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
)
355 /* Return a pointer to the bitmap storing bits affecting a block of pages */
356 static inline unsigned long *get_pageblock_bitmap(struct page
*page
,
359 #ifdef CONFIG_SPARSEMEM
360 return __pfn_to_section(pfn
)->pageblock_flags
;
362 return page_zone(page
)->pageblock_flags
;
363 #endif /* CONFIG_SPARSEMEM */
366 static inline int pfn_to_bitidx(struct page
*page
, unsigned long pfn
)
368 #ifdef CONFIG_SPARSEMEM
369 pfn
&= (PAGES_PER_SECTION
-1);
370 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
372 pfn
= pfn
- round_down(page_zone(page
)->zone_start_pfn
, pageblock_nr_pages
);
373 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
374 #endif /* CONFIG_SPARSEMEM */
378 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
379 * @page: The page within the block of interest
380 * @pfn: The target page frame number
381 * @end_bitidx: The last bit of interest to retrieve
382 * @mask: mask of bits that the caller is interested in
384 * Return: pageblock_bits flags
386 static __always_inline
unsigned long __get_pfnblock_flags_mask(struct page
*page
,
388 unsigned long end_bitidx
,
391 unsigned long *bitmap
;
392 unsigned long bitidx
, word_bitidx
;
395 bitmap
= get_pageblock_bitmap(page
, pfn
);
396 bitidx
= pfn_to_bitidx(page
, pfn
);
397 word_bitidx
= bitidx
/ BITS_PER_LONG
;
398 bitidx
&= (BITS_PER_LONG
-1);
400 word
= bitmap
[word_bitidx
];
401 bitidx
+= end_bitidx
;
402 return (word
>> (BITS_PER_LONG
- bitidx
- 1)) & mask
;
405 unsigned long get_pfnblock_flags_mask(struct page
*page
, unsigned long pfn
,
406 unsigned long end_bitidx
,
409 return __get_pfnblock_flags_mask(page
, pfn
, end_bitidx
, mask
);
412 static __always_inline
int get_pfnblock_migratetype(struct page
*page
, unsigned long pfn
)
414 return __get_pfnblock_flags_mask(page
, pfn
, PB_migrate_end
, MIGRATETYPE_MASK
);
418 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
419 * @page: The page within the block of interest
420 * @flags: The flags to set
421 * @pfn: The target page frame number
422 * @end_bitidx: The last bit of interest
423 * @mask: mask of bits that the caller is interested in
425 void set_pfnblock_flags_mask(struct page
*page
, unsigned long flags
,
427 unsigned long end_bitidx
,
430 unsigned long *bitmap
;
431 unsigned long bitidx
, word_bitidx
;
432 unsigned long old_word
, word
;
434 BUILD_BUG_ON(NR_PAGEBLOCK_BITS
!= 4);
436 bitmap
= get_pageblock_bitmap(page
, pfn
);
437 bitidx
= pfn_to_bitidx(page
, pfn
);
438 word_bitidx
= bitidx
/ BITS_PER_LONG
;
439 bitidx
&= (BITS_PER_LONG
-1);
441 VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page
), pfn
), page
);
443 bitidx
+= end_bitidx
;
444 mask
<<= (BITS_PER_LONG
- bitidx
- 1);
445 flags
<<= (BITS_PER_LONG
- bitidx
- 1);
447 word
= READ_ONCE(bitmap
[word_bitidx
]);
449 old_word
= cmpxchg(&bitmap
[word_bitidx
], word
, (word
& ~mask
) | flags
);
450 if (word
== old_word
)
456 void set_pageblock_migratetype(struct page
*page
, int migratetype
)
458 if (unlikely(page_group_by_mobility_disabled
&&
459 migratetype
< MIGRATE_PCPTYPES
))
460 migratetype
= MIGRATE_UNMOVABLE
;
462 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
463 PB_migrate
, PB_migrate_end
);
466 #ifdef CONFIG_DEBUG_VM
467 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
471 unsigned long pfn
= page_to_pfn(page
);
472 unsigned long sp
, start_pfn
;
475 seq
= zone_span_seqbegin(zone
);
476 start_pfn
= zone
->zone_start_pfn
;
477 sp
= zone
->spanned_pages
;
478 if (!zone_spans_pfn(zone
, pfn
))
480 } while (zone_span_seqretry(zone
, seq
));
483 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
484 pfn
, zone_to_nid(zone
), zone
->name
,
485 start_pfn
, start_pfn
+ sp
);
490 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
492 if (!pfn_valid_within(page_to_pfn(page
)))
494 if (zone
!= page_zone(page
))
500 * Temporary debugging check for pages not lying within a given zone.
502 static int bad_range(struct zone
*zone
, struct page
*page
)
504 if (page_outside_zone_boundaries(zone
, page
))
506 if (!page_is_consistent(zone
, page
))
512 static inline int bad_range(struct zone
*zone
, struct page
*page
)
518 static void bad_page(struct page
*page
, const char *reason
,
519 unsigned long bad_flags
)
521 static unsigned long resume
;
522 static unsigned long nr_shown
;
523 static unsigned long nr_unshown
;
525 /* Don't complain about poisoned pages */
526 if (PageHWPoison(page
)) {
527 page_mapcount_reset(page
); /* remove PageBuddy */
532 * Allow a burst of 60 reports, then keep quiet for that minute;
533 * or allow a steady drip of one report per second.
535 if (nr_shown
== 60) {
536 if (time_before(jiffies
, resume
)) {
542 "BUG: Bad page state: %lu messages suppressed\n",
549 resume
= jiffies
+ 60 * HZ
;
551 pr_alert("BUG: Bad page state in process %s pfn:%05lx\n",
552 current
->comm
, page_to_pfn(page
));
553 __dump_page(page
, reason
);
554 bad_flags
&= page
->flags
;
556 pr_alert("bad because of flags: %#lx(%pGp)\n",
557 bad_flags
, &bad_flags
);
558 dump_page_owner(page
);
563 /* Leave bad fields for debug, except PageBuddy could make trouble */
564 page_mapcount_reset(page
); /* remove PageBuddy */
565 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
569 * Higher-order pages are called "compound pages". They are structured thusly:
571 * The first PAGE_SIZE page is called the "head page" and have PG_head set.
573 * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
574 * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
576 * The first tail page's ->compound_dtor holds the offset in array of compound
577 * page destructors. See compound_page_dtors.
579 * The first tail page's ->compound_order holds the order of allocation.
580 * This usage means that zero-order pages may not be compound.
583 void free_compound_page(struct page
*page
)
585 __free_pages_ok(page
, compound_order(page
));
588 void prep_compound_page(struct page
*page
, unsigned int order
)
591 int nr_pages
= 1 << order
;
593 set_compound_page_dtor(page
, COMPOUND_PAGE_DTOR
);
594 set_compound_order(page
, order
);
596 for (i
= 1; i
< nr_pages
; i
++) {
597 struct page
*p
= page
+ i
;
598 set_page_count(p
, 0);
599 p
->mapping
= TAIL_MAPPING
;
600 set_compound_head(p
, page
);
602 atomic_set(compound_mapcount_ptr(page
), -1);
605 #ifdef CONFIG_DEBUG_PAGEALLOC
606 unsigned int _debug_guardpage_minorder
;
607 bool _debug_pagealloc_enabled __read_mostly
608 = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT
);
609 EXPORT_SYMBOL(_debug_pagealloc_enabled
);
610 bool _debug_guardpage_enabled __read_mostly
;
612 static int __init
early_debug_pagealloc(char *buf
)
616 return kstrtobool(buf
, &_debug_pagealloc_enabled
);
618 early_param("debug_pagealloc", early_debug_pagealloc
);
620 static bool need_debug_guardpage(void)
622 /* If we don't use debug_pagealloc, we don't need guard page */
623 if (!debug_pagealloc_enabled())
629 static void init_debug_guardpage(void)
631 if (!debug_pagealloc_enabled())
634 _debug_guardpage_enabled
= true;
637 struct page_ext_operations debug_guardpage_ops
= {
638 .need
= need_debug_guardpage
,
639 .init
= init_debug_guardpage
,
642 static int __init
debug_guardpage_minorder_setup(char *buf
)
646 if (kstrtoul(buf
, 10, &res
) < 0 || res
> MAX_ORDER
/ 2) {
647 pr_err("Bad debug_guardpage_minorder value\n");
650 _debug_guardpage_minorder
= res
;
651 pr_info("Setting debug_guardpage_minorder to %lu\n", res
);
654 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup
);
656 static inline void set_page_guard(struct zone
*zone
, struct page
*page
,
657 unsigned int order
, int migratetype
)
659 struct page_ext
*page_ext
;
661 if (!debug_guardpage_enabled())
664 page_ext
= lookup_page_ext(page
);
665 __set_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
667 INIT_LIST_HEAD(&page
->lru
);
668 set_page_private(page
, order
);
669 /* Guard pages are not available for any usage */
670 __mod_zone_freepage_state(zone
, -(1 << order
), migratetype
);
673 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
674 unsigned int order
, int migratetype
)
676 struct page_ext
*page_ext
;
678 if (!debug_guardpage_enabled())
681 page_ext
= lookup_page_ext(page
);
682 __clear_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
684 set_page_private(page
, 0);
685 if (!is_migrate_isolate(migratetype
))
686 __mod_zone_freepage_state(zone
, (1 << order
), migratetype
);
689 struct page_ext_operations debug_guardpage_ops
= { NULL
, };
690 static inline void set_page_guard(struct zone
*zone
, struct page
*page
,
691 unsigned int order
, int migratetype
) {}
692 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
693 unsigned int order
, int migratetype
) {}
696 static inline void set_page_order(struct page
*page
, unsigned int order
)
698 set_page_private(page
, order
);
699 __SetPageBuddy(page
);
702 static inline void rmv_page_order(struct page
*page
)
704 __ClearPageBuddy(page
);
705 set_page_private(page
, 0);
709 * This function checks whether a page is free && is the buddy
710 * we can do coalesce a page and its buddy if
711 * (a) the buddy is not in a hole &&
712 * (b) the buddy is in the buddy system &&
713 * (c) a page and its buddy have the same order &&
714 * (d) a page and its buddy are in the same zone.
