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/kmemcheck.h>
28 #include <linux/kasan.h>
29 #include <linux/module.h>
30 #include <linux/suspend.h>
31 #include <linux/pagevec.h>
32 #include <linux/blkdev.h>
33 #include <linux/slab.h>
34 #include <linux/ratelimit.h>
35 #include <linux/oom.h>
36 #include <linux/notifier.h>
37 #include <linux/topology.h>
38 #include <linux/sysctl.h>
39 #include <linux/cpu.h>
40 #include <linux/cpuset.h>
41 #include <linux/memory_hotplug.h>
42 #include <linux/nodemask.h>
43 #include <linux/vmalloc.h>
44 #include <linux/vmstat.h>
45 #include <linux/mempolicy.h>
46 #include <linux/memremap.h>
47 #include <linux/stop_machine.h>
48 #include <linux/sort.h>
49 #include <linux/pfn.h>
50 #include <linux/backing-dev.h>
51 #include <linux/fault-inject.h>
52 #include <linux/page-isolation.h>
53 #include <linux/page_ext.h>
54 #include <linux/debugobjects.h>
55 #include <linux/kmemleak.h>
56 #include <linux/compaction.h>
57 #include <trace/events/kmem.h>
58 #include <linux/prefetch.h>
59 #include <linux/mm_inline.h>
60 #include <linux/migrate.h>
61 #include <linux/page_ext.h>
62 #include <linux/hugetlb.h>
63 #include <linux/sched/rt.h>
64 #include <linux/page_owner.h>
65 #include <linux/kthread.h>
67 #include <asm/sections.h>
68 #include <asm/tlbflush.h>
69 #include <asm/div64.h>
72 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
73 static DEFINE_MUTEX(pcp_batch_high_lock);
74 #define MIN_PERCPU_PAGELIST_FRACTION (8)
76 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
77 DEFINE_PER_CPU(int, numa_node);
78 EXPORT_PER_CPU_SYMBOL(numa_node);
81 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
83 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
84 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
85 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
86 * defined in <linux/topology.h>.
88 DEFINE_PER_CPU(int, _numa_mem_); /* Kernel "local memory" node */
89 EXPORT_PER_CPU_SYMBOL(_numa_mem_);
90 int _node_numa_mem_[MAX_NUMNODES];
94 * Array of node states.
96 nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
97 [N_POSSIBLE] = NODE_MASK_ALL,
98 [N_ONLINE] = { { [0] = 1UL } },
100 [N_NORMAL_MEMORY] = { { [0] = 1UL } },
101 #ifdef CONFIG_HIGHMEM
102 [N_HIGH_MEMORY] = { { [0] = 1UL } },
104 #ifdef CONFIG_MOVABLE_NODE
105 [N_MEMORY] = { { [0] = 1UL } },
107 [N_CPU] = { { [0] = 1UL } },
110 EXPORT_SYMBOL(node_states);
112 /* Protect totalram_pages and zone->managed_pages */
113 static DEFINE_SPINLOCK(managed_page_count_lock);
115 unsigned long totalram_pages __read_mostly;
116 unsigned long totalreserve_pages __read_mostly;
117 unsigned long totalcma_pages __read_mostly;
119 int percpu_pagelist_fraction;
120 gfp_t gfp_allowed_mask __read_mostly = GFP_BOOT_MASK;
123 * A cached value of the page's pageblock's migratetype, used when the page is
124 * put on a pcplist. Used to avoid the pageblock migratetype lookup when
125 * freeing from pcplists in most cases, at the cost of possibly becoming stale.
126 * Also the migratetype set in the page does not necessarily match the pcplist
127 * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
128 * other index - this ensures that it will be put on the correct CMA freelist.
130 static inline int get_pcppage_migratetype(struct page *page)
135 static inline void set_pcppage_migratetype(struct page *page, int migratetype)
137 page->index = migratetype;
140 #ifdef CONFIG_PM_SLEEP
142 * The following functions are used by the suspend/hibernate code to temporarily
143 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
144 * while devices are suspended. To avoid races with the suspend/hibernate code,
145 * they should always be called with pm_mutex held (gfp_allowed_mask also should
146 * only be modified with pm_mutex held, unless the suspend/hibernate code is
147 * guaranteed not to run in parallel with that modification).
150 static gfp_t saved_gfp_mask;
152 void pm_restore_gfp_mask(void)
154 WARN_ON(!mutex_is_locked(&pm_mutex));
155 if (saved_gfp_mask) {
156 gfp_allowed_mask = saved_gfp_mask;
161 void pm_restrict_gfp_mask(void)
163 WARN_ON(!mutex_is_locked(&pm_mutex));
164 WARN_ON(saved_gfp_mask);
165 saved_gfp_mask = gfp_allowed_mask;
166 gfp_allowed_mask &= ~(__GFP_IO | __GFP_FS);
169 bool pm_suspended_storage(void)
171 if ((gfp_allowed_mask & (__GFP_IO | __GFP_FS)) == (__GFP_IO | __GFP_FS))
175 #endif /* CONFIG_PM_SLEEP */
177 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
178 unsigned int pageblock_order __read_mostly;
181 static void __free_pages_ok(struct page *page, unsigned int order);
184 * results with 256, 32 in the lowmem_reserve sysctl:
185 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
186 * 1G machine -> (16M dma, 784M normal, 224M high)
187 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
188 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
189 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
191 * TBD: should special case ZONE_DMA32 machines here - in those we normally
192 * don't need any ZONE_NORMAL reservation
194 int sysctl_lowmem_reserve_ratio[MAX_NR_ZONES-1] = {
195 #ifdef CONFIG_ZONE_DMA
198 #ifdef CONFIG_ZONE_DMA32
201 #ifdef CONFIG_HIGHMEM
207 EXPORT_SYMBOL(totalram_pages);
209 static char * const zone_names[MAX_NR_ZONES] = {
210 #ifdef CONFIG_ZONE_DMA
213 #ifdef CONFIG_ZONE_DMA32
217 #ifdef CONFIG_HIGHMEM
221 #ifdef CONFIG_ZONE_DEVICE
226 char * const migratetype_names[MIGRATE_TYPES] = {
234 #ifdef CONFIG_MEMORY_ISOLATION
239 compound_page_dtor * const compound_page_dtors[] = {
242 #ifdef CONFIG_HUGETLB_PAGE
245 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
250 int min_free_kbytes = 1024;
251 int user_min_free_kbytes = -1;
252 int watermark_scale_factor = 10;
254 static unsigned long __meminitdata nr_kernel_pages;
255 static unsigned long __meminitdata nr_all_pages;
256 static unsigned long __meminitdata dma_reserve;
258 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
259 static unsigned long __meminitdata arch_zone_lowest_possible_pfn[MAX_NR_ZONES];
260 static unsigned long __meminitdata arch_zone_highest_possible_pfn[MAX_NR_ZONES];
261 static unsigned long __initdata required_kernelcore;
262 static unsigned long __initdata required_movablecore;
263 static unsigned long __meminitdata zone_movable_pfn[MAX_NUMNODES];
264 static bool mirrored_kernelcore;
266 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
268 EXPORT_SYMBOL(movable_zone);
269 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
272 int nr_node_ids __read_mostly = MAX_NUMNODES;
273 int nr_online_nodes __read_mostly = 1;
274 EXPORT_SYMBOL(nr_node_ids);
275 EXPORT_SYMBOL(nr_online_nodes);
278 int page_group_by_mobility_disabled __read_mostly;
280 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
281 static inline void reset_deferred_meminit(pg_data_t *pgdat)
283 pgdat->first_deferred_pfn = ULONG_MAX;
286 /* Returns true if the struct page for the pfn is uninitialised */
287 static inline bool __meminit early_page_uninitialised(unsigned long pfn)
289 if (pfn >= NODE_DATA(early_pfn_to_nid(pfn))->first_deferred_pfn)
295 static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid)
297 if (pfn >= NODE_DATA(nid)->first_deferred_pfn)
304 * Returns false when the remaining initialisation should be deferred until
305 * later in the boot cycle when it can be parallelised.
307 static inline bool update_defer_init(pg_data_t *pgdat,
308 unsigned long pfn, unsigned long zone_end,
309 unsigned long *nr_initialised)
311 unsigned long max_initialise;
313 /* Always populate low zones for address-contrained allocations */
314 if (zone_end < pgdat_end_pfn(pgdat))
317 * Initialise at least 2G of a node but also take into account that
318 * two large system hashes that can take up 1GB for 0.25TB/node.
320 max_initialise = max(2UL << (30 - PAGE_SHIFT),
321 (pgdat->node_spanned_pages >> 8));
324 if ((*nr_initialised > max_initialise) &&
325 (pfn & (PAGES_PER_SECTION - 1)) == 0) {
326 pgdat->first_deferred_pfn = pfn;
333 static inline void reset_deferred_meminit(pg_data_t *pgdat)
337 static inline bool early_page_uninitialised(unsigned long pfn)
342 static inline bool early_page_nid_uninitialised(unsigned long pfn, int nid)
347 static inline bool update_defer_init(pg_data_t *pgdat,
348 unsigned long pfn, unsigned long zone_end,
349 unsigned long *nr_initialised)
356 void set_pageblock_migratetype(struct page *page, int migratetype)
358 if (unlikely(page_group_by_mobility_disabled &&
359 migratetype < MIGRATE_PCPTYPES))
360 migratetype = MIGRATE_UNMOVABLE;
362 set_pageblock_flags_group(page, (unsigned long)migratetype,
363 PB_migrate, PB_migrate_end);
366 #ifdef CONFIG_DEBUG_VM
367 static int page_outside_zone_boundaries(struct zone *zone, struct page *page)
371 unsigned long pfn = page_to_pfn(page);
372 unsigned long sp, start_pfn;
375 seq = zone_span_seqbegin(zone);
376 start_pfn = zone->zone_start_pfn;
377 sp = zone->spanned_pages;
378 if (!zone_spans_pfn(zone, pfn))
380 } while (zone_span_seqretry(zone, seq));
383 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
384 pfn, zone_to_nid(zone), zone->name,
385 start_pfn, start_pfn + sp);
390 static int page_is_consistent(struct zone *zone, struct page *page)
392 if (!pfn_valid_within(page_to_pfn(page)))
394 if (zone != page_zone(page))
400 * Temporary debugging check for pages not lying within a given zone.
402 static int bad_range(struct zone *zone, struct page *page)
404 if (page_outside_zone_boundaries(zone, page))
406 if (!page_is_consistent(zone, page))
412 static inline int bad_range(struct zone *zone, struct page *page)
418 static void bad_page(struct page *page, const char *reason,
419 unsigned long bad_flags)
421 static unsigned long resume;
422 static unsigned long nr_shown;
423 static unsigned long nr_unshown;
425 /* Don't complain about poisoned pages */
426 if (PageHWPoison(page)) {
427 page_mapcount_reset(page); /* remove PageBuddy */
432 * Allow a burst of 60 reports, then keep quiet for that minute;
433 * or allow a steady drip of one report per second.
435 if (nr_shown == 60) {
436 if (time_before(jiffies, resume)) {
442 "BUG: Bad page state: %lu messages suppressed\n",
449 resume = jiffies + 60 * HZ;
451 pr_alert("BUG: Bad page state in process %s pfn:%05lx\n",
452 current->comm, page_to_pfn(page));
453 __dump_page(page, reason);
454 bad_flags &= page->flags;
456 pr_alert("bad because of flags: %#lx(%pGp)\n",
457 bad_flags, &bad_flags);
458 dump_page_owner(page);
463 /* Leave bad fields for debug, except PageBuddy could make trouble */
464 page_mapcount_reset(page); /* remove PageBuddy */
465 add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
469 * Higher-order pages are called "compound pages". They are structured thusly:
471 * The first PAGE_SIZE page is called the "head page" and have PG_head set.
473 * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
474 * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
476 * The first tail page's ->compound_dtor holds the offset in array of compound
477 * page destructors. See compound_page_dtors.
479 * The first tail page's ->compound_order holds the order of allocation.
480 * This usage means that zero-order pages may not be compound.
483 void free_compound_page(struct page *page)
485 __free_pages_ok(page, compound_order(page));
488 void prep_compound_page(struct page *page, unsigned int order)
491 int nr_pages = 1 << order;
493 set_compound_page_dtor(page, COMPOUND_PAGE_DTOR);
494 set_compound_order(page, order);
496 for (i = 1; i < nr_pages; i++) {
497 struct page *p = page + i;
498 set_page_count(p, 0);
499 p->mapping = TAIL_MAPPING;
500 set_compound_head(p, page);
502 atomic_set(compound_mapcount_ptr(page), -1);
505 #ifdef CONFIG_DEBUG_PAGEALLOC
506 unsigned int _debug_guardpage_minorder;
507 bool _debug_pagealloc_enabled __read_mostly
508 = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT);
509 EXPORT_SYMBOL(_debug_pagealloc_enabled);
510 bool _debug_guardpage_enabled __read_mostly;
512 static int __init early_debug_pagealloc(char *buf)
517 if (strcmp(buf, "on") == 0)
518 _debug_pagealloc_enabled = true;
520 if (strcmp(buf, "off") == 0)
521 _debug_pagealloc_enabled = false;
525 early_param("debug_pagealloc", early_debug_pagealloc);
527 static bool need_debug_guardpage(void)
529 /* If we don't use debug_pagealloc, we don't need guard page */
530 if (!debug_pagealloc_enabled())
536 static void init_debug_guardpage(void)
538 if (!debug_pagealloc_enabled())
541 _debug_guardpage_enabled = true;
544 struct page_ext_operations debug_guardpage_ops = {
545 .need = need_debug_guardpage,
546 .init = init_debug_guardpage,
549 static int __init debug_guardpage_minorder_setup(char *buf)
553 if (kstrtoul(buf, 10, &res) < 0 || res > MAX_ORDER / 2) {
554 pr_err("Bad debug_guardpage_minorder value\n");
557 _debug_guardpage_minorder = res;
558 pr_info("Setting debug_guardpage_minorder to %lu\n", res);
561 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup);
563 static inline void set_page_guard(struct zone *zone, struct page *page,
564 unsigned int order, int migratetype)
566 struct page_ext *page_ext;
568 if (!debug_guardpage_enabled())
571 page_ext = lookup_page_ext(page);
572 __set_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
574 INIT_LIST_HEAD(&page->lru);
575 set_page_private(page, order);
576 /* Guard pages are not available for any usage */
577 __mod_zone_freepage_state(zone, -(1 << order), migratetype);
580 static inline void clear_page_guard(struct zone *zone, struct page *page,
581 unsigned int order, int migratetype)
583 struct page_ext *page_ext;
585 if (!debug_guardpage_enabled())
588 page_ext = lookup_page_ext(page);
589 __clear_bit(PAGE_EXT_DEBUG_GUARD, &page_ext->flags);
591 set_page_private(page, 0);
592 if (!is_migrate_isolate(migratetype))
593 __mod_zone_freepage_state(zone, (1 << order), migratetype);
596 struct page_ext_operations debug_guardpage_ops = { NULL, };
597 static inline void set_page_guard(struct zone *zone, struct page *page,
598 unsigned int order, int migratetype) {}
599 static inline void clear_page_guard(struct zone *zone, struct page *page,
600 unsigned int order, int migratetype) {}
603 static inline void set_page_order(struct page *page, unsigned int order)
605 set_page_private(page, order);
606 __SetPageBuddy(page);
609 static inline void rmv_page_order(struct page *page)
611 __ClearPageBuddy(page);
612 set_page_private(page, 0);
616 * This function checks whether a page is free && is the buddy
617 * we can do coalesce a page and its buddy if
618 * (a) the buddy is not in a hole &&
619 * (b) the buddy is in the buddy system &&
620 * (c) a page and its buddy have the same order &&
621 * (d) a page and its buddy are in the same zone.
623 * For recording whether a page is in the buddy system, we set ->_mapcount
624 * PAGE_BUDDY_MAPCOUNT_VALUE.
625 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
626 * serialized by zone->lock.
628 * For recording page's order, we use page_private(page).
630 static inline int page_is_buddy(struct page *page, struct page *buddy,
633 if (!pfn_valid_within(page_to_pfn(buddy)))
636 if (page_is_guard(buddy) && page_order(buddy) == order) {
637 if (page_zone_id(page) != page_zone_id(buddy))
640 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
645 if (PageBuddy(buddy) && page_order(buddy) == order) {
647 * zone check is done late to avoid uselessly
648 * calculating zone/node ids for pages that could
651 if (page_zone_id(page) != page_zone_id(buddy))
654 VM_BUG_ON_PAGE(page_count(buddy) != 0, buddy);
662 * Freeing function for a buddy system allocator.
664 * The concept of a buddy system is to maintain direct-mapped table
665 * (containing bit values) for memory blocks of various "orders".
666 * The bottom level table contains the map for the smallest allocatable
667 * units of memory (here, pages), and each level above it describes
668 * pairs of units from the levels below, hence, "buddies".
669 * At a high level, all that happens here is marking the table entry
670 * at the bottom level available, and propagating the changes upward
671 * as necessary, plus some accounting needed to play nicely with other
672 * parts of the VM system.
673 * At each level, we keep a list of pages, which are heads of continuous
674 * free pages of length of (1 << order) and marked with _mapcount
675 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
677 * So when we are allocating or freeing one, we can derive the state of the
678 * other. That is, if we allocate a small block, and both were
679 * free, the remainder of the region must be split into blocks.
680 * If a block is freed, and its buddy is also free, then this
681 * triggers coalescing into a block of larger size.
686 static inline void __free_one_page(struct page *page,
688 struct zone *zone, unsigned int order,
691 unsigned long page_idx;
692 unsigned long combined_idx;
693 unsigned long uninitialized_var(buddy_idx);
695 unsigned int max_order;
697 max_order = min_t(unsigned int, MAX_ORDER, pageblock_order + 1);
699 VM_BUG_ON(!zone_is_initialized(zone));
700 VM_BUG_ON_PAGE(page->flags & PAGE_FLAGS_CHECK_AT_PREP, page);
702 VM_BUG_ON(migratetype == -1);
703 if (likely(!is_migrate_isolate(migratetype)))
704 __mod_zone_freepage_state(zone, 1 << order, migratetype);
706 page_idx = pfn & ((1 << MAX_ORDER) - 1);
708 VM_BUG_ON_PAGE(page_idx & ((1 << order) - 1), page);
709 VM_BUG_ON_PAGE(bad_range(zone, page), page);
712 while (order < max_order - 1) {
713 buddy_idx = __find_buddy_index(page_idx, order);
714 buddy = page + (buddy_idx - page_idx);
715 if (!page_is_buddy(page, buddy, order))
718 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
719 * merge with it and move up one order.
721 if (page_is_guard(buddy)) {
722 clear_page_guard(zone, buddy, order, migratetype);
724 list_del(&buddy->lru);
725 zone->free_area[order].nr_free--;
726 rmv_page_order(buddy);
728 combined_idx = buddy_idx & page_idx;
729 page = page + (combined_idx - page_idx);
730 page_idx = combined_idx;
733 if (max_order < MAX_ORDER) {
734 /* If we are here, it means order is >= pageblock_order.
735 * We want to prevent merge between freepages on isolate
736 * pageblock and normal pageblock. Without this, pageblock
737 * isolation could cause incorrect freepage or CMA accounting.
739 * We don't want to hit this code for the more frequent
742 if (unlikely(has_isolate_pageblock(zone))) {
745 buddy_idx = __find_buddy_index(page_idx, order);
746 buddy = page + (buddy_idx - page_idx);
747 buddy_mt = get_pageblock_migratetype(buddy);
749 if (migratetype != buddy_mt
750 && (is_migrate_isolate(migratetype) ||
751 is_migrate_isolate(buddy_mt)))
755 goto continue_merging;
759 set_page_order(page, order);
762 * If this is not the largest possible page, check if the buddy
763 * of the next-highest order is free. If it is, it's possible
764 * that pages are being freed that will coalesce soon. In case,
765 * that is happening, add the free page to the tail of the list
766 * so it's less likely to be used soon and more likely to be merged
767 * as a higher order page
769 if ((order < MAX_ORDER-2) && pfn_valid_within(page_to_pfn(buddy))) {
770 struct page *higher_page, *higher_buddy;
771 combined_idx = buddy_idx & page_idx;
772 higher_page = page + (combined_idx - page_idx);
773 buddy_idx = __find_buddy_index(combined_idx, order + 1);
774 higher_buddy = higher_page + (buddy_idx - combined_idx);
775 if (page_is_buddy(higher_page, higher_buddy, order + 1)) {
776 list_add_tail(&page->lru,
777 &zone->free_area[order].free_list[migratetype]);
782 list_add(&page->lru, &zone->free_area[order].free_list[migratetype]);
784 zone->free_area[order].nr_free++;
788 * A bad page could be due to a number of fields. Instead of multiple branches,
789 * try and check multiple fields with one check. The caller must do a detailed
790 * check if necessary.
792 static inline bool page_expected_state(struct page *page,
793 unsigned long check_flags)
795 if (unlikely(atomic_read(&page->_mapcount) != -1))
798 if (unlikely((unsigned long)page->mapping |
799 page_ref_count(page) |
801 (unsigned long)page->mem_cgroup |
803 (page->flags & check_flags)))
809 static void free_pages_check_bad(struct page *page)
811 const char *bad_reason;
812 unsigned long bad_flags;
817 if (unlikely(atomic_read(&page->_mapcount) != -1))
818 bad_reason = "nonzero mapcount";
819 if (unlikely(page->mapping != NULL))
820 bad_reason = "non-NULL mapping";
821 if (unlikely(page_ref_count(page) != 0))
822 bad_reason = "nonzero _refcount";
823 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
824 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
825 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
828 if (unlikely(page->mem_cgroup))
829 bad_reason = "page still charged to cgroup";
831 bad_page(page, bad_reason, bad_flags);
834 static inline int free_pages_check(struct page *page)
836 if (likely(page_expected_state(page, PAGE_FLAGS_CHECK_AT_FREE)))
839 /* Something has gone sideways, find it */
840 free_pages_check_bad(page);
845 * Frees a number of pages from the PCP lists
846 * Assumes all pages on list are in same zone, and of same order.
847 * count is the number of pages to free.
849 * If the zone was previously in an "all pages pinned" state then look to
850 * see if this freeing clears that state.
852 * And clear the zone's pages_scanned counter, to hold off the "all pages are
853 * pinned" detection logic.
