(pgdat)
7325 pgdat->min_unmapped_pages = 0;
7328 zone->zone_pgdat->min_unmapped_pages += (zone->managed_pages *
7329 sysctl_min_unmapped_ratio) / 100;
7333 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
7334 void __user *buffer, size_t *length, loff_t *ppos)
7338 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7342 setup_min_unmapped_ratio();
7347 static void setup_min_slab_ratio(void)
7352 for_each_online_pgdat(pgdat)
7353 pgdat->min_slab_pages = 0;
7356 zone->zone_pgdat->min_slab_pages += (zone->managed_pages *
7357 sysctl_min_slab_ratio) / 100;
7360 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
7361 void __user *buffer, size_t *length, loff_t *ppos)
7365 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
7369 setup_min_slab_ratio();
7376 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
7377 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
7378 * whenever sysctl_lowmem_reserve_ratio changes.
7380 * The reserve ratio obviously has absolutely no relation with the
7381 * minimum watermarks. The lowmem reserve ratio can only make sense
7382 * if in function of the boot time zone sizes.
7384 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
7385 void __user *buffer, size_t *length, loff_t *ppos)
7387 proc_dointvec_minmax(table, write, buffer, length, ppos);
7388 setup_per_zone_lowmem_reserve();
7393 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
7394 * cpu. It is the fraction of total pages in each zone that a hot per cpu
7395 * pagelist can have before it gets flushed back to buddy allocator.
7397 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
7398 void __user *buffer, size_t *length, loff_t *ppos)
7401 int old_percpu_pagelist_fraction;
7404 mutex_lock(&pcp_batch_high_lock);
7405 old_percpu_pagelist_fraction = percpu_pagelist_fraction;
7407 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
7408 if (!write || ret < 0)
7411 /* Sanity checking to avoid pcp imbalance */
7412 if (percpu_pagelist_fraction &&
7413 percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
7414 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
7420 if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
7423 for_each_populated_zone(zone) {
7426 for_each_possible_cpu(cpu)
7427 pageset_set_high_and_batch(zone,
7428 per_cpu_ptr(zone->pageset, cpu));
7431 mutex_unlock(&pcp_batch_high_lock);
7436 int hashdist = HASHDIST_DEFAULT;
7438 static int __init set_hashdist(char *str)
7442 hashdist = simple_strtoul(str, &str, 0);
7445 __setup("hashdist=", set_hashdist);
7448 #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES
7450 * Returns the number of pages that arch has reserved but
7451 * is not known to alloc_large_system_hash().
7453 static unsigned long __init arch_reserved_kernel_pages(void)
7460 * Adaptive scale is meant to reduce sizes of hash tables on large memory
7461 * machines. As memory size is increased the scale is also increased but at
7462 * slower pace. Starting from ADAPT_SCALE_BASE (64G), every time memory
7463 * quadruples the scale is increased by one, which means the size of hash table
7464 * only doubles, instead of quadrupling as well.
7465 * Because 32-bit systems cannot have large physical memory, where this scaling
7466 * makes sense, it is disabled on such platforms.