716 * For recording whether a page is in the buddy system, we set ->_mapcount
717 * PAGE_BUDDY_MAPCOUNT_VALUE.
718 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
719 * serialized by zone->lock.
721 * For recording page's order, we use page_private(page).
723 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
726 if (!pfn_valid_within(page_to_pfn(buddy
)))
729 if (page_is_guard(buddy
) && page_order(buddy
) == order
) {
730 if (page_zone_id(page
) != page_zone_id(buddy
))
733 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
738 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
740 * zone check is done late to avoid uselessly
741 * calculating zone/node ids for pages that could
744 if (page_zone_id(page
) != page_zone_id(buddy
))
747 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
755 * Freeing function for a buddy system allocator.
757 * The concept of a buddy system is to maintain direct-mapped table
758 * (containing bit values) for memory blocks of various "orders".
759 * The bottom level table contains the map for the smallest allocatable
760 * units of memory (here, pages), and each level above it describes
761 * pairs of units from the levels below, hence, "buddies".
762 * At a high level, all that happens here is marking the table entry
763 * at the bottom level available, and propagating the changes upward
764 * as necessary, plus some accounting needed to play nicely with other
765 * parts of the VM system.
766 * At each level, we keep a list of pages, which are heads of continuous
767 * free pages of length of (1 << order) and marked with _mapcount
768 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
770 * So when we are allocating or freeing one, we can derive the state of the
771 * other. That is, if we allocate a small block, and both were
772 * free, the remainder of the region must be split into blocks.
773 * If a block is freed, and its buddy is also free, then this
774 * triggers coalescing into a block of larger size.
779 static inline void __free_one_page(struct page
*page
,
781 struct zone
*zone
, unsigned int order
,
784 unsigned long page_idx
;
785 unsigned long combined_idx
;
786 unsigned long uninitialized_var(buddy_idx
);
788 unsigned int max_order
;
790 max_order
= min_t(unsigned int, MAX_ORDER
, pageblock_order
+ 1);
792 VM_BUG_ON(!zone_is_initialized(zone
));
793 VM_BUG_ON_PAGE(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
, page
);
795 VM_BUG_ON(migratetype
== -1);
796 if (likely(!is_migrate_isolate(migratetype
)))
797 __mod_zone_freepage_state(zone
, 1 << order
, migratetype
);
799 page_idx
= pfn
& ((1 << MAX_ORDER
) - 1);
801 VM_BUG_ON_PAGE(page_idx
& ((1 << order
) - 1), page
);
802 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
805 while (order
< max_order
- 1) {
806 buddy_idx
= __find_buddy_index(page_idx
, order
);
807 buddy
= page
+ (buddy_idx
- page_idx
);
808 if (!page_is_buddy(page
, buddy
, order
))
811 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
812 * merge with it and move up one order.
814 if (page_is_guard(buddy
)) {
815 clear_page_guard(zone
, buddy
, order
, migratetype
);
817 list_del(&buddy
->lru
);
818 zone
->free_area
[order
].nr_free
--;
819 rmv_page_order(buddy
);
821 combined_idx
= buddy_idx
& page_idx
;
822 page
= page
+ (combined_idx
- page_idx
);
823 page_idx
= combined_idx
;
826 if (max_order
< MAX_ORDER
) {
827 /* If we are here, it means order is >= pageblock_order.
828 * We want to prevent merge between freepages on isolate
829 * pageblock and normal pageblock. Without this, pageblock
830 * isolation could cause incorrect freepage or CMA accounting.
832 * We don't want to hit this code for the more frequent
835 if (unlikely(has_isolate_pageblock(zone
))) {
838 buddy_idx
= __find_buddy_index(page_idx
, order
);
839 buddy
= page
+ (buddy_idx
- page_idx
);
840 buddy_mt
= get_pageblock_migratetype(buddy
);
842 if (migratetype
!= buddy_mt
843 && (is_migrate_isolate(migratetype
) ||
844 is_migrate_isolate(buddy_mt
)))
848 goto continue_merging
;
852 set_page_order(page
, order
);
855 * If this is not the largest possible page, check if the buddy
856 * of the next-highest order is free. If it is, it's possible
857 * that pages are being freed that will coalesce soon. In case,
858 * that is happening, add the free page to the tail of the list
859 * so it's less likely to be used soon and more likely to be merged
860 * as a higher order page
862 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
863 struct page
*higher_page
, *higher_buddy
;
864 combined_idx
= buddy_idx
& page_idx
;
865 higher_page
= page
+ (combined_idx
- page_idx
);
866 buddy_idx
= __find_buddy_index(combined_idx
, order
+ 1);
867 higher_buddy
= higher_page
+ (buddy_idx
- combined_idx
);
868 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
869 list_add_tail(&page
->lru
,
870 &zone
->free_area
[order
].free_list
[migratetype
]);
875 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
877 zone
->free_area
[order
].nr_free
++;
881 * A bad page could be due to a number of fields. Instead of multiple branches,
882 * try and check multiple fields with one check. The caller must do a detailed
883 * check if necessary.
885 static inline bool page_expected_state(struct page
*page
,
886 unsigned long check_flags
)
888 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
891 if (unlikely((unsigned long)page
->mapping
|
892 page_ref_count(page
) |
894 (unsigned long)page
->mem_cgroup
|
896 (page
->flags
& check_flags
)))
902 static void free_pages_check_bad(struct page
*page
)
904 const char *bad_reason
;
905 unsigned long bad_flags
;
910 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
911 bad_reason
= "nonzero mapcount";
912 if (unlikely(page
->mapping
!= NULL
))
913 bad_reason
= "non-NULL mapping";
914 if (unlikely(page_ref_count(page
) != 0))
915 bad_reason
= "nonzero _refcount";
916 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
)) {
917 bad_reason
= "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
918 bad_flags
= PAGE_FLAGS_CHECK_AT_FREE
;
921 if (unlikely(page
->mem_cgroup
))
922 bad_reason
= "page still charged to cgroup";
924 bad_page(page
, bad_reason
, bad_flags
);
927 static inline int free_pages_check(struct page
*page
)
929 if (likely(page_expected_state(page
, PAGE_FLAGS_CHECK_AT_FREE
)))
932 /* Something has gone sideways, find it */
933 free_pages_check_bad(page
);
937 static int free_tail_pages_check(struct page
*head_page
, struct page
*page
)
942 * We rely page->lru.next never has bit 0 set, unless the page
943 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
945 BUILD_BUG_ON((unsigned long)LIST_POISON1
& 1);
947 if (!IS_ENABLED(CONFIG_DEBUG_VM
)) {
951 switch (page
- head_page
) {
953 /* the first tail page: ->mapping is compound_mapcount() */
954 if (unlikely(compound_mapcount(page
))) {
955 bad_page(page
, "nonzero compound_mapcount", 0);
961 * the second tail page: ->mapping is
962 * page_deferred_list().next -- ignore value.
966 if (page
->mapping
!= TAIL_MAPPING
) {
967 bad_page(page
, "corrupted mapping in tail page", 0);
972 if (unlikely(!PageTail(page
))) {
973 bad_page(page
, "PageTail not set", 0);
976 if (unlikely(compound_head(page
) != head_page
)) {
977 bad_page(page
, "compound_head not consistent", 0);
982 page
->mapping
= NULL
;
983 clear_compound_head(page
);
987 static __always_inline
bool free_pages_prepare(struct page
*page
,
988 unsigned int order
, bool check_free
)
992 VM_BUG_ON_PAGE(PageTail(page
), page
);
994 trace_mm_page_free(page
, order
);
995 kmemcheck_free_shadow(page
, order
);
998 * Check tail pages before head page information is cleared to
999 * avoid checking PageCompound for order-0 pages.