855 static void free_pcppages_bulk(struct zone *zone, int count,
856 struct per_cpu_pages *pcp)
860 unsigned long nr_scanned;
861 bool isolated_pageblocks;
863 spin_lock(&zone->lock);
864 isolated_pageblocks = has_isolate_pageblock(zone);
865 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
867 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
871 struct list_head *list;
874 * Remove pages from lists in a round-robin fashion. A
875 * batch_free count is maintained that is incremented when an
876 * empty list is encountered. This is so more pages are freed
877 * off fuller lists instead of spinning excessively around empty
882 if (++migratetype == MIGRATE_PCPTYPES)
884 list = &pcp->lists[migratetype];
885 } while (list_empty(list));
887 /* This is the only non-empty list. Free them all. */
888 if (batch_free == MIGRATE_PCPTYPES)
892 int mt; /* migratetype of the to-be-freed page */
894 page = list_last_entry(list, struct page, lru);
895 /* must delete as __free_one_page list manipulates */
896 list_del(&page->lru);
898 mt = get_pcppage_migratetype(page);
899 /* MIGRATE_ISOLATE page should not go to pcplists */
900 VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
901 /* Pageblock could have been isolated meanwhile */
902 if (unlikely(isolated_pageblocks))
903 mt = get_pageblock_migratetype(page);
905 __free_one_page(page, page_to_pfn(page), zone, 0, mt);
906 trace_mm_page_pcpu_drain(page, 0, mt);
907 } while (--count && --batch_free && !list_empty(list));
909 spin_unlock(&zone->lock);
912 static void free_one_page(struct zone *zone,
913 struct page *page, unsigned long pfn,
917 unsigned long nr_scanned;
918 spin_lock(&zone->lock);
919 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
921 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
923 if (unlikely(has_isolate_pageblock(zone) ||
924 is_migrate_isolate(migratetype))) {
925 migratetype = get_pfnblock_migratetype(page, pfn);
927 __free_one_page(page, pfn, zone, order, migratetype);
928 spin_unlock(&zone->lock);
931 static int free_tail_pages_check(struct page *head_page, struct page *page)
936 * We rely page->lru.next never has bit 0 set, unless the page
937 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
939 BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
941 if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
945 switch (page - head_page) {
947 /* the first tail page: ->mapping is compound_mapcount() */
948 if (unlikely(compound_mapcount(page))) {
949 bad_page(page, "nonzero compound_mapcount", 0);
955 * the second tail page: ->mapping is
956 * page_deferred_list().next -- ignore value.
960 if (page->mapping != TAIL_MAPPING) {
961 bad_page(page, "corrupted mapping in tail page", 0);
966 if (unlikely(!PageTail(page))) {
967 bad_page(page, "PageTail not set", 0);
970 if (unlikely(compound_head(page) != head_page)) {
971 bad_page(page, "compound_head not consistent", 0);
976 page->mapping = NULL;
977 clear_compound_head(page);
981 static void __meminit __init_single_page(struct page *page, unsigned long pfn,
982 unsigned long zone, int nid)
984 set_page_links(page, zone, nid, pfn);
985 init_page_count(page);
986 page_mapcount_reset(page);
987 page_cpupid_reset_last(page);
989 INIT_LIST_HEAD(&page->lru);
990 #ifdef WANT_PAGE_VIRTUAL
991 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
992 if (!is_highmem_idx(zone))
993 set_page_address(page, __va(pfn << PAGE_SHIFT));
997 static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone,
1000 return __init_single_page(pfn_to_page(pfn), pfn, zone, nid);
1003 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1004 static void init_reserved_page(unsigned long pfn)
1009 if (!early_page_uninitialised(pfn))
1012 nid = early_pfn_to_nid(pfn);
1013 pgdat = NODE_DATA(nid);
1015 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1016 struct zone *zone = &pgdat->node_zones[zid];
1018 if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
1021 __init_single_pfn(pfn, zid, nid);
1024 static inline void init_reserved_page(unsigned long pfn)
1027 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1030 * Initialised pages do not have PageReserved set. This function is
1031 * called for each range allocated by the bootmem allocator and
1032 * marks the pages PageReserved. The remaining valid pages are later
1033 * sent to the buddy page allocator.
1035 void __meminit reserve_bootmem_region(unsigned long start, unsigned long end)
1037 unsigned long start_pfn = PFN_DOWN(start);
1038 unsigned long end_pfn = PFN_UP(end);
1040 for (; start_pfn < end_pfn; start_pfn++) {
1041 if (pfn_valid(start_pfn)) {
1042 struct page *page = pfn_to_page(start_pfn);
1044 init_reserved_page(start_pfn);
1046 /* Avoid false-positive PageTail() */
1047 INIT_LIST_HEAD(&page->lru);
1049 SetPageReserved(page);
1054 static bool free_pages_prepare(struct page *page, unsigned int order)
1058 VM_BUG_ON_PAGE(PageTail(page), page);
1060 trace_mm_page_free(page, order);
1061 kmemcheck_free_shadow(page, order);
1062 kasan_free_pages(page, order);
1065 * Check tail pages before head page information is cleared to
1066 * avoid checking PageCompound for order-0 pages.
1068 if (unlikely(order)) {
1069 bool compound = PageCompound(page);
1072 VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
1074 for (i = 1; i < (1 << order); i++) {
1076 bad += free_tail_pages_check(page, page + i);
1077 if (unlikely(free_pages_check(page + i))) {
1081 (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1084 if (PageAnonHead(page))
1085 page->mapping = NULL;
1086 bad += free_pages_check(page);
1090 page_cpupid_reset_last(page);
1091 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
1092 reset_page_owner(page, order);
1094 if (!PageHighMem(page)) {
1095 debug_check_no_locks_freed(page_address(page),
1096 PAGE_SIZE << order);
1097 debug_check_no_obj_freed(page_address(page),
1098 PAGE_SIZE << order);
1100 arch_free_page(page, order);
1101 kernel_poison_pages(page, 1 << order, 0);
1102 kernel_map_pages(page, 1 << order, 0);
1107 static void __free_pages_ok(struct page *page, unsigned int order)
1109 unsigned long flags;
1111 unsigned long pfn = page_to_pfn(page);
1113 if (!free_pages_prepare(page, order))
1116 migratetype = get_pfnblock_migratetype(page, pfn);
1117 local_irq_save(flags);
1118 __count_vm_events(PGFREE, 1 << order);
1119 free_one_page(page_zone(page), page, pfn, order, migratetype);
1120 local_irq_restore(flags);
1123 static void __init __free_pages_boot_core(struct page *page, unsigned int order)
1125 unsigned int nr_pages = 1 << order;
1126 struct page *p = page;
1130 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1132 __ClearPageReserved(p);
1133 set_page_count(p, 0);
1135 __ClearPageReserved(p);
1136 set_page_count(p, 0);
1138 page_zone(page)->managed_pages += nr_pages;
1139 set_page_refcounted(page);
1140 __free_pages(page, order);
1143 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1144 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1146 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1148 int __meminit early_pfn_to_nid(unsigned long pfn)
1150 static DEFINE_SPINLOCK(early_pfn_lock);
1153 spin_lock(&early_pfn_lock);
1154 nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
1157 spin_unlock(&early_pfn_lock);
1163 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1164 static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1165 struct mminit_pfnnid_cache *state)
1169 nid = __early_pfn_to_nid(pfn, state);
1170 if (nid >= 0 && nid != node)
1175 /* Only safe to use early in boot when initialisation is single-threaded */
1176 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1178 return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1183 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1187 static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1188 struct mminit_pfnnid_cache *state)
1195 void __init __free_pages_bootmem(struct page *page, unsigned long pfn,
1198 if (early_page_uninitialised(pfn))
1200 return __free_pages_boot_core(page, order);
1204 * Check that the whole (or subset of) a pageblock given by the interval of
1205 * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1206 * with the migration of free compaction scanner. The scanners then need to
1207 * use only pfn_valid_within() check for arches that allow holes within
1210 * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1212 * It's possible on some configurations to have a setup like node0 node1 node0
1213 * i.e. it's possible that all pages within a zones range of pages do not
1214 * belong to a single zone. We assume that a border between node0 and node1
1215 * can occur within a single pageblock, but not a node0 node1 node0
1216 * interleaving within a single pageblock. It is therefore sufficient to check
1217 * the first and last page of a pageblock and avoid checking each individual
1218 * page in a pageblock.
1220 struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
1221 unsigned long end_pfn, struct zone *zone)
1223 struct page *start_page;
1224 struct page *end_page;
1226 /* end_pfn is one past the range we are checking */
1229 if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
1232 start_page = pfn_to_page(start_pfn);
1234 if (page_zone(start_page) != zone)
1237 end_page = pfn_to_page(end_pfn);
1239 /* This gives a shorter code than deriving page_zone(end_page) */
1240 if (page_zone_id(start_page) != page_zone_id(end_page))
1246 void set_zone_contiguous(struct zone *zone)
1248 unsigned long block_start_pfn = zone->zone_start_pfn;
1249 unsigned long block_end_pfn;
1251 block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages);
1252 for (; block_start_pfn < zone_end_pfn(zone);
1253 block_start_pfn = block_end_pfn,
1254 block_end_pfn += pageblock_nr_pages) {
1256 block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
1258 if (!__pageblock_pfn_to_page(block_start_pfn,
1259 block_end_pfn, zone))
1263 /* We confirm that there is no hole */
1264 zone->contiguous = true;
1267 void clear_zone_contiguous(struct zone *zone)
1269 zone->contiguous = false;
1272 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1273 static void __init deferred_free_range(struct page *page,
1274 unsigned long pfn, int nr_pages)
1281 /* Free a large naturally-aligned chunk if possible */
1282 if (nr_pages == MAX_ORDER_NR_PAGES &&
1283 (pfn & (MAX_ORDER_NR_PAGES-1)) == 0) {
1284 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1285 __free_pages_boot_core(page, MAX_ORDER-1);
1289 for (i = 0; i < nr_pages; i++, page++)
1290 __free_pages_boot_core(page, 0);
1293 /* Completion tracking for deferred_init_memmap() threads */
1294 static atomic_t pgdat_init_n_undone __initdata;
1295 static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1297 static inline void __init pgdat_init_report_one_done(void)
1299 if (atomic_dec_and_test(&pgdat_init_n_undone))
1300 complete(&pgdat_init_all_done_comp);
1303 /* Initialise remaining memory on a node */
1304 static int __init deferred_init_memmap(void *data)
1306 pg_data_t *pgdat = data;
1307 int nid = pgdat->node_id;
1308 struct mminit_pfnnid_cache nid_init_state = { };
1309 unsigned long start = jiffies;
1310 unsigned long nr_pages = 0;
1311 unsigned long walk_start, walk_end;
1314 unsigned long first_init_pfn = pgdat->first_deferred_pfn;
1315 const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
1317 if (first_init_pfn == ULONG_MAX) {
1318 pgdat_init_report_one_done();
1322 /* Bind memory initialisation thread to a local node if possible */
1323 if (!cpumask_empty(cpumask))
1324 set_cpus_allowed_ptr(current, cpumask);
1326 /* Sanity check boundaries */
1327 BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1328 BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1329 pgdat->first_deferred_pfn = ULONG_MAX;
1331 /* Only the highest zone is deferred so find it */
1332 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1333 zone = pgdat->node_zones + zid;
1334 if (first_init_pfn < zone_end_pfn(zone))
1338 for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) {
1339 unsigned long pfn, end_pfn;
1340 struct page *page = NULL;
1341 struct page *free_base_page = NULL;
1342 unsigned long free_base_pfn = 0;
1345 end_pfn = min(walk_end, zone_end_pfn(zone));
1346 pfn = first_init_pfn;
1347 if (pfn < walk_start)
1349 if (pfn < zone->zone_start_pfn)
1350 pfn = zone->zone_start_pfn;
1352 for (; pfn < end_pfn; pfn++) {
1353 if (!pfn_valid_within(pfn))
1357 * Ensure pfn_valid is checked every
1358 * MAX_ORDER_NR_PAGES for memory holes
1360 if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
1361 if (!pfn_valid(pfn)) {
1367 if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) {
1372 /* Minimise pfn page lookups and scheduler checks */
1373 if (page && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) {
1376 nr_pages += nr_to_free;
1377 deferred_free_range(free_base_page,
1378 free_base_pfn, nr_to_free);
1379 free_base_page = NULL;
1380 free_base_pfn = nr_to_free = 0;
1382 page = pfn_to_page(pfn);
1387 VM_BUG_ON(page_zone(page) != zone);
1391 __init_single_page(page, pfn, zid, nid);
1392 if (!free_base_page) {
1393 free_base_page = page;
1394 free_base_pfn = pfn;
1399 /* Where possible, batch up pages for a single free */
1402 /* Free the current block of pages to allocator */
1403 nr_pages += nr_to_free;
1404 deferred_free_range(free_base_page, free_base_pfn,
1406 free_base_page = NULL;
1407 free_base_pfn = nr_to_free = 0;
1410 first_init_pfn = max(end_pfn, first_init_pfn);
1413 /* Sanity check that the next zone really is unpopulated */
1414 WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1416 pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
1417 jiffies_to_msecs(jiffies - start));
1419 pgdat_init_report_one_done();
1422 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1424 void __init page_alloc_init_late(void)
1428 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1431 /* There will be num_node_state(N_MEMORY) threads */
1432 atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
1433 for_each_node_state(nid, N_MEMORY) {
1434 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1437 /* Block until all are initialised */
1438 wait_for_completion(&pgdat_init_all_done_comp);
1440 /* Reinit limits that are based on free pages after the kernel is up */
1441 files_maxfiles_init();
1444 for_each_populated_zone(zone)
1445 set_zone_contiguous(zone);
1449 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1450 void __init init_cma_reserved_pageblock(struct page *page)
1452 unsigned i = pageblock_nr_pages;
1453 struct page *p = page;
1456 __ClearPageReserved(p);
1457 set_page_count(p, 0);
1460 set_pageblock_migratetype(page, MIGRATE_CMA);
1462 if (pageblock_order >= MAX_ORDER) {
1463 i = pageblock_nr_pages;
1466 set_page_refcounted(p);
1467 __free_pages(p, MAX_ORDER - 1);
1468 p += MAX_ORDER_NR_PAGES;
1469 } while (i -= MAX_ORDER_NR_PAGES);
1471 set_page_refcounted(page);
1472 __free_pages(page, pageblock_order);
1475 adjust_managed_page_count(page, pageblock_nr_pages);
1480 * The order of subdivision here is critical for the IO subsystem.
1481 * Please do not alter this order without good reasons and regression
1482 * testing. Specifically, as large blocks of memory are subdivided,
1483 * the order in which smaller blocks are delivered depends on the order
1484 * they're subdivided in this function. This is the primary factor
1485 * influencing the order in which pages are delivered to the IO
1486 * subsystem according to empirical testing, and this is also justified
1487 * by considering the behavior of a buddy system containing a single
1488 * large block of memory acted on by a series of small allocations.
1489 * This behavior is a critical factor in sglist merging's success.
1493 static inline void expand(struct zone *zone, struct page *page,
1494 int low, int high, struct free_area *area,
1497 unsigned long size = 1 << high;
1499 while (high > low) {
1503 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
1505 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) &&
1506 debug_guardpage_enabled() &&
1507 high < debug_guardpage_minorder()) {
1509 * Mark as guard pages (or page), that will allow to
1510 * merge back to allocator when buddy will be freed.
1511 * Corresponding page table entries will not be touched,
1512 * pages will stay not present in virtual address space
1514 set_page_guard(zone, &page[size], high, migratetype);
1517 list_add(&page[size].lru, &area->free_list[migratetype]);
1519 set_page_order(&page[size], high);
1524 * This page is about to be returned from the page allocator
1526 static inline int check_new_page(struct page *page)
1528 const char *bad_reason;
1529 unsigned long bad_flags;
1531 if (page_expected_state(page, PAGE_FLAGS_CHECK_AT_PREP|__PG_HWPOISON))
1536 if (unlikely(atomic_read(&page->_mapcount) != -1))
1537 bad_reason = "nonzero mapcount";
1538 if (unlikely(page->mapping != NULL))
1539 bad_reason = "non-NULL mapping";
1540 if (unlikely(page_ref_count(page) != 0))
1541 bad_reason = "nonzero _count";
1542 if (unlikely(page->flags & __PG_HWPOISON)) {
1543 bad_reason = "HWPoisoned (hardware-corrupted)";
1544 bad_flags = __PG_HWPOISON;
1546 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1547 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1548 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1551 if (unlikely(page->mem_cgroup))
1552 bad_reason = "page still charged to cgroup";
1554 if (unlikely(bad_reason)) {
1555 bad_page(page, bad_reason, bad_flags);
1561 static inline bool free_pages_prezeroed(bool poisoned)
1563 return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) &&
1564 page_poisoning_enabled() && poisoned;
1567 static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
1568 unsigned int alloc_flags)
1571 bool poisoned = true;
1573 for (i = 0; i < (1 << order); i++) {
1574 struct page *p = page + i;
1575 if (unlikely(check_new_page(p)))
1578 poisoned &= page_is_poisoned(p);
1581 set_page_private(page, 0);
1582 set_page_refcounted(page);
1584 arch_alloc_page(page, order);
1585 kernel_map_pages(page, 1 << order, 1);
1586 kernel_poison_pages(page, 1 << order, 1);
1587 kasan_alloc_pages(page, order);
1589 if (!free_pages_prezeroed(poisoned) && (gfp_flags & __GFP_ZERO))
1590 for (i = 0; i < (1 << order); i++)
1591 clear_highpage(page + i);
1593 if (order && (gfp_flags & __GFP_COMP))
1594 prep_compound_page(page, order);
1596 set_page_owner(page, order, gfp_flags);
1599 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1600 * allocate the page. The expectation is that the caller is taking
1601 * steps that will free more memory. The caller should avoid the page
1602 * being used for !PFMEMALLOC purposes.
1604 if (alloc_flags & ALLOC_NO_WATERMARKS)
1605 set_page_pfmemalloc(page);
1607 clear_page_pfmemalloc(page);
1613 * Go through the free lists for the given migratetype and remove
1614 * the smallest available page from the freelists
1617 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
1620 unsigned int current_order;
1621 struct free_area *area;
1624 /* Find a page of the appropriate size in the preferred list */
1625 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1626 area = &(zone->free_area[current_order]);
1627 page = list_first_entry_or_null(&area->free_list[migratetype],
1631 list_del(&page->lru);
1632 rmv_page_order(page);
1634 expand(zone, page, order, current_order, area, migratetype);
1635 set_pcppage_migratetype(page, migratetype);
1644 * This array describes the order lists are fallen back to when
1645 * the free lists for the desirable migrate type are depleted
1647 static int fallbacks[MIGRATE_TYPES][4] = {
1648 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
1649 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
1650 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
1652 [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */
1654 #ifdef CONFIG_MEMORY_ISOLATION
1655 [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */
1660 static struct page *__rmqueue_cma_fallback(struct zone *zone,
1663 return __rmqueue_smallest(zone, order, MIGRATE_CMA);
1666 static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1667 unsigned int order) { return NULL; }
1671 * Move the free pages in a range to the free lists of the requested type.
1672 * Note that start_page and end_pages are not aligned on a pageblock
1673 * boundary. If alignment is required, use move_freepages_block()
1675 int move_freepages(struct zone *zone,
1676 struct page *start_page, struct page *end_page,
1681 int pages_moved = 0;
1683 #ifndef CONFIG_HOLES_IN_ZONE
1685 * page_zone is not safe to call in this context when
1686 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1687 * anyway as we check zone boundaries in move_freepages_block().
1688 * Remove at a later date when no bug reports exist related to
1689 * grouping pages by mobility
1691 VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
1694 for (page = start_page; page <= end_page;) {
1695 /* Make sure we are not inadvertently changing nodes */
1696 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
1698 if (!pfn_valid_within(page_to_pfn(page))) {
1703 if (!PageBuddy(page)) {
1708 order = page_order(page);
1709 list_move(&page->lru,
1710 &zone->free_area[order].free_list[migratetype]);
1712 pages_moved += 1 << order;
1718 int move_freepages_block(struct zone *zone, struct page *page,
1721 unsigned long start_pfn, end_pfn;
1722 struct page *start_page, *end_page;
1724 start_pfn = page_to_pfn(page);
1725 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
1726 start_page = pfn_to_page(start_pfn);
1727 end_page = start_page + pageblock_nr_pages - 1;
1728 end_pfn = start_pfn + pageblock_nr_pages - 1;
1730 /* Do not cross zone boundaries */
1731 if (!zone_spans_pfn(zone, start_pfn))
1733 if (!zone_spans_pfn(zone, end_pfn))
1736 return move_freepages(zone, start_page, end_page, migratetype);
1739 static void change_pageblock_range(struct page *pageblock_page,
1740 int start_order, int migratetype)
1742 int nr_pageblocks = 1 << (start_order - pageblock_order);
1744 while (nr_pageblocks--) {
1745 set_pageblock_migratetype(pageblock_page, migratetype);
1746 pageblock_page += pageblock_nr_pages;
1751 * When we are falling back to another migratetype during allocation, try to
1752 * steal extra free pages from the same pageblocks to satisfy further
1753 * allocations, instead of polluting multiple pageblocks.
1755 * If we are stealing a relatively large buddy page, it is likely there will
1756 * be more free pages in the pageblock, so try to steal them all. For
1757 * reclaimable and unmovable allocations, we steal regardless of page size,
1758 * as fragmentation caused by those allocations polluting movable pageblocks
1759 * is worse than movable allocations stealing from unmovable and reclaimable
1762 static bool can_steal_fallback(unsigned int order, int start_mt)
1765 * Leaving this order check is intended, although there is
1766 * relaxed order check in next check. The reason is that
1767 * we can actually steal whole pageblock if this condition met,
1768 * but, below check doesn't guarantee it and that is just heuristic
1769 * so could be changed anytime.
1771 if (order >= pageblock_order)
1774 if (order >= pageblock_order / 2 ||
1775 start_mt == MIGRATE_RECLAIMABLE ||
1776 start_mt == MIGRATE_UNMOVABLE ||
1777 page_group_by_mobility_disabled)
1784 * This function implements actual steal behaviour. If order is large enough,
1785 * we can steal whole pageblock. If not, we first move freepages in this
1786 * pageblock and check whether half of pages are moved or not. If half of
1787 * pages are moved, we can change migratetype of pageblock and permanently
1788 * use it's pages as requested migratetype in the future.
1790 static void steal_suitable_fallback(struct zone *zone, struct page *page,
1793 unsigned int current_order = page_order(page);
1796 /* Take ownership for orders >= pageblock_order */
1797 if (current_order >= pageblock_order) {
1798 change_pageblock_range(page, current_order, start_type);
1802 pages = move_freepages_block(zone, page, start_type);
1804 /* Claim the whole block if over half of it is free */
1805 if (pages >= (1 << (pageblock_order-1)) ||
1806 page_group_by_mobility_disabled)
1807 set_pageblock_migratetype(page, start_type);
1811 * Check whether there is a suitable fallback freepage with requested order.
1812 * If only_stealable is true, this function returns fallback_mt only if
1813 * we can steal other freepages all together. This would help to reduce
1814 * fragmentation due to mixed migratetype pages in one pageblock.
1816 int find_suitable_fallback(struct free_area *area, unsigned int order,
1817 int migratetype, bool only_stealable, bool *can_steal)
1822 if (area->nr_free == 0)
1827 fallback_mt = fallbacks[migratetype][i];
1828 if (fallback_mt == MIGRATE_TYPES)
1831 if (list_empty(&area->free_list[fallback_mt]))
1834 if (can_steal_fallback(order, migratetype))
1837 if (!only_stealable)
1848 * Reserve a pageblock for exclusive use of high-order atomic allocations if
1849 * there are no empty page blocks that contain a page with a suitable order
1851 static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
1852 unsigned int alloc_order)
1855 unsigned long max_managed, flags;
1858 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
1859 * Check is race-prone but harmless.
