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++;
787 static inline int free_pages_check(struct page *page)
789 const char *bad_reason = NULL;
790 unsigned long bad_flags = 0;
792 if (unlikely(atomic_read(&page->_mapcount) != -1))
793 bad_reason = "nonzero mapcount";
794 if (unlikely(page->mapping != NULL))
795 bad_reason = "non-NULL mapping";
796 if (unlikely(page_ref_count(page) != 0))
797 bad_reason = "nonzero _refcount";
798 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_FREE)) {
799 bad_reason = "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
800 bad_flags = PAGE_FLAGS_CHECK_AT_FREE;
803 if (unlikely(page->mem_cgroup))
804 bad_reason = "page still charged to cgroup";
806 if (unlikely(bad_reason)) {
807 bad_page(page, bad_reason, bad_flags);
810 page_cpupid_reset_last(page);
811 if (page->flags & PAGE_FLAGS_CHECK_AT_PREP)
812 page->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
817 * Frees a number of pages from the PCP lists
818 * Assumes all pages on list are in same zone, and of same order.
819 * count is the number of pages to free.
821 * If the zone was previously in an "all pages pinned" state then look to
822 * see if this freeing clears that state.
824 * And clear the zone's pages_scanned counter, to hold off the "all pages are
825 * pinned" detection logic.
827 static void free_pcppages_bulk(struct zone *zone, int count,
828 struct per_cpu_pages *pcp)
833 unsigned long nr_scanned;
835 spin_lock(&zone->lock);
836 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
838 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
842 struct list_head *list;
845 * Remove pages from lists in a round-robin fashion. A
846 * batch_free count is maintained that is incremented when an
847 * empty list is encountered. This is so more pages are freed
848 * off fuller lists instead of spinning excessively around empty
853 if (++migratetype == MIGRATE_PCPTYPES)
855 list = &pcp->lists[migratetype];
856 } while (list_empty(list));
858 /* This is the only non-empty list. Free them all. */
859 if (batch_free == MIGRATE_PCPTYPES)
860 batch_free = to_free;
863 int mt; /* migratetype of the to-be-freed page */
865 page = list_last_entry(list, struct page, lru);
866 /* must delete as __free_one_page list manipulates */
867 list_del(&page->lru);
869 mt = get_pcppage_migratetype(page);
870 /* MIGRATE_ISOLATE page should not go to pcplists */
871 VM_BUG_ON_PAGE(is_migrate_isolate(mt), page);
872 /* Pageblock could have been isolated meanwhile */
873 if (unlikely(has_isolate_pageblock(zone)))
874 mt = get_pageblock_migratetype(page);
876 __free_one_page(page, page_to_pfn(page), zone, 0, mt);
877 trace_mm_page_pcpu_drain(page, 0, mt);
878 } while (--to_free && --batch_free && !list_empty(list));
880 spin_unlock(&zone->lock);
883 static void free_one_page(struct zone *zone,
884 struct page *page, unsigned long pfn,
888 unsigned long nr_scanned;
889 spin_lock(&zone->lock);
890 nr_scanned = zone_page_state(zone, NR_PAGES_SCANNED);
892 __mod_zone_page_state(zone, NR_PAGES_SCANNED, -nr_scanned);
894 if (unlikely(has_isolate_pageblock(zone) ||
895 is_migrate_isolate(migratetype))) {
896 migratetype = get_pfnblock_migratetype(page, pfn);
898 __free_one_page(page, pfn, zone, order, migratetype);
899 spin_unlock(&zone->lock);
902 static int free_tail_pages_check(struct page *head_page, struct page *page)
907 * We rely page->lru.next never has bit 0 set, unless the page
908 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
910 BUILD_BUG_ON((unsigned long)LIST_POISON1 & 1);
912 if (!IS_ENABLED(CONFIG_DEBUG_VM)) {
916 switch (page - head_page) {
918 /* the first tail page: ->mapping is compound_mapcount() */
919 if (unlikely(compound_mapcount(page))) {
920 bad_page(page, "nonzero compound_mapcount", 0);
926 * the second tail page: ->mapping is
927 * page_deferred_list().next -- ignore value.
931 if (page->mapping != TAIL_MAPPING) {
932 bad_page(page, "corrupted mapping in tail page", 0);
937 if (unlikely(!PageTail(page))) {
938 bad_page(page, "PageTail not set", 0);
941 if (unlikely(compound_head(page) != head_page)) {
942 bad_page(page, "compound_head not consistent", 0);
947 page->mapping = NULL;
948 clear_compound_head(page);
952 static void __meminit __init_single_page(struct page *page, unsigned long pfn,
953 unsigned long zone, int nid)
955 set_page_links(page, zone, nid, pfn);
956 init_page_count(page);
957 page_mapcount_reset(page);
958 page_cpupid_reset_last(page);
960 INIT_LIST_HEAD(&page->lru);
961 #ifdef WANT_PAGE_VIRTUAL
962 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
963 if (!is_highmem_idx(zone))
964 set_page_address(page, __va(pfn << PAGE_SHIFT));
968 static void __meminit __init_single_pfn(unsigned long pfn, unsigned long zone,
971 return __init_single_page(pfn_to_page(pfn), pfn, zone, nid);
974 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
975 static void init_reserved_page(unsigned long pfn)
980 if (!early_page_uninitialised(pfn))
983 nid = early_pfn_to_nid(pfn);
984 pgdat = NODE_DATA(nid);
986 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
987 struct zone *zone = &pgdat->node_zones[zid];
989 if (pfn >= zone->zone_start_pfn && pfn < zone_end_pfn(zone))
992 __init_single_pfn(pfn, zid, nid);
995 static inline void init_reserved_page(unsigned long pfn)
998 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1001 * Initialised pages do not have PageReserved set. This function is
1002 * called for each range allocated by the bootmem allocator and
1003 * marks the pages PageReserved. The remaining valid pages are later
1004 * sent to the buddy page allocator.
1006 void __meminit reserve_bootmem_region(unsigned long start, unsigned long end)
1008 unsigned long start_pfn = PFN_DOWN(start);
1009 unsigned long end_pfn = PFN_UP(end);
1011 for (; start_pfn < end_pfn; start_pfn++) {
1012 if (pfn_valid(start_pfn)) {
1013 struct page *page = pfn_to_page(start_pfn);
1015 init_reserved_page(start_pfn);
1017 /* Avoid false-positive PageTail() */
1018 INIT_LIST_HEAD(&page->lru);
1020 SetPageReserved(page);
1025 static bool free_pages_prepare(struct page *page, unsigned int order)
1027 bool compound = PageCompound(page);
1030 VM_BUG_ON_PAGE(PageTail(page), page);
1031 VM_BUG_ON_PAGE(compound && compound_order(page) != order, page);
1033 trace_mm_page_free(page, order);
1034 kmemcheck_free_shadow(page, order);
1035 kasan_free_pages(page, order);
1038 page->mapping = NULL;
1039 bad += free_pages_check(page);
1040 for (i = 1; i < (1 << order); i++) {
1042 bad += free_tail_pages_check(page, page + i);
1043 bad += free_pages_check(page + i);
1048 reset_page_owner(page, order);
1050 if (!PageHighMem(page)) {
1051 debug_check_no_locks_freed(page_address(page),
1052 PAGE_SIZE << order);
1053 debug_check_no_obj_freed(page_address(page),
1054 PAGE_SIZE << order);
1056 arch_free_page(page, order);
1057 kernel_poison_pages(page, 1 << order, 0);
1058 kernel_map_pages(page, 1 << order, 0);
1063 static void __free_pages_ok(struct page *page, unsigned int order)
1065 unsigned long flags;
1067 unsigned long pfn = page_to_pfn(page);
1069 if (!free_pages_prepare(page, order))
1072 migratetype = get_pfnblock_migratetype(page, pfn);
1073 local_irq_save(flags);
1074 __count_vm_events(PGFREE, 1 << order);
1075 free_one_page(page_zone(page), page, pfn, order, migratetype);
1076 local_irq_restore(flags);
1079 static void __init __free_pages_boot_core(struct page *page, unsigned int order)
1081 unsigned int nr_pages = 1 << order;
1082 struct page *p = page;
1086 for (loop = 0; loop < (nr_pages - 1); loop++, p++) {
1088 __ClearPageReserved(p);
1089 set_page_count(p, 0);
1091 __ClearPageReserved(p);
1092 set_page_count(p, 0);
1094 page_zone(page)->managed_pages += nr_pages;
1095 set_page_refcounted(page);
1096 __free_pages(page, order);
1099 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1100 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1102 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata;
1104 int __meminit early_pfn_to_nid(unsigned long pfn)
1106 static DEFINE_SPINLOCK(early_pfn_lock);
1109 spin_lock(&early_pfn_lock);
1110 nid = __early_pfn_to_nid(pfn, &early_pfnnid_cache);
1113 spin_unlock(&early_pfn_lock);
1119 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1120 static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1121 struct mminit_pfnnid_cache *state)
1125 nid = __early_pfn_to_nid(pfn, state);
1126 if (nid >= 0 && nid != node)
1131 /* Only safe to use early in boot when initialisation is single-threaded */
1132 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1134 return meminit_pfn_in_nid(pfn, node, &early_pfnnid_cache);
1139 static inline bool __meminit early_pfn_in_nid(unsigned long pfn, int node)
1143 static inline bool __meminit meminit_pfn_in_nid(unsigned long pfn, int node,
1144 struct mminit_pfnnid_cache *state)
1151 void __init __free_pages_bootmem(struct page *page, unsigned long pfn,
1154 if (early_page_uninitialised(pfn))
1156 return __free_pages_boot_core(page, order);
1160 * Check that the whole (or subset of) a pageblock given by the interval of
1161 * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1162 * with the migration of free compaction scanner. The scanners then need to
1163 * use only pfn_valid_within() check for arches that allow holes within
1166 * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1168 * It's possible on some configurations to have a setup like node0 node1 node0
1169 * i.e. it's possible that all pages within a zones range of pages do not
1170 * belong to a single zone. We assume that a border between node0 and node1
1171 * can occur within a single pageblock, but not a node0 node1 node0
1172 * interleaving within a single pageblock. It is therefore sufficient to check
1173 * the first and last page of a pageblock and avoid checking each individual
1174 * page in a pageblock.
1176 struct page *__pageblock_pfn_to_page(unsigned long start_pfn,
1177 unsigned long end_pfn, struct zone *zone)
1179 struct page *start_page;
1180 struct page *end_page;
1182 /* end_pfn is one past the range we are checking */
1185 if (!pfn_valid(start_pfn) || !pfn_valid(end_pfn))
1188 start_page = pfn_to_page(start_pfn);
1190 if (page_zone(start_page) != zone)
1193 end_page = pfn_to_page(end_pfn);
1195 /* This gives a shorter code than deriving page_zone(end_page) */
1196 if (page_zone_id(start_page) != page_zone_id(end_page))
1202 void set_zone_contiguous(struct zone *zone)
1204 unsigned long block_start_pfn = zone->zone_start_pfn;
1205 unsigned long block_end_pfn;
1207 block_end_pfn = ALIGN(block_start_pfn + 1, pageblock_nr_pages);
1208 for (; block_start_pfn < zone_end_pfn(zone);
1209 block_start_pfn = block_end_pfn,
1210 block_end_pfn += pageblock_nr_pages) {
1212 block_end_pfn = min(block_end_pfn, zone_end_pfn(zone));
1214 if (!__pageblock_pfn_to_page(block_start_pfn,
1215 block_end_pfn, zone))
1219 /* We confirm that there is no hole */
1220 zone->contiguous = true;
1223 void clear_zone_contiguous(struct zone *zone)
1225 zone->contiguous = false;
1228 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1229 static void __init deferred_free_range(struct page *page,
1230 unsigned long pfn, int nr_pages)
1237 /* Free a large naturally-aligned chunk if possible */
1238 if (nr_pages == MAX_ORDER_NR_PAGES &&
1239 (pfn & (MAX_ORDER_NR_PAGES-1)) == 0) {
1240 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
1241 __free_pages_boot_core(page, MAX_ORDER-1);
1245 for (i = 0; i < nr_pages; i++, page++)
1246 __free_pages_boot_core(page, 0);
1249 /* Completion tracking for deferred_init_memmap() threads */
1250 static atomic_t pgdat_init_n_undone __initdata;
1251 static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
1253 static inline void __init pgdat_init_report_one_done(void)
1255 if (atomic_dec_and_test(&pgdat_init_n_undone))
1256 complete(&pgdat_init_all_done_comp);
1259 /* Initialise remaining memory on a node */
1260 static int __init deferred_init_memmap(void *data)
1262 pg_data_t *pgdat = data;
1263 int nid = pgdat->node_id;
1264 struct mminit_pfnnid_cache nid_init_state = { };
1265 unsigned long start = jiffies;
1266 unsigned long nr_pages = 0;
1267 unsigned long walk_start, walk_end;
1270 unsigned long first_init_pfn = pgdat->first_deferred_pfn;
1271 const struct cpumask *cpumask = cpumask_of_node(pgdat->node_id);
1273 if (first_init_pfn == ULONG_MAX) {
1274 pgdat_init_report_one_done();
1278 /* Bind memory initialisation thread to a local node if possible */
1279 if (!cpumask_empty(cpumask))
1280 set_cpus_allowed_ptr(current, cpumask);
1282 /* Sanity check boundaries */
1283 BUG_ON(pgdat->first_deferred_pfn < pgdat->node_start_pfn);
1284 BUG_ON(pgdat->first_deferred_pfn > pgdat_end_pfn(pgdat));
1285 pgdat->first_deferred_pfn = ULONG_MAX;
1287 /* Only the highest zone is deferred so find it */
1288 for (zid = 0; zid < MAX_NR_ZONES; zid++) {
1289 zone = pgdat->node_zones + zid;
1290 if (first_init_pfn < zone_end_pfn(zone))
1294 for_each_mem_pfn_range(i, nid, &walk_start, &walk_end, NULL) {
1295 unsigned long pfn, end_pfn;
1296 struct page *page = NULL;
1297 struct page *free_base_page = NULL;
1298 unsigned long free_base_pfn = 0;
1301 end_pfn = min(walk_end, zone_end_pfn(zone));
1302 pfn = first_init_pfn;
1303 if (pfn < walk_start)
1305 if (pfn < zone->zone_start_pfn)
1306 pfn = zone->zone_start_pfn;
1308 for (; pfn < end_pfn; pfn++) {
1309 if (!pfn_valid_within(pfn))
1313 * Ensure pfn_valid is checked every
1314 * MAX_ORDER_NR_PAGES for memory holes
1316 if ((pfn & (MAX_ORDER_NR_PAGES - 1)) == 0) {
1317 if (!pfn_valid(pfn)) {
1323 if (!meminit_pfn_in_nid(pfn, nid, &nid_init_state)) {
1328 /* Minimise pfn page lookups and scheduler checks */
1329 if (page && (pfn & (MAX_ORDER_NR_PAGES - 1)) != 0) {
1332 nr_pages += nr_to_free;
1333 deferred_free_range(free_base_page,
1334 free_base_pfn, nr_to_free);
1335 free_base_page = NULL;
1336 free_base_pfn = nr_to_free = 0;
1338 page = pfn_to_page(pfn);
1343 VM_BUG_ON(page_zone(page) != zone);
1347 __init_single_page(page, pfn, zid, nid);
1348 if (!free_base_page) {
1349 free_base_page = page;
1350 free_base_pfn = pfn;
1355 /* Where possible, batch up pages for a single free */
1358 /* Free the current block of pages to allocator */
1359 nr_pages += nr_to_free;
1360 deferred_free_range(free_base_page, free_base_pfn,
1362 free_base_page = NULL;
1363 free_base_pfn = nr_to_free = 0;
1366 first_init_pfn = max(end_pfn, first_init_pfn);
1369 /* Sanity check that the next zone really is unpopulated */
1370 WARN_ON(++zid < MAX_NR_ZONES && populated_zone(++zone));
1372 pr_info("node %d initialised, %lu pages in %ums\n", nid, nr_pages,
1373 jiffies_to_msecs(jiffies - start));
1375 pgdat_init_report_one_done();
1378 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1380 void __init page_alloc_init_late(void)
1384 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1387 /* There will be num_node_state(N_MEMORY) threads */
1388 atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
1389 for_each_node_state(nid, N_MEMORY) {
1390 kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
1393 /* Block until all are initialised */
1394 wait_for_completion(&pgdat_init_all_done_comp);
1396 /* Reinit limits that are based on free pages after the kernel is up */
1397 files_maxfiles_init();
1400 for_each_populated_zone(zone)
1401 set_zone_contiguous(zone);
1405 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1406 void __init init_cma_reserved_pageblock(struct page *page)
1408 unsigned i = pageblock_nr_pages;
1409 struct page *p = page;
1412 __ClearPageReserved(p);
1413 set_page_count(p, 0);
1416 set_pageblock_migratetype(page, MIGRATE_CMA);
1418 if (pageblock_order >= MAX_ORDER) {
1419 i = pageblock_nr_pages;
1422 set_page_refcounted(p);
1423 __free_pages(p, MAX_ORDER - 1);
1424 p += MAX_ORDER_NR_PAGES;
1425 } while (i -= MAX_ORDER_NR_PAGES);
1427 set_page_refcounted(page);
1428 __free_pages(page, pageblock_order);
1431 adjust_managed_page_count(page, pageblock_nr_pages);
1436 * The order of subdivision here is critical for the IO subsystem.
1437 * Please do not alter this order without good reasons and regression
1438 * testing. Specifically, as large blocks of memory are subdivided,
1439 * the order in which smaller blocks are delivered depends on the order
1440 * they're subdivided in this function. This is the primary factor
1441 * influencing the order in which pages are delivered to the IO
1442 * subsystem according to empirical testing, and this is also justified
1443 * by considering the behavior of a buddy system containing a single
1444 * large block of memory acted on by a series of small allocations.
1445 * This behavior is a critical factor in sglist merging's success.
1449 static inline void expand(struct zone *zone, struct page *page,
1450 int low, int high, struct free_area *area,
1453 unsigned long size = 1 << high;
1455 while (high > low) {
1459 VM_BUG_ON_PAGE(bad_range(zone, &page[size]), &page[size]);
1461 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC) &&
1462 debug_guardpage_enabled() &&
1463 high < debug_guardpage_minorder()) {
1465 * Mark as guard pages (or page), that will allow to
1466 * merge back to allocator when buddy will be freed.
1467 * Corresponding page table entries will not be touched,
1468 * pages will stay not present in virtual address space
1470 set_page_guard(zone, &page[size], high, migratetype);
1473 list_add(&page[size].lru, &area->free_list[migratetype]);
1475 set_page_order(&page[size], high);
1480 * This page is about to be returned from the page allocator
1482 static inline int check_new_page(struct page *page)
1484 const char *bad_reason = NULL;
1485 unsigned long bad_flags = 0;
1487 if (unlikely(atomic_read(&page->_mapcount) != -1))
1488 bad_reason = "nonzero mapcount";
1489 if (unlikely(page->mapping != NULL))
1490 bad_reason = "non-NULL mapping";
1491 if (unlikely(page_ref_count(page) != 0))
1492 bad_reason = "nonzero _count";
1493 if (unlikely(page->flags & __PG_HWPOISON)) {
1494 bad_reason = "HWPoisoned (hardware-corrupted)";
1495 bad_flags = __PG_HWPOISON;
1497 if (unlikely(page->flags & PAGE_FLAGS_CHECK_AT_PREP)) {
1498 bad_reason = "PAGE_FLAGS_CHECK_AT_PREP flag set";
1499 bad_flags = PAGE_FLAGS_CHECK_AT_PREP;
1502 if (unlikely(page->mem_cgroup))
1503 bad_reason = "page still charged to cgroup";
1505 if (unlikely(bad_reason)) {
1506 bad_page(page, bad_reason, bad_flags);
1512 static inline bool free_pages_prezeroed(bool poisoned)
1514 return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO) &&
1515 page_poisoning_enabled() && poisoned;
1518 static int prep_new_page(struct page *page, unsigned int order, gfp_t gfp_flags,
1522 bool poisoned = true;
1524 for (i = 0; i < (1 << order); i++) {
1525 struct page *p = page + i;
1526 if (unlikely(check_new_page(p)))
1529 poisoned &= page_is_poisoned(p);
1532 set_page_private(page, 0);
1533 set_page_refcounted(page);
1535 arch_alloc_page(page, order);
1536 kernel_map_pages(page, 1 << order, 1);
1537 kernel_poison_pages(page, 1 << order, 1);
1538 kasan_alloc_pages(page, order);
1540 if (!free_pages_prezeroed(poisoned) && (gfp_flags & __GFP_ZERO))
1541 for (i = 0; i < (1 << order); i++)
1542 clear_highpage(page + i);
1544 if (order && (gfp_flags & __GFP_COMP))
1545 prep_compound_page(page, order);
1547 set_page_owner(page, order, gfp_flags);
1550 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1551 * allocate the page. The expectation is that the caller is taking
1552 * steps that will free more memory. The caller should avoid the page
1553 * being used for !PFMEMALLOC purposes.
1555 if (alloc_flags & ALLOC_NO_WATERMARKS)
1556 set_page_pfmemalloc(page);
1558 clear_page_pfmemalloc(page);
1564 * Go through the free lists for the given migratetype and remove
1565 * the smallest available page from the freelists
1568 struct page *__rmqueue_smallest(struct zone *zone, unsigned int order,
1571 unsigned int current_order;
1572 struct free_area *area;
1575 /* Find a page of the appropriate size in the preferred list */
1576 for (current_order = order; current_order < MAX_ORDER; ++current_order) {
1577 area = &(zone->free_area[current_order]);
1578 page = list_first_entry_or_null(&area->free_list[migratetype],
1582 list_del(&page->lru);
1583 rmv_page_order(page);
1585 expand(zone, page, order, current_order, area, migratetype);
1586 set_pcppage_migratetype(page, migratetype);
1595 * This array describes the order lists are fallen back to when
1596 * the free lists for the desirable migrate type are depleted
1598 static int fallbacks[MIGRATE_TYPES][4] = {
1599 [MIGRATE_UNMOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
1600 [MIGRATE_RECLAIMABLE] = { MIGRATE_UNMOVABLE, MIGRATE_MOVABLE, MIGRATE_TYPES },
1601 [MIGRATE_MOVABLE] = { MIGRATE_RECLAIMABLE, MIGRATE_UNMOVABLE, MIGRATE_TYPES },
1603 [MIGRATE_CMA] = { MIGRATE_TYPES }, /* Never used */
1605 #ifdef CONFIG_MEMORY_ISOLATION
1606 [MIGRATE_ISOLATE] = { MIGRATE_TYPES }, /* Never used */
1611 static struct page *__rmqueue_cma_fallback(struct zone *zone,
1614 return __rmqueue_smallest(zone, order, MIGRATE_CMA);
1617 static inline struct page *__rmqueue_cma_fallback(struct zone *zone,
1618 unsigned int order) { return NULL; }
1622 * Move the free pages in a range to the free lists of the requested type.