1001 if (unlikely(order
)) {
1002 bool compound
= PageCompound(page
);
1005 VM_BUG_ON_PAGE(compound
&& compound_order(page
) != order
, page
);
1007 for (i
= 1; i
< (1 << order
); i
++) {
1009 bad
+= free_tail_pages_check(page
, page
+ i
);
1010 if (unlikely(free_pages_check(page
+ i
))) {
1014 (page
+ i
)->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
1017 if (PageAnonHead(page
))
1018 page
->mapping
= NULL
;
1020 bad
+= free_pages_check(page
);
1024 page_cpupid_reset_last(page
);
1025 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
1026 reset_page_owner(page
, order
);
1028 if (!PageHighMem(page
)) {
1029 debug_check_no_locks_freed(page_address(page
),
1030 PAGE_SIZE
<< order
);
1031 debug_check_no_obj_freed(page_address(page
),
1032 PAGE_SIZE
<< order
);
1034 arch_free_page(page
, order
);
1035 kernel_poison_pages(page
, 1 << order
, 0);
1036 kernel_map_pages(page
, 1 << order
, 0);
1037 kasan_free_pages(page
, order
);
1042 #ifdef CONFIG_DEBUG_VM
1043 static inline bool free_pcp_prepare(struct page
*page
)
1045 return free_pages_prepare(page
, 0, true);
1048 static inline bool bulkfree_pcp_prepare(struct page
*page
)
1053 static bool free_pcp_prepare(struct page
*page
)
1055 return free_pages_prepare(page
, 0, false);
1058 static bool bulkfree_pcp_prepare(struct page
*page
)
1060 return free_pages_check(page
);
1062 #endif /* CONFIG_DEBUG_VM */
1065 * Frees a number of pages from the PCP lists
1066 * Assumes all pages on list are in same zone, and of same order.
1067 * count is the number of pages to free.
1069 * If the zone was previously in an "all pages pinned" state then look to
1070 * see if this freeing clears that state.
1072 * And clear the zone's pages_scanned counter, to hold off the "all pages are
1073 * pinned" detection logic.
1075 static void free_pcppages_bulk(struct zone
*zone
, int count
,
1076 struct per_cpu_pages
*pcp
)
1078 int migratetype
= 0;
1080 unsigned long nr_scanned
;
1081 bool isolated_pageblocks
;
1083 spin_lock(&zone
->lock
);
1084 isolated_pageblocks
= has_isolate_pageblock(zone
);
1085 nr_scanned
= zone_page_state(zone
, NR_PAGES_SCANNED
);
1087 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, -nr_scanned
);
1091 struct list_head
*list
;
1094 * Remove pages from lists in a round-robin fashion. A
1095 * batch_free count is maintained that is incremented when an
1096 * empty list is encountered. This is so more pages are freed
1097 * off fuller lists instead of spinning excessively around empty
1102 if (++migratetype
== MIGRATE_PCPTYPES
)
1104 list
= &pcp
->lists
[migratetype
];
1105 } while (list_empty(list
));
1107 /* This is the only non-empty list. Free them all. */
1108 if (batch_free
== MIGRATE_PCPTYPES
)
1112 int mt
; /* migratetype of the to-be-freed page */
1114 page
= list_last_entry(list
, struct page
, lru
);
1115 /* must delete as __free_one_page list manipulates */
1116 list_del(&page
->lru
);
1118 mt
= get_pcppage_migratetype(page
);
1119 /* MIGRATE_ISOLATE page should not go to pcplists */
1120 VM_BUG_ON_PAGE(is_migrate_isolate(mt
), page
);
1121 /* Pageblock could have been isolated meanwhile */
1122 if (unlikely(isolated_pageblocks
))
1123 mt
= get_pageblock_migratetype(page
);
1125 if (bulkfree_pcp_prepare(page
))
1128 __free_one_page(page
, page_to_pfn(page
), zone
, 0, mt
);
1129 trace_mm_page_pcpu_drain(page
, 0, mt
);
1130 } while (--count
&& --batch_free
&& !list_empty(list
));
1132 spin_unlock(&zone
->lock
);
1135 static void free_one_page(struct zone
*zone
,
1136 struct page
*page
, unsigned long pfn
,
1140 unsigned long nr_scanned
;
1141 spin_lock(&zone
->lock
);
1142 nr_scanned
= zone_page_state(zone
, NR_PAGES_SCANNED
);
1144 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, -nr_scanned
);
1146 if (unlikely(has_isolate_pageblock(zone
) ||
1147 is_migrate_isolate(migratetype
))) {
1148 migratetype
= get_pfnblock_migratetype(page
, pfn
);
1150 __free_one_page(page
, pfn
, zone
, order
, migratetype
);
1151 spin_unlock(&zone
->lock
);
1154 static void __meminit
__init_single_page(struct page
*page
, unsigned long pfn
,
1155 unsigned long zone
, int nid
)
1157 set_page_links(page
, zone
, nid
, pfn
);
1158 init_page_count(page
);
1159 page_mapcount_reset(page
);
1160 page_cpupid_reset_last(page
);
1162 INIT_LIST_HEAD(&page
->lru
);
1163 #ifdef WANT_PAGE_VIRTUAL
1164 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1165 if (!is_highmem_idx(zone
))
1166 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
1170 static void __meminit
__init_single_pfn(unsigned long pfn
, unsigned long zone
,
1173 return __init_single_page(pfn_to_page(pfn
), pfn
, zone
, nid
);
1176 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1177 static void init_reserved_page(unsigned long pfn
)
1182 if (!early_page_uninitialised(pfn
))
1185 nid
= early_pfn_to_nid(pfn
);
1186 pgdat
= NODE_DATA(nid
);
1188 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
1189 struct zone
*zone
= &pgdat
->node_zones
[zid
];
1191 if (pfn
>= zone
->zone_start_pfn
&& pfn
< zone_end_pfn(zone
))
1194 __init_single_pfn(pfn
, zid
, nid
);
1197 static inline void init_reserved_page(unsigned long pfn
)
1200 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1203 * Initialised pages do not have PageReserved set. This function is
1204 * called for each range allocated by the bootmem allocator and
1205 * marks the pages PageReserved. The remaining valid pages are later
1206 * sent to the buddy page allocator.
1208 void __meminit
reserve_bootmem_region(phys_addr_t start
, phys_addr_t end
)
1210 unsigned long start_pfn
= PFN_DOWN(start
);
1211 unsigned long end_pfn
= PFN_UP(end
);
1213 for (; start_pfn
< end_pfn
; start_pfn
++) {
1214 if (pfn_valid(start_pfn
)) {
1215 struct page
*page
= pfn_to_page(start_pfn
);
1217 init_reserved_page(start_pfn
);
1219 /* Avoid false-positive PageTail() */
1220 INIT_LIST_HEAD(&page
->lru
);
1222 SetPageReserved(page
);
1227 static void __free_pages_ok(struct page
*page
, unsigned int order
)
1229 unsigned long flags
;
1231 unsigned long pfn
= page_to_pfn(page
);
1233 if (!free_pages_prepare(page
, order
, true))
1236 migratetype
= get_pfnblock_migratetype(page
, pfn
);
1237 local_irq_save(flags
);
1238 __count_vm_events(PGFREE
, 1 << order
);
1239 free_one_page(page_zone(page
), page
, pfn
, order
, migratetype
);
1240 local_irq_restore(flags
);
1243 static void __init
__free_pages_boot_core(struct page
*page
, unsigned int order
)
1245 unsigned int nr_pages
= 1 << order
;
1246 struct page
*p
= page
;
1250 for (loop
= 0; loop
< (nr_pages
- 1); loop
++, p
++) {
1252 __ClearPageReserved(p
);
1253 set_page_count(p
, 0);
1255 __ClearPageReserved(p
);
1256 set_page_count(p
, 0);
1258 page_zone(page
)->managed_pages
+= nr_pages
;
1259 set_page_refcounted(page
);
1260 __free_pages(page
, order
);
1263 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1264 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1266 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata
;
1268 int __meminit
early_pfn_to_nid(unsigned long pfn
)
1270 static DEFINE_SPINLOCK(early_pfn_lock
);
1273 spin_lock(&early_pfn_lock
);
1274 nid
= __early_pfn_to_nid(pfn
, &early_pfnnid_cache
);
1277 spin_unlock(&early_pfn_lock
);
1283 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1284 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1285 struct mminit_pfnnid_cache
*state
)
1289 nid
= __early_pfn_to_nid(pfn
, state
);
1290 if (nid
>= 0 && nid
!= node
)
1295 /* Only safe to use early in boot when initialisation is single-threaded */
1296 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1298 return meminit_pfn_in_nid(pfn
, node
, &early_pfnnid_cache
);
1303 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1307 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1308 struct mminit_pfnnid_cache
*state
)
1315 void __init
__free_pages_bootmem(struct page
*page
, unsigned long pfn
,
1318 if (early_page_uninitialised(pfn
))
1320 return __free_pages_boot_core(page
, order
);
1324 * Check that the whole (or subset of) a pageblock given by the interval of
1325 * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1326 * with the migration of free compaction scanner. The scanners then need to
1327 * use only pfn_valid_within() check for arches that allow holes within
1330 * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1332 * It's possible on some configurations to have a setup like node0 node1 node0
1333 * i.e. it's possible that all pages within a zones range of pages do not
1334 * belong to a single zone. We assume that a border between node0 and node1
1335 * can occur within a single pageblock, but not a node0 node1 node0
1336 * interleaving within a single pageblock. It is therefore sufficient to check
1337 * the first and last page of a pageblock and avoid checking each individual
1338 * page in a pageblock.