1861 max_managed = (zone->managed_pages / 100) + pageblock_nr_pages;
1862 if (zone->nr_reserved_highatomic >= max_managed)
1865 spin_lock_irqsave(&zone->lock, flags);
1867 /* Recheck the nr_reserved_highatomic limit under the lock */
1868 if (zone->nr_reserved_highatomic >= max_managed)
1872 mt = get_pageblock_migratetype(page);
1873 if (mt != MIGRATE_HIGHATOMIC &&
1874 !is_migrate_isolate(mt) && !is_migrate_cma(mt)) {
1875 zone->nr_reserved_highatomic += pageblock_nr_pages;
1876 set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
1877 move_freepages_block(zone, page, MIGRATE_HIGHATOMIC);
1881 spin_unlock_irqrestore(&zone->lock, flags);
1885 * Used when an allocation is about to fail under memory pressure. This
1886 * potentially hurts the reliability of high-order allocations when under
1887 * intense memory pressure but failed atomic allocations should be easier
1888 * to recover from than an OOM.
1890 static void unreserve_highatomic_pageblock(const struct alloc_context *ac)
1892 struct zonelist *zonelist = ac->zonelist;
1893 unsigned long flags;
1899 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
1901 /* Preserve at least one pageblock */
1902 if (zone->nr_reserved_highatomic <= pageblock_nr_pages)
1905 spin_lock_irqsave(&zone->lock, flags);
1906 for (order = 0; order < MAX_ORDER; order++) {
1907 struct free_area *area = &(zone->free_area[order]);
1909 page = list_first_entry_or_null(
1910 &area->free_list[MIGRATE_HIGHATOMIC],
1916 * It should never happen but changes to locking could
1917 * inadvertently allow a per-cpu drain to add pages
1918 * to MIGRATE_HIGHATOMIC while unreserving so be safe
1919 * and watch for underflows.
1921 zone->nr_reserved_highatomic -= min(pageblock_nr_pages,
1922 zone->nr_reserved_highatomic);
1925 * Convert to ac->migratetype and avoid the normal
1926 * pageblock stealing heuristics. Minimally, the caller
1927 * is doing the work and needs the pages. More
1928 * importantly, if the block was always converted to
1929 * MIGRATE_UNMOVABLE or another type then the number
1930 * of pageblocks that cannot be completely freed
1933 set_pageblock_migratetype(page, ac->migratetype);
1934 move_freepages_block(zone, page, ac->migratetype);
1935 spin_unlock_irqrestore(&zone->lock, flags);
1938 spin_unlock_irqrestore(&zone->lock, flags);
1942 /* Remove an element from the buddy allocator from the fallback list */
1943 static inline struct page *
1944 __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
1946 struct free_area *area;
1947 unsigned int current_order;
1952 /* Find the largest possible block of pages in the other list */
1953 for (current_order = MAX_ORDER-1;
1954 current_order >= order && current_order <= MAX_ORDER-1;
1956 area = &(zone->free_area[current_order]);
1957 fallback_mt = find_suitable_fallback(area, current_order,
1958 start_migratetype, false, &can_steal);
1959 if (fallback_mt == -1)
1962 page = list_first_entry(&area->free_list[fallback_mt],
1965 steal_suitable_fallback(zone, page, start_migratetype);
1967 /* Remove the page from the freelists */
1969 list_del(&page->lru);
1970 rmv_page_order(page);
1972 expand(zone, page, order, current_order, area,
1975 * The pcppage_migratetype may differ from pageblock's
1976 * migratetype depending on the decisions in
1977 * find_suitable_fallback(). This is OK as long as it does not
1978 * differ for MIGRATE_CMA pageblocks. Those can be used as
1979 * fallback only via special __rmqueue_cma_fallback() function
1981 set_pcppage_migratetype(page, start_migratetype);
1983 trace_mm_page_alloc_extfrag(page, order, current_order,
1984 start_migratetype, fallback_mt);
1993 * Do the hard work of removing an element from the buddy allocator.
1994 * Call me with the zone->lock already held.
1996 static struct page *__rmqueue(struct zone *zone, unsigned int order,
2001 page = __rmqueue_smallest(zone, order, migratetype);
2002 if (unlikely(!page)) {
2003 if (migratetype == MIGRATE_MOVABLE)
2004 page = __rmqueue_cma_fallback(zone, order);
2007 page = __rmqueue_fallback(zone, order, migratetype);
2010 trace_mm_page_alloc_zone_locked(page, order, migratetype);
2015 * Obtain a specified number of elements from the buddy allocator, all under
2016 * a single hold of the lock, for efficiency. Add them to the supplied list.
2017 * Returns the number of new pages which were placed at *list.
2019 static int rmqueue_bulk(struct zone *zone, unsigned int order,
2020 unsigned long count, struct list_head *list,
2021 int migratetype, bool cold)
2025 spin_lock(&zone->lock);
2026 for (i = 0; i < count; ++i) {
2027 struct page *page = __rmqueue(zone, order, migratetype);
2028 if (unlikely(page == NULL))
2032 * Split buddy pages returned by expand() are received here
2033 * in physical page order. The page is added to the callers and
2034 * list and the list head then moves forward. From the callers
2035 * perspective, the linked list is ordered by page number in
2036 * some conditions. This is useful for IO devices that can
2037 * merge IO requests if the physical pages are ordered
2041 list_add(&page->lru, list);
2043 list_add_tail(&page->lru, list);
2045 if (is_migrate_cma(get_pcppage_migratetype(page)))
2046 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
2049 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
2050 spin_unlock(&zone->lock);
2056 * Called from the vmstat counter updater to drain pagesets of this
2057 * currently executing processor on remote nodes after they have
2060 * Note that this function must be called with the thread pinned to
2061 * a single processor.
2063 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
2065 unsigned long flags;
2066 int to_drain, batch;
2068 local_irq_save(flags);
2069 batch = READ_ONCE(pcp->batch);
2070 to_drain = min(pcp->count, batch);
2072 free_pcppages_bulk(zone, to_drain, pcp);
2073 pcp->count -= to_drain;
2075 local_irq_restore(flags);
2080 * Drain pcplists of the indicated processor and zone.
2082 * The processor must either be the current processor and the
2083 * thread pinned to the current processor or a processor that
2086 static void drain_pages_zone(unsigned int cpu, struct zone *zone)
2088 unsigned long flags;
2089 struct per_cpu_pageset *pset;
2090 struct per_cpu_pages *pcp;
2092 local_irq_save(flags);
2093 pset = per_cpu_ptr(zone->pageset, cpu);
2097 free_pcppages_bulk(zone, pcp->count, pcp);
2100 local_irq_restore(flags);
2104 * Drain pcplists of all zones on the indicated processor.
2106 * The processor must either be the current processor and the
2107 * thread pinned to the current processor or a processor that
2110 static void drain_pages(unsigned int cpu)
2114 for_each_populated_zone(zone) {
2115 drain_pages_zone(cpu, zone);
2120 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
2122 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2123 * the single zone's pages.
2125 void drain_local_pages(struct zone *zone)
2127 int cpu = smp_processor_id();
2130 drain_pages_zone(cpu, zone);
2136 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2138 * When zone parameter is non-NULL, spill just the single zone's pages.
2140 * Note that this code is protected against sending an IPI to an offline
2141 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
2142 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
2143 * nothing keeps CPUs from showing up after we populated the cpumask and
2144 * before the call to on_each_cpu_mask().
2146 void drain_all_pages(struct zone *zone)
2151 * Allocate in the BSS so we wont require allocation in
2152 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2154 static cpumask_t cpus_with_pcps;
2157 * We don't care about racing with CPU hotplug event
2158 * as offline notification will cause the notified
2159 * cpu to drain that CPU pcps and on_each_cpu_mask
2160 * disables preemption as part of its processing
2162 for_each_online_cpu(cpu) {
2163 struct per_cpu_pageset *pcp;
2165 bool has_pcps = false;
2168 pcp = per_cpu_ptr(zone->pageset, cpu);
2172 for_each_populated_zone(z) {
2173 pcp = per_cpu_ptr(z->pageset, cpu);
2174 if (pcp->pcp.count) {
2182 cpumask_set_cpu(cpu, &cpus_with_pcps);
2184 cpumask_clear_cpu(cpu, &cpus_with_pcps);
2186 on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages,
2190 #ifdef CONFIG_HIBERNATION
2192 void mark_free_pages(struct zone *zone)
2194 unsigned long pfn, max_zone_pfn;
2195 unsigned long flags;
2196 unsigned int order, t;
2199 if (zone_is_empty(zone))
2202 spin_lock_irqsave(&zone->lock, flags);
2204 max_zone_pfn = zone_end_pfn(zone);
2205 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
2206 if (pfn_valid(pfn)) {
2207 page = pfn_to_page(pfn);
2209 if (page_zone(page) != zone)
2212 if (!swsusp_page_is_forbidden(page))
2213 swsusp_unset_page_free(page);
2216 for_each_migratetype_order(order, t) {
2217 list_for_each_entry(page,
2218 &zone->free_area[order].free_list[t], lru) {
2221 pfn = page_to_pfn(page);
2222 for (i = 0; i < (1UL << order); i++)
2223 swsusp_set_page_free(pfn_to_page(pfn + i));
2226 spin_unlock_irqrestore(&zone->lock, flags);
2228 #endif /* CONFIG_PM */
2231 * Free a 0-order page
2232 * cold == true ? free a cold page : free a hot page
2234 void free_hot_cold_page(struct page *page, bool cold)
2236 struct zone *zone = page_zone(page);
2237 struct per_cpu_pages *pcp;
2238 unsigned long flags;
2239 unsigned long pfn = page_to_pfn(page);
2242 if (!free_pages_prepare(page, 0))
2245 migratetype = get_pfnblock_migratetype(page, pfn);
2246 set_pcppage_migratetype(page, migratetype);
2247 local_irq_save(flags);
2248 __count_vm_event(PGFREE);
2251 * We only track unmovable, reclaimable and movable on pcp lists.
2252 * Free ISOLATE pages back to the allocator because they are being
2253 * offlined but treat RESERVE as movable pages so we can get those
2254 * areas back if necessary. Otherwise, we may have to free
2255 * excessively into the page allocator
2257 if (migratetype >= MIGRATE_PCPTYPES) {
2258 if (unlikely(is_migrate_isolate(migratetype))) {
2259 free_one_page(zone, page, pfn, 0, migratetype);
2262 migratetype = MIGRATE_MOVABLE;
2265 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2267 list_add(&page->lru, &pcp->lists[migratetype]);
2269 list_add_tail(&page->lru, &pcp->lists[migratetype]);
2271 if (pcp->count >= pcp->high) {
2272 unsigned long batch = READ_ONCE(pcp->batch);
2273 free_pcppages_bulk(zone, batch, pcp);
2274 pcp->count -= batch;
2278 local_irq_restore(flags);
2282 * Free a list of 0-order pages
2284 void free_hot_cold_page_list(struct list_head *list, bool cold)
2286 struct page *page, *next;
2288 list_for_each_entry_safe(page, next, list, lru) {
2289 trace_mm_page_free_batched(page, cold);
2290 free_hot_cold_page(page, cold);
2295 * split_page takes a non-compound higher-order page, and splits it into
2296 * n (1<<order) sub-pages: page[0..n]
2297 * Each sub-page must be freed individually.
2299 * Note: this is probably too low level an operation for use in drivers.
2300 * Please consult with lkml before using this in your driver.
2302 void split_page(struct page *page, unsigned int order)
2307 VM_BUG_ON_PAGE(PageCompound(page), page);
2308 VM_BUG_ON_PAGE(!page_count(page), page);
2310 #ifdef CONFIG_KMEMCHECK
2312 * Split shadow pages too, because free(page[0]) would
2313 * otherwise free the whole shadow.
2315 if (kmemcheck_page_is_tracked(page))
2316 split_page(virt_to_page(page[0].shadow), order);
2319 gfp_mask = get_page_owner_gfp(page);
2320 set_page_owner(page, 0, gfp_mask);
2321 for (i = 1; i < (1 << order); i++) {
2322 set_page_refcounted(page + i);
2323 set_page_owner(page + i, 0, gfp_mask);
2326 EXPORT_SYMBOL_GPL(split_page);
2328 int __isolate_free_page(struct page *page, unsigned int order)
2330 unsigned long watermark;
2334 BUG_ON(!PageBuddy(page));
2336 zone = page_zone(page);
2337 mt = get_pageblock_migratetype(page);
2339 if (!is_migrate_isolate(mt)) {
2340 /* Obey watermarks as if the page was being allocated */
2341 watermark = low_wmark_pages(zone) + (1 << order);
2342 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
2345 __mod_zone_freepage_state(zone, -(1UL << order), mt);
2348 /* Remove page from free list */
2349 list_del(&page->lru);
2350 zone->free_area[order].nr_free--;
2351 rmv_page_order(page);
2353 set_page_owner(page, order, __GFP_MOVABLE);
2355 /* Set the pageblock if the isolated page is at least a pageblock */
2356 if (order >= pageblock_order - 1) {
2357 struct page *endpage = page + (1 << order) - 1;
2358 for (; page < endpage; page += pageblock_nr_pages) {
2359 int mt = get_pageblock_migratetype(page);
2360 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
2361 set_pageblock_migratetype(page,
2367 return 1UL << order;
2371 * Similar to split_page except the page is already free. As this is only
2372 * being used for migration, the migratetype of the block also changes.
2373 * As this is called with interrupts disabled, the caller is responsible
2374 * for calling arch_alloc_page() and kernel_map_page() after interrupts
2377 * Note: this is probably too low level an operation for use in drivers.
2378 * Please consult with lkml before using this in your driver.
2380 int split_free_page(struct page *page)
2385 order = page_order(page);
2387 nr_pages = __isolate_free_page(page, order);
2391 /* Split into individual pages */
2392 set_page_refcounted(page);
2393 split_page(page, order);
2398 * Update NUMA hit/miss statistics
2400 * Must be called with interrupts disabled.
2402 * When __GFP_OTHER_NODE is set assume the node of the preferred
2403 * zone is the local node. This is useful for daemons who allocate
2404 * memory on behalf of other processes.
2406 static inline void zone_statistics(struct zone *preferred_zone, struct zone *z,
2410 int local_nid = numa_node_id();
2411 enum zone_stat_item local_stat = NUMA_LOCAL;
2413 if (unlikely(flags & __GFP_OTHER_NODE)) {
2414 local_stat = NUMA_OTHER;
2415 local_nid = preferred_zone->node;
2418 if (z->node == local_nid) {
2419 __inc_zone_state(z, NUMA_HIT);
2420 __inc_zone_state(z, local_stat);
2422 __inc_zone_state(z, NUMA_MISS);
2423 __inc_zone_state(preferred_zone, NUMA_FOREIGN);
2429 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2432 struct page *buffered_rmqueue(struct zone *preferred_zone,
2433 struct zone *zone, unsigned int order,
2434 gfp_t gfp_flags, unsigned int alloc_flags,
2437 unsigned long flags;
2439 bool cold = ((gfp_flags & __GFP_COLD) != 0);
2441 if (likely(order == 0)) {
2442 struct per_cpu_pages *pcp;
2443 struct list_head *list;
2445 local_irq_save(flags);
2446 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2447 list = &pcp->lists[migratetype];
2448 if (list_empty(list)) {
2449 pcp->count += rmqueue_bulk(zone, 0,
2452 if (unlikely(list_empty(list)))
2457 page = list_last_entry(list, struct page, lru);
2459 page = list_first_entry(list, struct page, lru);
2461 __dec_zone_state(zone, NR_ALLOC_BATCH);
2462 list_del(&page->lru);
2466 * We most definitely don't want callers attempting to
2467 * allocate greater than order-1 page units with __GFP_NOFAIL.
2469 WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
2470 spin_lock_irqsave(&zone->lock, flags);
2473 if (alloc_flags & ALLOC_HARDER) {
2474 page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
2476 trace_mm_page_alloc_zone_locked(page, order, migratetype);
2479 page = __rmqueue(zone, order, migratetype);
2480 spin_unlock(&zone->lock);
2483 __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
2484 __mod_zone_freepage_state(zone, -(1 << order),
2485 get_pcppage_migratetype(page));
2488 if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 &&
2489 !test_bit(ZONE_FAIR_DEPLETED, &zone->flags))
2490 set_bit(ZONE_FAIR_DEPLETED, &zone->flags);
2492 __count_zone_vm_events(PGALLOC, zone, 1 << order);
2493 zone_statistics(preferred_zone, zone, gfp_flags);
2494 local_irq_restore(flags);
2496 VM_BUG_ON_PAGE(bad_range(zone, page), page);
2500 local_irq_restore(flags);
2504 #ifdef CONFIG_FAIL_PAGE_ALLOC
2507 struct fault_attr attr;
2509 bool ignore_gfp_highmem;
2510 bool ignore_gfp_reclaim;
2512 } fail_page_alloc = {
2513 .attr = FAULT_ATTR_INITIALIZER,
2514 .ignore_gfp_reclaim = true,
2515 .ignore_gfp_highmem = true,
2519 static int __init setup_fail_page_alloc(char *str)
2521 return setup_fault_attr(&fail_page_alloc.attr, str);
2523 __setup("fail_page_alloc=", setup_fail_page_alloc);
2525 static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
2527 if (order < fail_page_alloc.min_order)
2529 if (gfp_mask & __GFP_NOFAIL)
2531 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
2533 if (fail_page_alloc.ignore_gfp_reclaim &&
2534 (gfp_mask & __GFP_DIRECT_RECLAIM))
2537 return should_fail(&fail_page_alloc.attr, 1 << order);
2540 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2542 static int __init fail_page_alloc_debugfs(void)
2544 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
2547 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
2548 &fail_page_alloc.attr);
2550 return PTR_ERR(dir);
2552 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
2553 &fail_page_alloc.ignore_gfp_reclaim))
2555 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
2556 &fail_page_alloc.ignore_gfp_highmem))
2558 if (!debugfs_create_u32("min-order", mode, dir,
2559 &fail_page_alloc.min_order))
2564 debugfs_remove_recursive(dir);
2569 late_initcall(fail_page_alloc_debugfs);
2571 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2573 #else /* CONFIG_FAIL_PAGE_ALLOC */
2575 static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
2580 #endif /* CONFIG_FAIL_PAGE_ALLOC */
2583 * Return true if free base pages are above 'mark'. For high-order checks it
2584 * will return true of the order-0 watermark is reached and there is at least
2585 * one free page of a suitable size. Checking now avoids taking the zone lock
2586 * to check in the allocation paths if no pages are free.
2588 static bool __zone_watermark_ok(struct zone *z, unsigned int order,
2589 unsigned long mark, int classzone_idx,
2590 unsigned int alloc_flags,
2595 const bool alloc_harder = (alloc_flags & ALLOC_HARDER);
2597 /* free_pages may go negative - that's OK */
2598 free_pages -= (1 << order) - 1;
2600 if (alloc_flags & ALLOC_HIGH)
2604 * If the caller does not have rights to ALLOC_HARDER then subtract
2605 * the high-atomic reserves. This will over-estimate the size of the
2606 * atomic reserve but it avoids a search.
2608 if (likely(!alloc_harder))
2609 free_pages -= z->nr_reserved_highatomic;
2614 /* If allocation can't use CMA areas don't use free CMA pages */
2615 if (!(alloc_flags & ALLOC_CMA))
2616 free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES);
2620 * Check watermarks for an order-0 allocation request. If these
2621 * are not met, then a high-order request also cannot go ahead
2622 * even if a suitable page happened to be free.
2624 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
2627 /* If this is an order-0 request then the watermark is fine */
2631 /* For a high-order request, check at least one suitable page is free */
2632 for (o = order; o < MAX_ORDER; o++) {
2633 struct free_area *area = &z->free_area[o];
2642 for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
2643 if (!list_empty(&area->free_list[mt]))
2648 if ((alloc_flags & ALLOC_CMA) &&
2649 !list_empty(&area->free_list[MIGRATE_CMA])) {
2657 bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
2658 int classzone_idx, unsigned int alloc_flags)
2660 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2661 zone_page_state(z, NR_FREE_PAGES));
2664 static inline bool zone_watermark_fast(struct zone *z, unsigned int order,
2665 unsigned long mark, int classzone_idx, unsigned int alloc_flags)
2667 long free_pages = zone_page_state(z, NR_FREE_PAGES);
2671 /* If allocation can't use CMA areas don't use free CMA pages */
2672 if (!(alloc_flags & ALLOC_CMA))
2673 cma_pages = zone_page_state(z, NR_FREE_CMA_PAGES);
2677 * Fast check for order-0 only. If this fails then the reserves
2678 * need to be calculated. There is a corner case where the check
2679 * passes but only the high-order atomic reserve are free. If
2680 * the caller is !atomic then it'll uselessly search the free
2681 * list. That corner case is then slower but it is harmless.
2683 if (!order && (free_pages - cma_pages) > mark + z->lowmem_reserve[classzone_idx])
2686 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2690 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
2691 unsigned long mark, int classzone_idx)
2693 long free_pages = zone_page_state(z, NR_FREE_PAGES);
2695 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
2696 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
2698 return __zone_watermark_ok(z, order, mark, classzone_idx, 0,
2703 static bool zone_local(struct zone *local_zone, struct zone *zone)
2705 return local_zone->node == zone->node;
2708 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2710 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
2713 #else /* CONFIG_NUMA */
2714 static bool zone_local(struct zone *local_zone, struct zone *zone)
2719 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2723 #endif /* CONFIG_NUMA */
2725 static void reset_alloc_batches(struct zone *preferred_zone)
2727 struct zone *zone = preferred_zone->zone_pgdat->node_zones;
2730 mod_zone_page_state(zone, NR_ALLOC_BATCH,
2731 high_wmark_pages(zone) - low_wmark_pages(zone) -
2732 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
2733 clear_bit(ZONE_FAIR_DEPLETED, &zone->flags);
2734 } while (zone++ != preferred_zone);
2738 * get_page_from_freelist goes through the zonelist trying to allocate
2741 static struct page *
2742 get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
2743 const struct alloc_context *ac)
2745 struct zoneref *z = ac->preferred_zoneref;
2747 bool fair_skipped = false;
2748 bool apply_fair = (alloc_flags & ALLOC_FAIR);
2752 * Scan zonelist, looking for a zone with enough free.
2753 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2755 for_next_zone_zonelist_nodemask(zone, z, ac->zonelist, ac->high_zoneidx,
2760 if (cpusets_enabled() &&
2761 (alloc_flags & ALLOC_CPUSET) &&
2762 !cpuset_zone_allowed(zone, gfp_mask))
2765 * Distribute pages in proportion to the individual
2766 * zone size to ensure fair page aging. The zone a
2767 * page was allocated in should have no effect on the
2768 * time the page has in memory before being reclaimed.
2771 if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) {
2772 fair_skipped = true;
2775 if (!zone_local(ac->preferred_zoneref->zone, zone)) {
2782 * When allocating a page cache page for writing, we
2783 * want to get it from a zone that is within its dirty
2784 * limit, such that no single zone holds more than its
2785 * proportional share of globally allowed dirty pages.
2786 * The dirty limits take into account the zone's
2787 * lowmem reserves and high watermark so that kswapd
2788 * should be able to balance it without having to
2789 * write pages from its LRU list.
2791 * This may look like it could increase pressure on
2792 * lower zones by failing allocations in higher zones
2793 * before they are full. But the pages that do spill
2794 * over are limited as the lower zones are protected
2795 * by this very same mechanism. It should not become
2796 * a practical burden to them.