1623 * Note that start_page and end_pages are not aligned on a pageblock
1624 * boundary. If alignment is required, use move_freepages_block()
1626 int move_freepages(struct zone *zone,
1627 struct page *start_page, struct page *end_page,
1632 int pages_moved = 0;
1634 #ifndef CONFIG_HOLES_IN_ZONE
1636 * page_zone is not safe to call in this context when
1637 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1638 * anyway as we check zone boundaries in move_freepages_block().
1639 * Remove at a later date when no bug reports exist related to
1640 * grouping pages by mobility
1642 VM_BUG_ON(page_zone(start_page) != page_zone(end_page));
1645 for (page = start_page; page <= end_page;) {
1646 /* Make sure we are not inadvertently changing nodes */
1647 VM_BUG_ON_PAGE(page_to_nid(page) != zone_to_nid(zone), page);
1649 if (!pfn_valid_within(page_to_pfn(page))) {
1654 if (!PageBuddy(page)) {
1659 order = page_order(page);
1660 list_move(&page->lru,
1661 &zone->free_area[order].free_list[migratetype]);
1663 pages_moved += 1 << order;
1669 int move_freepages_block(struct zone *zone, struct page *page,
1672 unsigned long start_pfn, end_pfn;
1673 struct page *start_page, *end_page;
1675 start_pfn = page_to_pfn(page);
1676 start_pfn = start_pfn & ~(pageblock_nr_pages-1);
1677 start_page = pfn_to_page(start_pfn);
1678 end_page = start_page + pageblock_nr_pages - 1;
1679 end_pfn = start_pfn + pageblock_nr_pages - 1;
1681 /* Do not cross zone boundaries */
1682 if (!zone_spans_pfn(zone, start_pfn))
1684 if (!zone_spans_pfn(zone, end_pfn))
1687 return move_freepages(zone, start_page, end_page, migratetype);
1690 static void change_pageblock_range(struct page *pageblock_page,
1691 int start_order, int migratetype)
1693 int nr_pageblocks = 1 << (start_order - pageblock_order);
1695 while (nr_pageblocks--) {
1696 set_pageblock_migratetype(pageblock_page, migratetype);
1697 pageblock_page += pageblock_nr_pages;
1702 * When we are falling back to another migratetype during allocation, try to
1703 * steal extra free pages from the same pageblocks to satisfy further
1704 * allocations, instead of polluting multiple pageblocks.
1706 * If we are stealing a relatively large buddy page, it is likely there will
1707 * be more free pages in the pageblock, so try to steal them all. For
1708 * reclaimable and unmovable allocations, we steal regardless of page size,
1709 * as fragmentation caused by those allocations polluting movable pageblocks
1710 * is worse than movable allocations stealing from unmovable and reclaimable
1713 static bool can_steal_fallback(unsigned int order, int start_mt)
1716 * Leaving this order check is intended, although there is
1717 * relaxed order check in next check. The reason is that
1718 * we can actually steal whole pageblock if this condition met,
1719 * but, below check doesn't guarantee it and that is just heuristic
1720 * so could be changed anytime.
1722 if (order >= pageblock_order)
1725 if (order >= pageblock_order / 2 ||
1726 start_mt == MIGRATE_RECLAIMABLE ||
1727 start_mt == MIGRATE_UNMOVABLE ||
1728 page_group_by_mobility_disabled)
1735 * This function implements actual steal behaviour. If order is large enough,
1736 * we can steal whole pageblock. If not, we first move freepages in this
1737 * pageblock and check whether half of pages are moved or not. If half of
1738 * pages are moved, we can change migratetype of pageblock and permanently
1739 * use it's pages as requested migratetype in the future.
1741 static void steal_suitable_fallback(struct zone *zone, struct page *page,
1744 unsigned int current_order = page_order(page);
1747 /* Take ownership for orders >= pageblock_order */
1748 if (current_order >= pageblock_order) {
1749 change_pageblock_range(page, current_order, start_type);
1753 pages = move_freepages_block(zone, page, start_type);
1755 /* Claim the whole block if over half of it is free */
1756 if (pages >= (1 << (pageblock_order-1)) ||
1757 page_group_by_mobility_disabled)
1758 set_pageblock_migratetype(page, start_type);
1762 * Check whether there is a suitable fallback freepage with requested order.
1763 * If only_stealable is true, this function returns fallback_mt only if
1764 * we can steal other freepages all together. This would help to reduce
1765 * fragmentation due to mixed migratetype pages in one pageblock.
1767 int find_suitable_fallback(struct free_area *area, unsigned int order,
1768 int migratetype, bool only_stealable, bool *can_steal)
1773 if (area->nr_free == 0)
1778 fallback_mt = fallbacks[migratetype][i];
1779 if (fallback_mt == MIGRATE_TYPES)
1782 if (list_empty(&area->free_list[fallback_mt]))
1785 if (can_steal_fallback(order, migratetype))
1788 if (!only_stealable)
1799 * Reserve a pageblock for exclusive use of high-order atomic allocations if
1800 * there are no empty page blocks that contain a page with a suitable order
1802 static void reserve_highatomic_pageblock(struct page *page, struct zone *zone,
1803 unsigned int alloc_order)
1806 unsigned long max_managed, flags;
1809 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
1810 * Check is race-prone but harmless.
1812 max_managed = (zone->managed_pages / 100) + pageblock_nr_pages;
1813 if (zone->nr_reserved_highatomic >= max_managed)
1816 spin_lock_irqsave(&zone->lock, flags);
1818 /* Recheck the nr_reserved_highatomic limit under the lock */
1819 if (zone->nr_reserved_highatomic >= max_managed)
1823 mt = get_pageblock_migratetype(page);
1824 if (mt != MIGRATE_HIGHATOMIC &&
1825 !is_migrate_isolate(mt) && !is_migrate_cma(mt)) {
1826 zone->nr_reserved_highatomic += pageblock_nr_pages;
1827 set_pageblock_migratetype(page, MIGRATE_HIGHATOMIC);
1828 move_freepages_block(zone, page, MIGRATE_HIGHATOMIC);
1832 spin_unlock_irqrestore(&zone->lock, flags);
1836 * Used when an allocation is about to fail under memory pressure. This
1837 * potentially hurts the reliability of high-order allocations when under
1838 * intense memory pressure but failed atomic allocations should be easier
1839 * to recover from than an OOM.
1841 static void unreserve_highatomic_pageblock(const struct alloc_context *ac)
1843 struct zonelist *zonelist = ac->zonelist;
1844 unsigned long flags;
1850 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
1852 /* Preserve at least one pageblock */
1853 if (zone->nr_reserved_highatomic <= pageblock_nr_pages)
1856 spin_lock_irqsave(&zone->lock, flags);
1857 for (order = 0; order < MAX_ORDER; order++) {
1858 struct free_area *area = &(zone->free_area[order]);
1860 page = list_first_entry_or_null(
1861 &area->free_list[MIGRATE_HIGHATOMIC],
1867 * It should never happen but changes to locking could
1868 * inadvertently allow a per-cpu drain to add pages
1869 * to MIGRATE_HIGHATOMIC while unreserving so be safe
1870 * and watch for underflows.
1872 zone->nr_reserved_highatomic -= min(pageblock_nr_pages,
1873 zone->nr_reserved_highatomic);
1876 * Convert to ac->migratetype and avoid the normal
1877 * pageblock stealing heuristics. Minimally, the caller
1878 * is doing the work and needs the pages. More
1879 * importantly, if the block was always converted to
1880 * MIGRATE_UNMOVABLE or another type then the number
1881 * of pageblocks that cannot be completely freed
1884 set_pageblock_migratetype(page, ac->migratetype);
1885 move_freepages_block(zone, page, ac->migratetype);
1886 spin_unlock_irqrestore(&zone->lock, flags);
1889 spin_unlock_irqrestore(&zone->lock, flags);
1893 /* Remove an element from the buddy allocator from the fallback list */
1894 static inline struct page *
1895 __rmqueue_fallback(struct zone *zone, unsigned int order, int start_migratetype)
1897 struct free_area *area;
1898 unsigned int current_order;
1903 /* Find the largest possible block of pages in the other list */
1904 for (current_order = MAX_ORDER-1;
1905 current_order >= order && current_order <= MAX_ORDER-1;
1907 area = &(zone->free_area[current_order]);
1908 fallback_mt = find_suitable_fallback(area, current_order,
1909 start_migratetype, false, &can_steal);
1910 if (fallback_mt == -1)
1913 page = list_first_entry(&area->free_list[fallback_mt],
1916 steal_suitable_fallback(zone, page, start_migratetype);
1918 /* Remove the page from the freelists */
1920 list_del(&page->lru);
1921 rmv_page_order(page);
1923 expand(zone, page, order, current_order, area,
1926 * The pcppage_migratetype may differ from pageblock's
1927 * migratetype depending on the decisions in
1928 * find_suitable_fallback(). This is OK as long as it does not
1929 * differ for MIGRATE_CMA pageblocks. Those can be used as
1930 * fallback only via special __rmqueue_cma_fallback() function
1932 set_pcppage_migratetype(page, start_migratetype);
1934 trace_mm_page_alloc_extfrag(page, order, current_order,
1935 start_migratetype, fallback_mt);
1944 * Do the hard work of removing an element from the buddy allocator.
1945 * Call me with the zone->lock already held.
1947 static struct page *__rmqueue(struct zone *zone, unsigned int order,
1952 page = __rmqueue_smallest(zone, order, migratetype);
1953 if (unlikely(!page)) {
1954 if (migratetype == MIGRATE_MOVABLE)
1955 page = __rmqueue_cma_fallback(zone, order);
1958 page = __rmqueue_fallback(zone, order, migratetype);
1961 trace_mm_page_alloc_zone_locked(page, order, migratetype);
1966 * Obtain a specified number of elements from the buddy allocator, all under
1967 * a single hold of the lock, for efficiency. Add them to the supplied list.
1968 * Returns the number of new pages which were placed at *list.
1970 static int rmqueue_bulk(struct zone *zone, unsigned int order,
1971 unsigned long count, struct list_head *list,
1972 int migratetype, bool cold)
1976 spin_lock(&zone->lock);
1977 for (i = 0; i < count; ++i) {
1978 struct page *page = __rmqueue(zone, order, migratetype);
1979 if (unlikely(page == NULL))
1983 * Split buddy pages returned by expand() are received here
1984 * in physical page order. The page is added to the callers and
1985 * list and the list head then moves forward. From the callers
1986 * perspective, the linked list is ordered by page number in
1987 * some conditions. This is useful for IO devices that can
1988 * merge IO requests if the physical pages are ordered
1992 list_add(&page->lru, list);
1994 list_add_tail(&page->lru, list);
1996 if (is_migrate_cma(get_pcppage_migratetype(page)))
1997 __mod_zone_page_state(zone, NR_FREE_CMA_PAGES,
2000 __mod_zone_page_state(zone, NR_FREE_PAGES, -(i << order));
2001 spin_unlock(&zone->lock);
2007 * Called from the vmstat counter updater to drain pagesets of this
2008 * currently executing processor on remote nodes after they have
2011 * Note that this function must be called with the thread pinned to
2012 * a single processor.
2014 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp)
2016 unsigned long flags;
2017 int to_drain, batch;
2019 local_irq_save(flags);
2020 batch = READ_ONCE(pcp->batch);
2021 to_drain = min(pcp->count, batch);
2023 free_pcppages_bulk(zone, to_drain, pcp);
2024 pcp->count -= to_drain;
2026 local_irq_restore(flags);
2031 * Drain pcplists of the indicated processor and zone.
2033 * The processor must either be the current processor and the
2034 * thread pinned to the current processor or a processor that
2037 static void drain_pages_zone(unsigned int cpu, struct zone *zone)
2039 unsigned long flags;
2040 struct per_cpu_pageset *pset;
2041 struct per_cpu_pages *pcp;
2043 local_irq_save(flags);
2044 pset = per_cpu_ptr(zone->pageset, cpu);
2048 free_pcppages_bulk(zone, pcp->count, pcp);
2051 local_irq_restore(flags);
2055 * Drain pcplists of all zones on the indicated processor.
2057 * The processor must either be the current processor and the
2058 * thread pinned to the current processor or a processor that
2061 static void drain_pages(unsigned int cpu)
2065 for_each_populated_zone(zone) {
2066 drain_pages_zone(cpu, zone);
2071 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
2073 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2074 * the single zone's pages.
2076 void drain_local_pages(struct zone *zone)
2078 int cpu = smp_processor_id();
2081 drain_pages_zone(cpu, zone);
2087 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2089 * When zone parameter is non-NULL, spill just the single zone's pages.
2091 * Note that this code is protected against sending an IPI to an offline
2092 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
2093 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
2094 * nothing keeps CPUs from showing up after we populated the cpumask and
2095 * before the call to on_each_cpu_mask().
2097 void drain_all_pages(struct zone *zone)
2102 * Allocate in the BSS so we wont require allocation in
2103 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2105 static cpumask_t cpus_with_pcps;
2108 * We don't care about racing with CPU hotplug event
2109 * as offline notification will cause the notified
2110 * cpu to drain that CPU pcps and on_each_cpu_mask
2111 * disables preemption as part of its processing
2113 for_each_online_cpu(cpu) {
2114 struct per_cpu_pageset *pcp;
2116 bool has_pcps = false;
2119 pcp = per_cpu_ptr(zone->pageset, cpu);
2123 for_each_populated_zone(z) {
2124 pcp = per_cpu_ptr(z->pageset, cpu);
2125 if (pcp->pcp.count) {
2133 cpumask_set_cpu(cpu, &cpus_with_pcps);
2135 cpumask_clear_cpu(cpu, &cpus_with_pcps);
2137 on_each_cpu_mask(&cpus_with_pcps, (smp_call_func_t) drain_local_pages,
2141 #ifdef CONFIG_HIBERNATION
2143 void mark_free_pages(struct zone *zone)
2145 unsigned long pfn, max_zone_pfn;
2146 unsigned long flags;
2147 unsigned int order, t;
2150 if (zone_is_empty(zone))
2153 spin_lock_irqsave(&zone->lock, flags);
2155 max_zone_pfn = zone_end_pfn(zone);
2156 for (pfn = zone->zone_start_pfn; pfn < max_zone_pfn; pfn++)
2157 if (pfn_valid(pfn)) {
2158 page = pfn_to_page(pfn);
2160 if (page_zone(page) != zone)
2163 if (!swsusp_page_is_forbidden(page))
2164 swsusp_unset_page_free(page);
2167 for_each_migratetype_order(order, t) {
2168 list_for_each_entry(page,
2169 &zone->free_area[order].free_list[t], lru) {
2172 pfn = page_to_pfn(page);
2173 for (i = 0; i < (1UL << order); i++)
2174 swsusp_set_page_free(pfn_to_page(pfn + i));
2177 spin_unlock_irqrestore(&zone->lock, flags);
2179 #endif /* CONFIG_PM */
2182 * Free a 0-order page
2183 * cold == true ? free a cold page : free a hot page
2185 void free_hot_cold_page(struct page *page, bool cold)
2187 struct zone *zone = page_zone(page);
2188 struct per_cpu_pages *pcp;
2189 unsigned long flags;
2190 unsigned long pfn = page_to_pfn(page);
2193 if (!free_pages_prepare(page, 0))
2196 migratetype = get_pfnblock_migratetype(page, pfn);
2197 set_pcppage_migratetype(page, migratetype);
2198 local_irq_save(flags);
2199 __count_vm_event(PGFREE);
2202 * We only track unmovable, reclaimable and movable on pcp lists.
2203 * Free ISOLATE pages back to the allocator because they are being
2204 * offlined but treat RESERVE as movable pages so we can get those
2205 * areas back if necessary. Otherwise, we may have to free
2206 * excessively into the page allocator
2208 if (migratetype >= MIGRATE_PCPTYPES) {
2209 if (unlikely(is_migrate_isolate(migratetype))) {
2210 free_one_page(zone, page, pfn, 0, migratetype);
2213 migratetype = MIGRATE_MOVABLE;
2216 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2218 list_add(&page->lru, &pcp->lists[migratetype]);
2220 list_add_tail(&page->lru, &pcp->lists[migratetype]);
2222 if (pcp->count >= pcp->high) {
2223 unsigned long batch = READ_ONCE(pcp->batch);
2224 free_pcppages_bulk(zone, batch, pcp);
2225 pcp->count -= batch;
2229 local_irq_restore(flags);
2233 * Free a list of 0-order pages
2235 void free_hot_cold_page_list(struct list_head *list, bool cold)
2237 struct page *page, *next;
2239 list_for_each_entry_safe(page, next, list, lru) {
2240 trace_mm_page_free_batched(page, cold);
2241 free_hot_cold_page(page, cold);
2246 * split_page takes a non-compound higher-order page, and splits it into
2247 * n (1<<order) sub-pages: page[0..n]
2248 * Each sub-page must be freed individually.
2250 * Note: this is probably too low level an operation for use in drivers.
2251 * Please consult with lkml before using this in your driver.
2253 void split_page(struct page *page, unsigned int order)
2258 VM_BUG_ON_PAGE(PageCompound(page), page);
2259 VM_BUG_ON_PAGE(!page_count(page), page);
2261 #ifdef CONFIG_KMEMCHECK
2263 * Split shadow pages too, because free(page[0]) would
2264 * otherwise free the whole shadow.
2266 if (kmemcheck_page_is_tracked(page))
2267 split_page(virt_to_page(page[0].shadow), order);
2270 gfp_mask = get_page_owner_gfp(page);
2271 set_page_owner(page, 0, gfp_mask);
2272 for (i = 1; i < (1 << order); i++) {
2273 set_page_refcounted(page + i);
2274 set_page_owner(page + i, 0, gfp_mask);
2277 EXPORT_SYMBOL_GPL(split_page);
2279 int __isolate_free_page(struct page *page, unsigned int order)
2281 unsigned long watermark;
2285 BUG_ON(!PageBuddy(page));
2287 zone = page_zone(page);
2288 mt = get_pageblock_migratetype(page);
2290 if (!is_migrate_isolate(mt)) {
2291 /* Obey watermarks as if the page was being allocated */
2292 watermark = low_wmark_pages(zone) + (1 << order);
2293 if (!zone_watermark_ok(zone, 0, watermark, 0, 0))
2296 __mod_zone_freepage_state(zone, -(1UL << order), mt);
2299 /* Remove page from free list */
2300 list_del(&page->lru);
2301 zone->free_area[order].nr_free--;
2302 rmv_page_order(page);
2304 set_page_owner(page, order, __GFP_MOVABLE);
2306 /* Set the pageblock if the isolated page is at least a pageblock */
2307 if (order >= pageblock_order - 1) {
2308 struct page *endpage = page + (1 << order) - 1;
2309 for (; page < endpage; page += pageblock_nr_pages) {
2310 int mt = get_pageblock_migratetype(page);
2311 if (!is_migrate_isolate(mt) && !is_migrate_cma(mt))
2312 set_pageblock_migratetype(page,
2318 return 1UL << order;
2322 * Similar to split_page except the page is already free. As this is only
2323 * being used for migration, the migratetype of the block also changes.
2324 * As this is called with interrupts disabled, the caller is responsible
2325 * for calling arch_alloc_page() and kernel_map_page() after interrupts
2328 * Note: this is probably too low level an operation for use in drivers.
2329 * Please consult with lkml before using this in your driver.
2331 int split_free_page(struct page *page)
2336 order = page_order(page);
2338 nr_pages = __isolate_free_page(page, order);
2342 /* Split into individual pages */
2343 set_page_refcounted(page);
2344 split_page(page, order);
2349 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2352 struct page *buffered_rmqueue(struct zone *preferred_zone,
2353 struct zone *zone, unsigned int order,
2354 gfp_t gfp_flags, int alloc_flags, int migratetype)
2356 unsigned long flags;
2358 bool cold = ((gfp_flags & __GFP_COLD) != 0);
2360 if (likely(order == 0)) {
2361 struct per_cpu_pages *pcp;
2362 struct list_head *list;
2364 local_irq_save(flags);
2365 pcp = &this_cpu_ptr(zone->pageset)->pcp;
2366 list = &pcp->lists[migratetype];
2367 if (list_empty(list)) {
2368 pcp->count += rmqueue_bulk(zone, 0,
2371 if (unlikely(list_empty(list)))
2376 page = list_last_entry(list, struct page, lru);
2378 page = list_first_entry(list, struct page, lru);
2380 list_del(&page->lru);
2384 * We most definitely don't want callers attempting to
2385 * allocate greater than order-1 page units with __GFP_NOFAIL.
2387 WARN_ON_ONCE((gfp_flags & __GFP_NOFAIL) && (order > 1));
2388 spin_lock_irqsave(&zone->lock, flags);
2391 if (alloc_flags & ALLOC_HARDER) {
2392 page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
2394 trace_mm_page_alloc_zone_locked(page, order, migratetype);
2397 page = __rmqueue(zone, order, migratetype);
2398 spin_unlock(&zone->lock);
2401 __mod_zone_freepage_state(zone, -(1 << order),
2402 get_pcppage_migratetype(page));
2405 __mod_zone_page_state(zone, NR_ALLOC_BATCH, -(1 << order));
2406 if (atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]) <= 0 &&
2407 !test_bit(ZONE_FAIR_DEPLETED, &zone->flags))
2408 set_bit(ZONE_FAIR_DEPLETED, &zone->flags);
2410 __count_zone_vm_events(PGALLOC, zone, 1 << order);
2411 zone_statistics(preferred_zone, zone, gfp_flags);
2412 local_irq_restore(flags);
2414 VM_BUG_ON_PAGE(bad_range(zone, page), page);
2418 local_irq_restore(flags);
2422 #ifdef CONFIG_FAIL_PAGE_ALLOC
2425 struct fault_attr attr;
2427 bool ignore_gfp_highmem;
2428 bool ignore_gfp_reclaim;
2430 } fail_page_alloc = {
2431 .attr = FAULT_ATTR_INITIALIZER,
2432 .ignore_gfp_reclaim = true,
2433 .ignore_gfp_highmem = true,
2437 static int __init setup_fail_page_alloc(char *str)
2439 return setup_fault_attr(&fail_page_alloc.attr, str);
2441 __setup("fail_page_alloc=", setup_fail_page_alloc);
2443 static bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
2445 if (order < fail_page_alloc.min_order)
2447 if (gfp_mask & __GFP_NOFAIL)
2449 if (fail_page_alloc.ignore_gfp_highmem && (gfp_mask & __GFP_HIGHMEM))
2451 if (fail_page_alloc.ignore_gfp_reclaim &&
2452 (gfp_mask & __GFP_DIRECT_RECLAIM))
2455 return should_fail(&fail_page_alloc.attr, 1 << order);
2458 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2460 static int __init fail_page_alloc_debugfs(void)
2462 umode_t mode = S_IFREG | S_IRUSR | S_IWUSR;
2465 dir = fault_create_debugfs_attr("fail_page_alloc", NULL,
2466 &fail_page_alloc.attr);
2468 return PTR_ERR(dir);
2470 if (!debugfs_create_bool("ignore-gfp-wait", mode, dir,
2471 &fail_page_alloc.ignore_gfp_reclaim))
2473 if (!debugfs_create_bool("ignore-gfp-highmem", mode, dir,
2474 &fail_page_alloc.ignore_gfp_highmem))
2476 if (!debugfs_create_u32("min-order", mode, dir,
2477 &fail_page_alloc.min_order))
2482 debugfs_remove_recursive(dir);
2487 late_initcall(fail_page_alloc_debugfs);
2489 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2491 #else /* CONFIG_FAIL_PAGE_ALLOC */
2493 static inline bool should_fail_alloc_page(gfp_t gfp_mask, unsigned int order)
2498 #endif /* CONFIG_FAIL_PAGE_ALLOC */
2501 * Return true if free base pages are above 'mark'. For high-order checks it
2502 * will return true of the order-0 watermark is reached and there is at least
2503 * one free page of a suitable size. Checking now avoids taking the zone lock
2504 * to check in the allocation paths if no pages are free.