7468 #if __BITS_PER_LONG > 32
7469 #define ADAPT_SCALE_BASE (64ul << 30)
7470 #define ADAPT_SCALE_SHIFT 2
7471 #define ADAPT_SCALE_NPAGES (ADAPT_SCALE_BASE >> PAGE_SHIFT)
7475 * allocate a large system hash table from bootmem
7476 * - it is assumed that the hash table must contain an exact power-of-2
7477 * quantity of entries
7478 * - limit is the number of hash buckets, not the total allocation size
7480 void *__init alloc_large_system_hash(const char *tablename,
7481 unsigned long bucketsize,
7482 unsigned long numentries,
7485 unsigned int *_hash_shift,
7486 unsigned int *_hash_mask,
7487 unsigned long low_limit,
7488 unsigned long high_limit)
7490 unsigned long long max = high_limit;
7491 unsigned long log2qty, size;
7495 /* allow the kernel cmdline to have a say */
7497 /* round applicable memory size up to nearest megabyte */
7498 numentries = nr_kernel_pages;
7499 numentries -= arch_reserved_kernel_pages();
7501 /* It isn't necessary when PAGE_SIZE >= 1MB */
7502 if (PAGE_SHIFT < 20)
7503 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
7505 #if __BITS_PER_LONG > 32
7507 unsigned long adapt;
7509 for (adapt = ADAPT_SCALE_NPAGES; adapt < numentries;
7510 adapt <<= ADAPT_SCALE_SHIFT)
7515 /* limit to 1 bucket per 2^scale bytes of low memory */
7516 if (scale > PAGE_SHIFT)
7517 numentries >>= (scale - PAGE_SHIFT);
7519 numentries <<= (PAGE_SHIFT - scale);
7521 /* Make sure we've got at least a 0-order allocation.. */
7522 if (unlikely(flags & HASH_SMALL)) {
7523 /* Makes no sense without HASH_EARLY */
7524 WARN_ON(!(flags & HASH_EARLY));
7525 if (!(numentries >> *_hash_shift)) {
7526 numentries = 1UL << *_hash_shift;
7527 BUG_ON(!numentries);
7529 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
7530 numentries = PAGE_SIZE / bucketsize;
7532 numentries = roundup_pow_of_two(numentries);
7534 /* limit allocation size to 1/16 total memory by default */
7536 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
7537 do_div(max, bucketsize);
7539 max = min(max, 0x80000000ULL);
7541 if (numentries < low_limit)
7542 numentries = low_limit;
7543 if (numentries > max)
7546 log2qty = ilog2(numentries);
7548 gfp_flags = (flags & HASH_ZERO) ? GFP_ATOMIC | __GFP_ZERO : GFP_ATOMIC;
7550 size = bucketsize << log2qty;
7551 if (flags & HASH_EARLY) {
7552 if (flags & HASH_ZERO)
7553 table = memblock_virt_alloc_nopanic(size, 0);
7555 table = memblock_virt_alloc_raw(size, 0);
7556 } else if (hashdist) {
7557 table = __vmalloc(size, gfp_flags, PAGE_KERNEL);
7560 * If bucketsize is not a power-of-two, we may free
7561 * some pages at the end of hash table which
7562 * alloc_pages_exact() automatically does
7564 if (get_order(size) < MAX_ORDER) {
7565 table = alloc_pages_exact(size, gfp_flags);
7566 kmemleak_alloc(table, size, 1, gfp_flags);
7569 } while (!table && size > PAGE_SIZE && --log2qty);
7572 panic("Failed to allocate %s hash table\n", tablename);
7574 pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n",
7575 tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size);
7578 *_hash_shift = log2qty;
7580 *_hash_mask = (1 << log2qty) - 1;
7586 * This function checks whether pageblock includes unmovable pages or not.
7587 * If @count is not zero, it is okay to include less @count unmovable pages
7589 * PageLRU check without isolation or lru_lock could race so that
7590 * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable
7591 * check without lock_page also may miss some movable non-lru pages at
7592 * race condition. So you can't expect this function should be exact.
7594 bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
7596 bool skip_hwpoisoned_pages)
7598 unsigned long pfn, iter, found;
7601 * TODO we could make this much more efficient by not checking every
7602 * page in the range if we know all of them are in MOVABLE_ZONE and
7603 * that the movable zone guarantees that pages are migratable but
7604 * the later is not the case right now unfortunatelly. E.g. movablecore
7605 * can still lead to having bootmem allocations in zone_movable.
7609 * CMA allocations (alloc_contig_range) really need to mark isolate
7610 * CMA pageblocks even when they are not movable in fact so consider
7611 * them movable here.