2798 * XXX: For now, allow allocations to potentially
2799 * exceed the per-zone dirty limit in the slowpath
2800 * (spread_dirty_pages unset) before going into reclaim,
2801 * which is important when on a NUMA setup the allowed
2802 * zones are together not big enough to reach the
2803 * global limit. The proper fix for these situations
2804 * will require awareness of zones in the
2805 * dirty-throttling and the flusher threads.
2807 if (ac->spread_dirty_pages && !zone_dirty_ok(zone))
2810 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2811 if (!zone_watermark_fast(zone, order, mark,
2812 ac_classzone_idx(ac), alloc_flags)) {
2815 /* Checked here to keep the fast path fast */
2816 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2817 if (alloc_flags & ALLOC_NO_WATERMARKS)
2820 if (zone_reclaim_mode == 0 ||
2821 !zone_allows_reclaim(ac->preferred_zoneref->zone, zone))
2824 ret = zone_reclaim(zone, gfp_mask, order);
2826 case ZONE_RECLAIM_NOSCAN:
2829 case ZONE_RECLAIM_FULL:
2830 /* scanned but unreclaimable */
2833 /* did we reclaim enough */
2834 if (zone_watermark_ok(zone, order, mark,
2835 ac_classzone_idx(ac), alloc_flags))
2843 page = buffered_rmqueue(ac->preferred_zoneref->zone, zone, order,
2844 gfp_mask, alloc_flags, ac->migratetype);
2846 if (prep_new_page(page, order, gfp_mask, alloc_flags))
2850 * If this is a high-order atomic allocation then check
2851 * if the pageblock should be reserved for the future
2853 if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
2854 reserve_highatomic_pageblock(page, zone, order);
2861 * The first pass makes sure allocations are spread fairly within the
2862 * local node. However, the local node might have free pages left
2863 * after the fairness batches are exhausted, and remote zones haven't
2864 * even been considered yet. Try once more without fairness, and
2865 * include remote zones now, before entering the slowpath and waking
2866 * kswapd: prefer spilling to a remote zone over swapping locally.
2871 fair_skipped = false;
2872 reset_alloc_batches(ac->preferred_zoneref->zone);
2880 * Large machines with many possible nodes should not always dump per-node
2881 * meminfo in irq context.
2883 static inline bool should_suppress_show_mem(void)
2888 ret = in_interrupt();
2893 static DEFINE_RATELIMIT_STATE(nopage_rs,
2894 DEFAULT_RATELIMIT_INTERVAL,
2895 DEFAULT_RATELIMIT_BURST);
2897 void warn_alloc_failed(gfp_t gfp_mask, unsigned int order, const char *fmt, ...)
2899 unsigned int filter = SHOW_MEM_FILTER_NODES;
2901 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2902 debug_guardpage_minorder() > 0)
2906 * This documents exceptions given to allocations in certain
2907 * contexts that are allowed to allocate outside current's set
2910 if (!(gfp_mask & __GFP_NOMEMALLOC))
2911 if (test_thread_flag(TIF_MEMDIE) ||
2912 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2913 filter &= ~SHOW_MEM_FILTER_NODES;
2914 if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
2915 filter &= ~SHOW_MEM_FILTER_NODES;
2918 struct va_format vaf;
2921 va_start(args, fmt);
2926 pr_warn("%pV", &vaf);
2931 pr_warn("%s: page allocation failure: order:%u, mode:%#x(%pGg)\n",
2932 current->comm, order, gfp_mask, &gfp_mask);
2934 if (!should_suppress_show_mem())
2938 static inline struct page *
2939 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
2940 const struct alloc_context *ac, unsigned long *did_some_progress)
2942 struct oom_control oc = {
2943 .zonelist = ac->zonelist,
2944 .nodemask = ac->nodemask,
2945 .gfp_mask = gfp_mask,
2950 *did_some_progress = 0;
2953 * Acquire the oom lock. If that fails, somebody else is
2954 * making progress for us.
2956 if (!mutex_trylock(&oom_lock)) {
2957 *did_some_progress = 1;
2958 schedule_timeout_uninterruptible(1);
2963 * Go through the zonelist yet one more time, keep very high watermark
2964 * here, this is only to catch a parallel oom killing, we must fail if
2965 * we're still under heavy pressure.
2967 page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
2968 ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
2972 if (!(gfp_mask & __GFP_NOFAIL)) {
2973 /* Coredumps can quickly deplete all memory reserves */
2974 if (current->flags & PF_DUMPCORE)
2976 /* The OOM killer will not help higher order allocs */
2977 if (order > PAGE_ALLOC_COSTLY_ORDER)
2979 /* The OOM killer does not needlessly kill tasks for lowmem */
2980 if (ac->high_zoneidx < ZONE_NORMAL)
2982 if (pm_suspended_storage())
2985 * XXX: GFP_NOFS allocations should rather fail than rely on
2986 * other request to make a forward progress.
2987 * We are in an unfortunate situation where out_of_memory cannot
2988 * do much for this context but let's try it to at least get
2989 * access to memory reserved if the current task is killed (see
2990 * out_of_memory). Once filesystems are ready to handle allocation
2991 * failures more gracefully we should just bail out here.
2994 /* The OOM killer may not free memory on a specific node */
2995 if (gfp_mask & __GFP_THISNODE)
2998 /* Exhausted what can be done so it's blamo time */
2999 if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
3000 *did_some_progress = 1;
3002 if (gfp_mask & __GFP_NOFAIL) {
3003 page = get_page_from_freelist(gfp_mask, order,
3004 ALLOC_NO_WATERMARKS|ALLOC_CPUSET, ac);
3006 * fallback to ignore cpuset restriction if our nodes
3010 page = get_page_from_freelist(gfp_mask, order,
3011 ALLOC_NO_WATERMARKS, ac);
3015 mutex_unlock(&oom_lock);
3019 #ifdef CONFIG_COMPACTION
3020 /* Try memory compaction for high-order allocations before reclaim */
3021 static struct page *
3022 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
3023 unsigned int alloc_flags, const struct alloc_context *ac,
3024 enum migrate_mode mode, int *contended_compaction,
3025 bool *deferred_compaction)
3027 unsigned long compact_result;
3033 current->flags |= PF_MEMALLOC;
3034 compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
3035 mode, contended_compaction);
3036 current->flags &= ~PF_MEMALLOC;
3038 switch (compact_result) {
3039 case COMPACT_DEFERRED:
3040 *deferred_compaction = true;
3042 case COMPACT_SKIPPED:
3049 * At least in one zone compaction wasn't deferred or skipped, so let's
3050 * count a compaction stall
3052 count_vm_event(COMPACTSTALL);
3054 page = get_page_from_freelist(gfp_mask, order,
3055 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
3058 struct zone *zone = page_zone(page);
3060 zone->compact_blockskip_flush = false;
3061 compaction_defer_reset(zone, order, true);
3062 count_vm_event(COMPACTSUCCESS);
3067 * It's bad if compaction run occurs and fails. The most likely reason
3068 * is that pages exist, but not enough to satisfy watermarks.
3070 count_vm_event(COMPACTFAIL);
3077 static inline struct page *
3078 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
3079 unsigned int alloc_flags, const struct alloc_context *ac,
3080 enum migrate_mode mode, int *contended_compaction,
3081 bool *deferred_compaction)
3085 #endif /* CONFIG_COMPACTION */
3087 /* Perform direct synchronous page reclaim */
3089 __perform_reclaim(gfp_t gfp_mask, unsigned int order,
3090 const struct alloc_context *ac)
3092 struct reclaim_state reclaim_state;
3097 /* We now go into synchronous reclaim */
3098 cpuset_memory_pressure_bump();
3099 current->flags |= PF_MEMALLOC;
3100 lockdep_set_current_reclaim_state(gfp_mask);
3101 reclaim_state.reclaimed_slab = 0;
3102 current->reclaim_state = &reclaim_state;
3104 progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
3107 current->reclaim_state = NULL;
3108 lockdep_clear_current_reclaim_state();
3109 current->flags &= ~PF_MEMALLOC;
3116 /* The really slow allocator path where we enter direct reclaim */
3117 static inline struct page *
3118 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
3119 unsigned int alloc_flags, const struct alloc_context *ac,
3120 unsigned long *did_some_progress)
3122 struct page *page = NULL;
3123 bool drained = false;
3125 *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
3126 if (unlikely(!(*did_some_progress)))
3130 page = get_page_from_freelist(gfp_mask, order,
3131 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
3134 * If an allocation failed after direct reclaim, it could be because
3135 * pages are pinned on the per-cpu lists or in high alloc reserves.
3136 * Shrink them them and try again
3138 if (!page && !drained) {
3139 unreserve_highatomic_pageblock(ac);
3140 drain_all_pages(NULL);
3148 static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
3153 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
3154 ac->high_zoneidx, ac->nodemask)
3155 wakeup_kswapd(zone, order, ac_classzone_idx(ac));
3158 static inline unsigned int
3159 gfp_to_alloc_flags(gfp_t gfp_mask)
3161 unsigned int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
3163 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
3164 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
3167 * The caller may dip into page reserves a bit more if the caller
3168 * cannot run direct reclaim, or if the caller has realtime scheduling
3169 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
3170 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
3172 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
3174 if (gfp_mask & __GFP_ATOMIC) {
3176 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
3177 * if it can't schedule.
3179 if (!(gfp_mask & __GFP_NOMEMALLOC))
3180 alloc_flags |= ALLOC_HARDER;
3182 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
3183 * comment for __cpuset_node_allowed().
3185 alloc_flags &= ~ALLOC_CPUSET;
3186 } else if (unlikely(rt_task(current)) && !in_interrupt())
3187 alloc_flags |= ALLOC_HARDER;
3189 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
3190 if (gfp_mask & __GFP_MEMALLOC)
3191 alloc_flags |= ALLOC_NO_WATERMARKS;
3192 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
3193 alloc_flags |= ALLOC_NO_WATERMARKS;
3194 else if (!in_interrupt() &&
3195 ((current->flags & PF_MEMALLOC) ||
3196 unlikely(test_thread_flag(TIF_MEMDIE))))
3197 alloc_flags |= ALLOC_NO_WATERMARKS;
3200 if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
3201 alloc_flags |= ALLOC_CMA;
3206 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
3208 return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
3211 static inline bool is_thp_gfp_mask(gfp_t gfp_mask)
3213 return (gfp_mask & (GFP_TRANSHUGE | __GFP_KSWAPD_RECLAIM)) == GFP_TRANSHUGE;
3216 static inline struct page *
3217 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
3218 struct alloc_context *ac)
3220 bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
3221 struct page *page = NULL;
3222 unsigned int alloc_flags;
3223 unsigned long pages_reclaimed = 0;
3224 unsigned long did_some_progress;
3225 enum migrate_mode migration_mode = MIGRATE_ASYNC;
3226 bool deferred_compaction = false;
3227 int contended_compaction = COMPACT_CONTENDED_NONE;
3230 * In the slowpath, we sanity check order to avoid ever trying to
3231 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3232 * be using allocators in order of preference for an area that is
3235 if (order >= MAX_ORDER) {
3236 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
3241 * We also sanity check to catch abuse of atomic reserves being used by
3242 * callers that are not in atomic context.
3244 if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
3245 (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
3246 gfp_mask &= ~__GFP_ATOMIC;
3249 if (gfp_mask & __GFP_KSWAPD_RECLAIM)
3250 wake_all_kswapds(order, ac);
3253 * OK, we're below the kswapd watermark and have kicked background
3254 * reclaim. Now things get more complex, so set up alloc_flags according
3255 * to how we want to proceed.
3257 alloc_flags = gfp_to_alloc_flags(gfp_mask);
3259 /* This is the last chance, in general, before the goto nopage. */
3260 page = get_page_from_freelist(gfp_mask, order,
3261 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
3265 /* Allocate without watermarks if the context allows */
3266 if (alloc_flags & ALLOC_NO_WATERMARKS) {
3268 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3269 * the allocation is high priority and these type of
3270 * allocations are system rather than user orientated
3272 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
3273 page = get_page_from_freelist(gfp_mask, order,
3274 ALLOC_NO_WATERMARKS, ac);
3279 /* Caller is not willing to reclaim, we can't balance anything */
3280 if (!can_direct_reclaim) {
3282 * All existing users of the __GFP_NOFAIL are blockable, so warn
3283 * of any new users that actually allow this type of allocation
3286 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
3290 /* Avoid recursion of direct reclaim */
3291 if (current->flags & PF_MEMALLOC) {
3293 * __GFP_NOFAIL request from this context is rather bizarre
3294 * because we cannot reclaim anything and only can loop waiting
3295 * for somebody to do a work for us.
3297 if (WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
3304 /* Avoid allocations with no watermarks from looping endlessly */
3305 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
3309 * Try direct compaction. The first pass is asynchronous. Subsequent
3310 * attempts after direct reclaim are synchronous
3312 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
3314 &contended_compaction,
3315 &deferred_compaction);
3319 /* Checks for THP-specific high-order allocations */
3320 if (is_thp_gfp_mask(gfp_mask)) {
3322 * If compaction is deferred for high-order allocations, it is
3323 * because sync compaction recently failed. If this is the case
3324 * and the caller requested a THP allocation, we do not want
3325 * to heavily disrupt the system, so we fail the allocation
3326 * instead of entering direct reclaim.
3328 if (deferred_compaction)
3332 * In all zones where compaction was attempted (and not
3333 * deferred or skipped), lock contention has been detected.
3334 * For THP allocation we do not want to disrupt the others
3335 * so we fallback to base pages instead.
3337 if (contended_compaction == COMPACT_CONTENDED_LOCK)
3341 * If compaction was aborted due to need_resched(), we do not
3342 * want to further increase allocation latency, unless it is
3343 * khugepaged trying to collapse.
3345 if (contended_compaction == COMPACT_CONTENDED_SCHED
3346 && !(current->flags & PF_KTHREAD))
3351 * It can become very expensive to allocate transparent hugepages at
3352 * fault, so use asynchronous memory compaction for THP unless it is
3353 * khugepaged trying to collapse.
3355 if (!is_thp_gfp_mask(gfp_mask) || (current->flags & PF_KTHREAD))
3356 migration_mode = MIGRATE_SYNC_LIGHT;
3358 /* Try direct reclaim and then allocating */
3359 page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
3360 &did_some_progress);
3364 /* Do not loop if specifically requested */
3365 if (gfp_mask & __GFP_NORETRY)
3368 /* Keep reclaiming pages as long as there is reasonable progress */
3369 pages_reclaimed += did_some_progress;
3370 if ((did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) ||
3371 ((gfp_mask & __GFP_REPEAT) && pages_reclaimed < (1 << order))) {
3372 /* Wait for some write requests to complete then retry */
3373 wait_iff_congested(ac->preferred_zoneref->zone, BLK_RW_ASYNC, HZ/50);
3377 /* Reclaim has failed us, start killing things */
3378 page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
3382 /* Retry as long as the OOM killer is making progress */
3383 if (did_some_progress)
3388 * High-order allocations do not necessarily loop after
3389 * direct reclaim and reclaim/compaction depends on compaction
3390 * being called after reclaim so call directly if necessary
3392 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags,
3394 &contended_compaction,
3395 &deferred_compaction);
3399 warn_alloc_failed(gfp_mask, order, NULL);
3405 * This is the 'heart' of the zoned buddy allocator.
3408 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
3409 struct zonelist *zonelist, nodemask_t *nodemask)
3412 unsigned int cpuset_mems_cookie;
3413 unsigned int alloc_flags = ALLOC_WMARK_LOW|ALLOC_FAIR;
3414 gfp_t alloc_mask = gfp_mask; /* The gfp_t that was actually used for allocation */
3415 struct alloc_context ac = {
3416 .high_zoneidx = gfp_zone(gfp_mask),
3417 .zonelist = zonelist,
3418 .nodemask = nodemask,
3419 .migratetype = gfpflags_to_migratetype(gfp_mask),
3422 if (cpusets_enabled()) {
3423 alloc_mask |= __GFP_HARDWALL;
3424 alloc_flags |= ALLOC_CPUSET;
3426 ac.nodemask = &cpuset_current_mems_allowed;
3429 gfp_mask &= gfp_allowed_mask;
3431 lockdep_trace_alloc(gfp_mask);
3433 might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
3435 if (should_fail_alloc_page(gfp_mask, order))
3439 * Check the zones suitable for the gfp_mask contain at least one
3440 * valid zone. It's possible to have an empty zonelist as a result
3441 * of __GFP_THISNODE and a memoryless node
3443 if (unlikely(!zonelist->_zonerefs->zone))
3446 if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE)
3447 alloc_flags |= ALLOC_CMA;
3450 cpuset_mems_cookie = read_mems_allowed_begin();
3452 /* Dirty zone balancing only done in the fast path */
3453 ac.spread_dirty_pages = (gfp_mask & __GFP_WRITE);
3455 /* The preferred zone is used for statistics later */
3456 ac.preferred_zoneref = first_zones_zonelist(ac.zonelist,
3457 ac.high_zoneidx, ac.nodemask);
3458 if (!ac.preferred_zoneref) {
3463 /* First allocation attempt */
3464 page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
3469 * Runtime PM, block IO and its error handling path can deadlock
3470 * because I/O on the device might not complete.
3472 alloc_mask = memalloc_noio_flags(gfp_mask);
3473 ac.spread_dirty_pages = false;
3475 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
3479 * When updating a task's mems_allowed, it is possible to race with
3480 * parallel threads in such a way that an allocation can fail while
3481 * the mask is being updated. If a page allocation is about to fail,
3482 * check if the cpuset changed during allocation and if so, retry.
3484 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie))) {
3485 alloc_mask = gfp_mask;
3490 if (kmemcheck_enabled && page)
3491 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
3493 trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
3497 EXPORT_SYMBOL(__alloc_pages_nodemask);
3500 * Common helper functions.
3502 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
3507 * __get_free_pages() returns a 32-bit address, which cannot represent
3510 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
3512 page = alloc_pages(gfp_mask, order);
3515 return (unsigned long) page_address(page);
3517 EXPORT_SYMBOL(__get_free_pages);
3519 unsigned long get_zeroed_page(gfp_t gfp_mask)
3521 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
3523 EXPORT_SYMBOL(get_zeroed_page);
3525 void __free_pages(struct page *page, unsigned int order)
3527 if (put_page_testzero(page)) {
3529 free_hot_cold_page(page, false);
3531 __free_pages_ok(page, order);
3535 EXPORT_SYMBOL(__free_pages);
3537 void free_pages(unsigned long addr, unsigned int order)
3540 VM_BUG_ON(!virt_addr_valid((void *)addr));
3541 __free_pages(virt_to_page((void *)addr), order);
3545 EXPORT_SYMBOL(free_pages);
3549 * An arbitrary-length arbitrary-offset area of memory which resides
3550 * within a 0 or higher order page. Multiple fragments within that page
3551 * are individually refcounted, in the page's reference counter.
3553 * The page_frag functions below provide a simple allocation framework for
3554 * page fragments. This is used by the network stack and network device
3555 * drivers to provide a backing region of memory for use as either an
3556 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3558 static struct page *__page_frag_refill(struct page_frag_cache *nc,
3561 struct page *page = NULL;
3562 gfp_t gfp = gfp_mask;
3564 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3565 gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
3567 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
3568 PAGE_FRAG_CACHE_MAX_ORDER);
3569 nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
3571 if (unlikely(!page))
3572 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
3574 nc->va = page ? page_address(page) : NULL;
3579 void *__alloc_page_frag(struct page_frag_cache *nc,
3580 unsigned int fragsz, gfp_t gfp_mask)
3582 unsigned int size = PAGE_SIZE;
3586 if (unlikely(!nc->va)) {
3588 page = __page_frag_refill(nc, gfp_mask);
3592 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3593 /* if size can vary use size else just use PAGE_SIZE */
3596 /* Even if we own the page, we do not use atomic_set().
3597 * This would break get_page_unless_zero() users.
3599 page_ref_add(page, size - 1);
3601 /* reset page count bias and offset to start of new frag */
3602 nc->pfmemalloc = page_is_pfmemalloc(page);
3603 nc->pagecnt_bias = size;
3607 offset = nc->offset - fragsz;
3608 if (unlikely(offset < 0)) {
3609 page = virt_to_page(nc->va);
3611 if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
3614 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3615 /* if size can vary use size else just use PAGE_SIZE */
3618 /* OK, page count is 0, we can safely set it */
3619 set_page_count(page, size);
3621 /* reset page count bias and offset to start of new frag */
3622 nc->pagecnt_bias = size;
3623 offset = size - fragsz;
3627 nc->offset = offset;
3629 return nc->va + offset;
3631 EXPORT_SYMBOL(__alloc_page_frag);
3634 * Frees a page fragment allocated out of either a compound or order 0 page.
3636 void __free_page_frag(void *addr)
3638 struct page *page = virt_to_head_page(addr);
3640 if (unlikely(put_page_testzero(page)))
3641 __free_pages_ok(page, compound_order(page));
3643 EXPORT_SYMBOL(__free_page_frag);
3646 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
3647 * of the current memory cgroup if __GFP_ACCOUNT is set, other than that it is
3648 * equivalent to alloc_pages.
3650 * It should be used when the caller would like to use kmalloc, but since the
3651 * allocation is large, it has to fall back to the page allocator.
3653 struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order)
3657 page = alloc_pages(gfp_mask, order);
3658 if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) {
3659 __free_pages(page, order);
3665 struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
3669 page = alloc_pages_node(nid, gfp_mask, order);
3670 if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) {
3671 __free_pages(page, order);
3678 * __free_kmem_pages and free_kmem_pages will free pages allocated with
3681 void __free_kmem_pages(struct page *page, unsigned int order)
3683 memcg_kmem_uncharge(page, order);
3684 __free_pages(page, order);
3687 void free_kmem_pages(unsigned long addr, unsigned int order)
3690 VM_BUG_ON(!virt_addr_valid((void *)addr));
3691 __free_kmem_pages(virt_to_page((void *)addr), order);
3695 static void *make_alloc_exact(unsigned long addr, unsigned int order,
3699 unsigned long alloc_end = addr + (PAGE_SIZE << order);
3700 unsigned long used = addr + PAGE_ALIGN(size);
3702 split_page(virt_to_page((void *)addr), order);
3703 while (used < alloc_end) {
3708 return (void *)addr;
3712 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3713 * @size: the number of bytes to allocate
3714 * @gfp_mask: GFP flags for the allocation
3716 * This function is similar to alloc_pages(), except that it allocates the
3717 * minimum number of pages to satisfy the request. alloc_pages() can only
3718 * allocate memory in power-of-two pages.
3720 * This function is also limited by MAX_ORDER.
3722 * Memory allocated by this function must be released by free_pages_exact().
3724 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
3726 unsigned int order = get_order(size);
3729 addr = __get_free_pages(gfp_mask, order);
3730 return make_alloc_exact(addr, order, size);
3732 EXPORT_SYMBOL(alloc_pages_exact);
3735 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3737 * @nid: the preferred node ID where memory should be allocated
3738 * @size: the number of bytes to allocate
3739 * @gfp_mask: GFP flags for the allocation
3741 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3744 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
3746 unsigned int order = get_order(size);
3747 struct page *p = alloc_pages_node(nid, gfp_mask, order);
3750 return make_alloc_exact((unsigned long)page_address(p), order, size);
3754 * free_pages_exact - release memory allocated via alloc_pages_exact()
3755 * @virt: the value returned by alloc_pages_exact.
3756 * @size: size of allocation, same value as passed to alloc_pages_exact().