2506 static bool __zone_watermark_ok(struct zone *z, unsigned int order,
2507 unsigned long mark, int classzone_idx, int alloc_flags,
2512 const int alloc_harder = (alloc_flags & ALLOC_HARDER);
2514 /* free_pages may go negative - that's OK */
2515 free_pages -= (1 << order) - 1;
2517 if (alloc_flags & ALLOC_HIGH)
2521 * If the caller does not have rights to ALLOC_HARDER then subtract
2522 * the high-atomic reserves. This will over-estimate the size of the
2523 * atomic reserve but it avoids a search.
2525 if (likely(!alloc_harder))
2526 free_pages -= z->nr_reserved_highatomic;
2531 /* If allocation can't use CMA areas don't use free CMA pages */
2532 if (!(alloc_flags & ALLOC_CMA))
2533 free_pages -= zone_page_state(z, NR_FREE_CMA_PAGES);
2537 * Check watermarks for an order-0 allocation request. If these
2538 * are not met, then a high-order request also cannot go ahead
2539 * even if a suitable page happened to be free.
2541 if (free_pages <= min + z->lowmem_reserve[classzone_idx])
2544 /* If this is an order-0 request then the watermark is fine */
2548 /* For a high-order request, check at least one suitable page is free */
2549 for (o = order; o < MAX_ORDER; o++) {
2550 struct free_area *area = &z->free_area[o];
2559 for (mt = 0; mt < MIGRATE_PCPTYPES; mt++) {
2560 if (!list_empty(&area->free_list[mt]))
2565 if ((alloc_flags & ALLOC_CMA) &&
2566 !list_empty(&area->free_list[MIGRATE_CMA])) {
2574 bool zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark,
2575 int classzone_idx, int alloc_flags)
2577 return __zone_watermark_ok(z, order, mark, classzone_idx, alloc_flags,
2578 zone_page_state(z, NR_FREE_PAGES));
2581 bool zone_watermark_ok_safe(struct zone *z, unsigned int order,
2582 unsigned long mark, int classzone_idx)
2584 long free_pages = zone_page_state(z, NR_FREE_PAGES);
2586 if (z->percpu_drift_mark && free_pages < z->percpu_drift_mark)
2587 free_pages = zone_page_state_snapshot(z, NR_FREE_PAGES);
2589 return __zone_watermark_ok(z, order, mark, classzone_idx, 0,
2594 static bool zone_local(struct zone *local_zone, struct zone *zone)
2596 return local_zone->node == zone->node;
2599 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2601 return node_distance(zone_to_nid(local_zone), zone_to_nid(zone)) <
2604 #else /* CONFIG_NUMA */
2605 static bool zone_local(struct zone *local_zone, struct zone *zone)
2610 static bool zone_allows_reclaim(struct zone *local_zone, struct zone *zone)
2614 #endif /* CONFIG_NUMA */
2616 static void reset_alloc_batches(struct zone *preferred_zone)
2618 struct zone *zone = preferred_zone->zone_pgdat->node_zones;
2621 mod_zone_page_state(zone, NR_ALLOC_BATCH,
2622 high_wmark_pages(zone) - low_wmark_pages(zone) -
2623 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
2624 clear_bit(ZONE_FAIR_DEPLETED, &zone->flags);
2625 } while (zone++ != preferred_zone);
2629 * get_page_from_freelist goes through the zonelist trying to allocate
2632 static struct page *
2633 get_page_from_freelist(gfp_t gfp_mask, unsigned int order, int alloc_flags,
2634 const struct alloc_context *ac)
2636 struct zonelist *zonelist = ac->zonelist;
2638 struct page *page = NULL;
2640 int nr_fair_skipped = 0;
2641 bool zonelist_rescan;
2644 zonelist_rescan = false;
2647 * Scan zonelist, looking for a zone with enough free.
2648 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2650 for_each_zone_zonelist_nodemask(zone, z, zonelist, ac->high_zoneidx,
2654 if (cpusets_enabled() &&
2655 (alloc_flags & ALLOC_CPUSET) &&
2656 !cpuset_zone_allowed(zone, gfp_mask))
2659 * Distribute pages in proportion to the individual
2660 * zone size to ensure fair page aging. The zone a
2661 * page was allocated in should have no effect on the
2662 * time the page has in memory before being reclaimed.
2664 if (alloc_flags & ALLOC_FAIR) {
2665 if (!zone_local(ac->preferred_zone, zone))
2667 if (test_bit(ZONE_FAIR_DEPLETED, &zone->flags)) {
2673 * When allocating a page cache page for writing, we
2674 * want to get it from a zone that is within its dirty
2675 * limit, such that no single zone holds more than its
2676 * proportional share of globally allowed dirty pages.
2677 * The dirty limits take into account the zone's
2678 * lowmem reserves and high watermark so that kswapd
2679 * should be able to balance it without having to
2680 * write pages from its LRU list.
2682 * This may look like it could increase pressure on
2683 * lower zones by failing allocations in higher zones
2684 * before they are full. But the pages that do spill
2685 * over are limited as the lower zones are protected
2686 * by this very same mechanism. It should not become
2687 * a practical burden to them.
2689 * XXX: For now, allow allocations to potentially
2690 * exceed the per-zone dirty limit in the slowpath
2691 * (spread_dirty_pages unset) before going into reclaim,
2692 * which is important when on a NUMA setup the allowed
2693 * zones are together not big enough to reach the
2694 * global limit. The proper fix for these situations
2695 * will require awareness of zones in the
2696 * dirty-throttling and the flusher threads.
2698 if (ac->spread_dirty_pages && !zone_dirty_ok(zone))
2701 mark = zone->watermark[alloc_flags & ALLOC_WMARK_MASK];
2702 if (!zone_watermark_ok(zone, order, mark,
2703 ac->classzone_idx, alloc_flags)) {
2706 /* Checked here to keep the fast path fast */
2707 BUILD_BUG_ON(ALLOC_NO_WATERMARKS < NR_WMARK);
2708 if (alloc_flags & ALLOC_NO_WATERMARKS)
2711 if (zone_reclaim_mode == 0 ||
2712 !zone_allows_reclaim(ac->preferred_zone, zone))
2715 ret = zone_reclaim(zone, gfp_mask, order);
2717 case ZONE_RECLAIM_NOSCAN:
2720 case ZONE_RECLAIM_FULL:
2721 /* scanned but unreclaimable */
2724 /* did we reclaim enough */
2725 if (zone_watermark_ok(zone, order, mark,
2726 ac->classzone_idx, alloc_flags))
2734 page = buffered_rmqueue(ac->preferred_zone, zone, order,
2735 gfp_mask, alloc_flags, ac->migratetype);
2737 if (prep_new_page(page, order, gfp_mask, alloc_flags))
2741 * If this is a high-order atomic allocation then check
2742 * if the pageblock should be reserved for the future
2744 if (unlikely(order && (alloc_flags & ALLOC_HARDER)))
2745 reserve_highatomic_pageblock(page, zone, order);
2752 * The first pass makes sure allocations are spread fairly within the
2753 * local node. However, the local node might have free pages left
2754 * after the fairness batches are exhausted, and remote zones haven't
2755 * even been considered yet. Try once more without fairness, and
2756 * include remote zones now, before entering the slowpath and waking
2757 * kswapd: prefer spilling to a remote zone over swapping locally.
2759 if (alloc_flags & ALLOC_FAIR) {
2760 alloc_flags &= ~ALLOC_FAIR;
2761 if (nr_fair_skipped) {
2762 zonelist_rescan = true;
2763 reset_alloc_batches(ac->preferred_zone);
2765 if (nr_online_nodes > 1)
2766 zonelist_rescan = true;
2769 if (zonelist_rescan)
2776 * Large machines with many possible nodes should not always dump per-node
2777 * meminfo in irq context.
2779 static inline bool should_suppress_show_mem(void)
2784 ret = in_interrupt();
2789 static DEFINE_RATELIMIT_STATE(nopage_rs,
2790 DEFAULT_RATELIMIT_INTERVAL,
2791 DEFAULT_RATELIMIT_BURST);
2793 void warn_alloc_failed(gfp_t gfp_mask, unsigned int order, const char *fmt, ...)
2795 unsigned int filter = SHOW_MEM_FILTER_NODES;
2797 if ((gfp_mask & __GFP_NOWARN) || !__ratelimit(&nopage_rs) ||
2798 debug_guardpage_minorder() > 0)
2802 * This documents exceptions given to allocations in certain
2803 * contexts that are allowed to allocate outside current's set
2806 if (!(gfp_mask & __GFP_NOMEMALLOC))
2807 if (test_thread_flag(TIF_MEMDIE) ||
2808 (current->flags & (PF_MEMALLOC | PF_EXITING)))
2809 filter &= ~SHOW_MEM_FILTER_NODES;
2810 if (in_interrupt() || !(gfp_mask & __GFP_DIRECT_RECLAIM))
2811 filter &= ~SHOW_MEM_FILTER_NODES;
2814 struct va_format vaf;
2817 va_start(args, fmt);
2822 pr_warn("%pV", &vaf);
2827 pr_warn("%s: page allocation failure: order:%u, mode:%#x(%pGg)\n",
2828 current->comm, order, gfp_mask, &gfp_mask);
2830 if (!should_suppress_show_mem())
2834 static inline struct page *
2835 __alloc_pages_may_oom(gfp_t gfp_mask, unsigned int order,
2836 const struct alloc_context *ac, unsigned long *did_some_progress)
2838 struct oom_control oc = {
2839 .zonelist = ac->zonelist,
2840 .nodemask = ac->nodemask,
2841 .gfp_mask = gfp_mask,
2846 *did_some_progress = 0;
2849 * Acquire the oom lock. If that fails, somebody else is
2850 * making progress for us.
2852 if (!mutex_trylock(&oom_lock)) {
2853 *did_some_progress = 1;
2854 schedule_timeout_uninterruptible(1);
2859 * Go through the zonelist yet one more time, keep very high watermark
2860 * here, this is only to catch a parallel oom killing, we must fail if
2861 * we're still under heavy pressure.
2863 page = get_page_from_freelist(gfp_mask | __GFP_HARDWALL, order,
2864 ALLOC_WMARK_HIGH|ALLOC_CPUSET, ac);
2868 if (!(gfp_mask & __GFP_NOFAIL)) {
2869 /* Coredumps can quickly deplete all memory reserves */
2870 if (current->flags & PF_DUMPCORE)
2872 /* The OOM killer will not help higher order allocs */
2873 if (order > PAGE_ALLOC_COSTLY_ORDER)
2875 /* The OOM killer does not needlessly kill tasks for lowmem */
2876 if (ac->high_zoneidx < ZONE_NORMAL)
2878 /* The OOM killer does not compensate for IO-less reclaim */
2879 if (!(gfp_mask & __GFP_FS)) {
2881 * XXX: Page reclaim didn't yield anything,
2882 * and the OOM killer can't be invoked, but
2883 * keep looping as per tradition.
2885 * But do not keep looping if oom_killer_disable()
2886 * was already called, for the system is trying to
2887 * enter a quiescent state during suspend.
2889 *did_some_progress = !oom_killer_disabled;
2892 if (pm_suspended_storage())
2894 /* The OOM killer may not free memory on a specific node */
2895 if (gfp_mask & __GFP_THISNODE)
2898 /* Exhausted what can be done so it's blamo time */
2899 if (out_of_memory(&oc) || WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
2900 *did_some_progress = 1;
2902 if (gfp_mask & __GFP_NOFAIL) {
2903 page = get_page_from_freelist(gfp_mask, order,
2904 ALLOC_NO_WATERMARKS|ALLOC_CPUSET, ac);
2906 * fallback to ignore cpuset restriction if our nodes
2910 page = get_page_from_freelist(gfp_mask, order,
2911 ALLOC_NO_WATERMARKS, ac);
2915 mutex_unlock(&oom_lock);
2919 #ifdef CONFIG_COMPACTION
2920 /* Try memory compaction for high-order allocations before reclaim */
2921 static struct page *
2922 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2923 int alloc_flags, const struct alloc_context *ac,
2924 enum migrate_mode mode, int *contended_compaction,
2925 bool *deferred_compaction)
2927 unsigned long compact_result;
2933 current->flags |= PF_MEMALLOC;
2934 compact_result = try_to_compact_pages(gfp_mask, order, alloc_flags, ac,
2935 mode, contended_compaction);
2936 current->flags &= ~PF_MEMALLOC;
2938 switch (compact_result) {
2939 case COMPACT_DEFERRED:
2940 *deferred_compaction = true;
2942 case COMPACT_SKIPPED:
2949 * At least in one zone compaction wasn't deferred or skipped, so let's
2950 * count a compaction stall
2952 count_vm_event(COMPACTSTALL);
2954 page = get_page_from_freelist(gfp_mask, order,
2955 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
2958 struct zone *zone = page_zone(page);
2960 zone->compact_blockskip_flush = false;
2961 compaction_defer_reset(zone, order, true);
2962 count_vm_event(COMPACTSUCCESS);
2967 * It's bad if compaction run occurs and fails. The most likely reason
2968 * is that pages exist, but not enough to satisfy watermarks.
2970 count_vm_event(COMPACTFAIL);
2977 static inline struct page *
2978 __alloc_pages_direct_compact(gfp_t gfp_mask, unsigned int order,
2979 int alloc_flags, const struct alloc_context *ac,
2980 enum migrate_mode mode, int *contended_compaction,
2981 bool *deferred_compaction)
2985 #endif /* CONFIG_COMPACTION */
2987 /* Perform direct synchronous page reclaim */
2989 __perform_reclaim(gfp_t gfp_mask, unsigned int order,
2990 const struct alloc_context *ac)
2992 struct reclaim_state reclaim_state;
2997 /* We now go into synchronous reclaim */
2998 cpuset_memory_pressure_bump();
2999 current->flags |= PF_MEMALLOC;
3000 lockdep_set_current_reclaim_state(gfp_mask);
3001 reclaim_state.reclaimed_slab = 0;
3002 current->reclaim_state = &reclaim_state;
3004 progress = try_to_free_pages(ac->zonelist, order, gfp_mask,
3007 current->reclaim_state = NULL;
3008 lockdep_clear_current_reclaim_state();
3009 current->flags &= ~PF_MEMALLOC;
3016 /* The really slow allocator path where we enter direct reclaim */
3017 static inline struct page *
3018 __alloc_pages_direct_reclaim(gfp_t gfp_mask, unsigned int order,
3019 int alloc_flags, const struct alloc_context *ac,
3020 unsigned long *did_some_progress)
3022 struct page *page = NULL;
3023 bool drained = false;
3025 *did_some_progress = __perform_reclaim(gfp_mask, order, ac);
3026 if (unlikely(!(*did_some_progress)))
3030 page = get_page_from_freelist(gfp_mask, order,
3031 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
3034 * If an allocation failed after direct reclaim, it could be because
3035 * pages are pinned on the per-cpu lists or in high alloc reserves.
3036 * Shrink them them and try again
3038 if (!page && !drained) {
3039 unreserve_highatomic_pageblock(ac);
3040 drain_all_pages(NULL);
3048 static void wake_all_kswapds(unsigned int order, const struct alloc_context *ac)
3053 for_each_zone_zonelist_nodemask(zone, z, ac->zonelist,
3054 ac->high_zoneidx, ac->nodemask)
3055 wakeup_kswapd(zone, order, zone_idx(ac->preferred_zone));
3059 gfp_to_alloc_flags(gfp_t gfp_mask)
3061 int alloc_flags = ALLOC_WMARK_MIN | ALLOC_CPUSET;
3063 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
3064 BUILD_BUG_ON(__GFP_HIGH != (__force gfp_t) ALLOC_HIGH);
3067 * The caller may dip into page reserves a bit more if the caller
3068 * cannot run direct reclaim, or if the caller has realtime scheduling
3069 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
3070 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
3072 alloc_flags |= (__force int) (gfp_mask & __GFP_HIGH);
3074 if (gfp_mask & __GFP_ATOMIC) {
3076 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
3077 * if it can't schedule.
3079 if (!(gfp_mask & __GFP_NOMEMALLOC))
3080 alloc_flags |= ALLOC_HARDER;
3082 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
3083 * comment for __cpuset_node_allowed().
3085 alloc_flags &= ~ALLOC_CPUSET;
3086 } else if (unlikely(rt_task(current)) && !in_interrupt())
3087 alloc_flags |= ALLOC_HARDER;
3089 if (likely(!(gfp_mask & __GFP_NOMEMALLOC))) {
3090 if (gfp_mask & __GFP_MEMALLOC)
3091 alloc_flags |= ALLOC_NO_WATERMARKS;
3092 else if (in_serving_softirq() && (current->flags & PF_MEMALLOC))
3093 alloc_flags |= ALLOC_NO_WATERMARKS;
3094 else if (!in_interrupt() &&
3095 ((current->flags & PF_MEMALLOC) ||
3096 unlikely(test_thread_flag(TIF_MEMDIE))))
3097 alloc_flags |= ALLOC_NO_WATERMARKS;
3100 if (gfpflags_to_migratetype(gfp_mask) == MIGRATE_MOVABLE)
3101 alloc_flags |= ALLOC_CMA;
3106 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask)
3108 return !!(gfp_to_alloc_flags(gfp_mask) & ALLOC_NO_WATERMARKS);
3111 static inline bool is_thp_gfp_mask(gfp_t gfp_mask)
3113 return (gfp_mask & (GFP_TRANSHUGE | __GFP_KSWAPD_RECLAIM)) == GFP_TRANSHUGE;
3116 static inline struct page *
3117 __alloc_pages_slowpath(gfp_t gfp_mask, unsigned int order,
3118 struct alloc_context *ac)
3120 bool can_direct_reclaim = gfp_mask & __GFP_DIRECT_RECLAIM;
3121 struct page *page = NULL;
3123 unsigned long pages_reclaimed = 0;
3124 unsigned long did_some_progress;
3125 enum migrate_mode migration_mode = MIGRATE_ASYNC;
3126 bool deferred_compaction = false;
3127 int contended_compaction = COMPACT_CONTENDED_NONE;
3130 * In the slowpath, we sanity check order to avoid ever trying to
3131 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3132 * be using allocators in order of preference for an area that is
3135 if (order >= MAX_ORDER) {
3136 WARN_ON_ONCE(!(gfp_mask & __GFP_NOWARN));
3141 * We also sanity check to catch abuse of atomic reserves being used by
3142 * callers that are not in atomic context.
3144 if (WARN_ON_ONCE((gfp_mask & (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)) ==
3145 (__GFP_ATOMIC|__GFP_DIRECT_RECLAIM)))
3146 gfp_mask &= ~__GFP_ATOMIC;
3149 if (gfp_mask & __GFP_KSWAPD_RECLAIM)
3150 wake_all_kswapds(order, ac);
3153 * OK, we're below the kswapd watermark and have kicked background
3154 * reclaim. Now things get more complex, so set up alloc_flags according
3155 * to how we want to proceed.
3157 alloc_flags = gfp_to_alloc_flags(gfp_mask);
3160 * Find the true preferred zone if the allocation is unconstrained by
3163 if (!(alloc_flags & ALLOC_CPUSET) && !ac->nodemask) {
3164 struct zoneref *preferred_zoneref;
3165 preferred_zoneref = first_zones_zonelist(ac->zonelist,
3166 ac->high_zoneidx, NULL, &ac->preferred_zone);
3167 ac->classzone_idx = zonelist_zone_idx(preferred_zoneref);
3170 /* This is the last chance, in general, before the goto nopage. */
3171 page = get_page_from_freelist(gfp_mask, order,
3172 alloc_flags & ~ALLOC_NO_WATERMARKS, ac);
3176 /* Allocate without watermarks if the context allows */
3177 if (alloc_flags & ALLOC_NO_WATERMARKS) {
3179 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3180 * the allocation is high priority and these type of
3181 * allocations are system rather than user orientated
3183 ac->zonelist = node_zonelist(numa_node_id(), gfp_mask);
3184 page = get_page_from_freelist(gfp_mask, order,
3185 ALLOC_NO_WATERMARKS, ac);
3190 /* Caller is not willing to reclaim, we can't balance anything */
3191 if (!can_direct_reclaim) {
3193 * All existing users of the __GFP_NOFAIL are blockable, so warn
3194 * of any new users that actually allow this type of allocation
3197 WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL);
3201 /* Avoid recursion of direct reclaim */
3202 if (current->flags & PF_MEMALLOC) {
3204 * __GFP_NOFAIL request from this context is rather bizarre
3205 * because we cannot reclaim anything and only can loop waiting
3206 * for somebody to do a work for us.
3208 if (WARN_ON_ONCE(gfp_mask & __GFP_NOFAIL)) {
3215 /* Avoid allocations with no watermarks from looping endlessly */
3216 if (test_thread_flag(TIF_MEMDIE) && !(gfp_mask & __GFP_NOFAIL))
3220 * Try direct compaction. The first pass is asynchronous. Subsequent
3221 * attempts after direct reclaim are synchronous
3223 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags, ac,
3225 &contended_compaction,
3226 &deferred_compaction);
3230 /* Checks for THP-specific high-order allocations */
3231 if (is_thp_gfp_mask(gfp_mask)) {
3233 * If compaction is deferred for high-order allocations, it is
3234 * because sync compaction recently failed. If this is the case
3235 * and the caller requested a THP allocation, we do not want
3236 * to heavily disrupt the system, so we fail the allocation
3237 * instead of entering direct reclaim.
3239 if (deferred_compaction)
3243 * In all zones where compaction was attempted (and not
3244 * deferred or skipped), lock contention has been detected.
3245 * For THP allocation we do not want to disrupt the others
3246 * so we fallback to base pages instead.
3248 if (contended_compaction == COMPACT_CONTENDED_LOCK)
3252 * If compaction was aborted due to need_resched(), we do not
3253 * want to further increase allocation latency, unless it is
3254 * khugepaged trying to collapse.
3256 if (contended_compaction == COMPACT_CONTENDED_SCHED
3257 && !(current->flags & PF_KTHREAD))
3262 * It can become very expensive to allocate transparent hugepages at
3263 * fault, so use asynchronous memory compaction for THP unless it is
3264 * khugepaged trying to collapse.