7613 if (is_migrate_cma(migratetype) &&
7614 is_migrate_cma(get_pageblock_migratetype(page)))
7617 pfn = page_to_pfn(page);
7618 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
7619 unsigned long check = pfn + iter;
7621 if (!pfn_valid_within(check))
7624 page = pfn_to_page(check);
7626 if (PageReserved(page))
7630 * Hugepages are not in LRU lists, but they're movable.
7631 * We need not scan over tail pages bacause we don't
7632 * handle each tail page individually in migration.
7634 if (PageHuge(page)) {
7635 iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
7640 * We can't use page_count without pin a page
7641 * because another CPU can free compound page.
7642 * This check already skips compound tails of THP
7643 * because their page->_refcount is zero at all time.
7645 if (!page_ref_count(page)) {
7646 if (PageBuddy(page))
7647 iter += (1 << page_order(page)) - 1;
7652 * The HWPoisoned page may be not in buddy system, and
7653 * page_count() is not 0.
7655 if (skip_hwpoisoned_pages && PageHWPoison(page))
7658 if (__PageMovable(page))
7664 * If there are RECLAIMABLE pages, we need to check
7665 * it. But now, memory offline itself doesn't call
7666 * shrink_node_slabs() and it still to be fixed.
7669 * If the page is not RAM, page_count()should be 0.
7670 * we don't need more check. This is an _used_ not-movable page.
7672 * The problematic thing here is PG_reserved pages. PG_reserved
7673 * is set to both of a memory hole page and a _used_ kernel
7681 WARN_ON_ONCE(zone_idx(zone) == ZONE_MOVABLE);
7685 bool is_pageblock_removable_nolock(struct page *page)
7691 * We have to be careful here because we are iterating over memory
7692 * sections which are not zone aware so we might end up outside of
7693 * the zone but still within the section.
7694 * We have to take care about the node as well. If the node is offline
7695 * its NODE_DATA will be NULL - see page_zone.
7697 if (!node_online(page_to_nid(page)))
7700 zone = page_zone(page);
7701 pfn = page_to_pfn(page);
7702 if (!zone_spans_pfn(zone, pfn))
7705 return !has_unmovable_pages(zone, page, 0, MIGRATE_MOVABLE, true);
7708 #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
7710 static unsigned long pfn_max_align_down(unsigned long pfn)
7712 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
7713 pageblock_nr_pages) - 1);
7716 static unsigned long pfn_max_align_up(unsigned long pfn)
7718 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
7719 pageblock_nr_pages));
7722 /* [start, end) must belong to a single zone. */
7723 static int __alloc_contig_migrate_range(struct compact_control *cc,
7724 unsigned long start, unsigned long end)
7726 /* This function is based on compact_zone() from compaction.c. */
7727 unsigned long nr_reclaimed;
7728 unsigned long pfn = start;
7729 unsigned int tries = 0;
7734 while (pfn < end || !list_empty(&cc->migratepages)) {
7735 if (fatal_signal_pending(current)) {
7740 if (list_empty(&cc->migratepages)) {
7741 cc->nr_migratepages = 0;
7742 pfn = isolate_migratepages_range(cc, pfn, end);
7748 } else if (++tries == 5) {
7749 ret = ret < 0 ? ret : -EBUSY;
7753 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
7755 cc->nr_migratepages -= nr_reclaimed;
7757 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
7758 NULL, 0, cc->mode, MR_CONTIG_RANGE);
7761 putback_movable_pages(&cc->migratepages);
7768 * alloc_contig_range() -- tries to allocate given range of pages
7769 * @start: start PFN to allocate
7770 * @end: one-past-the-last PFN to allocate
7771 * @migratetype: migratetype of the underlaying pageblocks (either
7772 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
7773 * in range must have the same migratetype and it must
7774 * be either of the two.
7775 * @gfp_mask: GFP mask to use during compaction
7777 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
7778 * aligned. The PFN range must belong to a single zone.