1340 struct page
*__pageblock_pfn_to_page(unsigned long start_pfn
,
1341 unsigned long end_pfn
, struct zone
*zone
)
1343 struct page
*start_page
;
1344 struct page
*end_page
;
1346 /* end_pfn is one past the range we are checking */
1349 if (!pfn_valid(start_pfn
) || !pfn_valid(end_pfn
))
1352 start_page
= pfn_to_page(start_pfn
);
1354 if (page_zone(start_page
) != zone
)
1357 end_page
= pfn_to_page(end_pfn
);
1359 /* This gives a shorter code than deriving page_zone(end_page) */
1360 if (page_zone_id(start_page
) != page_zone_id(end_page
))
1366 void set_zone_contiguous(struct zone
*zone
)
1368 unsigned long block_start_pfn
= zone
->zone_start_pfn
;
1369 unsigned long block_end_pfn
;
1371 block_end_pfn
= ALIGN(block_start_pfn
+ 1, pageblock_nr_pages
);
1372 for (; block_start_pfn
< zone_end_pfn(zone
);
1373 block_start_pfn
= block_end_pfn
,
1374 block_end_pfn
+= pageblock_nr_pages
) {
1376 block_end_pfn
= min(block_end_pfn
, zone_end_pfn(zone
));
1378 if (!__pageblock_pfn_to_page(block_start_pfn
,
1379 block_end_pfn
, zone
))
1383 /* We confirm that there is no hole */
1384 zone
->contiguous
= true;
1387 void clear_zone_contiguous(struct zone
*zone
)
1389 zone
->contiguous
= false;
1392 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1393 static void __init
deferred_free_range(struct page
*page
,
1394 unsigned long pfn
, int nr_pages
)
1401 /* Free a large naturally-aligned chunk if possible */
1402 if (nr_pages
== MAX_ORDER_NR_PAGES
&&
1403 (pfn
& (MAX_ORDER_NR_PAGES
-1)) == 0) {
1404 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
1405 __free_pages_boot_core(page
, MAX_ORDER
-1);
1409 for (i
= 0; i
< nr_pages
; i
++, page
++)
1410 __free_pages_boot_core(page
, 0);
1413 /* Completion tracking for deferred_init_memmap() threads */
1414 static atomic_t pgdat_init_n_undone __initdata
;
1415 static __initdata
DECLARE_COMPLETION(pgdat_init_all_done_comp
);
1417 static inline void __init
pgdat_init_report_one_done(void)
1419 if (atomic_dec_and_test(&pgdat_init_n_undone
))
1420 complete(&pgdat_init_all_done_comp
);
1423 /* Initialise remaining memory on a node */
1424 static int __init
deferred_init_memmap(void *data
)
1426 pg_data_t
*pgdat
= data
;
1427 int nid
= pgdat
->node_id
;
1428 struct mminit_pfnnid_cache nid_init_state
= { };
1429 unsigned long start
= jiffies
;
1430 unsigned long nr_pages
= 0;
1431 unsigned long walk_start
, walk_end
;
1434 unsigned long first_init_pfn
= pgdat
->first_deferred_pfn
;
1435 const struct cpumask
*cpumask
= cpumask_of_node(pgdat
->node_id
);
1437 if (first_init_pfn
== ULONG_MAX
) {
1438 pgdat_init_report_one_done();
1442 /* Bind memory initialisation thread to a local node if possible */
1443 if (!cpumask_empty(cpumask
))
1444 set_cpus_allowed_ptr(current
, cpumask
);
1446 /* Sanity check boundaries */
1447 BUG_ON(pgdat
->first_deferred_pfn
< pgdat
->node_start_pfn
);
1448 BUG_ON(pgdat
->first_deferred_pfn
> pgdat_end_pfn(pgdat
));
1449 pgdat
->first_deferred_pfn
= ULONG_MAX
;
1451 /* Only the highest zone is deferred so find it */
1452 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
1453 zone
= pgdat
->node_zones
+ zid
;
1454 if (first_init_pfn
< zone_end_pfn(zone
))
1458 for_each_mem_pfn_range(i
, nid
, &walk_start
, &walk_end
, NULL
) {
1459 unsigned long pfn
, end_pfn
;
1460 struct page
*page
= NULL
;
1461 struct page
*free_base_page
= NULL
;
1462 unsigned long free_base_pfn
= 0;
1465 end_pfn
= min(walk_end
, zone_end_pfn(zone
));
1466 pfn
= first_init_pfn
;
1467 if (pfn
< walk_start
)
1469 if (pfn
< zone
->zone_start_pfn
)
1470 pfn
= zone
->zone_start_pfn
;
1472 for (; pfn
< end_pfn
; pfn
++) {
1473 if (!pfn_valid_within(pfn
))
1477 * Ensure pfn_valid is checked every
1478 * MAX_ORDER_NR_PAGES for memory holes
1480 if ((pfn
& (MAX_ORDER_NR_PAGES
- 1)) == 0) {
1481 if (!pfn_valid(pfn
)) {
1487 if (!meminit_pfn_in_nid(pfn
, nid
, &nid_init_state
)) {
1492 /* Minimise pfn page lookups and scheduler checks */
1493 if (page
&& (pfn
& (MAX_ORDER_NR_PAGES
- 1)) != 0) {
1496 nr_pages
+= nr_to_free
;
1497 deferred_free_range(free_base_page
,
1498 free_base_pfn
, nr_to_free
);
1499 free_base_page
= NULL
;
1500 free_base_pfn
= nr_to_free
= 0;
1502 page
= pfn_to_page(pfn
);
1507 VM_BUG_ON(page_zone(page
) != zone
);
1511 __init_single_page(page
, pfn
, zid
, nid
);
1512 if (!free_base_page
) {
1513 free_base_page
= page
;
1514 free_base_pfn
= pfn
;
1519 /* Where possible, batch up pages for a single free */
1522 /* Free the current block of pages to allocator */
1523 nr_pages
+= nr_to_free
;
1524 deferred_free_range(free_base_page
, free_base_pfn
,
1526 free_base_page
= NULL
;
1527 free_base_pfn
= nr_to_free
= 0;
1530 first_init_pfn
= max(end_pfn
, first_init_pfn
);
1533 /* Sanity check that the next zone really is unpopulated */
1534 WARN_ON(++zid
< MAX_NR_ZONES
&& populated_zone(++zone
));
1536 pr_info("node %d initialised, %lu pages in %ums\n", nid
, nr_pages
,
1537 jiffies_to_msecs(jiffies
- start
));
1539 pgdat_init_report_one_done();
1542 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1544 void __init
page_alloc_init_late(void)
1548 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1551 /* There will be num_node_state(N_MEMORY) threads */
1552 atomic_set(&pgdat_init_n_undone
, num_node_state(N_MEMORY
));
1553 for_each_node_state(nid
, N_MEMORY
) {
1554 kthread_run(deferred_init_memmap
, NODE_DATA(nid
), "pgdatinit%d", nid
);
1557 /* Block until all are initialised */
1558 wait_for_completion(&pgdat_init_all_done_comp
);
1560 /* Reinit limits that are based on free pages after the kernel is up */
1561 files_maxfiles_init();
1564 for_each_populated_zone(zone
)
1565 set_zone_contiguous(zone
);
1569 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1570 void __init
init_cma_reserved_pageblock(struct page
*page
)
1572 unsigned i
= pageblock_nr_pages
;
1573 struct page
*p
= page
;
1576 __ClearPageReserved(p
);
1577 set_page_count(p
, 0);
1580 set_pageblock_migratetype(page
, MIGRATE_CMA
);
1582 if (pageblock_order
>= MAX_ORDER
) {
1583 i
= pageblock_nr_pages
;
1586 set_page_refcounted(p
);
1587 __free_pages(p
, MAX_ORDER
- 1);
1588 p
+= MAX_ORDER_NR_PAGES
;
1589 } while (i
-= MAX_ORDER_NR_PAGES
);
1591 set_page_refcounted(page
);
1592 __free_pages(page
, pageblock_order
);
1595 adjust_managed_page_count(page
, pageblock_nr_pages
);
1600 * The order of subdivision here is critical for the IO subsystem.