3266 if (!is_thp_gfp_mask(gfp_mask) || (current->flags & PF_KTHREAD))
3267 migration_mode = MIGRATE_SYNC_LIGHT;
3269 /* Try direct reclaim and then allocating */
3270 page = __alloc_pages_direct_reclaim(gfp_mask, order, alloc_flags, ac,
3271 &did_some_progress);
3275 /* Do not loop if specifically requested */
3276 if (gfp_mask & __GFP_NORETRY)
3279 /* Keep reclaiming pages as long as there is reasonable progress */
3280 pages_reclaimed += did_some_progress;
3281 if ((did_some_progress && order <= PAGE_ALLOC_COSTLY_ORDER) ||
3282 ((gfp_mask & __GFP_REPEAT) && pages_reclaimed < (1 << order))) {
3283 /* Wait for some write requests to complete then retry */
3284 wait_iff_congested(ac->preferred_zone, BLK_RW_ASYNC, HZ/50);
3288 /* Reclaim has failed us, start killing things */
3289 page = __alloc_pages_may_oom(gfp_mask, order, ac, &did_some_progress);
3293 /* Retry as long as the OOM killer is making progress */
3294 if (did_some_progress)
3299 * High-order allocations do not necessarily loop after
3300 * direct reclaim and reclaim/compaction depends on compaction
3301 * being called after reclaim so call directly if necessary
3303 page = __alloc_pages_direct_compact(gfp_mask, order, alloc_flags,
3305 &contended_compaction,
3306 &deferred_compaction);
3310 warn_alloc_failed(gfp_mask, order, NULL);
3316 * This is the 'heart' of the zoned buddy allocator.
3319 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
3320 struct zonelist *zonelist, nodemask_t *nodemask)
3322 struct zoneref *preferred_zoneref;
3323 struct page *page = NULL;
3324 unsigned int cpuset_mems_cookie;
3325 int alloc_flags = ALLOC_WMARK_LOW|ALLOC_CPUSET|ALLOC_FAIR;
3326 gfp_t alloc_mask; /* The gfp_t that was actually used for allocation */
3327 struct alloc_context ac = {
3328 .high_zoneidx = gfp_zone(gfp_mask),
3329 .nodemask = nodemask,
3330 .migratetype = gfpflags_to_migratetype(gfp_mask),
3333 gfp_mask &= gfp_allowed_mask;
3335 lockdep_trace_alloc(gfp_mask);
3337 might_sleep_if(gfp_mask & __GFP_DIRECT_RECLAIM);
3339 if (should_fail_alloc_page(gfp_mask, order))
3343 * Check the zones suitable for the gfp_mask contain at least one
3344 * valid zone. It's possible to have an empty zonelist as a result
3345 * of __GFP_THISNODE and a memoryless node
3347 if (unlikely(!zonelist->_zonerefs->zone))
3350 if (IS_ENABLED(CONFIG_CMA) && ac.migratetype == MIGRATE_MOVABLE)
3351 alloc_flags |= ALLOC_CMA;
3354 cpuset_mems_cookie = read_mems_allowed_begin();
3356 /* We set it here, as __alloc_pages_slowpath might have changed it */
3357 ac.zonelist = zonelist;
3359 /* Dirty zone balancing only done in the fast path */
3360 ac.spread_dirty_pages = (gfp_mask & __GFP_WRITE);
3362 /* The preferred zone is used for statistics later */
3363 preferred_zoneref = first_zones_zonelist(ac.zonelist, ac.high_zoneidx,
3364 ac.nodemask ? : &cpuset_current_mems_allowed,
3365 &ac.preferred_zone);
3366 if (!ac.preferred_zone)
3368 ac.classzone_idx = zonelist_zone_idx(preferred_zoneref);
3370 /* First allocation attempt */
3371 alloc_mask = gfp_mask|__GFP_HARDWALL;
3372 page = get_page_from_freelist(alloc_mask, order, alloc_flags, &ac);
3373 if (unlikely(!page)) {
3375 * Runtime PM, block IO and its error handling path
3376 * can deadlock because I/O on the device might not
3379 alloc_mask = memalloc_noio_flags(gfp_mask);
3380 ac.spread_dirty_pages = false;
3382 page = __alloc_pages_slowpath(alloc_mask, order, &ac);
3385 if (kmemcheck_enabled && page)
3386 kmemcheck_pagealloc_alloc(page, order, gfp_mask);
3388 trace_mm_page_alloc(page, order, alloc_mask, ac.migratetype);
3392 * When updating a task's mems_allowed, it is possible to race with
3393 * parallel threads in such a way that an allocation can fail while
3394 * the mask is being updated. If a page allocation is about to fail,
3395 * check if the cpuset changed during allocation and if so, retry.
3397 if (unlikely(!page && read_mems_allowed_retry(cpuset_mems_cookie)))
3402 EXPORT_SYMBOL(__alloc_pages_nodemask);
3405 * Common helper functions.
3407 unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order)
3412 * __get_free_pages() returns a 32-bit address, which cannot represent
3415 VM_BUG_ON((gfp_mask & __GFP_HIGHMEM) != 0);
3417 page = alloc_pages(gfp_mask, order);
3420 return (unsigned long) page_address(page);
3422 EXPORT_SYMBOL(__get_free_pages);
3424 unsigned long get_zeroed_page(gfp_t gfp_mask)
3426 return __get_free_pages(gfp_mask | __GFP_ZERO, 0);
3428 EXPORT_SYMBOL(get_zeroed_page);
3430 void __free_pages(struct page *page, unsigned int order)
3432 if (put_page_testzero(page)) {
3434 free_hot_cold_page(page, false);
3436 __free_pages_ok(page, order);
3440 EXPORT_SYMBOL(__free_pages);
3442 void free_pages(unsigned long addr, unsigned int order)
3445 VM_BUG_ON(!virt_addr_valid((void *)addr));
3446 __free_pages(virt_to_page((void *)addr), order);
3450 EXPORT_SYMBOL(free_pages);
3454 * An arbitrary-length arbitrary-offset area of memory which resides
3455 * within a 0 or higher order page. Multiple fragments within that page
3456 * are individually refcounted, in the page's reference counter.
3458 * The page_frag functions below provide a simple allocation framework for
3459 * page fragments. This is used by the network stack and network device
3460 * drivers to provide a backing region of memory for use as either an
3461 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3463 static struct page *__page_frag_refill(struct page_frag_cache *nc,
3466 struct page *page = NULL;
3467 gfp_t gfp = gfp_mask;
3469 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3470 gfp_mask |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY |
3472 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask,
3473 PAGE_FRAG_CACHE_MAX_ORDER);
3474 nc->size = page ? PAGE_FRAG_CACHE_MAX_SIZE : PAGE_SIZE;
3476 if (unlikely(!page))
3477 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
3479 nc->va = page ? page_address(page) : NULL;
3484 void *__alloc_page_frag(struct page_frag_cache *nc,
3485 unsigned int fragsz, gfp_t gfp_mask)
3487 unsigned int size = PAGE_SIZE;
3491 if (unlikely(!nc->va)) {
3493 page = __page_frag_refill(nc, gfp_mask);
3497 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3498 /* if size can vary use size else just use PAGE_SIZE */
3501 /* Even if we own the page, we do not use atomic_set().
3502 * This would break get_page_unless_zero() users.
3504 page_ref_add(page, size - 1);
3506 /* reset page count bias and offset to start of new frag */
3507 nc->pfmemalloc = page_is_pfmemalloc(page);
3508 nc->pagecnt_bias = size;
3512 offset = nc->offset - fragsz;
3513 if (unlikely(offset < 0)) {
3514 page = virt_to_page(nc->va);
3516 if (!page_ref_sub_and_test(page, nc->pagecnt_bias))
3519 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3520 /* if size can vary use size else just use PAGE_SIZE */
3523 /* OK, page count is 0, we can safely set it */
3524 set_page_count(page, size);
3526 /* reset page count bias and offset to start of new frag */
3527 nc->pagecnt_bias = size;
3528 offset = size - fragsz;
3532 nc->offset = offset;
3534 return nc->va + offset;
3536 EXPORT_SYMBOL(__alloc_page_frag);
3539 * Frees a page fragment allocated out of either a compound or order 0 page.
3541 void __free_page_frag(void *addr)
3543 struct page *page = virt_to_head_page(addr);
3545 if (unlikely(put_page_testzero(page)))
3546 __free_pages_ok(page, compound_order(page));
3548 EXPORT_SYMBOL(__free_page_frag);
3551 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
3552 * of the current memory cgroup if __GFP_ACCOUNT is set, other than that it is
3553 * equivalent to alloc_pages.
3555 * It should be used when the caller would like to use kmalloc, but since the
3556 * allocation is large, it has to fall back to the page allocator.
3558 struct page *alloc_kmem_pages(gfp_t gfp_mask, unsigned int order)
3562 page = alloc_pages(gfp_mask, order);
3563 if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) {
3564 __free_pages(page, order);
3570 struct page *alloc_kmem_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
3574 page = alloc_pages_node(nid, gfp_mask, order);
3575 if (page && memcg_kmem_charge(page, gfp_mask, order) != 0) {
3576 __free_pages(page, order);
3583 * __free_kmem_pages and free_kmem_pages will free pages allocated with
3586 void __free_kmem_pages(struct page *page, unsigned int order)
3588 memcg_kmem_uncharge(page, order);
3589 __free_pages(page, order);
3592 void free_kmem_pages(unsigned long addr, unsigned int order)
3595 VM_BUG_ON(!virt_addr_valid((void *)addr));
3596 __free_kmem_pages(virt_to_page((void *)addr), order);
3600 static void *make_alloc_exact(unsigned long addr, unsigned int order,
3604 unsigned long alloc_end = addr + (PAGE_SIZE << order);
3605 unsigned long used = addr + PAGE_ALIGN(size);
3607 split_page(virt_to_page((void *)addr), order);
3608 while (used < alloc_end) {
3613 return (void *)addr;
3617 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3618 * @size: the number of bytes to allocate
3619 * @gfp_mask: GFP flags for the allocation
3621 * This function is similar to alloc_pages(), except that it allocates the
3622 * minimum number of pages to satisfy the request. alloc_pages() can only
3623 * allocate memory in power-of-two pages.
3625 * This function is also limited by MAX_ORDER.
3627 * Memory allocated by this function must be released by free_pages_exact().
3629 void *alloc_pages_exact(size_t size, gfp_t gfp_mask)
3631 unsigned int order = get_order(size);
3634 addr = __get_free_pages(gfp_mask, order);
3635 return make_alloc_exact(addr, order, size);
3637 EXPORT_SYMBOL(alloc_pages_exact);
3640 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3642 * @nid: the preferred node ID where memory should be allocated
3643 * @size: the number of bytes to allocate
3644 * @gfp_mask: GFP flags for the allocation
3646 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3649 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask)
3651 unsigned int order = get_order(size);
3652 struct page *p = alloc_pages_node(nid, gfp_mask, order);
3655 return make_alloc_exact((unsigned long)page_address(p), order, size);
3659 * free_pages_exact - release memory allocated via alloc_pages_exact()
3660 * @virt: the value returned by alloc_pages_exact.
3661 * @size: size of allocation, same value as passed to alloc_pages_exact().
3663 * Release the memory allocated by a previous call to alloc_pages_exact.
3665 void free_pages_exact(void *virt, size_t size)
3667 unsigned long addr = (unsigned long)virt;
3668 unsigned long end = addr + PAGE_ALIGN(size);
3670 while (addr < end) {
3675 EXPORT_SYMBOL(free_pages_exact);
3678 * nr_free_zone_pages - count number of pages beyond high watermark
3679 * @offset: The zone index of the highest zone
3681 * nr_free_zone_pages() counts the number of counts pages which are beyond the
3682 * high watermark within all zones at or below a given zone index. For each
3683 * zone, the number of pages is calculated as:
3684 * managed_pages - high_pages
3686 static unsigned long nr_free_zone_pages(int offset)
3691 /* Just pick one node, since fallback list is circular */
3692 unsigned long sum = 0;
3694 struct zonelist *zonelist = node_zonelist(numa_node_id(), GFP_KERNEL);
3696 for_each_zone_zonelist(zone, z, zonelist, offset) {
3697 unsigned long size = zone->managed_pages;
3698 unsigned long high = high_wmark_pages(zone);
3707 * nr_free_buffer_pages - count number of pages beyond high watermark
3709 * nr_free_buffer_pages() counts the number of pages which are beyond the high
3710 * watermark within ZONE_DMA and ZONE_NORMAL.
3712 unsigned long nr_free_buffer_pages(void)
3714 return nr_free_zone_pages(gfp_zone(GFP_USER));
3716 EXPORT_SYMBOL_GPL(nr_free_buffer_pages);
3719 * nr_free_pagecache_pages - count number of pages beyond high watermark
3721 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3722 * high watermark within all zones.
3724 unsigned long nr_free_pagecache_pages(void)
3726 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE));
3729 static inline void show_node(struct zone *zone)
3731 if (IS_ENABLED(CONFIG_NUMA))
3732 printk("Node %d ", zone_to_nid(zone));
3735 long si_mem_available(void)
3738 unsigned long pagecache;
3739 unsigned long wmark_low = 0;
3740 unsigned long pages[NR_LRU_LISTS];
3744 for (lru = LRU_BASE; lru < NR_LRU_LISTS; lru++)
3745 pages[lru] = global_page_state(NR_LRU_BASE + lru);
3748 wmark_low += zone->watermark[WMARK_LOW];
3751 * Estimate the amount of memory available for userspace allocations,
3752 * without causing swapping.
3754 available = global_page_state(NR_FREE_PAGES) - totalreserve_pages;
3757 * Not all the page cache can be freed, otherwise the system will
3758 * start swapping. Assume at least half of the page cache, or the
3759 * low watermark worth of cache, needs to stay.
3761 pagecache = pages[LRU_ACTIVE_FILE] + pages[LRU_INACTIVE_FILE];
3762 pagecache -= min(pagecache / 2, wmark_low);
3763 available += pagecache;
3766 * Part of the reclaimable slab consists of items that are in use,
3767 * and cannot be freed. Cap this estimate at the low watermark.
3769 available += global_page_state(NR_SLAB_RECLAIMABLE) -
3770 min(global_page_state(NR_SLAB_RECLAIMABLE) / 2, wmark_low);
3776 EXPORT_SYMBOL_GPL(si_mem_available);
3778 void si_meminfo(struct sysinfo *val)
3780 val->totalram = totalram_pages;
3781 val->sharedram = global_page_state(NR_SHMEM);
3782 val->freeram = global_page_state(NR_FREE_PAGES);
3783 val->bufferram = nr_blockdev_pages();
3784 val->totalhigh = totalhigh_pages;
3785 val->freehigh = nr_free_highpages();
3786 val->mem_unit = PAGE_SIZE;
3789 EXPORT_SYMBOL(si_meminfo);
3792 void si_meminfo_node(struct sysinfo *val, int nid)
3794 int zone_type; /* needs to be signed */
3795 unsigned long managed_pages = 0;
3796 unsigned long managed_highpages = 0;
3797 unsigned long free_highpages = 0;
3798 pg_data_t *pgdat = NODE_DATA(nid);
3800 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++)
3801 managed_pages += pgdat->node_zones[zone_type].managed_pages;
3802 val->totalram = managed_pages;
3803 val->sharedram = node_page_state(nid, NR_SHMEM);
3804 val->freeram = node_page_state(nid, NR_FREE_PAGES);
3805 #ifdef CONFIG_HIGHMEM
3806 for (zone_type = 0; zone_type < MAX_NR_ZONES; zone_type++) {
3807 struct zone *zone = &pgdat->node_zones[zone_type];
3809 if (is_highmem(zone)) {
3810 managed_highpages += zone->managed_pages;
3811 free_highpages += zone_page_state(zone, NR_FREE_PAGES);
3814 val->totalhigh = managed_highpages;
3815 val->freehigh = free_highpages;
3817 val->totalhigh = managed_highpages;
3818 val->freehigh = free_highpages;
3820 val->mem_unit = PAGE_SIZE;
3825 * Determine whether the node should be displayed or not, depending on whether
3826 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
3828 bool skip_free_areas_node(unsigned int flags, int nid)
3831 unsigned int cpuset_mems_cookie;
3833 if (!(flags & SHOW_MEM_FILTER_NODES))
3837 cpuset_mems_cookie = read_mems_allowed_begin();
3838 ret = !node_isset(nid, cpuset_current_mems_allowed);
3839 } while (read_mems_allowed_retry(cpuset_mems_cookie));
3844 #define K(x) ((x) << (PAGE_SHIFT-10))
3846 static void show_migration_types(unsigned char type)
3848 static const char types[MIGRATE_TYPES] = {
3849 [MIGRATE_UNMOVABLE] = 'U',
3850 [MIGRATE_MOVABLE] = 'M',
3851 [MIGRATE_RECLAIMABLE] = 'E',
3852 [MIGRATE_HIGHATOMIC] = 'H',
3854 [MIGRATE_CMA] = 'C',
3856 #ifdef CONFIG_MEMORY_ISOLATION
3857 [MIGRATE_ISOLATE] = 'I',
3860 char tmp[MIGRATE_TYPES + 1];
3864 for (i = 0; i < MIGRATE_TYPES; i++) {
3865 if (type & (1 << i))
3870 printk("(%s) ", tmp);
3874 * Show free area list (used inside shift_scroll-lock stuff)
3875 * We also calculate the percentage fragmentation. We do this by counting the
3876 * memory on each free list with the exception of the first item on the list.
3879 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
3882 void show_free_areas(unsigned int filter)
3884 unsigned long free_pcp = 0;
3888 for_each_populated_zone(zone) {
3889 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3892 for_each_online_cpu(cpu)
3893 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
3896 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3897 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3898 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
3899 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3900 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3901 " free:%lu free_pcp:%lu free_cma:%lu\n",
3902 global_page_state(NR_ACTIVE_ANON),
3903 global_page_state(NR_INACTIVE_ANON),
3904 global_page_state(NR_ISOLATED_ANON),
3905 global_page_state(NR_ACTIVE_FILE),
3906 global_page_state(NR_INACTIVE_FILE),
3907 global_page_state(NR_ISOLATED_FILE),
3908 global_page_state(NR_UNEVICTABLE),
3909 global_page_state(NR_FILE_DIRTY),
3910 global_page_state(NR_WRITEBACK),
3911 global_page_state(NR_UNSTABLE_NFS),
3912 global_page_state(NR_SLAB_RECLAIMABLE),
3913 global_page_state(NR_SLAB_UNRECLAIMABLE),
3914 global_page_state(NR_FILE_MAPPED),
3915 global_page_state(NR_SHMEM),
3916 global_page_state(NR_PAGETABLE),
3917 global_page_state(NR_BOUNCE),
3918 global_page_state(NR_FREE_PAGES),
3920 global_page_state(NR_FREE_CMA_PAGES));
3922 for_each_populated_zone(zone) {
3925 if (skip_free_areas_node(filter, zone_to_nid(zone)))
3929 for_each_online_cpu(cpu)
3930 free_pcp += per_cpu_ptr(zone->pageset, cpu)->pcp.count;
3938 " active_anon:%lukB"
3939 " inactive_anon:%lukB"
3940 " active_file:%lukB"
3941 " inactive_file:%lukB"
3942 " unevictable:%lukB"
3943 " isolated(anon):%lukB"
3944 " isolated(file):%lukB"
3952 " slab_reclaimable:%lukB"
3953 " slab_unreclaimable:%lukB"
3954 " kernel_stack:%lukB"
3961 " writeback_tmp:%lukB"
3962 " pages_scanned:%lu"
3963 " all_unreclaimable? %s"
3966 K(zone_page_state(zone, NR_FREE_PAGES)),
3967 K(min_wmark_pages(zone)),
3968 K(low_wmark_pages(zone)),
3969 K(high_wmark_pages(zone)),
3970 K(zone_page_state(zone, NR_ACTIVE_ANON)),
3971 K(zone_page_state(zone, NR_INACTIVE_ANON)),
3972 K(zone_page_state(zone, NR_ACTIVE_FILE)),
3973 K(zone_page_state(zone, NR_INACTIVE_FILE)),
3974 K(zone_page_state(zone, NR_UNEVICTABLE)),
3975 K(zone_page_state(zone, NR_ISOLATED_ANON)),
3976 K(zone_page_state(zone, NR_ISOLATED_FILE)),
3977 K(zone->present_pages),
3978 K(zone->managed_pages),
3979 K(zone_page_state(zone, NR_MLOCK)),
3980 K(zone_page_state(zone, NR_FILE_DIRTY)),
3981 K(zone_page_state(zone, NR_WRITEBACK)),
3982 K(zone_page_state(zone, NR_FILE_MAPPED)),
3983 K(zone_page_state(zone, NR_SHMEM)),
3984 K(zone_page_state(zone, NR_SLAB_RECLAIMABLE)),
3985 K(zone_page_state(zone, NR_SLAB_UNRECLAIMABLE)),
3986 zone_page_state(zone, NR_KERNEL_STACK) *
3988 K(zone_page_state(zone, NR_PAGETABLE)),
3989 K(zone_page_state(zone, NR_UNSTABLE_NFS)),
3990 K(zone_page_state(zone, NR_BOUNCE)),
3992 K(this_cpu_read(zone->pageset->pcp.count)),
3993 K(zone_page_state(zone, NR_FREE_CMA_PAGES)),
3994 K(zone_page_state(zone, NR_WRITEBACK_TEMP)),
3995 K(zone_page_state(zone, NR_PAGES_SCANNED)),
3996 (!zone_reclaimable(zone) ? "yes" : "no")
3998 printk("lowmem_reserve[]:");
3999 for (i = 0; i < MAX_NR_ZONES; i++)
4000 printk(" %ld", zone->lowmem_reserve[i]);
4004 for_each_populated_zone(zone) {
4006 unsigned long nr[MAX_ORDER], flags, total = 0;
4007 unsigned char types[MAX_ORDER];
4009 if (skip_free_areas_node(filter, zone_to_nid(zone)))
4012 printk("%s: ", zone->name);
4014 spin_lock_irqsave(&zone->lock, flags);
4015 for (order = 0; order < MAX_ORDER; order++) {
4016 struct free_area *area = &zone->free_area[order];
4019 nr[order] = area->nr_free;
4020 total += nr[order] << order;
4023 for (type = 0; type < MIGRATE_TYPES; type++) {
4024 if (!list_empty(&area->free_list[type]))
4025 types[order] |= 1 << type;
4028 spin_unlock_irqrestore(&zone->lock, flags);
4029 for (order = 0; order < MAX_ORDER; order++) {
4030 printk("%lu*%lukB ", nr[order], K(1UL) << order);
4032 show_migration_types(types[order]);
4034 printk("= %lukB\n", K(total));
4037 hugetlb_show_meminfo();
4039 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES));
4041 show_swap_cache_info();
4044 static void zoneref_set_zone(struct zone *zone, struct zoneref *zoneref)
4046 zoneref->zone = zone;
4047 zoneref->zone_idx = zone_idx(zone);
4051 * Builds allocation fallback zone lists.
4053 * Add all populated zones of a node to the zonelist.
4055 static int build_zonelists_node(pg_data_t *pgdat, struct zonelist *zonelist,
4059 enum zone_type zone_type = MAX_NR_ZONES;
4063 zone = pgdat->node_zones + zone_type;
4064 if (populated_zone(zone)) {
4065 zoneref_set_zone(zone,
4066 &zonelist->_zonerefs[nr_zones++]);
4067 check_highest_zone(zone_type);
4069 } while (zone_type);
4077 * 0 = automatic detection of better ordering.
4078 * 1 = order by ([node] distance, -zonetype)
4079 * 2 = order by (-zonetype, [node] distance)
4081 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
4082 * the same zonelist. So only NUMA can configure this param.