3758 * Release the memory allocated by a previous call to alloc_pages_exact.
3760 void free_pages_exact(void *virt, size_t size)
3762 unsigned long addr = (unsigned long)virt;
3763 unsigned long end = addr + PAGE_ALIGN(size);
3765 while (addr < end) {
3770 EXPORT_SYMBOL(free_pages_exact);
3773 * nr_free_zone_pages - count number of pages beyond high watermark
3774 * @offset: The zone index of the highest zone
3776 * nr_free_zone_pages() counts the number of counts pages which are beyond the
3777 * high watermark within all zones at or below a given zone index. For each
3778 * zone, the number of pages is calculated as:
3779 * managed_pages - high_pages
3781 static unsigned long nr_free_zone_pages(int offset)
3786 /* Just pick one node, since fallback list is circular */
3787 unsigned long sum = 0;
3789 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
3791 for_each_zone_zonelist(zone, z, zonelist, offset) {
3792 unsigned long size = zone->managed_pages;
3793 unsigned long high = high_wmark_pages(zone);
3802 * nr_free_buffer_pages - count number of pages beyond high watermark
3804 * nr_free_buffer_pages() counts the number of pages which are beyond the high
3805 * watermark within ZONE_DMA and ZONE_NORMAL.
3807 unsigned long nr_free_buffer_pages(void)
3809 return nr_free_zone_pages(gfp_zone(GFP_USER));
3811 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
3814 * nr_free_pagecache_pages - count number of pages beyond high watermark
3816 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3817 * high watermark within all zones.
3819 unsigned long nr_free_pagecache_pages(void)
3821 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
3824 static inline void show_node(struct zone *zone)
3826 if (IS_ENABLED(CONFIG_NUMA))
3827 printk("Node %d ", zone_to_nid(zone));
3830 long si_mem_available(void)
3833 unsigned long pagecache;
3834 unsigned long wmark_low = 0;
3835 unsigned long pages[NR_LRU_LISTS];
3839 for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++)
3840 pages[lru] = global_page_state(NR_LRU_BASE + lru);
3843 wmark_low += zone->watermark[WMARK_LOW];
3846 * Estimate the amount of memory available for userspace allocations,
3847 * without causing swapping.
3849 available = global_page_state(NR_FREE_PAGES) - totalreserve_pages;
3852 * Not all the page cache can be freed, otherwise the system will
3853 * start swapping. Assume at least half of the page cache, or the
3854 * low watermark worth of cache, needs to stay.
3856 pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE];
3857 pagecache -= min(pagecache / 2, wmark_low);
3858 available += pagecache;
3861 * Part of the reclaimable slab consists of items that are in use,
3862 * and cannot be freed. Cap this estimate at the low watermark.
3864 available += global_page_state(NR_SLAB_RECLAIMABLE) -
3865 min(global_page_state(NR_SLAB_RECLAIMABLE) / 2, wmark_low);
3871 EXPORT_SYMBOL_GPL(si_mem_available);
3873 void si_meminfo(struct sysinfo *val)
3875 val->totalram = totalram_pages;
3876 val->sharedram = global_page_state(NR_SHMEM);
3877 val->freeram = global_page_state(NR_FREE_PAGES);
3878 val->bufferram = nr_blockdev_pages();
3879 val->totalhigh = totalhigh_pages;
3880 val->freehigh = nr_free_highpages();
3881 val->mem_unit = PAGE_SIZE;
3884 EXPORT_SYMBOL(si_meminfo);
3887 void si_meminfo_node(struct sysinfo *val, int nid)
3889 int zone_type; /* needs to be signed */
3890 unsigned long managed_pages = 0;
3891 unsigned long managed_highpages = 0;
3892 unsigned long free_highpages = 0;
3893 pg_data_t *pgdat = NODE_DATA(nid);
3895 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
3896 managed_pages += pgdat->node_zones[zone_type].managed_pages;
3897 val->totalram = managed_pages;
3898 val->sharedram = node_page_state(nid, NR_SHMEM);
3899 val->freeram = node_page_state(nid, NR_FREE_PAGES);
3900 #ifdef CONFIG_HIGHMEM
3901 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3902 struct zone *zone = &pgdat->node_zones[zone_type];
3904 if (is_highmem(zone)) {
3905 managed_highpages += zone->managed_pages;
3906 free_highpages += zone_page_state(zone, NR_FREE_PAGES);
3909 val->totalhigh = managed_highpages;
3910 val->freehigh = free_highpages;
3912 val->totalhigh = managed_highpages;
3913 val->freehigh = free_highpages;
3915 val->mem_unit = PAGE_SIZE;
3920 * Determine whether the node should be displayed or not, depending on whether
3921 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
3923 bool skip_free_areas_node(unsigned int flags, int nid)
3926 unsigned int cpuset_mems_cookie;
3928 if (!(flags & SHOW_MEM_FILTER_NODES))
3932 cpuset_mems_cookie = read_mems_allowed_begin();
3933 ret = !node_isset(nid, cpuset_current_mems_allowed);
3934 } while (read_mems_allowed_retry(cpuset_mems_cookie));
3939 #define K(x) ((x) << (PAGE_SHIFT-10))
3941 static void show_migration_types(unsigned char type)
3943 static const char types[MIGRATE_TYPES] = {
3944 [MIGRATE_UNMOVABLE] = 'U',
3945 [MIGRATE_MOVABLE] = 'M',
3946 [MIGRATE_RECLAIMABLE] = 'E',
3947 [MIGRATE_HIGHATOMIC] = 'H',
3949 [MIGRATE_CMA] = 'C',
3951 #ifdef CONFIG_MEMORY_ISOLATION
3952 [MIGRATE_ISOLATE] = 'I',
3955 char tmp[MIGRATE_TYPES + 1];
3959 for (i = 0; i < MIGRATE_TYPES; i++) {
3960 if (type & (1 << i))
3965 printk("(%s) ", tmp);
3969 * Show free area list (used inside shift_scroll-lock stuff)
3970 * We also calculate the percentage fragmentation. We do this by counting the
3971 * memory on each free list with the exception of the first item on the list.
3974 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
3977 void show_free_areas(unsigned int filter)
3979 unsigned long free_pcp = 0;
3983 for_each_populated_zone(zone) {
3984 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3987 for_each_online_cpu(cpu)
3988 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
3991 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3992 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3993 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
3994 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3995 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3996 " free:%lu free_pcp:%lu free_cma:%lu\n",
3997 global_page_state(NR_ACTIVE_ANON),
3998 global_page_state(NR_INACTIVE_ANON),
3999 global_page_state(NR_ISOLATED_ANON),
4000 global_page_state(NR_ACTIVE_FILE),
4001 global_page_state(NR_INACTIVE_FILE),
4002 global_page_state(NR_ISOLATED_FILE),
4003 global_page_state(NR_UNEVICTABLE),
4004 global_page_state(NR_FILE_DIRTY),
4005 global_page_state(NR_WRITEBACK),
4006 global_page_state(NR_UNSTABLE_NFS),
4007 global_page_state(NR_SLAB_RECLAIMABLE),
4008 global_page_state(NR_SLAB_UNRECLAIMABLE),
4009 global_page_state(NR_FILE_MAPPED),
4010 global_page_state(NR_SHMEM),
4011 global_page_state(NR_PAGETABLE),
4012 global_page_state(NR_BOUNCE),
4013 global_page_state(NR_FREE_PAGES),
4015 global_page_state(NR_FREE_CMA_PAGES));
4017 for_each_populated_zone(zone) {
4020 if (skip_free_areas_node(filter, zone_to_nid(zone)))
4024 for_each_online_cpu(cpu)
4025 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
4033 " active_anon:%lukB"
4034 " inactive_anon:%lukB"
4035 " active_file:%lukB"
4036 " inactive_file:%lukB"
4037 " unevictable:%lukB"
4038 " isolated(anon):%lukB"
4039 " isolated(file):%lukB"
4047 " slab_reclaimable:%lukB"
4048 " slab_unreclaimable:%lukB"
4049 " kernel_stack:%lukB"
4056 " writeback_tmp:%lukB"
4057 " pages_scanned:%lu"
4058 " all_unreclaimable? %s"
4061 K(zone_page_state(zone, NR_FREE_PAGES)),
4062 K(min_wmark_pages(zone)),
4063 K(low_wmark_pages(zone)),
4064 K(high_wmark_pages(zone)),
4065 K(zone_page_state(zone, NR_ACTIVE_ANON)),
4066 K(zone_page_state(zone, NR_INACTIVE_ANON)),
4067 K(zone_page_state(zone, NR_ACTIVE_FILE)),
4068 K(zone_page_state(zone, NR_INACTIVE_FILE)),
4069 K(zone_page_state(zone, NR_UNEVICTABLE)),
4070 K(zone_page_state(zone, NR_ISOLATED_ANON)),
4071 K(zone_page_state(zone, NR_ISOLATED_FILE)),
4072 K(zone->present_pages),
4073 K(zone->managed_pages),
4074 K(zone_page_state(zone, NR_MLOCK)),
4075 K(zone_page_state(zone, NR_FILE_DIRTY)),
4076 K(zone_page_state(zone, NR_WRITEBACK)),
4077 K(zone_page_state(zone, NR_FILE_MAPPED)),
4078 K(zone_page_state(zone, NR_SHMEM)),
4079 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
4080 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
4081 zone_page_state(zone, NR_KERNEL_STACK) *
4083 K(zone_page_state(zone, NR_PAGETABLE)),
4084 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
4085 K(zone_page_state(zone, NR_BOUNCE)),
4087 K(this_cpu_read(zone->pageset->pcp.count)),
4088 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
4089 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
4090 K(zone_page_state(zone, NR_PAGES_SCANNED)),
4091 (!zone_reclaimable(zone) ? "yes" : "no")
4093 printk("lowmem_reserve[]:");
4094 for (i = 0; i < MAX_NR_ZONES; i++)
4095 printk(" %ld", zone->lowmem_reserve[i]);
4099 for_each_populated_zone(zone) {
4101 unsigned long nr[MAX_ORDER], flags, total = 0;
4102 unsigned char types[MAX_ORDER];
4104 if (skip_free_areas_node(filter, zone_to_nid(zone)))
4107 printk("%s: ", zone->name);
4109 spin_lock_irqsave(&zone->lock, flags);
4110 for (order = 0; order < MAX_ORDER; order++) {
4111 struct free_area *area = &zone->free_area[order];
4114 nr[order] = area->nr_free;
4115 total += nr[order] << order;
4118 for (type = 0; type < MIGRATE_TYPES; type++) {
4119 if (!list_empty(&area->free_list[type]))
4120 types[order] |= 1 << type;
4123 spin_unlock_irqrestore(&zone->lock, flags);
4124 for (order = 0; order < MAX_ORDER; order++) {
4125 printk("%lu*%lukB ", nr[order], K(1UL) << order);
4127 show_migration_types(types[order]);
4129 printk("= %lukB\n", K(total));
4132 hugetlb_show_meminfo();
4134 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
4136 show_swap_cache_info();
4139 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
4141 zoneref->zone = zone;
4142 zoneref->zone_idx = zone_idx(zone);
4146 * Builds allocation fallback zone lists.
4148 * Add all populated zones of a node to the zonelist.
4150 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
4154 enum zone_type zone_type = MAX_NR_ZONES;
4158 zone = pgdat->node_zones + zone_type;
4159 if (populated_zone(zone)) {
4160 zoneref_set_zone(zone,
4161 &zonelist->_zonerefs[nr_zones++]);
4162 check_highest_zone(zone_type);
4164 } while (zone_type);
4172 * 0 = automatic detection of better ordering.
4173 * 1 = order by ([node] distance, -zonetype)
4174 * 2 = order by (-zonetype, [node] distance)
4176 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
4177 * the same zonelist. So only NUMA can configure this param.
4179 #define ZONELIST_ORDER_DEFAULT 0
4180 #define ZONELIST_ORDER_NODE 1
4181 #define ZONELIST_ORDER_ZONE 2
4183 /* zonelist order in the kernel.
4184 * set_zonelist_order() will set this to NODE or ZONE.
4186 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
4187 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
4191 /* The value user specified ....changed by config */
4192 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
4193 /* string for sysctl */
4194 #define NUMA_ZONELIST_ORDER_LEN 16
4195 char numa_zonelist_order[16] = "default";
4198 * interface for configure zonelist ordering.
4199 * command line option "numa_zonelist_order"
4200 * = "[dD]efault - default, automatic configuration.
4201 * = "[nN]ode - order by node locality, then by zone within node
4202 * = "[zZ]one - order by zone, then by locality within zone
4205 static int __parse_numa_zonelist_order(char *s)
4207 if (*s == 'd' || *s == 'D') {
4208 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
4209 } else if (*s == 'n' || *s == 'N') {
4210 user_zonelist_order = ZONELIST_ORDER_NODE;
4211 } else if (*s == 'z' || *s == 'Z') {
4212 user_zonelist_order = ZONELIST_ORDER_ZONE;
4214 pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s);
4220 static __init int setup_numa_zonelist_order(char *s)
4227 ret = __parse_numa_zonelist_order(s);
4229 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
4233 early_param("numa_zonelist_order", setup_numa_zonelist_order);
4236 * sysctl handler for numa_zonelist_order
4238 int numa_zonelist_order_handler(struct ctl_table *table, int write,
4239 void __user *buffer, size_t *length,
4242 char saved_string[NUMA_ZONELIST_ORDER_LEN];
4244 static DEFINE_MUTEX(zl_order_mutex);
4246 mutex_lock(&zl_order_mutex);
4248 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
4252 strcpy(saved_string, (char *)table->data);
4254 ret = proc_dostring(table, write, buffer, length, ppos);
4258 int oldval = user_zonelist_order;
4260 ret = __parse_numa_zonelist_order((char *)table->data);
4263 * bogus value. restore saved string
4265 strncpy((char *)table->data, saved_string,
4266 NUMA_ZONELIST_ORDER_LEN);
4267 user_zonelist_order = oldval;
4268 } else if (oldval != user_zonelist_order) {
4269 mutex_lock(&zonelists_mutex);
4270 build_all_zonelists(NULL, NULL);
4271 mutex_unlock(&zonelists_mutex);
4275 mutex_unlock(&zl_order_mutex);
4280 #define MAX_NODE_LOAD (nr_online_nodes)
4281 static int node_load[MAX_NUMNODES];
4284 * find_next_best_node - find the next node that should appear in a given node's fallback list
4285 * @node: node whose fallback list we're appending
4286 * @used_node_mask: nodemask_t of already used nodes
4288 * We use a number of factors to determine which is the next node that should
4289 * appear on a given node's fallback list. The node should not have appeared
4290 * already in @node's fallback list, and it should be the next closest node
4291 * according to the distance array (which contains arbitrary distance values
4292 * from each node to each node in the system), and should also prefer nodes
4293 * with no CPUs, since presumably they'll have very little allocation pressure
4294 * on them otherwise.
4295 * It returns -1 if no node is found.
4297 static int find_next_best_node(int node, nodemask_t *used_node_mask)
4300 int min_val = INT_MAX;
4301 int best_node = NUMA_NO_NODE;
4302 const struct cpumask *tmp = cpumask_of_node(0);
4304 /* Use the local node if we haven't already */
4305 if (!node_isset(node, *used_node_mask)) {
4306 node_set(node, *used_node_mask);
4310 for_each_node_state(n, N_MEMORY) {
4312 /* Don't want a node to appear more than once */
4313 if (node_isset(n, *used_node_mask))
4316 /* Use the distance array to find the distance */
4317 val = node_distance(node, n);
4319 /* Penalize nodes under us ("prefer the next node") */
4322 /* Give preference to headless and unused nodes */
4323 tmp = cpumask_of_node(n);
4324 if (!cpumask_empty(tmp))
4325 val += PENALTY_FOR_NODE_WITH_CPUS;
4327 /* Slight preference for less loaded node */
4328 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
4329 val += node_load[n];
4331 if (val < min_val) {
4338 node_set(best_node, *used_node_mask);
4345 * Build zonelists ordered by node and zones within node.
4346 * This results in maximum locality--normal zone overflows into local
4347 * DMA zone, if any--but risks exhausting DMA zone.
4349 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
4352 struct zonelist *zonelist;
4354 zonelist = &pgdat->node_zonelists[0];
4355 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
4357 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4358 zonelist->_zonerefs[j].zone = NULL;
4359 zonelist->_zonerefs[j].zone_idx = 0;
4363 * Build gfp_thisnode zonelists
4365 static void build_thisnode_zonelists(pg_data_t *pgdat)
4368 struct zonelist *zonelist;
4370 zonelist = &pgdat->node_zonelists[1];
4371 j = build_zonelists_node(pgdat, zonelist, 0);
4372 zonelist->_zonerefs[j].zone = NULL;
4373 zonelist->_zonerefs[j].zone_idx = 0;
4377 * Build zonelists ordered by zone and nodes within zones.
4378 * This results in conserving DMA zone[s] until all Normal memory is
4379 * exhausted, but results in overflowing to remote node while memory
4380 * may still exist in local DMA zone.
4382 static int node_order[MAX_NUMNODES];
4384 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
4387 int zone_type; /* needs to be signed */
4389 struct zonelist *zonelist;
4391 zonelist = &pgdat->node_zonelists[0];
4393 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
4394 for (j = 0; j < nr_nodes; j++) {
4395 node = node_order[j];
4396 z = &NODE_DATA(node)->node_zones[zone_type];
4397 if (populated_zone(z)) {
4399 &zonelist->_zonerefs[pos++]);
4400 check_highest_zone(zone_type);
4404 zonelist->_zonerefs[pos].zone = NULL;
4405 zonelist->_zonerefs[pos].zone_idx = 0;
4408 #if defined(CONFIG_64BIT)
4410 * Devices that require DMA32/DMA are relatively rare and do not justify a
4411 * penalty to every machine in case the specialised case applies. Default
4412 * to Node-ordering on 64-bit NUMA machines
4414 static int default_zonelist_order(void)
4416 return ZONELIST_ORDER_NODE;
4420 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4421 * by the kernel. If processes running on node 0 deplete the low memory zone
4422 * then reclaim will occur more frequency increasing stalls and potentially
4423 * be easier to OOM if a large percentage of the zone is under writeback or
4424 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4425 * Hence, default to zone ordering on 32-bit.
4427 static int default_zonelist_order(void)
4429 return ZONELIST_ORDER_ZONE;
4431 #endif /* CONFIG_64BIT */
4433 static void set_zonelist_order(void)
4435 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
4436 current_zonelist_order = default_zonelist_order();
4438 current_zonelist_order = user_zonelist_order;
4441 static void build_zonelists(pg_data_t *pgdat)
4444 nodemask_t used_mask;
4445 int local_node, prev_node;
4446 struct zonelist *zonelist;
4447 unsigned int order = current_zonelist_order;
4449 /* initialize zonelists */
4450 for (i = 0; i < MAX_ZONELISTS; i++) {
4451 zonelist = pgdat->node_zonelists + i;
4452 zonelist->_zonerefs[0].zone = NULL;
4453 zonelist->_zonerefs[0].zone_idx = 0;
4456 /* NUMA-aware ordering of nodes */
4457 local_node = pgdat->node_id;
4458 load = nr_online_nodes;
4459 prev_node = local_node;
4460 nodes_clear(used_mask);
4462 memset(node_order, 0, sizeof(node_order));
4465 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
4467 * We don't want to pressure a particular node.
4468 * So adding penalty to the first node in same
4469 * distance group to make it round-robin.
4471 if (node_distance(local_node, node) !=
4472 node_distance(local_node, prev_node))
4473 node_load[node] = load;
4477 if (order == ZONELIST_ORDER_NODE)
4478 build_zonelists_in_node_order(pgdat, node);
4480 node_order[i++] = node; /* remember order */
4483 if (order == ZONELIST_ORDER_ZONE) {
4484 /* calculate node order -- i.e., DMA last! */
4485 build_zonelists_in_zone_order(pgdat, i);
4488 build_thisnode_zonelists(pgdat);
4491 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4493 * Return node id of node used for "local" allocations.
4494 * I.e., first node id of first zone in arg node's generic zonelist.
4495 * Used for initializing percpu 'numa_mem', which is used primarily
4496 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4498 int local_memory_node(int node)
4502 z = first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
4503 gfp_zone(GFP_KERNEL),
4505 return z->zone->node;
4509 #else /* CONFIG_NUMA */
4511 static void set_zonelist_order(void)
4513 current_zonelist_order = ZONELIST_ORDER_ZONE;
4516 static void build_zonelists(pg_data_t *pgdat)
4518 int node, local_node;
4520 struct zonelist *zonelist;
4522 local_node = pgdat->node_id;
4524 zonelist = &pgdat->node_zonelists[0];
4525 j = build_zonelists_node(pgdat, zonelist, 0);
4528 * Now we build the zonelist so that it contains the zones
4529 * of all the other nodes.
4530 * We don't want to pressure a particular node, so when
4531 * building the zones for node N, we make sure that the
4532 * zones coming right after the local ones are those from
4533 * node N+1 (modulo N)
4535 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
4536 if (!node_online(node))
4538 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4540 for (node = 0; node < local_node; node++) {
4541 if (!node_online(node))
4543 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4546 zonelist->_zonerefs[j].zone = NULL;
4547 zonelist->_zonerefs[j].zone_idx = 0;
4550 #endif /* CONFIG_NUMA */
4553 * Boot pageset table. One per cpu which is going to be used for all
4554 * zones and all nodes. The parameters will be set in such a way
4555 * that an item put on a list will immediately be handed over to
4556 * the buddy list. This is safe since pageset manipulation is done
4557 * with interrupts disabled.
4559 * The boot_pagesets must be kept even after bootup is complete for
4560 * unused processors and/or zones. They do play a role for bootstrapping
4561 * hotplugged processors.
4563 * zoneinfo_show() and maybe other functions do
4564 * not check if the processor is online before following the pageset pointer.
4565 * Other parts of the kernel may not check if the zone is available.
4567 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
4568 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
4569 static void setup_zone_pageset(struct zone *zone);
4572 * Global mutex to protect against size modification of zonelists
4573 * as well as to serialize pageset setup for the new populated zone.
4575 DEFINE_MUTEX(zonelists_mutex);
4577 /* return values int ....just for stop_machine() */
4578 static int __build_all_zonelists(void *data)
4582 pg_data_t *self = data;
4585 memset(node_load, 0, sizeof(node_load));
4588 if (self && !node_online(self->node_id)) {
4589 build_zonelists(self);
4592 for_each_online_node(nid) {
4593 pg_data_t *pgdat = NODE_DATA(nid);
4595 build_zonelists(pgdat);
4599 * Initialize the boot_pagesets that are going to be used
4600 * for bootstrapping processors. The real pagesets for
4601 * each zone will be allocated later when the per cpu
4602 * allocator is available.
4604 * boot_pagesets are used also for bootstrapping offline
4605 * cpus if the system is already booted because the pagesets
4606 * are needed to initialize allocators on a specific cpu too.
4607 * F.e. the percpu allocator needs the page allocator which
4608 * needs the percpu allocator in order to allocate its pagesets
4609 * (a chicken-egg dilemma).
4611 for_each_possible_cpu(cpu) {
4612 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
4614 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4616 * We now know the "local memory node" for each node--
4617 * i.e., the node of the first zone in the generic zonelist.