1601 * Please do not alter this order without good reasons and regression
1602 * testing. Specifically, as large blocks of memory are subdivided,
1603 * the order in which smaller blocks are delivered depends on the order
1604 * they're subdivided in this function. This is the primary factor
1605 * influencing the order in which pages are delivered to the IO
1606 * subsystem according to empirical testing, and this is also justified
1607 * by considering the behavior of a buddy system containing a single
1608 * large block of memory acted on by a series of small allocations.
1609 * This behavior is a critical factor in sglist merging's success.
1613 static inline void expand(struct zone
*zone
, struct page
*page
,
1614 int low
, int high
, struct free_area
*area
,
1617 unsigned long size
= 1 << high
;
1619 while (high
> low
) {
1623 VM_BUG_ON_PAGE(bad_range(zone
, &page
[size
]), &page
[size
]);
1625 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC
) &&
1626 debug_guardpage_enabled() &&
1627 high
< debug_guardpage_minorder()) {
1629 * Mark as guard pages (or page), that will allow to
1630 * merge back to allocator when buddy will be freed.
1631 * Corresponding page table entries will not be touched,
1632 * pages will stay not present in virtual address space
1634 set_page_guard(zone
, &page
[size
], high
, migratetype
);
1637 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
1639 set_page_order(&page
[size
], high
);
1643 static void check_new_page_bad(struct page
*page
)
1645 const char *bad_reason
= NULL
;
1646 unsigned long bad_flags
= 0;
1648 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
1649 bad_reason
= "nonzero mapcount";
1650 if (unlikely(page
->mapping
!= NULL
))
1651 bad_reason
= "non-NULL mapping";
1652 if (unlikely(page_ref_count(page
) != 0))
1653 bad_reason
= "nonzero _count";
1654 if (unlikely(page
->flags
& __PG_HWPOISON
)) {
1655 bad_reason
= "HWPoisoned (hardware-corrupted)";
1656 bad_flags
= __PG_HWPOISON
;
1658 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)) {
1659 bad_reason
= "PAGE_FLAGS_CHECK_AT_PREP flag set";
1660 bad_flags
= PAGE_FLAGS_CHECK_AT_PREP
;
1663 if (unlikely(page
->mem_cgroup
))
1664 bad_reason
= "page still charged to cgroup";
1666 bad_page(page
, bad_reason
, bad_flags
);
1670 * This page is about to be returned from the page allocator
1672 static inline int check_new_page(struct page
*page
)
1674 if (likely(page_expected_state(page
,
1675 PAGE_FLAGS_CHECK_AT_PREP
|__PG_HWPOISON
)))
1678 check_new_page_bad(page
);
1682 static inline bool free_pages_prezeroed(bool poisoned
)
1684 return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO
) &&
1685 page_poisoning_enabled() && poisoned
;
1688 #ifdef CONFIG_DEBUG_VM
1689 static bool check_pcp_refill(struct page
*page
)
1694 static bool check_new_pcp(struct page
*page
)
1696 return check_new_page(page
);
1699 static bool check_pcp_refill(struct page
*page
)
1701 return check_new_page(page
);
1703 static bool check_new_pcp(struct page
*page
)
1707 #endif /* CONFIG_DEBUG_VM */
1709 static bool check_new_pages(struct page
*page
, unsigned int order
)
1712 for (i
= 0; i
< (1 << order
); i
++) {
1713 struct page
*p
= page
+ i
;
1715 if (unlikely(check_new_page(p
)))
1722 static void prep_new_page(struct page
*page
, unsigned int order
, gfp_t gfp_flags
,
1723 unsigned int alloc_flags
)
1726 bool poisoned
= true;
1728 for (i
= 0; i
< (1 << order
); i
++) {
1729 struct page
*p
= page
+ i
;
1731 poisoned
&= page_is_poisoned(p
);
1734 set_page_private(page
, 0);
1735 set_page_refcounted(page
);
1737 arch_alloc_page(page
, order
);
1738 kernel_map_pages(page
, 1 << order
, 1);
1739 kernel_poison_pages(page
, 1 << order
, 1);
1740 kasan_alloc_pages(page
, order
);
1742 if (!free_pages_prezeroed(poisoned
) && (gfp_flags
& __GFP_ZERO
))
1743 for (i
= 0; i
< (1 << order
); i
++)
1744 clear_highpage(page
+ i
);
1746 if (order
&& (gfp_flags
& __GFP_COMP
))
1747 prep_compound_page(page
, order
);
1749 set_page_owner(page
, order
, gfp_flags
);
1752 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1753 * allocate the page. The expectation is that the caller is taking
1754 * steps that will free more memory. The caller should avoid the page
1755 * being used for !PFMEMALLOC purposes.
1757 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
1758 set_page_pfmemalloc(page
);
1760 clear_page_pfmemalloc(page
);
1764 * Go through the free lists for the given migratetype and remove
1765 * the smallest available page from the freelists
1768 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
1771 unsigned int current_order
;
1772 struct free_area
*area
;
1775 /* Find a page of the appropriate size in the preferred list */
1776 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
1777 area
= &(zone
->free_area
[current_order
]);
1778 page
= list_first_entry_or_null(&area
->free_list
[migratetype
],
1782 list_del(&page
->lru
);
1783 rmv_page_order(page
);
1785 expand(zone
, page
, order
, current_order
, area
, migratetype
);
1786 set_pcppage_migratetype(page
, migratetype
);
1795 * This array describes the order lists are fallen back to when
1796 * the free lists for the desirable migrate type are depleted
1798 static int fallbacks
[MIGRATE_TYPES
][4] = {
1799 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1800 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1801 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_TYPES
},
1803 [MIGRATE_CMA
] = { MIGRATE_TYPES
}, /* Never used */
1805 #ifdef CONFIG_MEMORY_ISOLATION
1806 [MIGRATE_ISOLATE
] = { MIGRATE_TYPES
}, /* Never used */
1811 static struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1814 return __rmqueue_smallest(zone
, order
, MIGRATE_CMA
);
1817 static inline struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1818 unsigned int order
) { return NULL
; }
1822 * Move the free pages in a range to the free lists of the requested type.
1823 * Note that start_page and end_pages are not aligned on a pageblock
1824 * boundary. If alignment is required, use move_freepages_block()
1826 int move_freepages(struct zone
*zone
,
1827 struct page
*start_page
, struct page
*end_page
,
1832 int pages_moved
= 0;
1834 #ifndef CONFIG_HOLES_IN_ZONE
1836 * page_zone is not safe to call in this context when
1837 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1838 * anyway as we check zone boundaries in move_freepages_block().
1839 * Remove at a later date when no bug reports exist related to
1840 * grouping pages by mobility
1842 VM_BUG_ON(page_zone(start_page
) != page_zone(end_page
));
1845 for (page
= start_page
; page
<= end_page
;) {
1846 /* Make sure we are not inadvertently changing nodes */
1847 VM_BUG_ON_PAGE(page_to_nid(page
) != zone_to_nid(zone
), page
);
1849 if (!pfn_valid_within(page_to_pfn(page
))) {
1854 if (!PageBuddy(page
)) {
1859 order
= page_order(page
);
1860 list_move(&page
->lru
,
1861 &zone
->free_area
[order
].free_list
[migratetype
]);
1863 pages_moved
+= 1 << order
;
1869 int move_freepages_block(struct zone
*zone
, struct page
*page
,
1872 unsigned long start_pfn
, end_pfn
;
1873 struct page
*start_page
, *end_page
;
1875 start_pfn
= page_to_pfn(page
);
1876 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
1877 start_page
= pfn_to_page(start_pfn
);
1878 end_page
= start_page
+ pageblock_nr_pages
- 1;
1879 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
1881 /* Do not cross zone boundaries */
1882 if (!zone_spans_pfn(zone
, start_pfn
))
1884 if (!zone_spans_pfn(zone
, end_pfn
))
1887 return move_freepages(zone
, start_page
, end_page
, migratetype
);
1890 static void change_pageblock_range(struct page
*pageblock_page
,
1891 int start_order
, int migratetype
)
1893 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
1895 while (nr_pageblocks
--) {
1896 set_pageblock_migratetype(pageblock_page
, migratetype
);
1897 pageblock_page
+= pageblock_nr_pages
;
1902 * When we are falling back to another migratetype during allocation, try to
1903 * steal extra free pages from the same pageblocks to satisfy further
1904 * allocations, instead of polluting multiple pageblocks.