4084 #define ZONELIST_ORDER_DEFAULT 0
4085 #define ZONELIST_ORDER_NODE 1
4086 #define ZONELIST_ORDER_ZONE 2
4088 /* zonelist order in the kernel.
4089 * set_zonelist_order() will set this to NODE or ZONE.
4091 static int current_zonelist_order = ZONELIST_ORDER_DEFAULT;
4092 static char zonelist_order_name[3][8] = {"Default", "Node", "Zone"};
4096 /* The value user specified ....changed by config */
4097 static int user_zonelist_order = ZONELIST_ORDER_DEFAULT;
4098 /* string for sysctl */
4099 #define NUMA_ZONELIST_ORDER_LEN 16
4100 char numa_zonelist_order[16] = "default";
4103 * interface for configure zonelist ordering.
4104 * command line option "numa_zonelist_order"
4105 * = "[dD]efault - default, automatic configuration.
4106 * = "[nN]ode - order by node locality, then by zone within node
4107 * = "[zZ]one - order by zone, then by locality within zone
4110 static int __parse_numa_zonelist_order(char *s)
4112 if (*s == 'd' || *s == 'D') {
4113 user_zonelist_order = ZONELIST_ORDER_DEFAULT;
4114 } else if (*s == 'n' || *s == 'N') {
4115 user_zonelist_order = ZONELIST_ORDER_NODE;
4116 } else if (*s == 'z' || *s == 'Z') {
4117 user_zonelist_order = ZONELIST_ORDER_ZONE;
4119 pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s);
4125 static __init int setup_numa_zonelist_order(char *s)
4132 ret = __parse_numa_zonelist_order(s);
4134 strlcpy(numa_zonelist_order, s, NUMA_ZONELIST_ORDER_LEN);
4138 early_param("numa_zonelist_order", setup_numa_zonelist_order);
4141 * sysctl handler for numa_zonelist_order
4143 int numa_zonelist_order_handler(struct ctl_table *table, int write,
4144 void __user *buffer, size_t *length,
4147 char saved_string[NUMA_ZONELIST_ORDER_LEN];
4149 static DEFINE_MUTEX(zl_order_mutex);
4151 mutex_lock(&zl_order_mutex);
4153 if (strlen((char *)table->data) >= NUMA_ZONELIST_ORDER_LEN) {
4157 strcpy(saved_string, (char *)table->data);
4159 ret = proc_dostring(table, write, buffer, length, ppos);
4163 int oldval = user_zonelist_order;
4165 ret = __parse_numa_zonelist_order((char *)table->data);
4168 * bogus value. restore saved string
4170 strncpy((char *)table->data, saved_string,
4171 NUMA_ZONELIST_ORDER_LEN);
4172 user_zonelist_order = oldval;
4173 } else if (oldval != user_zonelist_order) {
4174 mutex_lock(&zonelists_mutex);
4175 build_all_zonelists(NULL, NULL);
4176 mutex_unlock(&zonelists_mutex);
4180 mutex_unlock(&zl_order_mutex);
4185 #define MAX_NODE_LOAD (nr_online_nodes)
4186 static int node_load[MAX_NUMNODES];
4189 * find_next_best_node - find the next node that should appear in a given node's fallback list
4190 * @node: node whose fallback list we're appending
4191 * @used_node_mask: nodemask_t of already used nodes
4193 * We use a number of factors to determine which is the next node that should
4194 * appear on a given node's fallback list. The node should not have appeared
4195 * already in @node's fallback list, and it should be the next closest node
4196 * according to the distance array (which contains arbitrary distance values
4197 * from each node to each node in the system), and should also prefer nodes
4198 * with no CPUs, since presumably they'll have very little allocation pressure
4199 * on them otherwise.
4200 * It returns -1 if no node is found.
4202 static int find_next_best_node(int node, nodemask_t *used_node_mask)
4205 int min_val = INT_MAX;
4206 int best_node = NUMA_NO_NODE;
4207 const struct cpumask *tmp = cpumask_of_node(0);
4209 /* Use the local node if we haven't already */
4210 if (!node_isset(node, *used_node_mask)) {
4211 node_set(node, *used_node_mask);
4215 for_each_node_state(n, N_MEMORY) {
4217 /* Don't want a node to appear more than once */
4218 if (node_isset(n, *used_node_mask))
4221 /* Use the distance array to find the distance */
4222 val = node_distance(node, n);
4224 /* Penalize nodes under us ("prefer the next node") */
4227 /* Give preference to headless and unused nodes */
4228 tmp = cpumask_of_node(n);
4229 if (!cpumask_empty(tmp))
4230 val += PENALTY_FOR_NODE_WITH_CPUS;
4232 /* Slight preference for less loaded node */
4233 val *= (MAX_NODE_LOAD*MAX_NUMNODES);
4234 val += node_load[n];
4236 if (val < min_val) {
4243 node_set(best_node, *used_node_mask);
4250 * Build zonelists ordered by node and zones within node.
4251 * This results in maximum locality--normal zone overflows into local
4252 * DMA zone, if any--but risks exhausting DMA zone.
4254 static void build_zonelists_in_node_order(pg_data_t *pgdat, int node)
4257 struct zonelist *zonelist;
4259 zonelist = &pgdat->node_zonelists[0];
4260 for (j = 0; zonelist->_zonerefs[j].zone != NULL; j++)
4262 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4263 zonelist->_zonerefs[j].zone = NULL;
4264 zonelist->_zonerefs[j].zone_idx = 0;
4268 * Build gfp_thisnode zonelists
4270 static void build_thisnode_zonelists(pg_data_t *pgdat)
4273 struct zonelist *zonelist;
4275 zonelist = &pgdat->node_zonelists[1];
4276 j = build_zonelists_node(pgdat, zonelist, 0);
4277 zonelist->_zonerefs[j].zone = NULL;
4278 zonelist->_zonerefs[j].zone_idx = 0;
4282 * Build zonelists ordered by zone and nodes within zones.
4283 * This results in conserving DMA zone[s] until all Normal memory is
4284 * exhausted, but results in overflowing to remote node while memory
4285 * may still exist in local DMA zone.
4287 static int node_order[MAX_NUMNODES];
4289 static void build_zonelists_in_zone_order(pg_data_t *pgdat, int nr_nodes)
4292 int zone_type; /* needs to be signed */
4294 struct zonelist *zonelist;
4296 zonelist = &pgdat->node_zonelists[0];
4298 for (zone_type = MAX_NR_ZONES - 1; zone_type >= 0; zone_type--) {
4299 for (j = 0; j < nr_nodes; j++) {
4300 node = node_order[j];
4301 z = &NODE_DATA(node)->node_zones[zone_type];
4302 if (populated_zone(z)) {
4304 &zonelist->_zonerefs[pos++]);
4305 check_highest_zone(zone_type);
4309 zonelist->_zonerefs[pos].zone = NULL;
4310 zonelist->_zonerefs[pos].zone_idx = 0;
4313 #if defined(CONFIG_64BIT)
4315 * Devices that require DMA32/DMA are relatively rare and do not justify a
4316 * penalty to every machine in case the specialised case applies. Default
4317 * to Node-ordering on 64-bit NUMA machines
4319 static int default_zonelist_order(void)
4321 return ZONELIST_ORDER_NODE;
4325 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4326 * by the kernel. If processes running on node 0 deplete the low memory zone
4327 * then reclaim will occur more frequency increasing stalls and potentially
4328 * be easier to OOM if a large percentage of the zone is under writeback or
4329 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4330 * Hence, default to zone ordering on 32-bit.
4332 static int default_zonelist_order(void)
4334 return ZONELIST_ORDER_ZONE;
4336 #endif /* CONFIG_64BIT */
4338 static void set_zonelist_order(void)
4340 if (user_zonelist_order == ZONELIST_ORDER_DEFAULT)
4341 current_zonelist_order = default_zonelist_order();
4343 current_zonelist_order = user_zonelist_order;
4346 static void build_zonelists(pg_data_t *pgdat)
4349 nodemask_t used_mask;
4350 int local_node, prev_node;
4351 struct zonelist *zonelist;
4352 unsigned int order = current_zonelist_order;
4354 /* initialize zonelists */
4355 for (i = 0; i < MAX_ZONELISTS; i++) {
4356 zonelist = pgdat->node_zonelists + i;
4357 zonelist->_zonerefs[0].zone = NULL;
4358 zonelist->_zonerefs[0].zone_idx = 0;
4361 /* NUMA-aware ordering of nodes */
4362 local_node = pgdat->node_id;
4363 load = nr_online_nodes;
4364 prev_node = local_node;
4365 nodes_clear(used_mask);
4367 memset(node_order, 0, sizeof(node_order));
4370 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
4372 * We don't want to pressure a particular node.
4373 * So adding penalty to the first node in same
4374 * distance group to make it round-robin.
4376 if (node_distance(local_node, node) !=
4377 node_distance(local_node, prev_node))
4378 node_load[node] = load;
4382 if (order == ZONELIST_ORDER_NODE)
4383 build_zonelists_in_node_order(pgdat, node);
4385 node_order[i++] = node; /* remember order */
4388 if (order == ZONELIST_ORDER_ZONE) {
4389 /* calculate node order -- i.e., DMA last! */
4390 build_zonelists_in_zone_order(pgdat, i);
4393 build_thisnode_zonelists(pgdat);
4396 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4398 * Return node id of node used for "local" allocations.
4399 * I.e., first node id of first zone in arg node's generic zonelist.
4400 * Used for initializing percpu 'numa_mem', which is used primarily
4401 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4403 int local_memory_node(int node)
4407 (void)first_zones_zonelist(node_zonelist(node, GFP_KERNEL),
4408 gfp_zone(GFP_KERNEL),
4415 #else /* CONFIG_NUMA */
4417 static void set_zonelist_order(void)
4419 current_zonelist_order = ZONELIST_ORDER_ZONE;
4422 static void build_zonelists(pg_data_t *pgdat)
4424 int node, local_node;
4426 struct zonelist *zonelist;
4428 local_node = pgdat->node_id;
4430 zonelist = &pgdat->node_zonelists[0];
4431 j = build_zonelists_node(pgdat, zonelist, 0);
4434 * Now we build the zonelist so that it contains the zones
4435 * of all the other nodes.
4436 * We don't want to pressure a particular node, so when
4437 * building the zones for node N, we make sure that the
4438 * zones coming right after the local ones are those from
4439 * node N+1 (modulo N)
4441 for (node = local_node + 1; node < MAX_NUMNODES; node++) {
4442 if (!node_online(node))
4444 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4446 for (node = 0; node < local_node; node++) {
4447 if (!node_online(node))
4449 j = build_zonelists_node(NODE_DATA(node), zonelist, j);
4452 zonelist->_zonerefs[j].zone = NULL;
4453 zonelist->_zonerefs[j].zone_idx = 0;
4456 #endif /* CONFIG_NUMA */
4459 * Boot pageset table. One per cpu which is going to be used for all
4460 * zones and all nodes. The parameters will be set in such a way
4461 * that an item put on a list will immediately be handed over to
4462 * the buddy list. This is safe since pageset manipulation is done
4463 * with interrupts disabled.
4465 * The boot_pagesets must be kept even after bootup is complete for
4466 * unused processors and/or zones. They do play a role for bootstrapping
4467 * hotplugged processors.
4469 * zoneinfo_show() and maybe other functions do
4470 * not check if the processor is online before following the pageset pointer.
4471 * Other parts of the kernel may not check if the zone is available.
4473 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch);
4474 static DEFINE_PER_CPU(struct per_cpu_pageset, boot_pageset);
4475 static void setup_zone_pageset(struct zone *zone);
4478 * Global mutex to protect against size modification of zonelists
4479 * as well as to serialize pageset setup for the new populated zone.
4481 DEFINE_MUTEX(zonelists_mutex);
4483 /* return values int ....just for stop_machine() */
4484 static int __build_all_zonelists(void *data)
4488 pg_data_t *self = data;
4491 memset(node_load, 0, sizeof(node_load));
4494 if (self && !node_online(self->node_id)) {
4495 build_zonelists(self);
4498 for_each_online_node(nid) {
4499 pg_data_t *pgdat = NODE_DATA(nid);
4501 build_zonelists(pgdat);
4505 * Initialize the boot_pagesets that are going to be used
4506 * for bootstrapping processors. The real pagesets for
4507 * each zone will be allocated later when the per cpu
4508 * allocator is available.
4510 * boot_pagesets are used also for bootstrapping offline
4511 * cpus if the system is already booted because the pagesets
4512 * are needed to initialize allocators on a specific cpu too.
4513 * F.e. the percpu allocator needs the page allocator which
4514 * needs the percpu allocator in order to allocate its pagesets
4515 * (a chicken-egg dilemma).
4517 for_each_possible_cpu(cpu) {
4518 setup_pageset(&per_cpu(boot_pageset, cpu), 0);
4520 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4522 * We now know the "local memory node" for each node--
4523 * i.e., the node of the first zone in the generic zonelist.
4524 * Set up numa_mem percpu variable for on-line cpus. During
4525 * boot, only the boot cpu should be on-line; we'll init the
4526 * secondary cpus' numa_mem as they come on-line. During
4527 * node/memory hotplug, we'll fixup all on-line cpus.
4529 if (cpu_online(cpu))
4530 set_cpu_numa_mem(cpu, local_memory_node(cpu_to_node(cpu)));
4537 static noinline void __init
4538 build_all_zonelists_init(void)
4540 __build_all_zonelists(NULL);
4541 mminit_verify_zonelist();
4542 cpuset_init_current_mems_allowed();
4546 * Called with zonelists_mutex held always
4547 * unless system_state == SYSTEM_BOOTING.
4549 * __ref due to (1) call of __meminit annotated setup_zone_pageset
4550 * [we're only called with non-NULL zone through __meminit paths] and
4551 * (2) call of __init annotated helper build_all_zonelists_init
4552 * [protected by SYSTEM_BOOTING].
4554 void __ref build_all_zonelists(pg_data_t *pgdat, struct zone *zone)
4556 set_zonelist_order();
4558 if (system_state == SYSTEM_BOOTING) {
4559 build_all_zonelists_init();
4561 #ifdef CONFIG_MEMORY_HOTPLUG
4563 setup_zone_pageset(zone);
4565 /* we have to stop all cpus to guarantee there is no user
4567 stop_machine(__build_all_zonelists, pgdat, NULL);
4568 /* cpuset refresh routine should be here */
4570 vm_total_pages = nr_free_pagecache_pages();
4572 * Disable grouping by mobility if the number of pages in the
4573 * system is too low to allow the mechanism to work. It would be
4574 * more accurate, but expensive to check per-zone. This check is
4575 * made on memory-hotadd so a system can start with mobility
4576 * disabled and enable it later
4578 if (vm_total_pages < (pageblock_nr_pages * MIGRATE_TYPES))
4579 page_group_by_mobility_disabled = 1;
4581 page_group_by_mobility_disabled = 0;
4583 pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n",
4585 zonelist_order_name[current_zonelist_order],
4586 page_group_by_mobility_disabled ? "off" : "on",
4589 pr_info("Policy zone: %s\n", zone_names[policy_zone]);
4594 * Helper functions to size the waitqueue hash table.
4595 * Essentially these want to choose hash table sizes sufficiently
4596 * large so that collisions trying to wait on pages are rare.
4597 * But in fact, the number of active page waitqueues on typical
4598 * systems is ridiculously low, less than 200. So this is even
4599 * conservative, even though it seems large.
4601 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
4602 * waitqueues, i.e. the size of the waitq table given the number of pages.
4604 #define PAGES_PER_WAITQUEUE 256
4606 #ifndef CONFIG_MEMORY_HOTPLUG
4607 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4609 unsigned long size = 1;
4611 pages /= PAGES_PER_WAITQUEUE;
4613 while (size < pages)
4617 * Once we have dozens or even hundreds of threads sleeping
4618 * on IO we've got bigger problems than wait queue collision.
4619 * Limit the size of the wait table to a reasonable size.
4621 size = min(size, 4096UL);
4623 return max(size, 4UL);
4627 * A zone's size might be changed by hot-add, so it is not possible to determine
4628 * a suitable size for its wait_table. So we use the maximum size now.
4630 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
4632 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
4633 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4634 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
4636 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4637 * or more by the traditional way. (See above). It equals:
4639 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
4640 * ia64(16K page size) : = ( 8G + 4M)byte.
4641 * powerpc (64K page size) : = (32G +16M)byte.
4643 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages)
4650 * This is an integer logarithm so that shifts can be used later
4651 * to extract the more random high bits from the multiplicative
4652 * hash function before the remainder is taken.
4654 static inline unsigned long wait_table_bits(unsigned long size)
4660 * Initially all pages are reserved - free ones are freed
4661 * up by free_all_bootmem() once the early boot process is
4662 * done. Non-atomic initialization, single-pass.
4664 void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
4665 unsigned long start_pfn, enum memmap_context context)
4667 struct vmem_altmap *altmap = to_vmem_altmap(__pfn_to_phys(start_pfn));
4668 unsigned long end_pfn = start_pfn + size;
4669 pg_data_t *pgdat = NODE_DATA(nid);
4671 unsigned long nr_initialised = 0;
4672 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4673 struct memblock_region *r = NULL, *tmp;
4676 if (highest_memmap_pfn < end_pfn - 1)
4677 highest_memmap_pfn = end_pfn - 1;
4680 * Honor reservation requested by the driver for this ZONE_DEVICE
4683 if (altmap && start_pfn == altmap->base_pfn)
4684 start_pfn += altmap->reserve;
4686 for (pfn = start_pfn; pfn < end_pfn; pfn++) {
4688 * There can be holes in boot-time mem_map[]s handed to this
4689 * function. They do not exist on hotplugged memory.
4691 if (context != MEMMAP_EARLY)
4694 if (!early_pfn_valid(pfn))
4696 if (!early_pfn_in_nid(pfn, nid))
4698 if (!update_defer_init(pgdat, pfn, end_pfn, &nr_initialised))
4701 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4703 * If not mirrored_kernelcore and ZONE_MOVABLE exists, range
4704 * from zone_movable_pfn[nid] to end of each node should be
4705 * ZONE_MOVABLE not ZONE_NORMAL. skip it.
4707 if (!mirrored_kernelcore && zone_movable_pfn[nid])
4708 if (zone == ZONE_NORMAL && pfn >= zone_movable_pfn[nid])
4712 * Check given memblock attribute by firmware which can affect
4713 * kernel memory layout. If zone==ZONE_MOVABLE but memory is
4714 * mirrored, it's an overlapped memmap init. skip it.
4716 if (mirrored_kernelcore && zone == ZONE_MOVABLE) {
4717 if (!r || pfn >= memblock_region_memory_end_pfn(r)) {
4718 for_each_memblock(memory, tmp)
4719 if (pfn < memblock_region_memory_end_pfn(tmp))
4723 if (pfn >= memblock_region_memory_base_pfn(r) &&
4724 memblock_is_mirror(r)) {
4725 /* already initialized as NORMAL */
4726 pfn = memblock_region_memory_end_pfn(r);
4734 * Mark the block movable so that blocks are reserved for
4735 * movable at startup. This will force kernel allocations
4736 * to reserve their blocks rather than leaking throughout
4737 * the address space during boot when many long-lived
4738 * kernel allocations are made.
4740 * bitmap is created for zone's valid pfn range. but memmap
4741 * can be created for invalid pages (for alignment)
4742 * check here not to call set_pageblock_migratetype() against
4745 if (!(pfn & (pageblock_nr_pages - 1))) {
4746 struct page *page = pfn_to_page(pfn);
4748 __init_single_page(page, pfn, zone, nid);
4749 set_pageblock_migratetype(page, MIGRATE_MOVABLE);
4751 __init_single_pfn(pfn, zone, nid);
4756 static void __meminit zone_init_free_lists(struct zone *zone)
4758 unsigned int order, t;
4759 for_each_migratetype_order(order, t) {
4760 INIT_LIST_HEAD(&zone->free_area[order].free_list[t]);
4761 zone->free_area[order].nr_free = 0;
4765 #ifndef __HAVE_ARCH_MEMMAP_INIT
4766 #define memmap_init(size, nid, zone, start_pfn) \
4767 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4770 static int zone_batchsize(struct zone *zone)
4776 * The per-cpu-pages pools are set to around 1000th of the
4777 * size of the zone. But no more than 1/2 of a meg.
4779 * OK, so we don't know how big the cache is. So guess.
4781 batch = zone->managed_pages / 1024;
4782 if (batch * PAGE_SIZE > 512 * 1024)
4783 batch = (512 * 1024) / PAGE_SIZE;
4784 batch /= 4; /* We effectively *= 4 below */
4789 * Clamp the batch to a 2^n - 1 value. Having a power
4790 * of 2 value was found to be more likely to have
4791 * suboptimal cache aliasing properties in some cases.
4793 * For example if 2 tasks are alternately allocating
4794 * batches of pages, one task can end up with a lot
4795 * of pages of one half of the possible page colors
4796 * and the other with pages of the other colors.
4798 batch = rounddown_pow_of_two(batch + batch/2) - 1;
4803 /* The deferral and batching of frees should be suppressed under NOMMU
4806 * The problem is that NOMMU needs to be able to allocate large chunks
4807 * of contiguous memory as there's no hardware page translation to
4808 * assemble apparent contiguous memory from discontiguous pages.
4810 * Queueing large contiguous runs of pages for batching, however,
4811 * causes the pages to actually be freed in smaller chunks. As there
4812 * can be a significant delay between the individual batches being
4813 * recycled, this leads to the once large chunks of space being
4814 * fragmented and becoming unavailable for high-order allocations.
4821 * pcp->high and pcp->batch values are related and dependent on one another:
4822 * ->batch must never be higher then ->high.
4823 * The following function updates them in a safe manner without read side
4826 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4827 * those fields changing asynchronously (acording the the above rule).
4829 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4830 * outside of boot time (or some other assurance that no concurrent updaters
4833 static void pageset_update(struct per_cpu_pages *pcp, unsigned long high,
4834 unsigned long batch)
4836 /* start with a fail safe value for batch */
4840 /* Update high, then batch, in order */
4847 /* a companion to pageset_set_high() */
4848 static void pageset_set_batch(struct per_cpu_pageset *p, unsigned long batch)
4850 pageset_update(&p->pcp, 6 * batch, max(1UL, 1 * batch));
4853 static void pageset_init(struct per_cpu_pageset *p)
4855 struct per_cpu_pages *pcp;
4858 memset(p, 0, sizeof(*p));
4862 for (migratetype = 0; migratetype < MIGRATE_PCPTYPES; migratetype++)
4863 INIT_LIST_HEAD(&pcp->lists[migratetype]);
4866 static void setup_pageset(struct per_cpu_pageset *p, unsigned long batch)
4869 pageset_set_batch(p, batch);
4873 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
4874 * to the value high for the pageset p.