4618 * Set up numa_mem percpu variable for on-line cpus. During
4619 * boot, only the boot cpu should be on-line; we'll init the
4620 * secondary cpus' numa_mem as they come on-line. During
4621 * node/memory hotplug, we'll fixup all on-line cpus.
4623 if (cpu_online(cpu))
4624 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
4631 static noinline void __init
4632 build_all_zonelists_init(void)
4634 __build_all_zonelists(NULL);
4635 mminit_verify_zonelist();
4636 cpuset_init_current_mems_allowed();
4640 * Called with zonelists_mutex held always
4641 * unless system_state == SYSTEM_BOOTING.
4643 * __ref due to (1) call of __meminit annotated setup_zone_pageset
4644 * [we're only called with non-NULL zone through __meminit paths] and
4645 * (2) call of __init annotated helper build_all_zonelists_init
4646 * [protected by SYSTEM_BOOTING].
4648 void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
4650 set_zonelist_order();
4652 if (system_state == SYSTEM_BOOTING) {
4653 build_all_zonelists_init();
4655 #ifdef CONFIG_MEMORY_HOTPLUG
4657 setup_zone_pageset(zone);
4659 /* we have to stop all cpus to guarantee there is no user
4661 stop_machine(__build_all_zonelists, pgdat, NULL);
4662 /* cpuset refresh routine should be here */
4664 vm_total_pages = nr_free_pagecache_pages();
4666 * Disable grouping by mobility if the number of pages in the
4667 * system is too low to allow the mechanism to work. It would be
4668 * more accurate, but expensive to check per-zone. This check is
4669 * made on memory-hotadd so a system can start with mobility
4670 * disabled and enable it later
4672 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
4673 page_group_by_mobility_disabled = 1;
4675 page_group_by_mobility_disabled = 0;
4677 pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n",
4679 zonelist_order_name[current_zonelist_order],
4680 page_group_by_mobility_disabled ? "off" : "on",
4683 pr_info("Policy zone: %s\n", zone_names[policy_zone]);
4688 * Helper functions to size the waitqueue hash table.
4689 * Essentially these want to choose hash table sizes sufficiently
4690 * large so that collisions trying to wait on pages are rare.
4691 * But in fact, the number of active page waitqueues on typical
4692 * systems is ridiculously low, less than 200. So this is even
4693 * conservative, even though it seems large.
4695 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
4696 * waitqueues, i.e. the size of the waitq table given the number of pages.
4698 #define PAGES_PER_WAITQUEUE 256
4700 #ifndef CONFIG_MEMORY_HOTPLUG
4701 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4703 unsigned long size = 1;
4705 pages /= PAGES_PER_WAITQUEUE;
4707 while (size < pages)
4711 * Once we have dozens or even hundreds of threads sleeping
4712 * on IO we've got bigger problems than wait queue collision.
4713 * Limit the size of the wait table to a reasonable size.
4715 size = min(size, 4096UL);
4717 return max(size, 4UL);
4721 * A zone's size might be changed by hot-add, so it is not possible to determine
4722 * a suitable size for its wait_table. So we use the maximum size now.
4724 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
4726 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
4727 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4728 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
4730 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4731 * or more by the traditional way. (See above). It equals:
4733 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
4734 * ia64(16K page size) : = ( 8G + 4M)byte.
4735 * powerpc (64K page size) : = (32G +16M)byte.
4737 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4744 * This is an integer logarithm so that shifts can be used later
4745 * to extract the more random high bits from the multiplicative
4746 * hash function before the remainder is taken.
4748 static inline unsigned long wait_table_bits(unsigned long size)
4754 * Initially all pages are reserved - free ones are freed
4755 * up by free_all_bootmem() once the early boot process is
4756 * done. Non-atomic initialization, single-pass.
4758 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
4759 unsigned long start_pfn, enum memmap_context context)
4761 struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn));
4762 unsigned long end_pfn = start_pfn + size;
4763 pg_data_t *pgdat = NODE_DATA(nid);
4765 unsigned long nr_initialised = 0;
4766 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4767 struct memblock_region *r = NULL, *tmp;
4770 if (highest_memmap_pfn < end_pfn - 1)
4771 highest_memmap_pfn = end_pfn - 1;
4774 * Honor reservation requested by the driver for this ZONE_DEVICE
4777 if (altmap && start_pfn == altmap->base_pfn)
4778 start_pfn += altmap->reserve;
4780 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
4782 * There can be holes in boot-time mem_map[]s handed to this
4783 * function. They do not exist on hotplugged memory.
4785 if (context != MEMMAP_EARLY)
4788 if (!early_pfn_valid(pfn))
4790 if (!early_pfn_in_nid(pfn, nid))
4792 if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised))
4795 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4797 * If not mirrored_kernelcore and ZONE_MOVABLE exists, range
4798 * from zone_movable_pfn[nid] to end of each node should be
4799 * ZONE_MOVABLE not ZONE_NORMAL. skip it.
4801 if (!mirrored_kernelcore && zone_movable_pfn[nid])
4802 if (zone == ZONE_NORMAL && pfn >= zone_movable_pfn[nid])
4806 * Check given memblock attribute by firmware which can affect
4807 * kernel memory layout. If zone==ZONE_MOVABLE but memory is
4808 * mirrored, it's an overlapped memmap init. skip it.
4810 if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
4811 if (!r || pfn >= memblock_region_memory_end_pfn(r)) {
4812 for_each_memblock(memory, tmp)
4813 if (pfn < memblock_region_memory_end_pfn(tmp))
4817 if (pfn >= memblock_region_memory_base_pfn(r) &&
4818 memblock_is_mirror(r)) {
4819 /* already initialized as NORMAL */
4820 pfn = memblock_region_memory_end_pfn(r);
4828 * Mark the block movable so that blocks are reserved for
4829 * movable at startup. This will force kernel allocations
4830 * to reserve their blocks rather than leaking throughout
4831 * the address space during boot when many long-lived
4832 * kernel allocations are made.
4834 * bitmap is created for zone's valid pfn range. but memmap
4835 * can be created for invalid pages (for alignment)
4836 * check here not to call set_pageblock_migratetype() against
4839 if (!(pfn & (pageblock_nr_pages - 1))) {
4840 struct page *page = pfn_to_page(pfn);
4842 __init_single_page(page, pfn, zone, nid);
4843 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4845 __init_single_pfn(pfn, zone, nid);
4850 static void __meminit zone_init_free_lists(struct zone *zone)
4852 unsigned int order, t;
4853 for_each_migratetype_order(order, t) {
4854 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
4855 zone->free_area[order].nr_free = 0;
4859 #ifndef __HAVE_ARCH_MEMMAP_INIT
4860 #define memmap_init(size, nid, zone, start_pfn) \
4861 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4864 static int zone_batchsize(struct zone *zone)
4870 * The per-cpu-pages pools are set to around 1000th of the
4871 * size of the zone. But no more than 1/2 of a meg.
4873 * OK, so we don't know how big the cache is. So guess.
4875 batch = zone->managed_pages / 1024;
4876 if (batch * PAGE_SIZE > 512 * 1024)
4877 batch = (512 * 1024) / PAGE_SIZE;
4878 batch /= 4; /* We effectively *= 4 below */
4883 * Clamp the batch to a 2^n - 1 value. Having a power
4884 * of 2 value was found to be more likely to have
4885 * suboptimal cache aliasing properties in some cases.
4887 * For example if 2 tasks are alternately allocating
4888 * batches of pages, one task can end up with a lot
4889 * of pages of one half of the possible page colors
4890 * and the other with pages of the other colors.
4892 batch = rounddown_pow_of_two(batch + batch/2) - 1;
4897 /* The deferral and batching of frees should be suppressed under NOMMU
4900 * The problem is that NOMMU needs to be able to allocate large chunks
4901 * of contiguous memory as there's no hardware page translation to
4902 * assemble apparent contiguous memory from discontiguous pages.
4904 * Queueing large contiguous runs of pages for batching, however,
4905 * causes the pages to actually be freed in smaller chunks. As there
4906 * can be a significant delay between the individual batches being
4907 * recycled, this leads to the once large chunks of space being
4908 * fragmented and becoming unavailable for high-order allocations.
4915 * pcp->high and pcp->batch values are related and dependent on one another:
4916 * ->batch must never be higher then ->high.
4917 * The following function updates them in a safe manner without read side
4920 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4921 * those fields changing asynchronously (acording the the above rule).
4923 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4924 * outside of boot time (or some other assurance that no concurrent updaters
4927 static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4928 unsigned long batch)
4930 /* start with a fail safe value for batch */
4934 /* Update high, then batch, in order */
4941 /* a companion to pageset_set_high() */
4942 static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4944 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
4947 static void pageset_init(struct per_cpu_pageset *p)
4949 struct per_cpu_pages *pcp;
4952 memset(p, 0, sizeof(*p));
4956 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4957 INIT_LIST_HEAD(&pcp->lists[migratetype]);
4960 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4963 pageset_set_batch(p, batch);
4967 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
4968 * to the value high for the pageset p.
4970 static void pageset_set_high(struct per_cpu_pageset *p,
4973 unsigned long batch = max(1UL, high / 4);
4974 if ((high / 4) > (PAGE_SHIFT * 8))
4975 batch = PAGE_SHIFT * 8;
4977 pageset_update(&p->pcp, high, batch);
4980 static void pageset_set_high_and_batch(struct zone *zone,
4981 struct per_cpu_pageset *pcp)
4983 if (percpu_pagelist_fraction)
4984 pageset_set_high(pcp,
4985 (zone->managed_pages /
4986 percpu_pagelist_fraction));
4988 pageset_set_batch(pcp, zone_batchsize(zone));
4991 static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4993 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4996 pageset_set_high_and_batch(zone, pcp);
4999 static void __meminit setup_zone_pageset(struct zone *zone)
5002 zone->pageset = alloc_percpu(struct per_cpu_pageset);
5003 for_each_possible_cpu(cpu)
5004 zone_pageset_init(zone, cpu);
5008 * Allocate per cpu pagesets and initialize them.
5009 * Before this call only boot pagesets were available.
5011 void __init setup_per_cpu_pageset(void)
5015 for_each_populated_zone(zone)
5016 setup_zone_pageset(zone);
5019 static noinline __init_refok
5020 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
5026 * The per-page waitqueue mechanism uses hashed waitqueues
5029 zone->wait_table_hash_nr_entries =
5030 wait_table_hash_nr_entries(zone_size_pages);
5031 zone->wait_table_bits =
5032 wait_table_bits(zone->wait_table_hash_nr_entries);
5033 alloc_size = zone->wait_table_hash_nr_entries
5034 * sizeof(wait_queue_head_t);
5036 if (!slab_is_available()) {
5037 zone->wait_table = (wait_queue_head_t *)
5038 memblock_virt_alloc_node_nopanic(
5039 alloc_size, zone->zone_pgdat->node_id);
5042 * This case means that a zone whose size was 0 gets new memory
5043 * via memory hot-add.
5044 * But it may be the case that a new node was hot-added. In
5045 * this case vmalloc() will not be able to use this new node's
5046 * memory - this wait_table must be initialized to use this new
5047 * node itself as well.
5048 * To use this new node's memory, further consideration will be
5051 zone->wait_table = vmalloc(alloc_size);
5053 if (!zone->wait_table)
5056 for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
5057 init_waitqueue_head(zone->wait_table + i);
5062 static __meminit void zone_pcp_init(struct zone *zone)
5065 * per cpu subsystem is not up at this point. The following code
5066 * relies on the ability of the linker to provide the
5067 * offset of a (static) per cpu variable into the per cpu area.
5069 zone->pageset = &boot_pageset;
5071 if (populated_zone(zone))
5072 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
5073 zone->name, zone->present_pages,
5074 zone_batchsize(zone));
5077 int __meminit init_currently_empty_zone(struct zone *zone,
5078 unsigned long zone_start_pfn,
5081 struct pglist_data *pgdat = zone->zone_pgdat;
5083 ret = zone_wait_table_init(zone, size);
5086 pgdat->nr_zones = zone_idx(zone) + 1;
5088 zone->zone_start_pfn = zone_start_pfn;
5090 mminit_dprintk(MMINIT_TRACE, "memmap_init",
5091 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
5093 (unsigned long)zone_idx(zone),
5094 zone_start_pfn, (zone_start_pfn + size));
5096 zone_init_free_lists(zone);
5101 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5102 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
5105 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
5107 int __meminit __early_pfn_to_nid(unsigned long pfn,
5108 struct mminit_pfnnid_cache *state)
5110 unsigned long start_pfn, end_pfn;
5113 if (state->last_start <= pfn && pfn < state->last_end)
5114 return state->last_nid;
5116 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
5118 state->last_start = start_pfn;
5119 state->last_end = end_pfn;
5120 state->last_nid = nid;
5125 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
5128 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
5129 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
5130 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
5132 * If an architecture guarantees that all ranges registered contain no holes
5133 * and may be freed, this this function may be used instead of calling
5134 * memblock_free_early_nid() manually.
5136 void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
5138 unsigned long start_pfn, end_pfn;
5141 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
5142 start_pfn = min(start_pfn, max_low_pfn);
5143 end_pfn = min(end_pfn, max_low_pfn);
5145 if (start_pfn < end_pfn)
5146 memblock_free_early_nid(PFN_PHYS(start_pfn),
5147 (end_pfn - start_pfn) << PAGE_SHIFT,
5153 * sparse_memory_present_with_active_regions - Call memory_present for each active range
5154 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
5156 * If an architecture guarantees that all ranges registered contain no holes and may
5157 * be freed, this function may be used instead of calling memory_present() manually.
5159 void __init sparse_memory_present_with_active_regions(int nid)
5161 unsigned long start_pfn, end_pfn;
5164 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
5165 memory_present(this_nid, start_pfn, end_pfn);
5169 * get_pfn_range_for_nid - Return the start and end page frames for a node
5170 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
5171 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
5172 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
5174 * It returns the start and end page frame of a node based on information
5175 * provided by memblock_set_node(). If called for a node
5176 * with no available memory, a warning is printed and the start and end
5179 void __meminit get_pfn_range_for_nid(unsigned int nid,
5180 unsigned long *start_pfn, unsigned long *end_pfn)
5182 unsigned long this_start_pfn, this_end_pfn;
5188 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
5189 *start_pfn = min(*start_pfn, this_start_pfn);
5190 *end_pfn = max(*end_pfn, this_end_pfn);
5193 if (*start_pfn == -1UL)
5198 * This finds a zone that can be used for ZONE_MOVABLE pages. The
5199 * assumption is made that zones within a node are ordered in monotonic
5200 * increasing memory addresses so that the "highest" populated zone is used
5202 static void __init find_usable_zone_for_movable(void)
5205 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
5206 if (zone_index == ZONE_MOVABLE)
5209 if (arch_zone_highest_possible_pfn[zone_index] >
5210 arch_zone_lowest_possible_pfn[zone_index])
5214 VM_BUG_ON(zone_index == -1);
5215 movable_zone = zone_index;
5219 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
5220 * because it is sized independent of architecture. Unlike the other zones,
5221 * the starting point for ZONE_MOVABLE is not fixed. It may be different
5222 * in each node depending on the size of each node and how evenly kernelcore
5223 * is distributed. This helper function adjusts the zone ranges
5224 * provided by the architecture for a given node by using the end of the
5225 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5226 * zones within a node are in order of monotonic increases memory addresses
5228 static void __meminit adjust_zone_range_for_zone_movable(int nid,
5229 unsigned long zone_type,
5230 unsigned long node_start_pfn,
5231 unsigned long node_end_pfn,
5232 unsigned long *zone_start_pfn,
5233 unsigned long *zone_end_pfn)
5235 /* Only adjust if ZONE_MOVABLE is on this node */
5236 if (zone_movable_pfn[nid]) {
5237 /* Size ZONE_MOVABLE */
5238 if (zone_type == ZONE_MOVABLE) {
5239 *zone_start_pfn = zone_movable_pfn[nid];
5240 *zone_end_pfn = min(node_end_pfn,
5241 arch_zone_highest_possible_pfn[movable_zone]);
5243 /* Check if this whole range is within ZONE_MOVABLE */
5244 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
5245 *zone_start_pfn = *zone_end_pfn;
5250 * Return the number of pages a zone spans in a node, including holes
5251 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5253 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
5254 unsigned long zone_type,
5255 unsigned long node_start_pfn,
5256 unsigned long node_end_pfn,
5257 unsigned long *zone_start_pfn,
5258 unsigned long *zone_end_pfn,
5259 unsigned long *ignored)
5261 /* When hotadd a new node from cpu_up(), the node should be empty */
5262 if (!node_start_pfn && !node_end_pfn)
5265 /* Get the start and end of the zone */
5266 *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
5267 *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
5268 adjust_zone_range_for_zone_movable(nid, zone_type,
5269 node_start_pfn, node_end_pfn,
5270 zone_start_pfn, zone_end_pfn);
5272 /* Check that this node has pages within the zone's required range */
5273 if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
5276 /* Move the zone boundaries inside the node if necessary */
5277 *zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
5278 *zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
5280 /* Return the spanned pages */
5281 return *zone_end_pfn - *zone_start_pfn;
5285 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
5286 * then all holes in the requested range will be accounted for.
5288 unsigned long __meminit __absent_pages_in_range(int nid,
5289 unsigned long range_start_pfn,
5290 unsigned long range_end_pfn)
5292 unsigned long nr_absent = range_end_pfn - range_start_pfn;
5293 unsigned long start_pfn, end_pfn;
5296 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5297 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
5298 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
5299 nr_absent -= end_pfn - start_pfn;
5305 * absent_pages_in_range - Return number of page frames in holes within a range
5306 * @start_pfn: The start PFN to start searching for holes
5307 * @end_pfn: The end PFN to stop searching for holes
5309 * It returns the number of pages frames in memory holes within a range.
5311 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
5312 unsigned long end_pfn)
5314 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
5317 /* Return the number of page frames in holes in a zone on a node */
5318 static unsigned long __meminit zone_absent_pages_in_node(int nid,
5319 unsigned long zone_type,
5320 unsigned long node_start_pfn,
5321 unsigned long node_end_pfn,
5322 unsigned long *ignored)
5324 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
5325 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
5326 unsigned long zone_start_pfn, zone_end_pfn;
5327 unsigned long nr_absent;
5329 /* When hotadd a new node from cpu_up(), the node should be empty */
5330 if (!node_start_pfn && !node_end_pfn)
5333 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
5334 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
5336 adjust_zone_range_for_zone_movable(nid, zone_type,
5337 node_start_pfn, node_end_pfn,
5338 &zone_start_pfn, &zone_end_pfn);
5339 nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
5342 * ZONE_MOVABLE handling.
5343 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
5346 if (zone_movable_pfn[nid]) {
5347 if (mirrored_kernelcore) {
5348 unsigned long start_pfn, end_pfn;
5349 struct memblock_region *r;
5351 for_each_memblock(memory, r) {
5352 start_pfn = clamp(memblock_region_memory_base_pfn(r),
5353 zone_start_pfn, zone_end_pfn);
5354 end_pfn = clamp(memblock_region_memory_end_pfn(r),
5355 zone_start_pfn, zone_end_pfn);
5357 if (zone_type == ZONE_MOVABLE &&
5358 memblock_is_mirror(r))
5359 nr_absent += end_pfn - start_pfn;
5361 if (zone_type == ZONE_NORMAL &&
5362 !memblock_is_mirror(r))
5363 nr_absent += end_pfn - start_pfn;
5366 if (zone_type == ZONE_NORMAL)
5367 nr_absent += node_end_pfn - zone_movable_pfn[nid];
5374 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5375 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
5376 unsigned long zone_type,
5377 unsigned long node_start_pfn,
5378 unsigned long node_end_pfn,
5379 unsigned long *zone_start_pfn,
5380 unsigned long *zone_end_pfn,
5381 unsigned long *zones_size)
5385 *zone_start_pfn = node_start_pfn;
5386 for (zone = 0; zone < zone_type; zone++)
5387 *zone_start_pfn += zones_size[zone];
5389 *zone_end_pfn = *zone_start_pfn + zones_size[zone_type];
5391 return zones_size[zone_type];
5394 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
5395 unsigned long zone_type,
5396 unsigned long node_start_pfn,
5397 unsigned long node_end_pfn,
5398 unsigned long *zholes_size)
5403 return zholes_size[zone_type];
5406 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5408 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
5409 unsigned long node_start_pfn,
5410 unsigned long node_end_pfn,
5411 unsigned long *zones_size,
5412 unsigned long *zholes_size)
5414 unsigned long realtotalpages = 0, totalpages = 0;
5417 for (i = 0; i < MAX_NR_ZONES; i++) {
5418 struct zone *zone = pgdat->node_zones + i;
5419 unsigned long zone_start_pfn, zone_end_pfn;
5420 unsigned long size, real_size;
5422 size = zone_spanned_pages_in_node(pgdat->node_id, i,
5428 real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
5429 node_start_pfn, node_end_pfn,
5432 zone->zone_start_pfn = zone_start_pfn;
5434 zone->zone_start_pfn = 0;
5435 zone->spanned_pages = size;
5436 zone->present_pages = real_size;
5439 realtotalpages += real_size;
5442 pgdat->node_spanned_pages = totalpages;
5443 pgdat->node_present_pages = realtotalpages;
5444 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
5448 #ifndef CONFIG_SPARSEMEM
5450 * Calculate the size of the zone->blockflags rounded to an unsigned long
5451 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5452 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
5453 * round what is now in bits to nearest long in bits, then return it in
5456 static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
5458 unsigned long usemapsize;
5460 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
5461 usemapsize = roundup(zonesize, pageblock_nr_pages);
5462 usemapsize = usemapsize >> pageblock_order;
5463 usemapsize *= NR_PAGEBLOCK_BITS;
5464 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
5466 return usemapsize / 8;
5469 static void __init setup_usemap(struct pglist_data *pgdat,
5471 unsigned long zone_start_pfn,
5472 unsigned long zonesize)
5474 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
5475 zone->pageblock_flags = NULL;
5477 zone->pageblock_flags =
5478 memblock_virt_alloc_node_nopanic(usemapsize,
5482 static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
5483 unsigned long zone_start_pfn, unsigned long zonesize) {}
5484 #endif /* CONFIG_SPARSEMEM */
5486 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
5488 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
5489 void __paginginit set_pageblock_order(void)
5493 /* Check that pageblock_nr_pages has not already been setup */
5494 if (pageblock_order)
5497 if (HPAGE_SHIFT > PAGE_SHIFT)
5498 order = HUGETLB_PAGE_ORDER;
5500 order = MAX_ORDER - 1;
5503 * Assume the largest contiguous order of interest is a huge page.
5504 * This value may be variable depending on boot parameters on IA64 and
5507 pageblock_order = order;
5509 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5512 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
5513 * is unused as pageblock_order is set at compile-time. See
5514 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5517 void __paginginit set_pageblock_order(void)
5521 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5523 static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
5524 unsigned long present_pages)
5526 unsigned long pages = spanned_pages;
5529 * Provide a more accurate estimation if there are holes within
5530 * the zone and SPARSEMEM is in use. If there are holes within the
5531 * zone, each populated memory region may cost us one or two extra
5532 * memmap pages due to alignment because memmap pages for each
5533 * populated regions may not naturally algined on page boundary.