1906 * If we are stealing a relatively large buddy page, it is likely there will
1907 * be more free pages in the pageblock, so try to steal them all. For
1908 * reclaimable and unmovable allocations, we steal regardless of page size,
1909 * as fragmentation caused by those allocations polluting movable pageblocks
1910 * is worse than movable allocations stealing from unmovable and reclaimable
1913 static bool can_steal_fallback(unsigned int order
, int start_mt
)
1916 * Leaving this order check is intended, although there is
1917 * relaxed order check in next check. The reason is that
1918 * we can actually steal whole pageblock if this condition met,
1919 * but, below check doesn't guarantee it and that is just heuristic
1920 * so could be changed anytime.
1922 if (order
>= pageblock_order
)
1925 if (order
>= pageblock_order
/ 2 ||
1926 start_mt
== MIGRATE_RECLAIMABLE
||
1927 start_mt
== MIGRATE_UNMOVABLE
||
1928 page_group_by_mobility_disabled
)
1935 * This function implements actual steal behaviour. If order is large enough,
1936 * we can steal whole pageblock. If not, we first move freepages in this
1937 * pageblock and check whether half of pages are moved or not. If half of
1938 * pages are moved, we can change migratetype of pageblock and permanently
1939 * use it's pages as requested migratetype in the future.
1941 static void steal_suitable_fallback(struct zone
*zone
, struct page
*page
,
1944 unsigned int current_order
= page_order(page
);
1947 /* Take ownership for orders >= pageblock_order */
1948 if (current_order
>= pageblock_order
) {
1949 change_pageblock_range(page
, current_order
, start_type
);
1953 pages
= move_freepages_block(zone
, page
, start_type
);
1955 /* Claim the whole block if over half of it is free */
1956 if (pages
>= (1 << (pageblock_order
-1)) ||
1957 page_group_by_mobility_disabled
)
1958 set_pageblock_migratetype(page
, start_type
);
1962 * Check whether there is a suitable fallback freepage with requested order.
1963 * If only_stealable is true, this function returns fallback_mt only if
1964 * we can steal other freepages all together. This would help to reduce
1965 * fragmentation due to mixed migratetype pages in one pageblock.
1967 int find_suitable_fallback(struct free_area
*area
, unsigned int order
,
1968 int migratetype
, bool only_stealable
, bool *can_steal
)
1973 if (area
->nr_free
== 0)
1978 fallback_mt
= fallbacks
[migratetype
][i
];
1979 if (fallback_mt
== MIGRATE_TYPES
)
1982 if (list_empty(&area
->free_list
[fallback_mt
]))
1985 if (can_steal_fallback(order
, migratetype
))
1988 if (!only_stealable
)
1999 * Reserve a pageblock for exclusive use of high-order atomic allocations if
2000 * there are no empty page blocks that contain a page with a suitable order
2002 static void reserve_highatomic_pageblock(struct page
*page
, struct zone
*zone
,
2003 unsigned int alloc_order
)
2006 unsigned long max_managed
, flags
;
2009 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
2010 * Check is race-prone but harmless.
2012 max_managed
= (zone
->managed_pages
/ 100) + pageblock_nr_pages
;
2013 if (zone
->nr_reserved_highatomic
>= max_managed
)
2016 spin_lock_irqsave(&zone
->lock
, flags
);
2018 /* Recheck the nr_reserved_highatomic limit under the lock */
2019 if (zone
->nr_reserved_highatomic
>= max_managed
)
2023 mt
= get_pageblock_migratetype(page
);
2024 if (mt
!= MIGRATE_HIGHATOMIC
&&
2025 !is_migrate_isolate(mt
) && !is_migrate_cma(mt
)) {
2026 zone
->nr_reserved_highatomic
+= pageblock_nr_pages
;
2027 set_pageblock_migratetype(page
, MIGRATE_HIGHATOMIC
);
2028 move_freepages_block(zone
, page
, MIGRATE_HIGHATOMIC
);
2032 spin_unlock_irqrestore(&zone
->lock
, flags
);
2036 * Used when an allocation is about to fail under memory pressure. This
2037 * potentially hurts the reliability of high-order allocations when under
2038 * intense memory pressure but failed atomic allocations should be easier
2039 * to recover from than an OOM.
2041 static void unreserve_highatomic_pageblock(const struct alloc_context
*ac
)
2043 struct zonelist
*zonelist
= ac
->zonelist
;
2044 unsigned long flags
;
2050 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
, ac
->high_zoneidx
,
2052 /* Preserve at least one pageblock */
2053 if (zone
->nr_reserved_highatomic
<= pageblock_nr_pages
)
2056 spin_lock_irqsave(&zone
->lock
, flags
);
2057 for (order
= 0; order
< MAX_ORDER
; order
++) {
2058 struct free_area
*area
= &(zone
->free_area
[order
]);
2060 page
= list_first_entry_or_null(
2061 &area
->free_list
[MIGRATE_HIGHATOMIC
],
2067 * It should never happen but changes to locking could
2068 * inadvertently allow a per-cpu drain to add pages
2069 * to MIGRATE_HIGHATOMIC while unreserving so be safe
2070 * and watch for underflows.
2072 zone
->nr_reserved_highatomic
-= min(pageblock_nr_pages
,
2073 zone
->nr_reserved_highatomic
);
2076 * Convert to ac->migratetype and avoid the normal
2077 * pageblock stealing heuristics. Minimally, the caller
2078 * is doing the work and needs the pages. More
2079 * importantly, if the block was always converted to
2080 * MIGRATE_UNMOVABLE or another type then the number
2081 * of pageblocks that cannot be completely freed
2084 set_pageblock_migratetype(page
, ac
->migratetype
);
2085 move_freepages_block(zone
, page
, ac
->migratetype
);
2086 spin_unlock_irqrestore(&zone
->lock
, flags
);
2089 spin_unlock_irqrestore(&zone
->lock
, flags
);
2093 /* Remove an element from the buddy allocator from the fallback list */
2094 static inline struct page
*
2095 __rmqueue_fallback(struct zone
*zone
, unsigned int order
, int start_migratetype
)
2097 struct free_area
*area
;
2098 unsigned int current_order
;
2103 /* Find the largest possible block of pages in the other list */
2104 for (current_order
= MAX_ORDER
-1;
2105 current_order
>= order
&& current_order
<= MAX_ORDER
-1;
2107 area
= &(zone
->free_area
[current_order
]);
2108 fallback_mt
= find_suitable_fallback(area
, current_order
,
2109 start_migratetype
, false, &can_steal
);
2110 if (fallback_mt
== -1)
2113 page
= list_first_entry(&area
->free_list
[fallback_mt
],
2116 steal_suitable_fallback(zone
, page
, start_migratetype
);
2118 /* Remove the page from the freelists */
2120 list_del(&page
->lru
);
2121 rmv_page_order(page
);
2123 expand(zone
, page
, order
, current_order
, area
,
2126 * The pcppage_migratetype may differ from pageblock's
2127 * migratetype depending on the decisions in
2128 * find_suitable_fallback(). This is OK as long as it does not
2129 * differ for MIGRATE_CMA pageblocks. Those can be used as
2130 * fallback only via special __rmqueue_cma_fallback() function
2132 set_pcppage_migratetype(page
, start_migratetype
);
2134 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
2135 start_migratetype
, fallback_mt
);
2144 * Do the hard work of removing an element from the buddy allocator.
2145 * Call me with the zone->lock already held.
2147 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
2152 page
= __rmqueue_smallest(zone
, order
, migratetype
);
2153 if (unlikely(!page
)) {
2154 if (migratetype
== MIGRATE_MOVABLE
)
2155 page
= __rmqueue_cma_fallback(zone
, order
);
2158 page
= __rmqueue_fallback(zone
, order
, migratetype
);
2161 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
2166 * Obtain a specified number of elements from the buddy allocator, all under
2167 * a single hold of the lock, for efficiency. Add them to the supplied list.
2168 * Returns the number of new pages which were placed at *list.
2170 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
2171 unsigned long count
, struct list_head
*list
,
2172 int migratetype
, bool cold
)
2176 spin_lock(&zone
->lock
);
2177 for (i
= 0; i
< count
; ++i
) {
2178 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
2179 if (unlikely(page
== NULL
))
2182 if (unlikely(check_pcp_refill(page
)))
2186 * Split buddy pages returned by expand() are received here
2187 * in physical page order. The page is added to the callers and
2188 * list and the list head then moves forward. From the callers
2189 * perspective, the linked list is ordered by page number in
2190 * some conditions. This is useful for IO devices that can
2191 * merge IO requests if the physical pages are ordered
2195 list_add(&page
->lru
, list
);
2197 list_add_tail(&page
->lru
, list
);
2199 if (is_migrate_cma(get_pcppage_migratetype(page
)))
2200 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
,
2203 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
2204 spin_unlock(&zone
->lock
);
2210 * Called from the vmstat counter updater to drain pagesets of this
2211 * currently executing processor on remote nodes after they have
2214 * Note that this function must be called with the thread pinned to
2215 * a single processor.