4876 static void pageset_set_high(struct per_cpu_pageset *p,
4879 unsigned long batch = max(1UL, high / 4);
4880 if ((high / 4) > (PAGE_SHIFT * 8))
4881 batch = PAGE_SHIFT * 8;
4883 pageset_update(&p->pcp, high, batch);
4886 static void pageset_set_high_and_batch(struct zone *zone,
4887 struct per_cpu_pageset *pcp)
4889 if (percpu_pagelist_fraction)
4890 pageset_set_high(pcp,
4891 (zone->managed_pages /
4892 percpu_pagelist_fraction));
4894 pageset_set_batch(pcp, zone_batchsize(zone));
4897 static void __meminit zone_pageset_init(struct zone *zone, int cpu)
4899 struct per_cpu_pageset *pcp = per_cpu_ptr(zone->pageset, cpu);
4902 pageset_set_high_and_batch(zone, pcp);
4905 static void __meminit setup_zone_pageset(struct zone *zone)
4908 zone->pageset = alloc_percpu(struct per_cpu_pageset);
4909 for_each_possible_cpu(cpu)
4910 zone_pageset_init(zone, cpu);
4914 * Allocate per cpu pagesets and initialize them.
4915 * Before this call only boot pagesets were available.
4917 void __init setup_per_cpu_pageset(void)
4921 for_each_populated_zone(zone)
4922 setup_zone_pageset(zone);
4925 static noinline __init_refok
4926 int zone_wait_table_init(struct zone *zone, unsigned long zone_size_pages)
4932 * The per-page waitqueue mechanism uses hashed waitqueues
4935 zone->wait_table_hash_nr_entries =
4936 wait_table_hash_nr_entries(zone_size_pages);
4937 zone->wait_table_bits =
4938 wait_table_bits(zone->wait_table_hash_nr_entries);
4939 alloc_size = zone->wait_table_hash_nr_entries
4940 * sizeof(wait_queue_head_t);
4942 if (!slab_is_available()) {
4943 zone->wait_table = (wait_queue_head_t *)
4944 memblock_virt_alloc_node_nopanic(
4945 alloc_size, zone->zone_pgdat->node_id);
4948 * This case means that a zone whose size was 0 gets new memory
4949 * via memory hot-add.
4950 * But it may be the case that a new node was hot-added. In
4951 * this case vmalloc() will not be able to use this new node's
4952 * memory - this wait_table must be initialized to use this new
4953 * node itself as well.
4954 * To use this new node's memory, further consideration will be
4957 zone->wait_table = vmalloc(alloc_size);
4959 if (!zone->wait_table)
4962 for (i = 0; i < zone->wait_table_hash_nr_entries; ++i)
4963 init_waitqueue_head(zone->wait_table + i);
4968 static __meminit void zone_pcp_init(struct zone *zone)
4971 * per cpu subsystem is not up at this point. The following code
4972 * relies on the ability of the linker to provide the
4973 * offset of a (static) per cpu variable into the per cpu area.
4975 zone->pageset = &boot_pageset;
4977 if (populated_zone(zone))
4978 printk(KERN_DEBUG " %s zone: %lu pages, LIFO batch:%u\n",
4979 zone->name, zone->present_pages,
4980 zone_batchsize(zone));
4983 int __meminit init_currently_empty_zone(struct zone *zone,
4984 unsigned long zone_start_pfn,
4987 struct pglist_data *pgdat = zone->zone_pgdat;
4989 ret = zone_wait_table_init(zone, size);
4992 pgdat->nr_zones = zone_idx(zone) + 1;
4994 zone->zone_start_pfn = zone_start_pfn;
4996 mminit_dprintk(MMINIT_TRACE, "memmap_init",
4997 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4999 (unsigned long)zone_idx(zone),
5000 zone_start_pfn, (zone_start_pfn + size));
5002 zone_init_free_lists(zone);
5007 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5008 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
5011 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
5013 int __meminit __early_pfn_to_nid(unsigned long pfn,
5014 struct mminit_pfnnid_cache *state)
5016 unsigned long start_pfn, end_pfn;
5019 if (state->last_start <= pfn && pfn < state->last_end)
5020 return state->last_nid;
5022 nid = memblock_search_pfn_nid(pfn, &start_pfn, &end_pfn);
5024 state->last_start = start_pfn;
5025 state->last_end = end_pfn;
5026 state->last_nid = nid;
5031 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
5034 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
5035 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
5036 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
5038 * If an architecture guarantees that all ranges registered contain no holes
5039 * and may be freed, this this function may be used instead of calling
5040 * memblock_free_early_nid() manually.
5042 void __init free_bootmem_with_active_regions(int nid, unsigned long max_low_pfn)
5044 unsigned long start_pfn, end_pfn;
5047 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid) {
5048 start_pfn = min(start_pfn, max_low_pfn);
5049 end_pfn = min(end_pfn, max_low_pfn);
5051 if (start_pfn < end_pfn)
5052 memblock_free_early_nid(PFN_PHYS(start_pfn),
5053 (end_pfn - start_pfn) << PAGE_SHIFT,
5059 * sparse_memory_present_with_active_regions - Call memory_present for each active range
5060 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
5062 * If an architecture guarantees that all ranges registered contain no holes and may
5063 * be freed, this function may be used instead of calling memory_present() manually.
5065 void __init sparse_memory_present_with_active_regions(int nid)
5067 unsigned long start_pfn, end_pfn;
5070 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, &this_nid)
5071 memory_present(this_nid, start_pfn, end_pfn);
5075 * get_pfn_range_for_nid - Return the start and end page frames for a node
5076 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
5077 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
5078 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
5080 * It returns the start and end page frame of a node based on information
5081 * provided by memblock_set_node(). If called for a node
5082 * with no available memory, a warning is printed and the start and end
5085 void __meminit get_pfn_range_for_nid(unsigned int nid,
5086 unsigned long *start_pfn, unsigned long *end_pfn)
5088 unsigned long this_start_pfn, this_end_pfn;
5094 for_each_mem_pfn_range(i, nid, &this_start_pfn, &this_end_pfn, NULL) {
5095 *start_pfn = min(*start_pfn, this_start_pfn);
5096 *end_pfn = max(*end_pfn, this_end_pfn);
5099 if (*start_pfn == -1UL)
5104 * This finds a zone that can be used for ZONE_MOVABLE pages. The
5105 * assumption is made that zones within a node are ordered in monotonic
5106 * increasing memory addresses so that the "highest" populated zone is used
5108 static void __init find_usable_zone_for_movable(void)
5111 for (zone_index = MAX_NR_ZONES - 1; zone_index >= 0; zone_index--) {
5112 if (zone_index == ZONE_MOVABLE)
5115 if (arch_zone_highest_possible_pfn[zone_index] >
5116 arch_zone_lowest_possible_pfn[zone_index])
5120 VM_BUG_ON(zone_index == -1);
5121 movable_zone = zone_index;
5125 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
5126 * because it is sized independent of architecture. Unlike the other zones,
5127 * the starting point for ZONE_MOVABLE is not fixed. It may be different
5128 * in each node depending on the size of each node and how evenly kernelcore
5129 * is distributed. This helper function adjusts the zone ranges
5130 * provided by the architecture for a given node by using the end of the
5131 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5132 * zones within a node are in order of monotonic increases memory addresses
5134 static void __meminit adjust_zone_range_for_zone_movable(int nid,
5135 unsigned long zone_type,
5136 unsigned long node_start_pfn,
5137 unsigned long node_end_pfn,
5138 unsigned long *zone_start_pfn,
5139 unsigned long *zone_end_pfn)
5141 /* Only adjust if ZONE_MOVABLE is on this node */
5142 if (zone_movable_pfn[nid]) {
5143 /* Size ZONE_MOVABLE */
5144 if (zone_type == ZONE_MOVABLE) {
5145 *zone_start_pfn = zone_movable_pfn[nid];
5146 *zone_end_pfn = min(node_end_pfn,
5147 arch_zone_highest_possible_pfn[movable_zone]);
5149 /* Check if this whole range is within ZONE_MOVABLE */
5150 } else if (*zone_start_pfn >= zone_movable_pfn[nid])
5151 *zone_start_pfn = *zone_end_pfn;
5156 * Return the number of pages a zone spans in a node, including holes
5157 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5159 static unsigned long __meminit zone_spanned_pages_in_node(int nid,
5160 unsigned long zone_type,
5161 unsigned long node_start_pfn,
5162 unsigned long node_end_pfn,
5163 unsigned long *zone_start_pfn,
5164 unsigned long *zone_end_pfn,
5165 unsigned long *ignored)
5167 /* When hotadd a new node from cpu_up(), the node should be empty */
5168 if (!node_start_pfn && !node_end_pfn)
5171 /* Get the start and end of the zone */
5172 *zone_start_pfn = arch_zone_lowest_possible_pfn[zone_type];
5173 *zone_end_pfn = arch_zone_highest_possible_pfn[zone_type];
5174 adjust_zone_range_for_zone_movable(nid, zone_type,
5175 node_start_pfn, node_end_pfn,
5176 zone_start_pfn, zone_end_pfn);
5178 /* Check that this node has pages within the zone's required range */
5179 if (*zone_end_pfn < node_start_pfn || *zone_start_pfn > node_end_pfn)
5182 /* Move the zone boundaries inside the node if necessary */
5183 *zone_end_pfn = min(*zone_end_pfn, node_end_pfn);
5184 *zone_start_pfn = max(*zone_start_pfn, node_start_pfn);
5186 /* Return the spanned pages */
5187 return *zone_end_pfn - *zone_start_pfn;
5191 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
5192 * then all holes in the requested range will be accounted for.
5194 unsigned long __meminit __absent_pages_in_range(int nid,
5195 unsigned long range_start_pfn,
5196 unsigned long range_end_pfn)
5198 unsigned long nr_absent = range_end_pfn - range_start_pfn;
5199 unsigned long start_pfn, end_pfn;
5202 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5203 start_pfn = clamp(start_pfn, range_start_pfn, range_end_pfn);
5204 end_pfn = clamp(end_pfn, range_start_pfn, range_end_pfn);
5205 nr_absent -= end_pfn - start_pfn;
5211 * absent_pages_in_range - Return number of page frames in holes within a range
5212 * @start_pfn: The start PFN to start searching for holes
5213 * @end_pfn: The end PFN to stop searching for holes
5215 * It returns the number of pages frames in memory holes within a range.
5217 unsigned long __init absent_pages_in_range(unsigned long start_pfn,
5218 unsigned long end_pfn)
5220 return __absent_pages_in_range(MAX_NUMNODES, start_pfn, end_pfn);
5223 /* Return the number of page frames in holes in a zone on a node */
5224 static unsigned long __meminit zone_absent_pages_in_node(int nid,
5225 unsigned long zone_type,
5226 unsigned long node_start_pfn,
5227 unsigned long node_end_pfn,
5228 unsigned long *ignored)
5230 unsigned long zone_low = arch_zone_lowest_possible_pfn[zone_type];
5231 unsigned long zone_high = arch_zone_highest_possible_pfn[zone_type];
5232 unsigned long zone_start_pfn, zone_end_pfn;
5233 unsigned long nr_absent;
5235 /* When hotadd a new node from cpu_up(), the node should be empty */
5236 if (!node_start_pfn && !node_end_pfn)
5239 zone_start_pfn = clamp(node_start_pfn, zone_low, zone_high);
5240 zone_end_pfn = clamp(node_end_pfn, zone_low, zone_high);
5242 adjust_zone_range_for_zone_movable(nid, zone_type,
5243 node_start_pfn, node_end_pfn,
5244 &zone_start_pfn, &zone_end_pfn);
5245 nr_absent = __absent_pages_in_range(nid, zone_start_pfn, zone_end_pfn);
5248 * ZONE_MOVABLE handling.
5249 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
5252 if (zone_movable_pfn[nid]) {
5253 if (mirrored_kernelcore) {
5254 unsigned long start_pfn, end_pfn;
5255 struct memblock_region *r;
5257 for_each_memblock(memory, r) {
5258 start_pfn = clamp(memblock_region_memory_base_pfn(r),
5259 zone_start_pfn, zone_end_pfn);
5260 end_pfn = clamp(memblock_region_memory_end_pfn(r),
5261 zone_start_pfn, zone_end_pfn);
5263 if (zone_type == ZONE_MOVABLE &&
5264 memblock_is_mirror(r))
5265 nr_absent += end_pfn - start_pfn;
5267 if (zone_type == ZONE_NORMAL &&
5268 !memblock_is_mirror(r))
5269 nr_absent += end_pfn - start_pfn;
5272 if (zone_type == ZONE_NORMAL)
5273 nr_absent += node_end_pfn - zone_movable_pfn[nid];
5280 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5281 static inline unsigned long __meminit zone_spanned_pages_in_node(int nid,
5282 unsigned long zone_type,
5283 unsigned long node_start_pfn,
5284 unsigned long node_end_pfn,
5285 unsigned long *zone_start_pfn,
5286 unsigned long *zone_end_pfn,
5287 unsigned long *zones_size)
5291 *zone_start_pfn = node_start_pfn;
5292 for (zone = 0; zone < zone_type; zone++)
5293 *zone_start_pfn += zones_size[zone];
5295 *zone_end_pfn = *zone_start_pfn + zones_size[zone_type];
5297 return zones_size[zone_type];
5300 static inline unsigned long __meminit zone_absent_pages_in_node(int nid,
5301 unsigned long zone_type,
5302 unsigned long node_start_pfn,
5303 unsigned long node_end_pfn,
5304 unsigned long *zholes_size)
5309 return zholes_size[zone_type];
5312 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5314 static void __meminit calculate_node_totalpages(struct pglist_data *pgdat,
5315 unsigned long node_start_pfn,
5316 unsigned long node_end_pfn,
5317 unsigned long *zones_size,
5318 unsigned long *zholes_size)
5320 unsigned long realtotalpages = 0, totalpages = 0;
5323 for (i = 0; i < MAX_NR_ZONES; i++) {
5324 struct zone *zone = pgdat->node_zones + i;
5325 unsigned long zone_start_pfn, zone_end_pfn;
5326 unsigned long size, real_size;
5328 size = zone_spanned_pages_in_node(pgdat->node_id, i,
5334 real_size = size - zone_absent_pages_in_node(pgdat->node_id, i,
5335 node_start_pfn, node_end_pfn,
5338 zone->zone_start_pfn = zone_start_pfn;
5340 zone->zone_start_pfn = 0;
5341 zone->spanned_pages = size;
5342 zone->present_pages = real_size;
5345 realtotalpages += real_size;
5348 pgdat->node_spanned_pages = totalpages;
5349 pgdat->node_present_pages = realtotalpages;
5350 printk(KERN_DEBUG "On node %d totalpages: %lu\n", pgdat->node_id,
5354 #ifndef CONFIG_SPARSEMEM
5356 * Calculate the size of the zone->blockflags rounded to an unsigned long
5357 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5358 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
5359 * round what is now in bits to nearest long in bits, then return it in
5362 static unsigned long __init usemap_size(unsigned long zone_start_pfn, unsigned long zonesize)
5364 unsigned long usemapsize;
5366 zonesize += zone_start_pfn & (pageblock_nr_pages-1);
5367 usemapsize = roundup(zonesize, pageblock_nr_pages);
5368 usemapsize = usemapsize >> pageblock_order;
5369 usemapsize *= NR_PAGEBLOCK_BITS;
5370 usemapsize = roundup(usemapsize, 8 * sizeof(unsigned long));
5372 return usemapsize / 8;
5375 static void __init setup_usemap(struct pglist_data *pgdat,
5377 unsigned long zone_start_pfn,
5378 unsigned long zonesize)
5380 unsigned long usemapsize = usemap_size(zone_start_pfn, zonesize);
5381 zone->pageblock_flags = NULL;
5383 zone->pageblock_flags =
5384 memblock_virt_alloc_node_nopanic(usemapsize,
5388 static inline void setup_usemap(struct pglist_data *pgdat, struct zone *zone,
5389 unsigned long zone_start_pfn, unsigned long zonesize) {}
5390 #endif /* CONFIG_SPARSEMEM */
5392 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
5394 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
5395 void __paginginit set_pageblock_order(void)
5399 /* Check that pageblock_nr_pages has not already been setup */
5400 if (pageblock_order)
5403 if (HPAGE_SHIFT > PAGE_SHIFT)
5404 order = HUGETLB_PAGE_ORDER;
5406 order = MAX_ORDER - 1;
5409 * Assume the largest contiguous order of interest is a huge page.
5410 * This value may be variable depending on boot parameters on IA64 and
5413 pageblock_order = order;
5415 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5418 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
5419 * is unused as pageblock_order is set at compile-time. See
5420 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5423 void __paginginit set_pageblock_order(void)
5427 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5429 static unsigned long __paginginit calc_memmap_size(unsigned long spanned_pages,
5430 unsigned long present_pages)
5432 unsigned long pages = spanned_pages;
5435 * Provide a more accurate estimation if there are holes within
5436 * the zone and SPARSEMEM is in use. If there are holes within the
5437 * zone, each populated memory region may cost us one or two extra
5438 * memmap pages due to alignment because memmap pages for each
5439 * populated regions may not naturally algined on page boundary.
5440 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5442 if (spanned_pages > present_pages + (present_pages >> 4) &&
5443 IS_ENABLED(CONFIG_SPARSEMEM))
5444 pages = present_pages;
5446 return PAGE_ALIGN(pages * sizeof(struct page)) >> PAGE_SHIFT;
5450 * Set up the zone data structures:
5451 * - mark all pages reserved
5452 * - mark all memory queues empty
5453 * - clear the memory bitmaps
5455 * NOTE: pgdat should get zeroed by caller.
5457 static void __paginginit free_area_init_core(struct pglist_data *pgdat)
5460 int nid = pgdat->node_id;
5463 pgdat_resize_init(pgdat);
5464 #ifdef CONFIG_NUMA_BALANCING
5465 spin_lock_init(&pgdat->numabalancing_migrate_lock);
5466 pgdat->numabalancing_migrate_nr_pages = 0;
5467 pgdat->numabalancing_migrate_next_window = jiffies;
5469 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5470 spin_lock_init(&pgdat->split_queue_lock);
5471 INIT_LIST_HEAD(&pgdat->split_queue);
5472 pgdat->split_queue_len = 0;
5474 init_waitqueue_head(&pgdat->kswapd_wait);
5475 init_waitqueue_head(&pgdat->pfmemalloc_wait);
5476 #ifdef CONFIG_COMPACTION
5477 init_waitqueue_head(&pgdat->kcompactd_wait);
5479 pgdat_page_ext_init(pgdat);
5481 for (j = 0; j < MAX_NR_ZONES; j++) {
5482 struct zone *zone = pgdat->node_zones + j;
5483 unsigned long size, realsize, freesize, memmap_pages;
5484 unsigned long zone_start_pfn = zone->zone_start_pfn;
5486 size = zone->spanned_pages;
5487 realsize = freesize = zone->present_pages;
5490 * Adjust freesize so that it accounts for how much memory
5491 * is used by this zone for memmap. This affects the watermark
5492 * and per-cpu initialisations
5494 memmap_pages = calc_memmap_size(size, realsize);
5495 if (!is_highmem_idx(j)) {
5496 if (freesize >= memmap_pages) {
5497 freesize -= memmap_pages;
5500 " %s zone: %lu pages used for memmap\n",
5501 zone_names[j], memmap_pages);
5503 pr_warn(" %s zone: %lu pages exceeds freesize %lu\n",
5504 zone_names[j], memmap_pages, freesize);
5507 /* Account for reserved pages */
5508 if (j == 0 && freesize > dma_reserve) {
5509 freesize -= dma_reserve;
5510 printk(KERN_DEBUG " %s zone: %lu pages reserved\n",
5511 zone_names[0], dma_reserve);
5514 if (!is_highmem_idx(j))
5515 nr_kernel_pages += freesize;
5516 /* Charge for highmem memmap if there are enough kernel pages */
5517 else if (nr_kernel_pages > memmap_pages * 2)
5518 nr_kernel_pages -= memmap_pages;
5519 nr_all_pages += freesize;
5522 * Set an approximate value for lowmem here, it will be adjusted
5523 * when the bootmem allocator frees pages into the buddy system.
5524 * And all highmem pages will be managed by the buddy system.
5526 zone->managed_pages = is_highmem_idx(j) ? realsize : freesize;
5529 zone->min_unmapped_pages = (freesize*sysctl_min_unmapped_ratio)
5531 zone->min_slab_pages = (freesize * sysctl_min_slab_ratio) / 100;
5533 zone->name = zone_names[j];
5534 spin_lock_init(&zone->lock);
5535 spin_lock_init(&zone->lru_lock);
5536 zone_seqlock_init(zone);
5537 zone->zone_pgdat = pgdat;
5538 zone_pcp_init(zone);
5540 /* For bootup, initialized properly in watermark setup */
5541 mod_zone_page_state(zone, NR_ALLOC_BATCH, zone->managed_pages);
5543 lruvec_init(&zone->lruvec);
5547 set_pageblock_order();
5548 setup_usemap(pgdat, zone, zone_start_pfn, size);
5549 ret = init_currently_empty_zone(zone, zone_start_pfn, size);
5551 memmap_init(size, nid, j, zone_start_pfn);
5555 static void __init_refok alloc_node_mem_map(struct pglist_data *pgdat)
5557 unsigned long __maybe_unused start = 0;
5558 unsigned long __maybe_unused offset = 0;
5560 /* Skip empty nodes */
5561 if (!pgdat->node_spanned_pages)
5564 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5565 start = pgdat->node_start_pfn & ~(MAX_ORDER_NR_PAGES - 1);
5566 offset = pgdat->node_start_pfn - start;
5567 /* ia64 gets its own node_mem_map, before this, without bootmem */
5568 if (!pgdat->node_mem_map) {
5569 unsigned long size, end;
5573 * The zone's endpoints aren't required to be MAX_ORDER
5574 * aligned but the node_mem_map endpoints must be in order
5575 * for the buddy allocator to function correctly.
5577 end = pgdat_end_pfn(pgdat);
5578 end = ALIGN(end, MAX_ORDER_NR_PAGES);
5579 size = (end - start) * sizeof(struct page);
5580 map = alloc_remap(pgdat->node_id, size);
5582 map = memblock_virt_alloc_node_nopanic(size,
5584 pgdat->node_mem_map = map + offset;
5586 #ifndef CONFIG_NEED_MULTIPLE_NODES
5588 * With no DISCONTIG, the global mem_map is just set as node 0's
5590 if (pgdat == NODE_DATA(0)) {
5591 mem_map = NODE_DATA(0)->node_mem_map;
5592 #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
5593 if (page_to_pfn(mem_map) != pgdat->node_start_pfn)
5595 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5598 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
5601 void __paginginit free_area_init_node(int nid, unsigned long *zones_size,
5602 unsigned long node_start_pfn, unsigned long *zholes_size)
5604 pg_data_t *pgdat = NODE_DATA(nid);
5605 unsigned long start_pfn = 0;
5606 unsigned long end_pfn = 0;
5608 /* pg_data_t should be reset to zero when it's allocated */
5609 WARN_ON(pgdat->nr_zones || pgdat->classzone_idx);
5611 reset_deferred_meminit(pgdat);
5612 pgdat->node_id = nid;
5613 pgdat->node_start_pfn = node_start_pfn;
5614 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5615 get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
5616 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid,
5617 (u64)start_pfn << PAGE_SHIFT,
5618 end_pfn ? ((u64)end_pfn << PAGE_SHIFT) - 1 : 0);
5620 start_pfn = node_start_pfn;
5622 calculate_node_totalpages(pgdat, start_pfn, end_pfn,
5623 zones_size, zholes_size);
5625 alloc_node_mem_map(pgdat);
5626 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5627 printk(KERN_DEBUG "free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5628 nid, (unsigned long)pgdat,
5629 (unsigned long)pgdat->node_mem_map);
5632 free_area_init_core(pgdat);
5635 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5637 #if MAX_NUMNODES > 1
5639 * Figure out the number of possible node ids.