7780 * The first thing this routine does is attempt to MIGRATE_ISOLATE all
7781 * pageblocks in the range. Once isolated, the pageblocks should not
7782 * be modified by others.
7784 * Returns zero on success or negative error code. On success all
7785 * pages which PFN is in [start, end) are allocated for the caller and
7786 * need to be freed with free_contig_range().
7788 int alloc_contig_range(unsigned long start, unsigned long end,
7789 unsigned migratetype, gfp_t gfp_mask)
7791 unsigned long outer_start, outer_end;
7795 struct compact_control cc = {
7796 .nr_migratepages = 0,
7798 .zone = page_zone(pfn_to_page(start)),
7799 .mode = MIGRATE_SYNC,
7800 .ignore_skip_hint = true,
7801 .no_set_skip_hint = true,
7802 .gfp_mask = current_gfp_context(gfp_mask),
7804 INIT_LIST_HEAD(&cc.migratepages);
7807 * What we do here is we mark all pageblocks in range as
7808 * MIGRATE_ISOLATE. Because pageblock and max order pages may
7809 * have different sizes, and due to the way page allocator
7810 * work, we align the range to biggest of the two pages so
7811 * that page allocator won't try to merge buddies from
7812 * different pageblocks and change MIGRATE_ISOLATE to some
7813 * other migration type.
7815 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
7816 * migrate the pages from an unaligned range (ie. pages that
7817 * we are interested in). This will put all the pages in
7818 * range back to page allocator as MIGRATE_ISOLATE.
7820 * When this is done, we take the pages in range from page
7821 * allocator removing them from the buddy system. This way
7822 * page allocator will never consider using them.
7824 * This lets us mark the pageblocks back as
7825 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
7826 * aligned range but not in the unaligned, original range are
7827 * put back to page allocator so that buddy can use them.
7830 ret = start_isolate_page_range(pfn_max_align_down(start),
7831 pfn_max_align_up(end), migratetype,
7837 * In case of -EBUSY, we'd like to know which page causes problem.
7838 * So, just fall through. test_pages_isolated() has a tracepoint
7839 * which will report the busy page.
7841 * It is possible that busy pages could become available before
7842 * the call to test_pages_isolated, and the range will actually be
7843 * allocated. So, if we fall through be sure to clear ret so that
7844 * -EBUSY is not accidentally used or returned to caller.
7846 ret = __alloc_contig_migrate_range(&cc, start, end);
7847 if (ret && ret != -EBUSY)
7852 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
7853 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
7854 * more, all pages in [start, end) are free in page allocator.
7855 * What we are going to do is to allocate all pages from
7856 * [start, end) (that is remove them from page allocator).
7858 * The only problem is that pages at the beginning and at the
7859 * end of interesting range may be not aligned with pages that
7860 * page allocator holds, ie. they can be part of higher order
7861 * pages. Because of this, we reserve the bigger range and
7862 * once this is done free the pages we are not interested in.
7864 * We don't have to hold zone->lock here because the pages are
7865 * isolated thus they won't get removed from buddy.