2217 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
2219 unsigned long flags
;
2220 int to_drain
, batch
;
2222 local_irq_save(flags
);
2223 batch
= READ_ONCE(pcp
->batch
);
2224 to_drain
= min(pcp
->count
, batch
);
2226 free_pcppages_bulk(zone
, to_drain
, pcp
);
2227 pcp
->count
-= to_drain
;
2229 local_irq_restore(flags
);
2234 * Drain pcplists of the indicated processor and zone.
2236 * The processor must either be the current processor and the
2237 * thread pinned to the current processor or a processor that
2240 static void drain_pages_zone(unsigned int cpu
, struct zone
*zone
)
2242 unsigned long flags
;
2243 struct per_cpu_pageset
*pset
;
2244 struct per_cpu_pages
*pcp
;
2246 local_irq_save(flags
);
2247 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
2251 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
2254 local_irq_restore(flags
);
2258 * Drain pcplists of all zones on the indicated processor.
2260 * The processor must either be the current processor and the
2261 * thread pinned to the current processor or a processor that
2264 static void drain_pages(unsigned int cpu
)
2268 for_each_populated_zone(zone
) {
2269 drain_pages_zone(cpu
, zone
);
2274 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
2276 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2277 * the single zone's pages.
2279 void drain_local_pages(struct zone
*zone
)
2281 int cpu
= smp_processor_id();
2284 drain_pages_zone(cpu
, zone
);
2290 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2292 * When zone parameter is non-NULL, spill just the single zone's pages.
2294 * Note that this code is protected against sending an IPI to an offline
2295 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
2296 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
2297 * nothing keeps CPUs from showing up after we populated the cpumask and
2298 * before the call to on_each_cpu_mask().
2300 void drain_all_pages(struct zone
*zone
)
2305 * Allocate in the BSS so we wont require allocation in
2306 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2308 static cpumask_t cpus_with_pcps
;
2311 * We don't care about racing with CPU hotplug event
2312 * as offline notification will cause the notified
2313 * cpu to drain that CPU pcps and on_each_cpu_mask
2314 * disables preemption as part of its processing
2316 for_each_online_cpu(cpu
) {
2317 struct per_cpu_pageset
*pcp
;
2319 bool has_pcps
= false;
2322 pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
2326 for_each_populated_zone(z
) {
2327 pcp
= per_cpu_ptr(z
->pageset
, cpu
);
2328 if (pcp
->pcp
.count
) {
2336 cpumask_set_cpu(cpu
, &cpus_with_pcps
);
2338 cpumask_clear_cpu(cpu
, &cpus_with_pcps
);
2340 on_each_cpu_mask(&cpus_with_pcps
, (smp_call_func_t
) drain_local_pages
,
2344 #ifdef CONFIG_HIBERNATION
2346 void mark_free_pages(struct zone
*zone
)
2348 unsigned long pfn
, max_zone_pfn
;
2349 unsigned long flags
;
2350 unsigned int order
, t
;
2353 if (zone_is_empty(zone
))
2356 spin_lock_irqsave(&zone
->lock
, flags
);
2358 max_zone_pfn
= zone_end_pfn(zone
);
2359 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
2360 if (pfn_valid(pfn
)) {
2361 page
= pfn_to_page(pfn
);
2363 if (page_zone(page
) != zone
)
2366 if (!swsusp_page_is_forbidden(page
))
2367 swsusp_unset_page_free(page
);
2370 for_each_migratetype_order(order
, t
) {
2371 list_for_each_entry(page
,
2372 &zone
->free_area
[order
].free_list
[t
], lru
) {
2375 pfn
= page_to_pfn(page
);
2376 for (i
= 0; i
< (1UL << order
); i
++)
2377 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
2380 spin_unlock_irqrestore(&zone
->lock
, flags
);
2382 #endif /* CONFIG_PM */
2385 * Free a 0-order page
2386 * cold == true ? free a cold page : free a hot page
2388 void free_hot_cold_page(struct page
*page
, bool cold
)
2390 struct zone
*zone
= page_zone(page
);
2391 struct per_cpu_pages
*pcp
;
2392 unsigned long flags
;
2393 unsigned long pfn
= page_to_pfn(page
);
2396 if (!free_pcp_prepare(page
))
2399 migratetype
= get_pfnblock_migratetype(page
, pfn
);
2400 set_pcppage_migratetype(page
, migratetype
);
2401 local_irq_save(flags
);
2402 __count_vm_event(PGFREE
);
2405 * We only track unmovable, reclaimable and movable on pcp lists.
2406 * Free ISOLATE pages back to the allocator because they are being
2407 * offlined but treat RESERVE as movable pages so we can get those
2408 * areas back if necessary. Otherwise, we may have to free
2409 * excessively into the page allocator
2411 if (migratetype
>= MIGRATE_PCPTYPES
) {
2412 if (unlikely(is_migrate_isolate(migratetype
))) {
2413 free_one_page(zone
, page
, pfn
, 0, migratetype
);
2416 migratetype
= MIGRATE_MOVABLE
;
2419 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2421 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
2423 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
2425 if (pcp
->count
>= pcp
->high
) {
2426 unsigned long batch
= READ_ONCE(pcp
->batch
);
2427 free_pcppages_bulk(zone
, batch
, pcp
);
2428 pcp
->count
-= batch
;
2432 local_irq_restore(flags
);
2436 * Free a list of 0-order pages
2438 void free_hot_cold_page_list(struct list_head
*list
, bool cold
)
2440 struct page
*page
, *next
;
2442 list_for_each_entry_safe(page
, next
, list
, lru
) {
2443 trace_mm_page_free_batched(page
, cold
);
2444 free_hot_cold_page(page
, cold
);
2449 * split_page takes a non-compound higher-order page, and splits it into
2450 * n (1<<order) sub-pages: page[0..n]
2451 * Each sub-page must be freed individually.
2453 * Note: this is probably too low level an operation for use in drivers.
2454 * Please consult with lkml before using this in your driver.
2456 void split_page(struct page
*page
, unsigned int order
)
2461 VM_BUG_ON_PAGE(PageCompound(page
), page
);
2462 VM_BUG_ON_PAGE(!page_count(page
), page
);
2464 #ifdef CONFIG_KMEMCHECK
2466 * Split shadow pages too, because free(page[0]) would
2467 * otherwise free the whole shadow.
2469 if (kmemcheck_page_is_tracked(page
))
2470 split_page(virt_to_page(page
[0].shadow
), order
);
2473 gfp_mask
= get_page_owner_gfp(page
);
2474 set_page_owner(page
, 0, gfp_mask
);
2475 for (i
= 1; i
< (1 << order
); i
++) {
2476 set_page_refcounted(page
+ i
);
2477 set_page_owner(page
+ i
, 0, gfp_mask
);
2480 EXPORT_SYMBOL_GPL(split_page
);
2482 int __isolate_free_page(struct page
*page
, unsigned int order
)
2484 unsigned long watermark
;
2488 BUG_ON(!PageBuddy(page
));
2490 zone
= page_zone(page
);
2491 mt
= get_pageblock_migratetype(page
);
2493 if (!is_migrate_isolate(mt
)) {
2494 /* Obey watermarks as if the page was being allocated */
2495 watermark
= low_wmark_pages(zone
) + (1 << order
);
2496 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
2499 __mod_zone_freepage_state(zone
, -(1UL << order
), mt
);
2502 /* Remove page from free list */
2503 list_del(&page
->lru
);
2504 zone
->free_area
[order
].nr_free
--;
2505 rmv_page_order(page
);
2507 set_page_owner(page
, order
, __GFP_MOVABLE
);
2509 /* Set the pageblock if the isolated page is at least a pageblock */
2510 if (order
>= pageblock_order
- 1) {
2511 struct page
*endpage
= page
+ (1 << order
) - 1;
2512 for (; page
< endpage
; page
+= pageblock_nr_pages
) {
2513 int mt
= get_pageblock_migratetype(page
);
2514 if (!is_migrate_isolate(mt
) && !is_migrate_cma(mt
))
2515 set_pageblock_migratetype(page
,
2521 return 1UL << order
;
2525 * Similar to split_page except the page is already free. As this is only
2526 * being used for migration, the migratetype of the block also changes.
2527 * As this is called with interrupts disabled, the caller is responsible
2528 * for calling arch_alloc_page() and kernel_map_page() after interrupts
2531 * Note: this is probably too low level an operation for use in drivers.
2532 * Please consult with lkml before using this in your driver.