5641 void __init setup_nr_node_ids(void)
5643 unsigned int highest;
5645 highest = find_last_bit(node_possible_map.bits, MAX_NUMNODES);
5646 nr_node_ids = highest + 1;
5651 * node_map_pfn_alignment - determine the maximum internode alignment
5653 * This function should be called after node map is populated and sorted.
5654 * It calculates the maximum power of two alignment which can distinguish
5657 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5658 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
5659 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
5660 * shifted, 1GiB is enough and this function will indicate so.
5662 * This is used to test whether pfn -> nid mapping of the chosen memory
5663 * model has fine enough granularity to avoid incorrect mapping for the
5664 * populated node map.
5666 * Returns the determined alignment in pfn's. 0 if there is no alignment
5667 * requirement (single node).
5669 unsigned long __init node_map_pfn_alignment(void)
5671 unsigned long accl_mask = 0, last_end = 0;
5672 unsigned long start, end, mask;
5676 for_each_mem_pfn_range(i, MAX_NUMNODES, &start, &end, &nid) {
5677 if (!start || last_nid < 0 || last_nid == nid) {
5684 * Start with a mask granular enough to pin-point to the
5685 * start pfn and tick off bits one-by-one until it becomes
5686 * too coarse to separate the current node from the last.
5688 mask = ~((1 << __ffs(start)) - 1);
5689 while (mask && last_end <= (start & (mask << 1)))
5692 /* accumulate all internode masks */
5696 /* convert mask to number of pages */
5697 return ~accl_mask + 1;
5700 /* Find the lowest pfn for a node */
5701 static unsigned long __init find_min_pfn_for_node(int nid)
5703 unsigned long min_pfn = ULONG_MAX;
5704 unsigned long start_pfn;
5707 for_each_mem_pfn_range(i, nid, &start_pfn, NULL, NULL)
5708 min_pfn = min(min_pfn, start_pfn);
5710 if (min_pfn == ULONG_MAX) {
5711 pr_warn("Could not find start_pfn for node %d\n", nid);
5719 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5721 * It returns the minimum PFN based on information provided via
5722 * memblock_set_node().
5724 unsigned long __init find_min_pfn_with_active_regions(void)
5726 return find_min_pfn_for_node(MAX_NUMNODES);
5730 * early_calculate_totalpages()
5731 * Sum pages in active regions for movable zone.
5732 * Populate N_MEMORY for calculating usable_nodes.
5734 static unsigned long __init early_calculate_totalpages(void)
5736 unsigned long totalpages = 0;
5737 unsigned long start_pfn, end_pfn;
5740 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid) {
5741 unsigned long pages = end_pfn - start_pfn;
5743 totalpages += pages;
5745 node_set_state(nid, N_MEMORY);
5751 * Find the PFN the Movable zone begins in each node. Kernel memory
5752 * is spread evenly between nodes as long as the nodes have enough
5753 * memory. When they don't, some nodes will have more kernelcore than
5756 static void __init find_zone_movable_pfns_for_nodes(void)
5759 unsigned long usable_startpfn;
5760 unsigned long kernelcore_node, kernelcore_remaining;
5761 /* save the state before borrow the nodemask */
5762 nodemask_t saved_node_state = node_states[N_MEMORY];
5763 unsigned long totalpages = early_calculate_totalpages();
5764 int usable_nodes = nodes_weight(node_states[N_MEMORY]);
5765 struct memblock_region *r;
5767 /* Need to find movable_zone earlier when movable_node is specified. */
5768 find_usable_zone_for_movable();
5771 * If movable_node is specified, ignore kernelcore and movablecore
5774 if (movable_node_is_enabled()) {
5775 for_each_memblock(memory, r) {
5776 if (!memblock_is_hotpluggable(r))
5781 usable_startpfn = PFN_DOWN(r->base);
5782 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5783 min(usable_startpfn, zone_movable_pfn[nid]) :
5791 * If kernelcore=mirror is specified, ignore movablecore option
5793 if (mirrored_kernelcore) {
5794 bool mem_below_4gb_not_mirrored = false;
5796 for_each_memblock(memory, r) {
5797 if (memblock_is_mirror(r))
5802 usable_startpfn = memblock_region_memory_base_pfn(r);
5804 if (usable_startpfn < 0x100000) {
5805 mem_below_4gb_not_mirrored = true;
5809 zone_movable_pfn[nid] = zone_movable_pfn[nid] ?
5810 min(usable_startpfn, zone_movable_pfn[nid]) :
5814 if (mem_below_4gb_not_mirrored)
5815 pr_warn("This configuration results in unmirrored kernel memory.");
5821 * If movablecore=nn[KMG] was specified, calculate what size of
5822 * kernelcore that corresponds so that memory usable for
5823 * any allocation type is evenly spread. If both kernelcore
5824 * and movablecore are specified, then the value of kernelcore
5825 * will be used for required_kernelcore if it's greater than
5826 * what movablecore would have allowed.
5828 if (required_movablecore) {
5829 unsigned long corepages;
5832 * Round-up so that ZONE_MOVABLE is at least as large as what
5833 * was requested by the user
5835 required_movablecore =
5836 roundup(required_movablecore, MAX_ORDER_NR_PAGES);
5837 required_movablecore = min(totalpages, required_movablecore);
5838 corepages = totalpages - required_movablecore;
5840 required_kernelcore = max(required_kernelcore, corepages);
5844 * If kernelcore was not specified or kernelcore size is larger
5845 * than totalpages, there is no ZONE_MOVABLE.
5847 if (!required_kernelcore || required_kernelcore >= totalpages)
5850 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5851 usable_startpfn = arch_zone_lowest_possible_pfn[movable_zone];
5854 /* Spread kernelcore memory as evenly as possible throughout nodes */
5855 kernelcore_node = required_kernelcore / usable_nodes;
5856 for_each_node_state(nid, N_MEMORY) {
5857 unsigned long start_pfn, end_pfn;
5860 * Recalculate kernelcore_node if the division per node
5861 * now exceeds what is necessary to satisfy the requested
5862 * amount of memory for the kernel
5864 if (required_kernelcore < kernelcore_node)
5865 kernelcore_node = required_kernelcore / usable_nodes;
5868 * As the map is walked, we track how much memory is usable
5869 * by the kernel using kernelcore_remaining. When it is
5870 * 0, the rest of the node is usable by ZONE_MOVABLE
5872 kernelcore_remaining = kernelcore_node;
5874 /* Go through each range of PFNs within this node */
5875 for_each_mem_pfn_range(i, nid, &start_pfn, &end_pfn, NULL) {
5876 unsigned long size_pages;
5878 start_pfn = max(start_pfn, zone_movable_pfn[nid]);
5879 if (start_pfn >= end_pfn)
5882 /* Account for what is only usable for kernelcore */
5883 if (start_pfn < usable_startpfn) {
5884 unsigned long kernel_pages;
5885 kernel_pages = min(end_pfn, usable_startpfn)
5888 kernelcore_remaining -= min(kernel_pages,
5889 kernelcore_remaining);
5890 required_kernelcore -= min(kernel_pages,
5891 required_kernelcore);
5893 /* Continue if range is now fully accounted */
5894 if (end_pfn <= usable_startpfn) {
5897 * Push zone_movable_pfn to the end so
5898 * that if we have to rebalance
5899 * kernelcore across nodes, we will
5900 * not double account here
5902 zone_movable_pfn[nid] = end_pfn;
5905 start_pfn = usable_startpfn;
5909 * The usable PFN range for ZONE_MOVABLE is from
5910 * start_pfn->end_pfn. Calculate size_pages as the
5911 * number of pages used as kernelcore
5913 size_pages = end_pfn - start_pfn;
5914 if (size_pages > kernelcore_remaining)
5915 size_pages = kernelcore_remaining;
5916 zone_movable_pfn[nid] = start_pfn + size_pages;
5919 * Some kernelcore has been met, update counts and
5920 * break if the kernelcore for this node has been
5923 required_kernelcore -= min(required_kernelcore,
5925 kernelcore_remaining -= size_pages;
5926 if (!kernelcore_remaining)
5932 * If there is still required_kernelcore, we do another pass with one
5933 * less node in the count. This will push zone_movable_pfn[nid] further
5934 * along on the nodes that still have memory until kernelcore is
5938 if (usable_nodes && required_kernelcore > usable_nodes)
5942 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5943 for (nid = 0; nid < MAX_NUMNODES; nid++)
5944 zone_movable_pfn[nid] =
5945 roundup(zone_movable_pfn[nid], MAX_ORDER_NR_PAGES);
5948 /* restore the node_state */
5949 node_states[N_MEMORY] = saved_node_state;
5952 /* Any regular or high memory on that node ? */
5953 static void check_for_memory(pg_data_t *pgdat, int nid)
5955 enum zone_type zone_type;
5957 if (N_MEMORY == N_NORMAL_MEMORY)
5960 for (zone_type = 0; zone_type <= ZONE_MOVABLE - 1; zone_type++) {
5961 struct zone *zone = &pgdat->node_zones[zone_type];
5962 if (populated_zone(zone)) {
5963 node_set_state(nid, N_HIGH_MEMORY);
5964 if (N_NORMAL_MEMORY != N_HIGH_MEMORY &&
5965 zone_type <= ZONE_NORMAL)
5966 node_set_state(nid, N_NORMAL_MEMORY);
5973 * free_area_init_nodes - Initialise all pg_data_t and zone data
5974 * @max_zone_pfn: an array of max PFNs for each zone
5976 * This will call free_area_init_node() for each active node in the system.
5977 * Using the page ranges provided by memblock_set_node(), the size of each
5978 * zone in each node and their holes is calculated. If the maximum PFN
5979 * between two adjacent zones match, it is assumed that the zone is empty.
5980 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5981 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5982 * starts where the previous one ended. For example, ZONE_DMA32 starts
5983 * at arch_max_dma_pfn.
5985 void __init free_area_init_nodes(unsigned long *max_zone_pfn)
5987 unsigned long start_pfn, end_pfn;
5990 /* Record where the zone boundaries are */
5991 memset(arch_zone_lowest_possible_pfn, 0,
5992 sizeof(arch_zone_lowest_possible_pfn));
5993 memset(arch_zone_highest_possible_pfn, 0,
5994 sizeof(arch_zone_highest_possible_pfn));
5995 arch_zone_lowest_possible_pfn[0] = find_min_pfn_with_active_regions();
5996 arch_zone_highest_possible_pfn[0] = max_zone_pfn[0];
5997 for (i = 1; i < MAX_NR_ZONES; i++) {
5998 if (i == ZONE_MOVABLE)
6000 arch_zone_lowest_possible_pfn[i] =
6001 arch_zone_highest_possible_pfn[i-1];
6002 arch_zone_highest_possible_pfn[i] =
6003 max(max_zone_pfn[i], arch_zone_lowest_possible_pfn[i]);
6005 arch_zone_lowest_possible_pfn[ZONE_MOVABLE] = 0;
6006 arch_zone_highest_possible_pfn[ZONE_MOVABLE] = 0;
6008 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
6009 memset(zone_movable_pfn, 0, sizeof(zone_movable_pfn));
6010 find_zone_movable_pfns_for_nodes();
6012 /* Print out the zone ranges */
6013 pr_info("Zone ranges:\n");
6014 for (i = 0; i < MAX_NR_ZONES; i++) {
6015 if (i == ZONE_MOVABLE)
6017 pr_info(" %-8s ", zone_names[i]);
6018 if (arch_zone_lowest_possible_pfn[i] ==
6019 arch_zone_highest_possible_pfn[i])
6022 pr_cont("[mem %#018Lx-%#018Lx]\n",
6023 (u64)arch_zone_lowest_possible_pfn[i]
6025 ((u64)arch_zone_highest_possible_pfn[i]
6026 << PAGE_SHIFT) - 1);
6029 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
6030 pr_info("Movable zone start for each node\n");
6031 for (i = 0; i < MAX_NUMNODES; i++) {
6032 if (zone_movable_pfn[i])
6033 pr_info(" Node %d: %#018Lx\n", i,
6034 (u64)zone_movable_pfn[i] << PAGE_SHIFT);
6037 /* Print out the early node map */
6038 pr_info("Early memory node ranges\n");
6039 for_each_mem_pfn_range(i, MAX_NUMNODES, &start_pfn, &end_pfn, &nid)
6040 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid,
6041 (u64)start_pfn << PAGE_SHIFT,
6042 ((u64)end_pfn << PAGE_SHIFT) - 1);
6044 /* Initialise every node */
6045 mminit_verify_pageflags_layout();
6046 setup_nr_node_ids();
6047 for_each_online_node(nid) {
6048 pg_data_t *pgdat = NODE_DATA(nid);
6049 free_area_init_node(nid, NULL,
6050 find_min_pfn_for_node(nid), NULL);
6052 /* Any memory on that node */
6053 if (pgdat->node_present_pages)
6054 node_set_state(nid, N_MEMORY);
6055 check_for_memory(pgdat, nid);
6059 static int __init cmdline_parse_core(char *p, unsigned long *core)
6061 unsigned long long coremem;
6065 coremem = memparse(p, &p);
6066 *core = coremem >> PAGE_SHIFT;
6068 /* Paranoid check that UL is enough for the coremem value */
6069 WARN_ON((coremem >> PAGE_SHIFT) > ULONG_MAX);
6075 * kernelcore=size sets the amount of memory for use for allocations that
6076 * cannot be reclaimed or migrated.
6078 static int __init cmdline_parse_kernelcore(char *p)
6080 /* parse kernelcore=mirror */
6081 if (parse_option_str(p, "mirror")) {
6082 mirrored_kernelcore = true;
6086 return cmdline_parse_core(p, &required_kernelcore);
6090 * movablecore=size sets the amount of memory for use for allocations that
6091 * can be reclaimed or migrated.
6093 static int __init cmdline_parse_movablecore(char *p)
6095 return cmdline_parse_core(p, &required_movablecore);
6098 early_param("kernelcore", cmdline_parse_kernelcore);
6099 early_param("movablecore", cmdline_parse_movablecore);
6101 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6103 void adjust_managed_page_count(struct page *page, long count)
6105 spin_lock(&managed_page_count_lock);
6106 page_zone(page)->managed_pages += count;
6107 totalram_pages += count;
6108 #ifdef CONFIG_HIGHMEM
6109 if (PageHighMem(page))
6110 totalhigh_pages += count;
6112 spin_unlock(&managed_page_count_lock);
6114 EXPORT_SYMBOL(adjust_managed_page_count);
6116 unsigned long free_reserved_area(void *start, void *end, int poison, char *s)
6119 unsigned long pages = 0;
6121 start = (void *)PAGE_ALIGN((unsigned long)start);
6122 end = (void *)((unsigned long)end & PAGE_MASK);
6123 for (pos = start; pos < end; pos += PAGE_SIZE, pages++) {
6124 if ((unsigned int)poison <= 0xFF)
6125 memset(pos, poison, PAGE_SIZE);
6126 free_reserved_page(virt_to_page(pos));
6130 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
6131 s, pages << (PAGE_SHIFT - 10), start, end);
6135 EXPORT_SYMBOL(free_reserved_area);
6137 #ifdef CONFIG_HIGHMEM
6138 void free_highmem_page(struct page *page)
6140 __free_reserved_page(page);
6142 page_zone(page)->managed_pages++;
6148 void __init mem_init_print_info(const char *str)
6150 unsigned long physpages, codesize, datasize, rosize, bss_size;
6151 unsigned long init_code_size, init_data_size;
6153 physpages = get_num_physpages();
6154 codesize = _etext - _stext;
6155 datasize = _edata - _sdata;
6156 rosize = __end_rodata - __start_rodata;
6157 bss_size = __bss_stop - __bss_start;
6158 init_data_size = __init_end - __init_begin;
6159 init_code_size = _einittext - _sinittext;
6162 * Detect special cases and adjust section sizes accordingly:
6163 * 1) .init.* may be embedded into .data sections
6164 * 2) .init.text.* may be out of [__init_begin, __init_end],
6165 * please refer to arch/tile/kernel/vmlinux.lds.S.
6166 * 3) .rodata.* may be embedded into .text or .data sections.
6168 #define adj_init_size(start, end, size, pos, adj) \
6170 if (start <= pos && pos < end && size > adj) \
6174 adj_init_size(__init_begin, __init_end, init_data_size,
6175 _sinittext, init_code_size);
6176 adj_init_size(_stext, _etext, codesize, _sinittext, init_code_size);
6177 adj_init_size(_sdata, _edata, datasize, __init_begin, init_data_size);
6178 adj_init_size(_stext, _etext, codesize, __start_rodata, rosize);
6179 adj_init_size(_sdata, _edata, datasize, __start_rodata, rosize);
6181 #undef adj_init_size
6183 pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
6184 #ifdef CONFIG_HIGHMEM
6188 nr_free_pages() << (PAGE_SHIFT - 10),
6189 physpages << (PAGE_SHIFT - 10),
6190 codesize >> 10, datasize >> 10, rosize >> 10,
6191 (init_data_size + init_code_size) >> 10, bss_size >> 10,
6192 (physpages - totalram_pages - totalcma_pages) << (PAGE_SHIFT - 10),
6193 totalcma_pages << (PAGE_SHIFT - 10),
6194 #ifdef CONFIG_HIGHMEM
6195 totalhigh_pages << (PAGE_SHIFT - 10),
6197 str ? ", " : "", str ? str : "");
6201 * set_dma_reserve - set the specified number of pages reserved in the first zone
6202 * @new_dma_reserve: The number of pages to mark reserved
6204 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
6205 * In the DMA zone, a significant percentage may be consumed by kernel image
6206 * and other unfreeable allocations which can skew the watermarks badly. This
6207 * function may optionally be used to account for unfreeable pages in the
6208 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6209 * smaller per-cpu batchsize.
6211 void __init set_dma_reserve(unsigned long new_dma_reserve)
6213 dma_reserve = new_dma_reserve;
6216 void __init free_area_init(unsigned long *zones_size)
6218 free_area_init_node(0, zones_size,
6219 __pa(PAGE_OFFSET) >> PAGE_SHIFT, NULL);
6222 static int page_alloc_cpu_notify(struct notifier_block *self,
6223 unsigned long action, void *hcpu)
6225 int cpu = (unsigned long)hcpu;
6227 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) {
6228 lru_add_drain_cpu(cpu);
6232 * Spill the event counters of the dead processor
6233 * into the current processors event counters.
6234 * This artificially elevates the count of the current
6237 vm_events_fold_cpu(cpu);
6240 * Zero the differential counters of the dead processor
6241 * so that the vm statistics are consistent.
6243 * This is only okay since the processor is dead and cannot
6244 * race with what we are doing.
6246 cpu_vm_stats_fold(cpu);
6251 void __init page_alloc_init(void)
6253 hotcpu_notifier(page_alloc_cpu_notify, 0);
6257 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
6258 * or min_free_kbytes changes.
6260 static void calculate_totalreserve_pages(void)
6262 struct pglist_data *pgdat;
6263 unsigned long reserve_pages = 0;
6264 enum zone_type i, j;
6266 for_each_online_pgdat(pgdat) {
6267 for (i = 0; i < MAX_NR_ZONES; i++) {
6268 struct zone *zone = pgdat->node_zones + i;
6271 /* Find valid and maximum lowmem_reserve in the zone */
6272 for (j = i; j < MAX_NR_ZONES; j++) {
6273 if (zone->lowmem_reserve[j] > max)
6274 max = zone->lowmem_reserve[j];
6277 /* we treat the high watermark as reserved pages. */
6278 max += high_wmark_pages(zone);
6280 if (max > zone->managed_pages)
6281 max = zone->managed_pages;
6283 zone->totalreserve_pages = max;
6285 reserve_pages += max;
6288 totalreserve_pages = reserve_pages;
6292 * setup_per_zone_lowmem_reserve - called whenever
6293 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
6294 * has a correct pages reserved value, so an adequate number of
6295 * pages are left in the zone after a successful __alloc_pages().
6297 static void setup_per_zone_lowmem_reserve(void)
6299 struct pglist_data *pgdat;
6300 enum zone_type j, idx;
6302 for_each_online_pgdat(pgdat) {
6303 for (j = 0; j < MAX_NR_ZONES; j++) {
6304 struct zone *zone = pgdat->node_zones + j;
6305 unsigned long managed_pages = zone->managed_pages;
6307 zone->lowmem_reserve[j] = 0;
6311 struct zone *lower_zone;
6315 if (sysctl_lowmem_reserve_ratio[idx] < 1)
6316 sysctl_lowmem_reserve_ratio[idx] = 1;
6318 lower_zone = pgdat->node_zones + idx;
6319 lower_zone->lowmem_reserve[j] = managed_pages /
6320 sysctl_lowmem_reserve_ratio[idx];
6321 managed_pages += lower_zone->managed_pages;
6326 /* update totalreserve_pages */
6327 calculate_totalreserve_pages();
6330 static void __setup_per_zone_wmarks(void)
6332 unsigned long pages_min = min_free_kbytes >> (PAGE_SHIFT - 10);
6333 unsigned long lowmem_pages = 0;
6335 unsigned long flags;
6337 /* Calculate total number of !ZONE_HIGHMEM pages */
6338 for_each_zone(zone) {
6339 if (!is_highmem(zone))
6340 lowmem_pages += zone->managed_pages;
6343 for_each_zone(zone) {
6346 spin_lock_irqsave(&zone->lock, flags);
6347 tmp = (u64)pages_min * zone->managed_pages;
6348 do_div(tmp, lowmem_pages);
6349 if (is_highmem(zone)) {
6351 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6352 * need highmem pages, so cap pages_min to a small
6355 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
6356 * deltas control asynch page reclaim, and so should
6357 * not be capped for highmem.
6359 unsigned long min_pages;
6361 min_pages = zone->managed_pages / 1024;
6362 min_pages = clamp(min_pages, SWAP_CLUSTER_MAX, 128UL);
6363 zone->watermark[WMARK_MIN] = min_pages;
6366 * If it's a lowmem zone, reserve a number of pages
6367 * proportionate to the zone's size.
6369 zone->watermark[WMARK_MIN] = tmp;
6373 * Set the kswapd watermarks distance according to the
6374 * scale factor in proportion to available memory, but
6375 * ensure a minimum size on small systems.
6377 tmp = max_t(u64, tmp >> 2,
6378 mult_frac(zone->managed_pages,
6379 watermark_scale_factor, 10000));
6381 zone->watermark[WMARK_LOW] = min_wmark_pages(zone) + tmp;
6382 zone->watermark[WMARK_HIGH] = min_wmark_pages(zone) + tmp * 2;
6384 __mod_zone_page_state(zone, NR_ALLOC_BATCH,
6385 high_wmark_pages(zone) - low_wmark_pages(zone) -
6386 atomic_long_read(&zone->vm_stat[NR_ALLOC_BATCH]));
6388 spin_unlock_irqrestore(&zone->lock, flags);
6391 /* update totalreserve_pages */
6392 calculate_totalreserve_pages();
6396 * setup_per_zone_wmarks - called when min_free_kbytes changes
6397 * or when memory is hot-{added|removed}
6399 * Ensures that the watermark[min,low,high] values for each zone are set
6400 * correctly with respect to min_free_kbytes.