5534 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5536 if (spanned_pages > present_pages + (present_pages >> 4) &&
5537 IS_ENABLED(CONFIG_SPARSEMEM))
5538 pages = present_pages;
5540 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
5544 * Set up the zone data structures:
5545 * - mark all pages reserved
5546 * - mark all memory queues empty
5547 * - clear the memory bitmaps
5549 * NOTE: pgdat should get zeroed by caller.
5551 static void __paginginit free_area_init_core(struct pglist_data *pgdat)
5554 int nid = pgdat->node_id;
5557 pgdat_resize_init(pgdat);
5558 #ifdef CONFIG_NUMA_BALANCING
5559 spin_lock_init(&pgdat->numabalancing_migrate_lock);
5560 pgdat->numabalancing_migrate_nr_pages = 0;
5561 pgdat->numabalancing_migrate_next_window = jiffies;
5563 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5564 spin_lock_init(&pgdat->split_queue_lock);
5565 INIT_LIST_HEAD(&pgdat->split_queue);
5566 pgdat->split_queue_len = 0;
5568 init_waitqueue_head(&pgdat->kswapd_wait);
5569 init_waitqueue_head(&pgdat->pfmemalloc_wait);
5570 #ifdef CONFIG_COMPACTION
5571 init_waitqueue_head(&pgdat->kcompactd_wait);
5573 pgdat_page_ext_init(pgdat);
5575 for (j = 0; j < MAX_NR_ZONES; j++) {
5576 struct zone *zone = pgdat->node_zones + j;
5577 unsigned long size, realsize, freesize, memmap_pages;
5578 unsigned long zone_start_pfn = zone->zone_start_pfn;
5580 size = zone->spanned_pages;
5581 realsize = freesize = zone->present_pages;
5584 * Adjust freesize so that it accounts for how much memory
5585 * is used by this zone for memmap. This affects the watermark
5586 * and per-cpu initialisations
5588 memmap_pages = calc_memmap_size(size, realsize);
5589 if (!is_highmem_idx(j)) {
5590 if (freesize >= memmap_pages) {
5591 freesize -= memmap_pages;
5594 " %s zone: %lu pages used for memmap\n",
5595 zone_names[j], memmap_pages);
5597 pr_warn(" %s zone: %lu pages exceeds freesize %lu\n",
5598 zone_names[j], memmap_pages, freesize);
5601 /* Account for reserved pages */
5602 if (j == 0 && freesize > dma_reserve) {
5603 freesize -= dma_reserve;
5604 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
5605 zone_names[0], dma_reserve);
5608 if (!is_highmem_idx(j))
5609 nr_kernel_pages += freesize;
5610 /* Charge for highmem memmap if there are enough kernel pages */
5611 else if (nr_kernel_pages > memmap_pages * 2)
5612 nr_kernel_pages -= memmap_pages;
5613 nr_all_pages += freesize;
5616 * Set an approximate value for lowmem here, it will be adjusted
5617 * when the bootmem allocator frees pages into the buddy system.
5618 * And all highmem pages will be managed by the buddy system.
5620 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
5623 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
5625 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
5627 zone->name = zone_names[j];
5628 spin_lock_init(&zone->lock);
5629 spin_lock_init(&zone->lru_lock);
5630 zone_seqlock_init(zone);
5631 zone->zone_pgdat = pgdat;
5632 zone_pcp_init(zone);
5634 /* For bootup, initialized properly in watermark setup */
5635 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
5637 lruvec_init(&zone->lruvec);
5641 set_pageblock_order();
5642 setup_usemap(pgdat, zone, zone_start_pfn, size);
5643 ret = init_currently_empty_zone(zone, zone_start_pfn, size);
5645 memmap_init(size, nid, j, zone_start_pfn);
5649 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
5651 unsigned long __maybe_unused start = 0;
5652 unsigned long __maybe_unused offset = 0;
5654 /* Skip empty nodes */
5655 if (!pgdat->node_spanned_pages)
5658 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5659 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
5660 offset = pgdat->node_start_pfn - start;
5661 /* ia64 gets its own node_mem_map, before this, without bootmem */
5662 if (!pgdat->node_mem_map) {
5663 unsigned long size, end;
5667 * The zone's endpoints aren't required to be MAX_ORDER
5668 * aligned but the node_mem_map endpoints must be in order
5669 * for the buddy allocator to function correctly.
5671 end = pgdat_end_pfn(pgdat);
5672 end = ALIGN(end, MAX_ORDER_NR_PAGES);
5673 size = (end - start) * sizeof(struct page);
5674 map = alloc_remap(pgdat->node_id, size);
5676 map = memblock_virt_alloc_node_nopanic(size,
5678 pgdat->node_mem_map = map + offset;
5680 #ifndef CONFIG_NEED_MULTIPLE_NODES
5682 * With no DISCONTIG, the global mem_map is just set as node 0's
5684 if (pgdat == NODE_DATA(0)) {
5685 mem_map = NODE_DATA(0)->node_mem_map;
5686 #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
5687 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
5689 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5692 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
5695 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
5696 unsigned long node_start_pfn, unsigned long *zholes_size)
5698 pg_data_t *pgdat = NODE_DATA(nid);
5699 unsigned long start_pfn = 0;
5700 unsigned long end_pfn = 0;
5702 /* pg_data_t should be reset to zero when it's allocated */
5703 WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
5705 reset_deferred_meminit(pgdat);
5706 pgdat->node_id = nid;
5707 pgdat->node_start_pfn = node_start_pfn;
5708 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5709 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
5710 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
5711 (u64)start_pfn << PAGE_SHIFT,
5712 end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
5714 start_pfn = node_start_pfn;
5716 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
5717 zones_size, zholes_size);
5719 alloc_node_mem_map(pgdat);
5720 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5721 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5722 nid, (unsigned long)pgdat,
5723 (unsigned long)pgdat->node_mem_map);
5726 free_area_init_core(pgdat);
5729 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5731 #if MAX_NUMNODES > 1
5733 * Figure out the number of possible node ids.
5735 void __init setup_nr_node_ids(void)
5737 unsigned int highest;
5739 highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
5740 nr_node_ids = highest + 1;
5745 * node_map_pfn_alignment - determine the maximum internode alignment
5747 * This function should be called after node map is populated and sorted.
5748 * It calculates the maximum power of two alignment which can distinguish
5751 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5752 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
5753 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
5754 * shifted, 1GiB is enough and this function will indicate so.
5756 * This is used to test whether pfn -> nid mapping of the chosen memory
5757 * model has fine enough granularity to avoid incorrect mapping for the
5758 * populated node map.
5760 * Returns the determined alignment in pfn's. 0 if there is no alignment
5761 * requirement (single node).
5763 unsigned long __init node_map_pfn_alignment(void)
5765 unsigned long accl_mask = 0, last_end = 0;
5766 unsigned long start, end, mask;
5770 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
5771 if (!start || last_nid < 0 || last_nid == nid) {
5778 * Start with a mask granular enough to pin-point to the
5779 * start pfn and tick off bits one-by-one until it becomes
5780 * too coarse to separate the current node from the last.
5782 mask = ~((1 << __ffs(start)) - 1);
5783 while (mask && last_end <= (start & (mask << 1)))
5786 /* accumulate all internode masks */
5790 /* convert mask to number of pages */
5791 return ~accl_mask + 1;
5794 /* Find the lowest pfn for a node */
5795 static unsigned long __init find_min_pfn_for_node(int nid)
5797 unsigned long min_pfn = ULONG_MAX;
5798 unsigned long start_pfn;
5801 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5802 min_pfn = min(min_pfn, start_pfn);
5804 if (min_pfn == ULONG_MAX) {
5805 pr_warn("Could not find start_pfn for node %d\n", nid);
5813 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5815 * It returns the minimum PFN based on information provided via
5816 * memblock_set_node().
5818 unsigned long __init find_min_pfn_with_active_regions(void)
5820 return find_min_pfn_for_node(MAX_NUMNODES);
5824 * early_calculate_totalpages()
5825 * Sum pages in active regions for movable zone.
5826 * Populate N_MEMORY for calculating usable_nodes.
5828 static unsigned long __init early_calculate_totalpages(void)
5830 unsigned long totalpages = 0;
5831 unsigned long start_pfn, end_pfn;
5834 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5835 unsigned long pages = end_pfn - start_pfn;
5837 totalpages += pages;
5839 node_set_state(nid, N_MEMORY);
5845 * Find the PFN the Movable zone begins in each node. Kernel memory
5846 * is spread evenly between nodes as long as the nodes have enough
5847 * memory. When they don't, some nodes will have more kernelcore than
5850 static void __init find_zone_movable_pfns_for_nodes(void)
5853 unsigned long usable_startpfn;
5854 unsigned long kernelcore_node, kernelcore_remaining;
5855 /* save the state before borrow the nodemask */
5856 nodemask_t saved_node_state = node_states[N_MEMORY];
5857 unsigned long totalpages = early_calculate_totalpages();
5858 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
5859 struct memblock_region *r;
5861 /* Need to find movable_zone earlier when movable_node is specified. */
5862 find_usable_zone_for_movable();
5865 * If movable_node is specified, ignore kernelcore and movablecore
5868 if (movable_node_is_enabled()) {
5869 for_each_memblock(memory, r) {
5870 if (!memblock_is_hotpluggable(r))
5875 usable_startpfn = PFN_DOWN(r->base);
5876 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5877 min(usable_startpfn, zone_movable_pfn[nid]) :
5885 * If kernelcore=mirror is specified, ignore movablecore option
5887 if (mirrored_kernelcore) {
5888 bool mem_below_4gb_not_mirrored = false;
5890 for_each_memblock(memory, r) {
5891 if (memblock_is_mirror(r))
5896 usable_startpfn = memblock_region_memory_base_pfn(r);
5898 if (usable_startpfn < 0x100000) {
5899 mem_below_4gb_not_mirrored = true;
5903 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5904 min(usable_startpfn, zone_movable_pfn[nid]) :
5908 if (mem_below_4gb_not_mirrored)
5909 pr_warn("This configuration results in unmirrored kernel memory.");
5915 * If movablecore=nn[KMG] was specified, calculate what size of
5916 * kernelcore that corresponds so that memory usable for
5917 * any allocation type is evenly spread. If both kernelcore
5918 * and movablecore are specified, then the value of kernelcore
5919 * will be used for required_kernelcore if it's greater than
5920 * what movablecore would have allowed.
5922 if (required_movablecore) {
5923 unsigned long corepages;
5926 * Round-up so that ZONE_MOVABLE is at least as large as what
5927 * was requested by the user
5929 required_movablecore =
5930 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
5931 required_movablecore = min(totalpages, required_movablecore);
5932 corepages = totalpages - required_movablecore;
5934 required_kernelcore = max(required_kernelcore, corepages);
5938 * If kernelcore was not specified or kernelcore size is larger
5939 * than totalpages, there is no ZONE_MOVABLE.
5941 if (!required_kernelcore || required_kernelcore >= totalpages)
5944 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5945 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5948 /* Spread kernelcore memory as evenly as possible throughout nodes */
5949 kernelcore_node = required_kernelcore / usable_nodes;
5950 for_each_node_state(nid, N_MEMORY) {
5951 unsigned long start_pfn, end_pfn;
5954 * Recalculate kernelcore_node if the division per node
5955 * now exceeds what is necessary to satisfy the requested
5956 * amount of memory for the kernel
5958 if (required_kernelcore < kernelcore_node)
5959 kernelcore_node = required_kernelcore / usable_nodes;
5962 * As the map is walked, we track how much memory is usable
5963 * by the kernel using kernelcore_remaining. When it is
5964 * 0, the rest of the node is usable by ZONE_MOVABLE
5966 kernelcore_remaining = kernelcore_node;
5968 /* Go through each range of PFNs within this node */
5969 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5970 unsigned long size_pages;
5972 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
5973 if (start_pfn >= end_pfn)
5976 /* Account for what is only usable for kernelcore */
5977 if (start_pfn < usable_startpfn) {
5978 unsigned long kernel_pages;
5979 kernel_pages = min(end_pfn, usable_startpfn)
5982 kernelcore_remaining -= min(kernel_pages,
5983 kernelcore_remaining);
5984 required_kernelcore -= min(kernel_pages,
5985 required_kernelcore);
5987 /* Continue if range is now fully accounted */
5988 if (end_pfn <= usable_startpfn) {
5991 * Push zone_movable_pfn to the end so
5992 * that if we have to rebalance
5993 * kernelcore across nodes, we will
5994 * not double account here
5996 zone_movable_pfn[nid] = end_pfn;
5999 start_pfn = usable_startpfn;
6003 * The usable PFN range for ZONE_MOVABLE is from
6004 * start_pfn->end_pfn. Calculate size_pages as the
6005 * number of pages used as kernelcore
6007 size_pages = end_pfn - start_pfn;
6008 if (size_pages > kernelcore_remaining)
6009 size_pages = kernelcore_remaining;
6010 zone_movable_pfn[nid] = start_pfn + size_pages;
6013 * Some kernelcore has been met, update counts and
6014 * break if the kernelcore for this node has been
6017 required_kernelcore -= min(required_kernelcore,
6019 kernelcore_remaining -= size_pages;
6020 if (!kernelcore_remaining)
6026 * If there is still required_kernelcore, we do another pass with one
6027 * less node in the count. This will push zone_movable_pfn[nid] further
6028 * along on the nodes that still have memory until kernelcore is
6032 if (usable_nodes && required_kernelcore > usable_nodes)
6036 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
6037 for (nid = 0; nid < MAX_NUMNODES; nid++)
6038 zone_movable_pfn[nid] =
6039 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
6042 /* restore the node_state */
6043 node_states[N_MEMORY] = saved_node_state;
6046 /* Any regular or high memory on that node ? */
6047 static void check_for_memory(pg_data_t *pgdat, int nid)
6049 enum zone_type zone_type;
6051 if (N_MEMORY == N_NORMAL_MEMORY)
6054 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
6055 struct zone *zone = &pgdat->node_zones[zone_type];
6056 if (populated_zone(zone)) {
6057 node_set_state(nid, N_HIGH_MEMORY);
6058 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
6059 zone_type <= ZONE_NORMAL)
6060 node_set_state(nid, N_NORMAL_MEMORY);
6067 * free_area_init_nodes - Initialise all pg_data_t and zone data
6068 * @max_zone_pfn: an array of max PFNs for each zone
6070 * This will call free_area_init_node() for each active node in the system.
6071 * Using the page ranges provided by memblock_set_node(), the size of each
6072 * zone in each node and their holes is calculated. If the maximum PFN
6073 * between two adjacent zones match, it is assumed that the zone is empty.
6074 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
6075 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
6076 * starts where the previous one ended. For example, ZONE_DMA32 starts
6077 * at arch_max_dma_pfn.
6079 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
6081 unsigned long start_pfn, end_pfn;
6084 /* Record where the zone boundaries are */
6085 memset(arch_zone_lowest_possible_pfn, 0,
6086 sizeof(arch_zone_lowest_possible_pfn));
6087 memset(arch_zone_highest_possible_pfn, 0,
6088 sizeof(arch_zone_highest_possible_pfn));
6089 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
6090 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
6091 for (i = 1; i < MAX_NR_ZONES; i++) {
6092 if (i == ZONE_MOVABLE)
6094 arch_zone_lowest_possible_pfn[i] =
6095 arch_zone_highest_possible_pfn[i-1];
6096 arch_zone_highest_possible_pfn[i] =
6097 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
6099 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
6100 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
6102 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
6103 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
6104 find_zone_movable_pfns_for_nodes();
6106 /* Print out the zone ranges */
6107 pr_info("Zone ranges:\n");
6108 for (i = 0; i < MAX_NR_ZONES; i++) {
6109 if (i == ZONE_MOVABLE)
6111 pr_info(" %-8s ", zone_names[i]);
6112 if (arch_zone_lowest_possible_pfn[i] ==
6113 arch_zone_highest_possible_pfn[i])
6116 pr_cont("[mem %#018Lx-%#018Lx]\n",
6117 (u64)arch_zone_lowest_possible_pfn[i]
6119 ((u64)arch_zone_highest_possible_pfn[i]
6120 << PAGE_SHIFT) - 1);
6123 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
6124 pr_info("Movable zone start for each node\n");
6125 for (i = 0; i < MAX_NUMNODES; i++) {
6126 if (zone_movable_pfn[i])
6127 pr_info(" Node %d: %#018Lx\n", i,
6128 (u64)zone_movable_pfn[i] << PAGE_SHIFT);
6131 /* Print out the early node map */
6132 pr_info("Early memory node ranges\n");
6133 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
6134 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
6135 (u64)start_pfn << PAGE_SHIFT,
6136 ((u64)end_pfn << PAGE_SHIFT) - 1);
6138 /* Initialise every node */
6139 mminit_verify_pageflags_layout();
6140 setup_nr_node_ids();
6141 for_each_online_node(nid) {
6142 pg_data_t *pgdat = NODE_DATA(nid);
6143 free_area_init_node(nid, NULL,
6144 find_min_pfn_for_node(nid), NULL);
6146 /* Any memory on that node */
6147 if (pgdat->node_present_pages)
6148 node_set_state(nid, N_MEMORY);
6149 check_for_memory(pgdat, nid);
6153 static int __init cmdline_parse_core(char *p, unsigned long *core)
6155 unsigned long long coremem;
6159 coremem = memparse(p, &p);
6160 *core = coremem >> PAGE_SHIFT;
6162 /* Paranoid check that UL is enough for the coremem value */
6163 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
6169 * kernelcore=size sets the amount of memory for use for allocations that
6170 * cannot be reclaimed or migrated.
6172 static int __init cmdline_parse_kernelcore(char *p)
6174 /* parse kernelcore=mirror */
6175 if (parse_option_str(p, "mirror")) {
6176 mirrored_kernelcore = true;
6180 return cmdline_parse_core(p, &required_kernelcore);
6184 * movablecore=size sets the amount of memory for use for allocations that
6185 * can be reclaimed or migrated.
6187 static int __init cmdline_parse_movablecore(char *p)
6189 return cmdline_parse_core(p, &required_movablecore);
6192 early_param("kernelcore", cmdline_parse_kernelcore);
6193 early_param("movablecore", cmdline_parse_movablecore);
6195 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6197 void adjust_managed_page_count(struct page *page, long count)
6199 spin_lock(&managed_page_count_lock);
6200 page_zone(page)->managed_pages += count;
6201 totalram_pages += count;
6202 #ifdef CONFIG_HIGHMEM
6203 if (PageHighMem(page))
6204 totalhigh_pages += count;
6206 spin_unlock(&managed_page_count_lock);
6208 EXPORT_SYMBOL(adjust_managed_page_count);
6210 unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
6213 unsigned long pages = 0;
6215 start = (void *)PAGE_ALIGN((unsigned long)start);
6216 end = (void *)((unsigned long)end & PAGE_MASK);
6217 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
6218 if ((unsigned int)poison <= 0xFF)
6219 memset(pos, poison, PAGE_SIZE);
6220 free_reserved_page(virt_to_page(pos));
6224 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
6225 s, pages << (PAGE_SHIFT - 10), start, end);
6229 EXPORT_SYMBOL(free_reserved_area);
6231 #ifdef CONFIG_HIGHMEM
6232 void free_highmem_page(struct page *page)
6234 __free_reserved_page(page);
6236 page_zone(page)->managed_pages++;
6242 void __init mem_init_print_info(const char *str)
6244 unsigned long physpages, codesize, datasize, rosize, bss_size;
6245 unsigned long init_code_size, init_data_size;
6247 physpages = get_num_physpages();
6248 codesize = _etext - _stext;
6249 datasize = _edata - _sdata;
6250 rosize = __end_rodata - __start_rodata;
6251 bss_size = __bss_stop - __bss_start;
6252 init_data_size = __init_end - __init_begin;
6253 init_code_size = _einittext - _sinittext;
6256 * Detect special cases and adjust section sizes accordingly:
6257 * 1) .init.* may be embedded into .data sections
6258 * 2) .init.text.* may be out of [__init_begin, __init_end],
6259 * please refer to arch/tile/kernel/vmlinux.lds.S.
6260 * 3) .rodata.* may be embedded into .text or .data sections.
6262 #define adj_init_size(start, end, size, pos, adj) \
6264 if (start <= pos && pos < end && size > adj) \
6268 adj_init_size(__init_begin, __init_end, init_data_size,
6269 _sinittext, init_code_size);
6270 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
6271 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
6272 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
6273 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
6275 #undef adj_init_size
6277 pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
6278 #ifdef CONFIG_HIGHMEM
6282 nr_free_pages() << (PAGE_SHIFT - 10),
6283 physpages << (PAGE_SHIFT - 10),
6284 codesize >> 10, datasize >> 10, rosize >> 10,
6285 (init_data_size + init_code_size) >> 10, bss_size >> 10,
6286 (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10),
6287 totalcma_pages << (PAGE_SHIFT - 10),
6288 #ifdef CONFIG_HIGHMEM
6289 totalhigh_pages << (PAGE_SHIFT - 10),
6291 str ? ", " : "", str ? str : "");
6295 * set_dma_reserve - set the specified number of pages reserved in the first zone
6296 * @new_dma_reserve: The number of pages to mark reserved
6298 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
6299 * In the DMA zone, a significant percentage may be consumed by kernel image
6300 * and other unfreeable allocations which can skew the watermarks badly. This
6301 * function may optionally be used to account for unfreeable pages in the
6302 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6303 * smaller per-cpu batchsize.
6305 void __init set_dma_reserve(unsigned long new_dma_reserve)
6307 dma_reserve = new_dma_reserve;
6310 void __init free_area_init(unsigned long *zones_size)
6312 free_area_init_node(0, zones_size,
6313 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
6316 static int page_alloc_cpu_notify(struct notifier_block *self,
6317 unsigned long action, void *hcpu)
6319 int cpu = (unsigned long)hcpu;
6321 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
6322 lru_add_drain_cpu(cpu);
6326 * Spill the event counters of the dead processor
6327 * into the current processors event counters.
6328 * This artificially elevates the count of the current
6331 vm_events_fold_cpu(cpu);
6334 * Zero the differential counters of the dead processor
6335 * so that the vm statistics are consistent.
6337 * This is only okay since the processor is dead and cannot
6338 * race with what we are doing.
6340 cpu_vm_stats_fold(cpu);
6345 void __init page_alloc_init(void)
6347 hotcpu_notifier(page_alloc_cpu_notify, 0);
6351 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
6352 * or min_free_kbytes changes.
6354 static void calculate_totalreserve_pages(void)
6356 struct pglist_data *pgdat;
6357 unsigned long reserve_pages = 0;
6358 enum zone_type i, j;
6360 for_each_online_pgdat(pgdat) {
6361 for (i = 0; i < MAX_NR_ZONES; i++) {
6362 struct zone *zone = pgdat->node_zones + i;
6365 /* Find valid and maximum lowmem_reserve in the zone */
6366 for (j = i; j < MAX_NR_ZONES; j++) {
6367 if (zone->lowmem_reserve[j] > max)
6368 max = zone->lowmem_reserve[j];
6371 /* we treat the high watermark as reserved pages. */
6372 max += high_wmark_pages(zone);
6374 if (max > zone->managed_pages)
6375 max = zone->managed_pages;
6377 zone->totalreserve_pages = max;
6379 reserve_pages += max;
6382 totalreserve_pages = reserve_pages;
6386 * setup_per_zone_lowmem_reserve - called whenever
6387 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
6388 * has a correct pages reserved value, so an adequate number of
6389 * pages are left in the zone after a successful __alloc_pages().