7868 lru_add_drain_all();
7869 drain_all_pages(cc.zone);
7872 outer_start = start;
7873 while (!PageBuddy(pfn_to_page(outer_start))) {
7874 if (++order >= MAX_ORDER) {
7875 outer_start = start;
7878 outer_start &= ~0UL << order;
7881 if (outer_start != start) {
7882 order = page_order(pfn_to_page(outer_start));
7885 * outer_start page could be small order buddy page and
7886 * it doesn't include start page. Adjust outer_start
7887 * in this case to report failed page properly
7888 * on tracepoint in test_pages_isolated()
7890 if (outer_start + (1UL << order) <= start)
7891 outer_start = start;
7894 /* Make sure the range is really isolated. */
7895 if (test_pages_isolated(outer_start, end, false)) {
7896 pr_info_ratelimited("%s: [%lx, %lx) PFNs busy\n",
7897 __func__, outer_start, end);
7902 /* Grab isolated pages from freelists. */
7903 outer_end = isolate_freepages_range(&cc, outer_start, end);
7909 /* Free head and tail (if any) */
7910 if (start != outer_start)
7911 free_contig_range(outer_start, start - outer_start);
7912 if (end != outer_end)
7913 free_contig_range(end, outer_end - end);
7916 undo_isolate_page_range(pfn_max_align_down(start),
7917 pfn_max_align_up(end), migratetype);
7921 void free_contig_range(unsigned long pfn, unsigned nr_pages)
7923 unsigned int count = 0;
7925 for (; nr_pages--; pfn++) {
7926 struct page *page = pfn_to_page(pfn);
7928 count += page_count(page) != 1;
7931 WARN(count != 0, "%d pages are still in use!\n", count);
7935 #ifdef CONFIG_MEMORY_HOTPLUG
7937 * The zone indicated has a new number of managed_pages; batch sizes and percpu
7938 * page high values need to be recalulated.
7940 void __meminit zone_pcp_update(struct zone *zone)
7943 mutex_lock(&pcp_batch_high_lock);
7944 for_each_possible_cpu(cpu)
7945 pageset_set_high_and_batch(zone,
7946 per_cpu_ptr(zone->pageset, cpu));
7947 mutex_unlock(&pcp_batch_high_lock);
7951 void zone_pcp_reset(struct zone *zone)
7953 unsigned long flags;
7955 struct per_cpu_pageset *pset;
7957 /* avoid races with drain_pages() */
7958 local_irq_save(flags);
7959 if (zone->pageset != &boot_pageset) {
7960 for_each_online_cpu(cpu) {
7961 pset = per_cpu_ptr(zone->pageset, cpu);
7962 drain_zonestat(zone, pset);
7964 free_percpu(zone->pageset);
7965 zone->pageset = &boot_pageset;
7967 local_irq_restore(flags);
7970 #ifdef CONFIG_MEMORY_HOTREMOVE
7972 * All pages in the range must be in a single zone and isolated
7973 * before calling this.
7976 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
7980 unsigned int order, i;
7982 unsigned long flags;
7983 /* find the first valid pfn */
7984 for (pfn = start_pfn; pfn < end_pfn; pfn++)
7989 offline_mem_sections(pfn, end_pfn);
7990 zone = page_zone(pfn_to_page(pfn));
7991 spin_lock_irqsave(&zone->lock, flags);
7993 while (pfn < end_pfn) {
7994 if (!pfn_valid(pfn)) {
7998 page = pfn_to_page(pfn);
8000 * The HWPoisoned page may be not in buddy system, and
8001 * page_count() is not 0.
8003 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
8005 SetPageReserved(page);
8009 BUG_ON(page_count(page));
8010 BUG_ON(!PageBuddy(page));
8011 order = page_order(page);
8012 #ifdef CONFIG_DEBUG_VM
8013 pr_info("remove from free list %lx %d %lx\n",
8014 pfn, 1 << order, end_pfn);
8016 list_del(&page->lru);
8017 rmv_page_order(page);
8018 zone->free_area[order].nr_free--;
8019 for (i = 0; i < (1 << order); i++)
8020 SetPageReserved((page+i));
8021 pfn += (1 << order);
8023 spin_unlock_irqrestore(&zone->lock, flags);
8027 bool is_free_buddy_page(struct page *page)
8029 struct zone *zone = page_zone(page);
8030 unsigned long pfn = page_to_pfn(page);
8031 unsigned long flags;
8034 spin_lock_irqsave(&zone->lock, flags);
8035 for (order = 0; order < MAX_ORDER; order++) {
8036 struct page *page_head = page - (pfn & ((1 << order) - 1));
8038 if (PageBuddy(page_head) && page_order(page_head) >= order)
8041 spin_unlock_irqrestore(&zone->lock, flags);
8043 return order < MAX_ORDER;