6402 void setup_per_zone_wmarks(void)
6404 mutex_lock(&zonelists_mutex);
6405 __setup_per_zone_wmarks();
6406 mutex_unlock(&zonelists_mutex);
6410 * The inactive anon list should be small enough that the VM never has to
6411 * do too much work, but large enough that each inactive page has a chance
6412 * to be referenced again before it is swapped out.
6414 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
6415 * INACTIVE_ANON pages on this zone's LRU, maintained by the
6416 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
6417 * the anonymous pages are kept on the inactive list.
6420 * memory ratio inactive anon
6421 * -------------------------------------
6430 static void __meminit calculate_zone_inactive_ratio(struct zone *zone)
6432 unsigned int gb, ratio;
6434 /* Zone size in gigabytes */
6435 gb = zone->managed_pages >> (30 - PAGE_SHIFT);
6437 ratio = int_sqrt(10 * gb);
6441 zone->inactive_ratio = ratio;
6444 static void __meminit setup_per_zone_inactive_ratio(void)
6449 calculate_zone_inactive_ratio(zone);
6453 * Initialise min_free_kbytes.
6455 * For small machines we want it small (128k min). For large machines
6456 * we want it large (64MB max). But it is not linear, because network
6457 * bandwidth does not increase linearly with machine size. We use
6459 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
6460 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6476 int __meminit init_per_zone_wmark_min(void)
6478 unsigned long lowmem_kbytes;
6479 int new_min_free_kbytes;
6481 lowmem_kbytes = nr_free_buffer_pages() * (PAGE_SIZE >> 10);
6482 new_min_free_kbytes = int_sqrt(lowmem_kbytes * 16);
6484 if (new_min_free_kbytes > user_min_free_kbytes) {
6485 min_free_kbytes = new_min_free_kbytes;
6486 if (min_free_kbytes < 128)
6487 min_free_kbytes = 128;
6488 if (min_free_kbytes > 65536)
6489 min_free_kbytes = 65536;
6491 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6492 new_min_free_kbytes, user_min_free_kbytes);
6494 setup_per_zone_wmarks();
6495 refresh_zone_stat_thresholds();
6496 setup_per_zone_lowmem_reserve();
6497 setup_per_zone_inactive_ratio();
6500 core_initcall(init_per_zone_wmark_min)
6503 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
6504 * that we can call two helper functions whenever min_free_kbytes
6507 int min_free_kbytes_sysctl_handler(struct ctl_table *table, int write,
6508 void __user *buffer, size_t *length, loff_t *ppos)
6512 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6517 user_min_free_kbytes = min_free_kbytes;
6518 setup_per_zone_wmarks();
6523 int watermark_scale_factor_sysctl_handler(struct ctl_table *table, int write,
6524 void __user *buffer, size_t *length, loff_t *ppos)
6528 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6533 setup_per_zone_wmarks();
6539 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table *table, int write,
6540 void __user *buffer, size_t *length, loff_t *ppos)
6545 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6550 zone->min_unmapped_pages = (zone->managed_pages *
6551 sysctl_min_unmapped_ratio) / 100;
6555 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table *table, int write,
6556 void __user *buffer, size_t *length, loff_t *ppos)
6561 rc = proc_dointvec_minmax(table, write, buffer, length, ppos);
6566 zone->min_slab_pages = (zone->managed_pages *
6567 sysctl_min_slab_ratio) / 100;
6573 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6574 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6575 * whenever sysctl_lowmem_reserve_ratio changes.
6577 * The reserve ratio obviously has absolutely no relation with the
6578 * minimum watermarks. The lowmem reserve ratio can only make sense
6579 * if in function of the boot time zone sizes.
6581 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table *table, int write,
6582 void __user *buffer, size_t *length, loff_t *ppos)
6584 proc_dointvec_minmax(table, write, buffer, length, ppos);
6585 setup_per_zone_lowmem_reserve();
6590 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
6591 * cpu. It is the fraction of total pages in each zone that a hot per cpu
6592 * pagelist can have before it gets flushed back to buddy allocator.
6594 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table *table, int write,
6595 void __user *buffer, size_t *length, loff_t *ppos)
6598 int old_percpu_pagelist_fraction;
6601 mutex_lock(&pcp_batch_high_lock);
6602 old_percpu_pagelist_fraction = percpu_pagelist_fraction;
6604 ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
6605 if (!write || ret < 0)
6608 /* Sanity checking to avoid pcp imbalance */
6609 if (percpu_pagelist_fraction &&
6610 percpu_pagelist_fraction < MIN_PERCPU_PAGELIST_FRACTION) {
6611 percpu_pagelist_fraction = old_percpu_pagelist_fraction;
6617 if (percpu_pagelist_fraction == old_percpu_pagelist_fraction)
6620 for_each_populated_zone(zone) {
6623 for_each_possible_cpu(cpu)
6624 pageset_set_high_and_batch(zone,
6625 per_cpu_ptr(zone->pageset, cpu));
6628 mutex_unlock(&pcp_batch_high_lock);
6633 int hashdist = HASHDIST_DEFAULT;
6635 static int __init set_hashdist(char *str)
6639 hashdist = simple_strtoul(str, &str, 0);
6642 __setup("hashdist=", set_hashdist);
6646 * allocate a large system hash table from bootmem
6647 * - it is assumed that the hash table must contain an exact power-of-2
6648 * quantity of entries
6649 * - limit is the number of hash buckets, not the total allocation size
6651 void *__init alloc_large_system_hash(const char *tablename,
6652 unsigned long bucketsize,
6653 unsigned long numentries,
6656 unsigned int *_hash_shift,
6657 unsigned int *_hash_mask,
6658 unsigned long low_limit,
6659 unsigned long high_limit)
6661 unsigned long long max = high_limit;
6662 unsigned long log2qty, size;
6665 /* allow the kernel cmdline to have a say */
6667 /* round applicable memory size up to nearest megabyte */
6668 numentries = nr_kernel_pages;
6670 /* It isn't necessary when PAGE_SIZE >= 1MB */
6671 if (PAGE_SHIFT < 20)
6672 numentries = round_up(numentries, (1<<20)/PAGE_SIZE);
6674 /* limit to 1 bucket per 2^scale bytes of low memory */
6675 if (scale > PAGE_SHIFT)
6676 numentries >>= (scale - PAGE_SHIFT);
6678 numentries <<= (PAGE_SHIFT - scale);
6680 /* Make sure we've got at least a 0-order allocation.. */
6681 if (unlikely(flags & HASH_SMALL)) {
6682 /* Makes no sense without HASH_EARLY */
6683 WARN_ON(!(flags & HASH_EARLY));
6684 if (!(numentries >> *_hash_shift)) {
6685 numentries = 1UL << *_hash_shift;
6686 BUG_ON(!numentries);
6688 } else if (unlikely((numentries * bucketsize) < PAGE_SIZE))
6689 numentries = PAGE_SIZE / bucketsize;
6691 numentries = roundup_pow_of_two(numentries);
6693 /* limit allocation size to 1/16 total memory by default */
6695 max = ((unsigned long long)nr_all_pages << PAGE_SHIFT) >> 4;
6696 do_div(max, bucketsize);
6698 max = min(max, 0x80000000ULL);
6700 if (numentries < low_limit)
6701 numentries = low_limit;
6702 if (numentries > max)
6705 log2qty = ilog2(numentries);
6708 size = bucketsize << log2qty;
6709 if (flags & HASH_EARLY)
6710 table = memblock_virt_alloc_nopanic(size, 0);
6712 table = __vmalloc(size, GFP_ATOMIC, PAGE_KERNEL);
6715 * If bucketsize is not a power-of-two, we may free
6716 * some pages at the end of hash table which
6717 * alloc_pages_exact() automatically does
6719 if (get_order(size) < MAX_ORDER) {
6720 table = alloc_pages_exact(size, GFP_ATOMIC);
6721 kmemleak_alloc(table, size, 1, GFP_ATOMIC);
6724 } while (!table && size > PAGE_SIZE && --log2qty);
6727 panic("Failed to allocate %s hash table\n", tablename);
6729 pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n",
6730 tablename, 1UL << log2qty, ilog2(size) - PAGE_SHIFT, size);
6733 *_hash_shift = log2qty;
6735 *_hash_mask = (1 << log2qty) - 1;
6740 /* Return a pointer to the bitmap storing bits affecting a block of pages */
6741 static inline unsigned long *get_pageblock_bitmap(struct zone *zone,
6744 #ifdef CONFIG_SPARSEMEM
6745 return __pfn_to_section(pfn)->pageblock_flags;
6747 return zone->pageblock_flags;
6748 #endif /* CONFIG_SPARSEMEM */
6751 static inline int pfn_to_bitidx(struct zone *zone, unsigned long pfn)
6753 #ifdef CONFIG_SPARSEMEM
6754 pfn &= (PAGES_PER_SECTION-1);
6755 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
6757 pfn = pfn - round_down(zone->zone_start_pfn, pageblock_nr_pages);
6758 return (pfn >> pageblock_order) * NR_PAGEBLOCK_BITS;
6759 #endif /* CONFIG_SPARSEMEM */
6763 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
6764 * @page: The page within the block of interest
6765 * @pfn: The target page frame number
6766 * @end_bitidx: The last bit of interest to retrieve
6767 * @mask: mask of bits that the caller is interested in
6769 * Return: pageblock_bits flags
6771 unsigned long get_pfnblock_flags_mask(struct page *page, unsigned long pfn,
6772 unsigned long end_bitidx,
6776 unsigned long *bitmap;
6777 unsigned long bitidx, word_bitidx;
6780 zone = page_zone(page);
6781 bitmap = get_pageblock_bitmap(zone, pfn);
6782 bitidx = pfn_to_bitidx(zone, pfn);
6783 word_bitidx = bitidx / BITS_PER_LONG;
6784 bitidx &= (BITS_PER_LONG-1);
6786 word = bitmap[word_bitidx];
6787 bitidx += end_bitidx;
6788 return (word >> (BITS_PER_LONG - bitidx - 1)) & mask;
6792 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
6793 * @page: The page within the block of interest
6794 * @flags: The flags to set
6795 * @pfn: The target page frame number
6796 * @end_bitidx: The last bit of interest
6797 * @mask: mask of bits that the caller is interested in
6799 void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
6801 unsigned long end_bitidx,
6805 unsigned long *bitmap;
6806 unsigned long bitidx, word_bitidx;
6807 unsigned long old_word, word;
6809 BUILD_BUG_ON(NR_PAGEBLOCK_BITS != 4);
6811 zone = page_zone(page);
6812 bitmap = get_pageblock_bitmap(zone, pfn);
6813 bitidx = pfn_to_bitidx(zone, pfn);
6814 word_bitidx = bitidx / BITS_PER_LONG;
6815 bitidx &= (BITS_PER_LONG-1);
6817 VM_BUG_ON_PAGE(!zone_spans_pfn(zone, pfn), page);
6819 bitidx += end_bitidx;
6820 mask <<= (BITS_PER_LONG - bitidx - 1);
6821 flags <<= (BITS_PER_LONG - bitidx - 1);
6823 word = READ_ONCE(bitmap[word_bitidx]);
6825 old_word = cmpxchg(&bitmap[word_bitidx], word, (word & ~mask) | flags);
6826 if (word == old_word)
6833 * This function checks whether pageblock includes unmovable pages or not.
6834 * If @count is not zero, it is okay to include less @count unmovable pages
6836 * PageLRU check without isolation or lru_lock could race so that
6837 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6838 * expect this function should be exact.
6840 bool has_unmovable_pages(struct zone *zone, struct page *page, int count,
6841 bool skip_hwpoisoned_pages)
6843 unsigned long pfn, iter, found;
6847 * For avoiding noise data, lru_add_drain_all() should be called
6848 * If ZONE_MOVABLE, the zone never contains unmovable pages
6850 if (zone_idx(zone) == ZONE_MOVABLE)
6852 mt = get_pageblock_migratetype(page);
6853 if (mt == MIGRATE_MOVABLE || is_migrate_cma(mt))
6856 pfn = page_to_pfn(page);
6857 for (found = 0, iter = 0; iter < pageblock_nr_pages; iter++) {
6858 unsigned long check = pfn + iter;
6860 if (!pfn_valid_within(check))
6863 page = pfn_to_page(check);
6866 * Hugepages are not in LRU lists, but they're movable.
6867 * We need not scan over tail pages bacause we don't
6868 * handle each tail page individually in migration.
6870 if (PageHuge(page)) {
6871 iter = round_up(iter + 1, 1<<compound_order(page)) - 1;
6876 * We can't use page_count without pin a page
6877 * because another CPU can free compound page.
6878 * This check already skips compound tails of THP
6879 * because their page->_refcount is zero at all time.
6881 if (!page_ref_count(page)) {
6882 if (PageBuddy(page))
6883 iter += (1 << page_order(page)) - 1;
6888 * The HWPoisoned page may be not in buddy system, and
6889 * page_count() is not 0.
6891 if (skip_hwpoisoned_pages && PageHWPoison(page))
6897 * If there are RECLAIMABLE pages, we need to check
6898 * it. But now, memory offline itself doesn't call
6899 * shrink_node_slabs() and it still to be fixed.
6902 * If the page is not RAM, page_count()should be 0.
6903 * we don't need more check. This is an _used_ not-movable page.
6905 * The problematic thing here is PG_reserved pages. PG_reserved
6906 * is set to both of a memory hole page and a _used_ kernel
6915 bool is_pageblock_removable_nolock(struct page *page)
6921 * We have to be careful here because we are iterating over memory
6922 * sections which are not zone aware so we might end up outside of
6923 * the zone but still within the section.
6924 * We have to take care about the node as well. If the node is offline
6925 * its NODE_DATA will be NULL - see page_zone.
6927 if (!node_online(page_to_nid(page)))
6930 zone = page_zone(page);
6931 pfn = page_to_pfn(page);
6932 if (!zone_spans_pfn(zone, pfn))
6935 return !has_unmovable_pages(zone, page, 0, true);
6938 #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
6940 static unsigned long pfn_max_align_down(unsigned long pfn)
6942 return pfn & ~(max_t(unsigned long, MAX_ORDER_NR_PAGES,
6943 pageblock_nr_pages) - 1);
6946 static unsigned long pfn_max_align_up(unsigned long pfn)
6948 return ALIGN(pfn, max_t(unsigned long, MAX_ORDER_NR_PAGES,
6949 pageblock_nr_pages));
6952 /* [start, end) must belong to a single zone. */
6953 static int __alloc_contig_migrate_range(struct compact_control *cc,
6954 unsigned long start, unsigned long end)
6956 /* This function is based on compact_zone() from compaction.c. */
6957 unsigned long nr_reclaimed;
6958 unsigned long pfn = start;
6959 unsigned int tries = 0;
6964 while (pfn < end || !list_empty(&cc->migratepages)) {
6965 if (fatal_signal_pending(current)) {
6970 if (list_empty(&cc->migratepages)) {
6971 cc->nr_migratepages = 0;
6972 pfn = isolate_migratepages_range(cc, pfn, end);
6978 } else if (++tries == 5) {
6979 ret = ret < 0 ? ret : -EBUSY;
6983 nr_reclaimed = reclaim_clean_pages_from_list(cc->zone,
6985 cc->nr_migratepages -= nr_reclaimed;
6987 ret = migrate_pages(&cc->migratepages, alloc_migrate_target,
6988 NULL, 0, cc->mode, MR_CMA);
6991 putback_movable_pages(&cc->migratepages);
6998 * alloc_contig_range() -- tries to allocate given range of pages
6999 * @start: start PFN to allocate
7000 * @end: one-past-the-last PFN to allocate
7001 * @migratetype: migratetype of the underlaying pageblocks (either
7002 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
7003 * in range must have the same migratetype and it must
7004 * be either of the two.
7006 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
7007 * aligned, however it's the caller's responsibility to guarantee that
7008 * we are the only thread that changes migrate type of pageblocks the
7011 * The PFN range must belong to a single zone.
7013 * Returns zero on success or negative error code. On success all
7014 * pages which PFN is in [start, end) are allocated for the caller and
7015 * need to be freed with free_contig_range().
7017 int alloc_contig_range(unsigned long start, unsigned long end,
7018 unsigned migratetype)
7020 unsigned long outer_start, outer_end;
7024 struct compact_control cc = {
7025 .nr_migratepages = 0,
7027 .zone = page_zone(pfn_to_page(start)),
7028 .mode = MIGRATE_SYNC,
7029 .ignore_skip_hint = true,
7031 INIT_LIST_HEAD(&cc.migratepages);
7034 * What we do here is we mark all pageblocks in range as
7035 * MIGRATE_ISOLATE. Because pageblock and max order pages may
7036 * have different sizes, and due to the way page allocator
7037 * work, we align the range to biggest of the two pages so
7038 * that page allocator won't try to merge buddies from
7039 * different pageblocks and change MIGRATE_ISOLATE to some
7040 * other migration type.
7042 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
7043 * migrate the pages from an unaligned range (ie. pages that
7044 * we are interested in). This will put all the pages in
7045 * range back to page allocator as MIGRATE_ISOLATE.
7047 * When this is done, we take the pages in range from page
7048 * allocator removing them from the buddy system. This way
7049 * page allocator will never consider using them.
7051 * This lets us mark the pageblocks back as
7052 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
7053 * aligned range but not in the unaligned, original range are
7054 * put back to page allocator so that buddy can use them.
7057 ret = start_isolate_page_range(pfn_max_align_down(start),
7058 pfn_max_align_up(end), migratetype,
7064 * In case of -EBUSY, we'd like to know which page causes problem.
7065 * So, just fall through. We will check it in test_pages_isolated().
7067 ret = __alloc_contig_migrate_range(&cc, start, end);
7068 if (ret && ret != -EBUSY)
7072 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
7073 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
7074 * more, all pages in [start, end) are free in page allocator.
7075 * What we are going to do is to allocate all pages from
7076 * [start, end) (that is remove them from page allocator).
7078 * The only problem is that pages at the beginning and at the
7079 * end of interesting range may be not aligned with pages that
7080 * page allocator holds, ie. they can be part of higher order
7081 * pages. Because of this, we reserve the bigger range and
7082 * once this is done free the pages we are not interested in.
7084 * We don't have to hold zone->lock here because the pages are
7085 * isolated thus they won't get removed from buddy.
7088 lru_add_drain_all();
7089 drain_all_pages(cc.zone);
7092 outer_start = start;
7093 while (!PageBuddy(pfn_to_page(outer_start))) {
7094 if (++order >= MAX_ORDER) {
7095 outer_start = start;
7098 outer_start &= ~0UL << order;
7101 if (outer_start != start) {
7102 order = page_order(pfn_to_page(outer_start));
7105 * outer_start page could be small order buddy page and
7106 * it doesn't include start page. Adjust outer_start
7107 * in this case to report failed page properly
7108 * on tracepoint in test_pages_isolated()
7110 if (outer_start + (1UL << order) <= start)
7111 outer_start = start;
7114 /* Make sure the range is really isolated. */
7115 if (test_pages_isolated(outer_start, end, false)) {
7116 pr_info("%s: [%lx, %lx) PFNs busy\n",
7117 __func__, outer_start, end);
7122 /* Grab isolated pages from freelists. */
7123 outer_end = isolate_freepages_range(&cc, outer_start, end);
7129 /* Free head and tail (if any) */
7130 if (start != outer_start)
7131 free_contig_range(outer_start, start - outer_start);
7132 if (end != outer_end)
7133 free_contig_range(end, outer_end - end);
7136 undo_isolate_page_range(pfn_max_align_down(start),
7137 pfn_max_align_up(end), migratetype);
7141 void free_contig_range(unsigned long pfn, unsigned nr_pages)
7143 unsigned int count = 0;
7145 for (; nr_pages--; pfn++) {
7146 struct page *page = pfn_to_page(pfn);
7148 count += page_count(page) != 1;
7151 WARN(count != 0, "%d pages are still in use!\n", count);
7155 #ifdef CONFIG_MEMORY_HOTPLUG
7157 * The zone indicated has a new number of managed_pages; batch sizes and percpu
7158 * page high values need to be recalulated.
7160 void __meminit zone_pcp_update(struct zone *zone)
7163 mutex_lock(&pcp_batch_high_lock);
7164 for_each_possible_cpu(cpu)
7165 pageset_set_high_and_batch(zone,
7166 per_cpu_ptr(zone->pageset, cpu));
7167 mutex_unlock(&pcp_batch_high_lock);
7171 void zone_pcp_reset(struct zone *zone)
7173 unsigned long flags;
7175 struct per_cpu_pageset *pset;
7177 /* avoid races with drain_pages() */
7178 local_irq_save(flags);
7179 if (zone->pageset != &boot_pageset) {
7180 for_each_online_cpu(cpu) {
7181 pset = per_cpu_ptr(zone->pageset, cpu);
7182 drain_zonestat(zone, pset);
7184 free_percpu(zone->pageset);
7185 zone->pageset = &boot_pageset;
7187 local_irq_restore(flags);
7190 #ifdef CONFIG_MEMORY_HOTREMOVE
7192 * All pages in the range must be in a single zone and isolated
7193 * before calling this.
7196 __offline_isolated_pages(unsigned long start_pfn, unsigned long end_pfn)
7200 unsigned int order, i;
7202 unsigned long flags;
7203 /* find the first valid pfn */
7204 for (pfn = start_pfn; pfn < end_pfn; pfn++)
7209 zone = page_zone(pfn_to_page(pfn));
7210 spin_lock_irqsave(&zone->lock, flags);
7212 while (pfn < end_pfn) {
7213 if (!pfn_valid(pfn)) {
7217 page = pfn_to_page(pfn);
7219 * The HWPoisoned page may be not in buddy system, and
7220 * page_count() is not 0.
7222 if (unlikely(!PageBuddy(page) && PageHWPoison(page))) {
7224 SetPageReserved(page);
7228 BUG_ON(page_count(page));
7229 BUG_ON(!PageBuddy(page));
7230 order = page_order(page);
7231 #ifdef CONFIG_DEBUG_VM
7232 pr_info("remove from free list %lx %d %lx\n",
7233 pfn, 1 << order, end_pfn);
7235 list_del(&page->lru);
7236 rmv_page_order(page);
7237 zone->free_area[order].nr_free--;
7238 for (i = 0; i < (1 << order); i++)
7239 SetPageReserved((page+i));
7240 pfn += (1 << order);
7242 spin_unlock_irqrestore(&zone->lock, flags);
7246 bool is_free_buddy_page(struct page *page)
7248 struct zone *zone = page_zone(page);
7249 unsigned long pfn = page_to_pfn(page);
7250 unsigned long flags;
7253 spin_lock_irqsave(&zone->lock, flags);
7254 for (order = 0; order < MAX_ORDER; order++) {
7255 struct page *page_head = page - (pfn & ((1 << order) - 1));
7257 if (PageBuddy(page_head) && page_order(page_head) >= order)
7260 spin_unlock_irqrestore(&zone->lock, flags);
7262 return order < MAX_ORDER;