6391 static void setup_per_zone_lowmem_reserve(void)
6393 struct pglist_data *pgdat;
6394 enum zone_type j, idx;
6396 for_each_online_pgdat(pgdat) {
6397 for (j = 0; j < MAX_NR_ZONES; j++) {
6398 struct zone *zone = pgdat->node_zones + j;
6399 unsigned long managed_pages = zone->managed_pages;
6401 zone->lowmem_reserve[j] = 0;
6405 struct zone *lower_zone;
6409 if (sysctl_lowmem_reserve_ratio[idx] < 1)
6410 sysctl_lowmem_reserve_ratio[idx] = 1;
6412 lower_zone = pgdat->node_zones + idx;
6413 lower_zone->lowmem_reserve[j] = managed_pages /
6414 sysctl_lowmem_reserve_ratio[idx];
6415 managed_pages += lower_zone->managed_pages;
6420 /* update totalreserve_pages */
6421 calculate_totalreserve_pages();
6424 static void __setup_per_zone_wmarks(void)
6426 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
6427 unsigned long lowmem_pages = 0;
6429 unsigned long flags;
6431 /* Calculate total number of !ZONE_HIGHMEM pages */
6432 for_each_zone(zone) {
6433 if (!is_highmem(zone))
6434 lowmem_pages += zone->managed_pages;
6437 for_each_zone(zone) {
6440 spin_lock_irqsave(&zone->lock, flags);
6441 tmp = (u64)pages_min * zone->managed_pages;
6442 do_div(tmp, lowmem_pages);
6443 if (is_highmem(zone)) {
6445 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6446 * need highmem pages, so cap pages_min to a small
6449 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
6450 * deltas control asynch page reclaim, and so should
6451 * not be capped for highmem.
6453 unsigned long min_pages;
6455 min_pages = zone->managed_pages / 1024;
6456 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
6457 zone->watermark[WMARK_MIN] = min_pages;
6460 * If it's a lowmem zone, reserve a number of pages
6461 * proportionate to the zone's size.
6463 zone->watermark[WMARK_MIN] = tmp;
6467 * Set the kswapd watermarks distance according to the
6468 * scale factor in proportion to available memory, but
6469 * ensure a minimum size on small systems.
6471 tmp = max_t(u64, tmp >> 2,
6472 mult_frac(zone->managed_pages,
6473 watermark_scale_factor, 10000));
6475 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp;
6476 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2;
6478 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
6479 high_wmark_pages(zone) - low_wmark_pages(zone) -
6480 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
6482 spin_unlock_irqrestore(&zone->lock, flags);
6485 /* update totalreserve_pages */
6486 calculate_totalreserve_pages();
6490 * setup_per_zone_wmarks - called when min_free_kbytes changes
6491 * or when memory is hot-{added|removed}
6493 * Ensures that the watermark[min,low,high] values for each zone are set
6494 * correctly with respect to min_free_kbytes.
6496 void setup_per_zone_wmarks(void)
6498 mutex_lock(&zonelists_mutex);
6499 __setup_per_zone_wmarks();
6500 mutex_unlock(&zonelists_mutex);
6504 * The inactive anon list should be small enough that the VM never has to
6505 * do too much work, but large enough that each inactive page has a chance
6506 * to be referenced again before it is swapped out.
6508 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
6509 * INACTIVE_ANON pages on this zone's LRU, maintained by the
6510 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
6511 * the anonymous pages are kept on the inactive list.
6514 * memory ratio inactive anon
6515 * -------------------------------------
6524 static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
6526 unsigned int gb, ratio;
6528 /* Zone size in gigabytes */
6529 gb = zone->managed_pages >> (30 - PAGE_SHIFT);
6531 ratio = int_sqrt(10 * gb);
6535 zone->inactive_ratio = ratio;
6538 static void __meminit setup_per_zone_inactive_ratio(void)
6543 calculate_zone_inactive_ratio(zone);
6547 * Initialise min_free_kbytes.
6549 * For small machines we want it small (128k min). For large machines
6550 * we want it large (64MB max). But it is not linear, because network
6551 * bandwidth does not increase linearly with machine size. We use
6553 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
6554 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6570 int __meminit init_per_zone_wmark_min(void)
6572 unsigned long lowmem_kbytes;
6573 int new_min_free_kbytes;
6575 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
6576 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
6578 if (new_min_free_kbytes > user_min_free_kbytes) {
6579 min_free_kbytes = new_min_free_kbytes;
6580 if (min_free_kbytes < 128)
6581 min_free_kbytes = 128;
6582 if (min_free_kbytes > 65536)
6583 min_free_kbytes = 65536;
6585 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6586 new_min_free_kbytes, user_min_free_kbytes);
6588 setup_per_zone_wmarks();
6589 refresh_zone_stat_thresholds();
6590 setup_per_zone_lowmem_reserve();
6591 setup_per_zone_inactive_ratio();
6594 core_initcall(init_per_zone_wmark_min)
6597 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
6598 * that we can call two helper functions whenever min_free_kbytes
6601 int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
6602 void __user *buffer, size_t *length, loff_t *ppos)
6606 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6611 user_min_free_kbytes = min_free_kbytes;
6612 setup_per_zone_wmarks();
6617 int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write,
6618 void __user *buffer, size_t *length, loff_t *ppos)
6622 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6627 setup_per_zone_wmarks();
6633 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
6634 void __user *buffer, size_t *length, loff_t *ppos)
6639 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6644 zone->min_unmapped_pages = (zone->managed_pages *
6645 sysctl_min_unmapped_ratio) / 100;
6649 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
6650 void __user *buffer, size_t *length, loff_t *ppos)
6655 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6660 zone->min_slab_pages = (zone->managed_pages *
6661 sysctl_min_slab_ratio) / 100;
6667 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6668 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6669 * whenever sysctl_lowmem_reserve_ratio changes.
6671 * The reserve ratio obviously has absolutely no relation with the
6672 * minimum watermarks. The lowmem reserve ratio can only make sense
6673 * if in function of the boot time zone sizes.
6675 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
6676 void __user *buffer, size_t *length, loff_t *ppos)
6678 proc_dointvec_minmax(table, write, buffer, length, ppos);
6679 setup_per_zone_lowmem_reserve();
6684 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
6685 * cpu. It is the fraction of total pages in each zone that a hot per cpu
6686 * pagelist can have before it gets flushed back to buddy allocator.
6688 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
6689 void __user *buffer, size_t *length, loff_t *ppos)
6692 int old_percpu_pagelist_fraction;
6695 mutex_lock(&pcp_batch_high_lock);
6696 old_percpu_pagelist_fraction = percpu_pagelist_fraction;
6698 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
6699 if (!write || ret < 0)
6702 /* Sanity checking to avoid pcp imbalance */
6703 if (percpu_pagelist_fraction &&
6704 percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
6705 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
6711 if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
6714 for_each_populated_zone(zone) {
6717 for_each_possible_cpu(cpu)
6718 pageset_set_high_and_batch(zone,
6719 per_cpu_ptr(zone->pageset, cpu));
6722 mutex_unlock(&pcp_batch_high_lock);
6727 int hashdist = HASHDIST_DEFAULT;
6729 static int __init set_hashdist(char *str)
6733 hashdist = simple_strtoul(str, &str, 0);
6736 __setup("hashdist=", set_hashdist);
6740 * allocate a large system hash table from bootmem
6741 * - it is assumed that the hash table must contain an exact power-of-2
6742 * quantity of entries
6743 * - limit is the number of hash buckets, not the total allocation size
6745 void *__init alloc_large_system_hash(const char *tablename,
6746 unsigned long bucketsize,
6747 unsigned long numentries,
6750 unsigned int *_hash_shift,
6751 unsigned int *_hash_mask,
6752 unsigned long low_limit,
6753 unsigned long high_limit)
6755 unsigned long long max = high_limit;
6756 unsigned long log2qty, size;
6759 /* allow the kernel cmdline to have a say */
6761 /* round applicable memory size up to nearest megabyte */
6762 numentries = nr_kernel_pages;
6764 /* It isn't necessary when PAGE_SIZE >= 1MB */
6765 if (PAGE_SHIFT < 20)
6766 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
6768 /* limit to 1 bucket per 2^scale bytes of low memory */
6769 if (scale > PAGE_SHIFT)
6770 numentries >>= (scale - PAGE_SHIFT);
6772 numentries <<= (PAGE_SHIFT - scale);
6774 /* Make sure we've got at least a 0-order allocation.. */
6775 if (unlikely(flags & HASH_SMALL)) {
6776 /* Makes no sense without HASH_EARLY */
6777 WARN_ON(!(flags & HASH_EARLY));
6778 if (!(numentries >> *_hash_shift)) {
6779 numentries = 1UL << *_hash_shift;
6780 BUG_ON(!numentries);
6782 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
6783 numentries = PAGE_SIZE / bucketsize;
6785 numentries = roundup_pow_of_two(numentries);
6787 /* limit allocation size to 1/16 total memory by default */
6789 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
6790 do_div(max, bucketsize);
6792 max = min(max, 0x80000000ULL);
6794 if (numentries < low_limit)
6795 numentries = low_limit;
6796 if (numentries > max)
6799 log2qty = ilog2(numentries);
6802 size = bucketsize << log2qty;
6803 if (flags & HASH_EARLY)
6804 table = memblock_virt_alloc_nopanic(size, 0);
6806 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
6809 * If bucketsize is not a power-of-two, we may free
6810 * some pages at the end of hash table which
6811 * alloc_pages_exact() automatically does
6813 if (get_order(size) < MAX_ORDER) {
6814 table = alloc_pages_exact(size, GFP_ATOMIC);
6815 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
6818 } while (!table && size > PAGE_SIZE && --log2qty);
6821 panic("Failed to allocate %s hash table\n", tablename);
6823 pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n",
6824 tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size);
6827 *_hash_shift = log2qty;
6829 *_hash_mask = (1 << log2qty) - 1;
6834 /* Return a pointer to the bitmap storing bits affecting a block of pages */
6835 static inline unsigned long *get_pageblock_bitmap(struct page *page,
6838 #ifdef CONFIG_SPARSEMEM
6839 return __pfn_to_section(pfn)->pageblock_flags;
6841 return page_zone(page)->pageblock_flags;
6842 #endif /* CONFIG_SPARSEMEM */
6845 static inline int pfn_to_bitidx(struct page *page, unsigned long pfn)
6847 #ifdef CONFIG_SPARSEMEM
6848 pfn &= (PAGES_PER_SECTION-1);
6849 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
6851 pfn = pfn - round_down(page_zone(page)->zone_start_pfn, pageblock_nr_pages);
6852 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
6853 #endif /* CONFIG_SPARSEMEM */
6857 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
6858 * @page: The page within the block of interest
6859 * @pfn: The target page frame number
6860 * @end_bitidx: The last bit of interest to retrieve
6861 * @mask: mask of bits that the caller is interested in
6863 * Return: pageblock_bits flags
6865 unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
6866 unsigned long end_bitidx,
6869 unsigned long *bitmap;
6870 unsigned long bitidx, word_bitidx;
6873 bitmap = get_pageblock_bitmap(page, pfn);
6874 bitidx = pfn_to_bitidx(page, pfn);
6875 word_bitidx = bitidx / BITS_PER_LONG;
6876 bitidx &= (BITS_PER_LONG-1);
6878 word = bitmap[word_bitidx];
6879 bitidx += end_bitidx;
6880 return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
6884 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
6885 * @page: The page within the block of interest
6886 * @flags: The flags to set
6887 * @pfn: The target page frame number
6888 * @end_bitidx: The last bit of interest
6889 * @mask: mask of bits that the caller is interested in
6891 void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
6893 unsigned long end_bitidx,
6896 unsigned long *bitmap;
6897 unsigned long bitidx, word_bitidx;
6898 unsigned long old_word, word;
6900 BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
6902 bitmap = get_pageblock_bitmap(page, pfn);
6903 bitidx = pfn_to_bitidx(page, pfn);
6904 word_bitidx = bitidx / BITS_PER_LONG;
6905 bitidx &= (BITS_PER_LONG-1);
6907 VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page), pfn), page);
6909 bitidx += end_bitidx;
6910 mask <<= (BITS_PER_LONG - bitidx - 1);
6911 flags <<= (BITS_PER_LONG - bitidx - 1);
6913 word = READ_ONCE(bitmap[word_bitidx]);
6915 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
6916 if (word == old_word)
6923 * This function checks whether pageblock includes unmovable pages or not.
6924 * If @count is not zero, it is okay to include less @count unmovable pages
6926 * PageLRU check without isolation or lru_lock could race so that
6927 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6928 * expect this function should be exact.
6930 bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
6931 bool skip_hwpoisoned_pages)
6933 unsigned long pfn, iter, found;
6937 * For avoiding noise data, lru_add_drain_all() should be called
6938 * If ZONE_MOVABLE, the zone never contains unmovable pages
6940 if (zone_idx(zone) == ZONE_MOVABLE)
6942 mt = get_pageblock_migratetype(page);
6943 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
6946 pfn = page_to_pfn(page);
6947 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6948 unsigned long check = pfn + iter;
6950 if (!pfn_valid_within(check))
6953 page = pfn_to_page(check);
6956 * Hugepages are not in LRU lists, but they're movable.
6957 * We need not scan over tail pages bacause we don't
6958 * handle each tail page individually in migration.
6960 if (PageHuge(page)) {
6961 iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6966 * We can't use page_count without pin a page
6967 * because another CPU can free compound page.
6968 * This check already skips compound tails of THP
6969 * because their page->_refcount is zero at all time.
6971 if (!page_ref_count(page)) {
6972 if (PageBuddy(page))
6973 iter += (1 << page_order(page)) - 1;
6978 * The HWPoisoned page may be not in buddy system, and
6979 * page_count() is not 0.
6981 if (skip_hwpoisoned_pages && PageHWPoison(page))
6987 * If there are RECLAIMABLE pages, we need to check
6988 * it. But now, memory offline itself doesn't call
6989 * shrink_node_slabs() and it still to be fixed.
6992 * If the page is not RAM, page_count()should be 0.
6993 * we don't need more check. This is an _used_ not-movable page.
6995 * The problematic thing here is PG_reserved pages. PG_reserved
6996 * is set to both of a memory hole page and a _used_ kernel
7005 bool is_pageblock_removable_nolock(struct page *page)
7011 * We have to be careful here because we are iterating over memory
7012 * sections which are not zone aware so we might end up outside of
7013 * the zone but still within the section.
7014 * We have to take care about the node as well. If the node is offline
7015 * its NODE_DATA will be NULL - see page_zone.
7017 if (!node_online(page_to_nid(page)))
7020 zone = page_zone(page);
7021 pfn = page_to_pfn(page);
7022 if (!zone_spans_pfn(zone, pfn))
7025 return !has_unmovable_pages(zone, page, 0, true);
7028 #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
7030 static unsigned long pfn_max_align_down(unsigned long pfn)
7032 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
7033 pageblock_nr_pages) - 1);
7036 static unsigned long pfn_max_align_up(unsigned long pfn)
7038 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
7039 pageblock_nr_pages));
7042 /* [start, end) must belong to a single zone. */
7043 static int __alloc_contig_migrate_range(struct compact_control *cc,
7044 unsigned long start, unsigned long end)
7046 /* This function is based on compact_zone() from compaction.c. */
7047 unsigned long nr_reclaimed;
7048 unsigned long pfn = start;
7049 unsigned int tries = 0;
7054 while (pfn < end || !list_empty(&cc->migratepages)) {
7055 if (fatal_signal_pending(current)) {
7060 if (list_empty(&cc->migratepages)) {
7061 cc->nr_migratepages = 0;
7062 pfn = isolate_migratepages_range(cc, pfn, end);
7068 } else if (++tries == 5) {
7069 ret = ret < 0 ? ret : -EBUSY;
7073 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
7075 cc->nr_migratepages -= nr_reclaimed;
7077 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
7078 NULL, 0, cc->mode, MR_CMA);
7081 putback_movable_pages(&cc->migratepages);
7088 * alloc_contig_range() -- tries to allocate given range of pages
7089 * @start: start PFN to allocate
7090 * @end: one-past-the-last PFN to allocate
7091 * @migratetype: migratetype of the underlaying pageblocks (either
7092 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
7093 * in range must have the same migratetype and it must
7094 * be either of the two.
7096 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
7097 * aligned, however it's the caller's responsibility to guarantee that
7098 * we are the only thread that changes migrate type of pageblocks the
7101 * The PFN range must belong to a single zone.
7103 * Returns zero on success or negative error code. On success all
7104 * pages which PFN is in [start, end) are allocated for the caller and
7105 * need to be freed with free_contig_range().
7107 int alloc_contig_range(unsigned long start, unsigned long end,
7108 unsigned migratetype)
7110 unsigned long outer_start, outer_end;
7114 struct compact_control cc = {
7115 .nr_migratepages = 0,
7117 .zone = page_zone(pfn_to_page(start)),
7118 .mode = MIGRATE_SYNC,
7119 .ignore_skip_hint = true,
7121 INIT_LIST_HEAD(&cc.migratepages);
7124 * What we do here is we mark all pageblocks in range as
7125 * MIGRATE_ISOLATE. Because pageblock and max order pages may
7126 * have different sizes, and due to the way page allocator
7127 * work, we align the range to biggest of the two pages so
7128 * that page allocator won't try to merge buddies from
7129 * different pageblocks and change MIGRATE_ISOLATE to some
7130 * other migration type.
7132 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
7133 * migrate the pages from an unaligned range (ie. pages that
7134 * we are interested in). This will put all the pages in
7135 * range back to page allocator as MIGRATE_ISOLATE.
7137 * When this is done, we take the pages in range from page
7138 * allocator removing them from the buddy system. This way
7139 * page allocator will never consider using them.
7141 * This lets us mark the pageblocks back as
7142 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
7143 * aligned range but not in the unaligned, original range are
7144 * put back to page allocator so that buddy can use them.
7147 ret = start_isolate_page_range(pfn_max_align_down(start),
7148 pfn_max_align_up(end), migratetype,
7154 * In case of -EBUSY, we'd like to know which page causes problem.
7155 * So, just fall through. We will check it in test_pages_isolated().
7157 ret = __alloc_contig_migrate_range(&cc, start, end);
7158 if (ret && ret != -EBUSY)
7162 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
7163 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
7164 * more, all pages in [start, end) are free in page allocator.
7165 * What we are going to do is to allocate all pages from
7166 * [start, end) (that is remove them from page allocator).
7168 * The only problem is that pages at the beginning and at the
7169 * end of interesting range may be not aligned with pages that
7170 * page allocator holds, ie. they can be part of higher order
7171 * pages. Because of this, we reserve the bigger range and
7172 * once this is done free the pages we are not interested in.
7174 * We don't have to hold zone->lock here because the pages are
7175 * isolated thus they won't get removed from buddy.
7178 lru_add_drain_all();
7179 drain_all_pages(cc.zone);
7182 outer_start = start;
7183 while (!PageBuddy(pfn_to_page(outer_start))) {
7184 if (++order >= MAX_ORDER) {
7185 outer_start = start;
7188 outer_start &= ~0UL << order;
7191 if (outer_start != start) {
7192 order = page_order(pfn_to_page(outer_start));
7195 * outer_start page could be small order buddy page and
7196 * it doesn't include start page. Adjust outer_start
7197 * in this case to report failed page properly
7198 * on tracepoint in test_pages_isolated()
7200 if (outer_start + (1UL << order) <= start)
7201 outer_start = start;
7204 /* Make sure the range is really isolated. */
7205 if (test_pages_isolated(outer_start, end, false)) {
7206 pr_info("%s: [%lx, %lx) PFNs busy\n",
7207 __func__, outer_start, end);
7212 /* Grab isolated pages from freelists. */
7213 outer_end = isolate_freepages_range(&cc, outer_start, end);
7219 /* Free head and tail (if any) */
7220 if (start != outer_start)
7221 free_contig_range(outer_start, start - outer_start);
7222 if (end != outer_end)
7223 free_contig_range(end, outer_end - end);
7226 undo_isolate_page_range(pfn_max_align_down(start),
7227 pfn_max_align_up(end), migratetype);
7231 void free_contig_range(unsigned long pfn, unsigned nr_pages)
7233 unsigned int count = 0;
7235 for (; nr_pages--; pfn++) {
7236 struct page *page = pfn_to_page(pfn);
7238 count += page_count(page) != 1;
7241 WARN(count != 0, "%d pages are still in use!\n", count);
7245 #ifdef CONFIG_MEMORY_HOTPLUG
7247 * The zone indicated has a new number of managed_pages; batch sizes and percpu
7248 * page high values need to be recalulated.
7250 void __meminit zone_pcp_update(struct zone *zone)
7253 mutex_lock(&pcp_batch_high_lock);
7254 for_each_possible_cpu(cpu)
7255 pageset_set_high_and_batch(zone,
7256 per_cpu_ptr(zone->pageset, cpu));
7257 mutex_unlock(&pcp_batch_high_lock);
7261 void zone_pcp_reset(struct zone *zone)
7263 unsigned long flags;
7265 struct per_cpu_pageset *pset;
7267 /* avoid races with drain_pages() */
7268 local_irq_save(flags);
7269 if (zone->pageset != &boot_pageset) {
7270 for_each_online_cpu(cpu) {
7271 pset = per_cpu_ptr(zone->pageset, cpu);
7272 drain_zonestat(zone, pset);
7274 free_percpu(zone->pageset);
7275 zone->pageset = &boot_pageset;
7277 local_irq_restore(flags);
7280 #ifdef CONFIG_MEMORY_HOTREMOVE
7282 * All pages in the range must be in a single zone and isolated
7283 * before calling this.
7286 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
7290 unsigned int order, i;
7292 unsigned long flags;
7293 /* find the first valid pfn */
7294 for (pfn = start_pfn; pfn < end_pfn; pfn++)
7299 zone = page_zone(pfn_to_page(pfn));
7300 spin_lock_irqsave(&zone->lock, flags);
7302 while (pfn < end_pfn) {
7303 if (!pfn_valid(pfn)) {
7307 page = pfn_to_page(pfn);
7309 * The HWPoisoned page may be not in buddy system, and
7310 * page_count() is not 0.
7312 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
7314 SetPageReserved(page);
7318 BUG_ON(page_count(page));
7319 BUG_ON(!PageBuddy(page));
7320 order = page_order(page);
7321 #ifdef CONFIG_DEBUG_VM
7322 pr_info("remove from free list %lx %d %lx\n",
7323 pfn, 1 << order, end_pfn);
7325 list_del(&page->lru);
7326 rmv_page_order(page);
7327 zone->free_area[order].nr_free--;
7328 for (i = 0; i < (1 << order); i++)
7329 SetPageReserved((page+i));
7330 pfn += (1 << order);
7332 spin_unlock_irqrestore(&zone->lock, flags);
7336 bool is_free_buddy_page(struct page *page)
7338 struct zone *zone = page_zone(page);
7339 unsigned long pfn = page_to_pfn(page);
7340 unsigned long flags;
7343 spin_lock_irqsave(&zone->lock, flags);
7344 for (order = 0; order < MAX_ORDER; order++) {
7345 struct page *page_head = page - (pfn & ((1 << order) - 1));
7347 if (PageBuddy(page_head) && page_order(page_head) >= order)
7350 spin_unlock_irqrestore(&zone->lock, flags);
7352 return order < MAX_ORDER;