2534 int split_free_page(struct page
*page
)
2539 order
= page_order(page
);
2541 nr_pages
= __isolate_free_page(page
, order
);
2545 /* Split into individual pages */
2546 set_page_refcounted(page
);
2547 split_page(page
, order
);
2552 * Update NUMA hit/miss statistics
2554 * Must be called with interrupts disabled.
2556 * When __GFP_OTHER_NODE is set assume the node of the preferred
2557 * zone is the local node. This is useful for daemons who allocate
2558 * memory on behalf of other processes.
2560 static inline void zone_statistics(struct zone
*preferred_zone
, struct zone
*z
,
2564 int local_nid
= numa_node_id();
2565 enum zone_stat_item local_stat
= NUMA_LOCAL
;
2567 if (unlikely(flags
& __GFP_OTHER_NODE
)) {
2568 local_stat
= NUMA_OTHER
;
2569 local_nid
= preferred_zone
->node
;
2572 if (z
->node
== local_nid
) {
2573 __inc_zone_state(z
, NUMA_HIT
);
2574 __inc_zone_state(z
, local_stat
);
2576 __inc_zone_state(z
, NUMA_MISS
);
2577 __inc_zone_state(preferred_zone
, NUMA_FOREIGN
);
2583 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2586 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
2587 struct zone
*zone
, unsigned int order
,
2588 gfp_t gfp_flags
, unsigned int alloc_flags
,
2591 unsigned long flags
;
2593 bool cold
= ((gfp_flags
& __GFP_COLD
) != 0);
2595 if (likely(order
== 0)) {
2596 struct per_cpu_pages
*pcp
;
2597 struct list_head
*list
;
2599 local_irq_save(flags
);
2601 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2602 list
= &pcp
->lists
[migratetype
];
2603 if (list_empty(list
)) {
2604 pcp
->count
+= rmqueue_bulk(zone
, 0,
2607 if (unlikely(list_empty(list
)))
2612 page
= list_last_entry(list
, struct page
, lru
);
2614 page
= list_first_entry(list
, struct page
, lru
);
2615 } while (page
&& check_new_pcp(page
));
2617 __dec_zone_state(zone
, NR_ALLOC_BATCH
);
2618 list_del(&page
->lru
);
2622 * We most definitely don't want callers attempting to
2623 * allocate greater than order-1 page units with __GFP_NOFAIL.
2625 WARN_ON_ONCE((gfp_flags
& __GFP_NOFAIL
) && (order
> 1));
2626 spin_lock_irqsave(&zone
->lock
, flags
);
2630 if (alloc_flags
& ALLOC_HARDER
) {
2631 page
= __rmqueue_smallest(zone
, order
, MIGRATE_HIGHATOMIC
);
2633 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
2636 page
= __rmqueue(zone
, order
, migratetype
);
2637 } while (page
&& check_new_pages(page
, order
));
2638 spin_unlock(&zone
->lock
);
2641 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
, -(1 << order
));
2642 __mod_zone_freepage_state(zone
, -(1 << order
),
2643 get_pcppage_migratetype(page
));
2646 if (atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]) <= 0 &&
2647 !test_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
))
2648 set_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
);
2650 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
2651 zone_statistics(preferred_zone
, zone
, gfp_flags
);
2652 local_irq_restore(flags
);
2654 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
2658 local_irq_restore(flags
);
2662 #ifdef CONFIG_FAIL_PAGE_ALLOC
2665 struct fault_attr attr
;
2667 bool ignore_gfp_highmem
;
2668 bool ignore_gfp_reclaim
;
2670 } fail_page_alloc
= {
2671 .attr
= FAULT_ATTR_INITIALIZER
,
2672 .ignore_gfp_reclaim
= true,
2673 .ignore_gfp_highmem
= true,
2677 static int __init
setup_fail_page_alloc(char *str
)
2679 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
2681 __setup("fail_page_alloc=", setup_fail_page_alloc
);
2683 static bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2685 if (order
< fail_page_alloc
.min_order
)
2687 if (gfp_mask
& __GFP_NOFAIL
)
2689 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
2691 if (fail_page_alloc
.ignore_gfp_reclaim
&&
2692 (gfp_mask
& __GFP_DIRECT_RECLAIM
))
2695 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
2698 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2700 static int __init
fail_page_alloc_debugfs(void)
2702 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
2705 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
2706 &fail_page_alloc
.attr
);
2708 return PTR_ERR(dir
);
2710 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
2711 &fail_page_alloc
.ignore_gfp_reclaim
))
2713 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
2714 &fail_page_alloc
.ignore_gfp_highmem
))
2716 if (!debugfs_create_u32("min-order", mode
, dir
,
2717 &fail_page_alloc
.min_order
))
2722 debugfs_remove_recursive(dir
);
2727 late_initcall(fail_page_alloc_debugfs
);
2729 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2731 #else /* CONFIG_FAIL_PAGE_ALLOC */
2733 static inline bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2738 #endif /* CONFIG_FAIL_PAGE_ALLOC */
2741 * Return true if free base pages are above 'mark'. For high-order checks it
2742 * will return true of the order-0 watermark is reached and there is at least
2743 * one free page of a suitable size. Checking now avoids taking the zone lock
2744 * to check in the allocation paths if no pages are free.
2746 bool __zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
2747 int classzone_idx
, unsigned int alloc_flags
,
2752 const bool alloc_harder
= (alloc_flags
& ALLOC_HARDER
);
2754 /* free_pages may go negative - that's OK */
2755 free_pages
-= (1 << order
) - 1;
2757 if (alloc_flags
& ALLOC_HIGH
)
2761 * If the caller does not have rights to ALLOC_HARDER then subtract
2762 * the high-atomic reserves. This will over-estimate the size of the
2763 * atomic reserve but it avoids a search.
2765 if (likely(!alloc_harder
))
2766 free_pages
-= z
->nr_reserved_highatomic
;
2771 /* If allocation can't use CMA areas don't use free CMA pages */
2772 if (!(alloc_flags
& ALLOC_CMA
))
2773 free_pages
-= zone_page_state(z
, NR_FREE_CMA_PAGES
);
2777 * Check watermarks for an order-0 allocation request. If these
2778 * are not met, then a high-order request also cannot go ahead
2779 * even if a suitable page happened to be free.
2781 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
2784 /* If this is an order-0 request then the watermark is fine */
2788 /* For a high-order request, check at least one suitable page is free */
2789 for (o
= order
; o
< MAX_ORDER
; o
++) {
2790 struct free_area
*area
= &z
->free_area
[o
];
2799 for (mt
= 0; mt
< MIGRATE_PCPTYPES
; mt
++) {
2800 if (!list_empty(&area
->free_list
[mt
]))
2805 if ((alloc_flags
& ALLOC_CMA
) &&
2806 !list_empty(&area
->free_list
[MIGRATE_CMA
])) {
2814 bool zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
2815 int classzone_idx
, unsigned int alloc_flags
)
2817 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
2818 zone_page_state(z
, NR_FREE_PAGES
));
2821 static inline bool zone_watermark_fast(struct zone
*z
, unsigned int order
,
2822 unsigned long mark
, int classzone_idx
, unsigned int alloc_flags
)
2824 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
2828 /* If allocation can't use CMA areas don't use free CMA pages */
2829 if (!(alloc_flags
& ALLOC_CMA
))
2830 cma_pages
= zone_page_state(z
, NR_FREE_CMA_PAGES
);
2834 * Fast check for order-0 only. If this fails then the reserves
2835 * need to be calculated. There is a corner case where the check
2836 * passes but only the high-order atomic reserve are free. If
2837 * the caller is !atomic then it'll uselessly search the free
2838 * list. That corner case is then slower but it is harmless.
2840 if (!order
&& (free_pages
- cma_pages
) > mark
+ z
->lowmem_reserve
[classzone_idx
])
2843 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
2847 bool zone_watermark_ok_safe(struct zone
*z
, unsigned int order
,
2848 unsigned long mark
, int classzone_idx
)
2850 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
2852 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
2853 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
2855 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, 0,
2860 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
2862 return local_zone
->node
== zone
->node
;
2865 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2867 return node_distance(zone_to_nid(local_zone
), zone_to_nid(zone
)) <
2870 #else /* CONFIG_NUMA */
2871 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
2876 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2880 #endif /* CONFIG_NUMA */
2882 static void reset_alloc_batches(struct zone
*preferred_zone
)
2884 struct zone
*zone
= preferred_zone
->zone_pgdat
->node_zones
;
2887 mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
2888 high_wmark_pages(zone
) - low_wmark_pages(zone
) -
2889 atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]));
2890 clear_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
);
2891 } while (zone
++ != preferred_zone
);
2895 * get_page_from_freelist goes through the zonelist trying to allocate
2898 static struct page
*
2899 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
, int alloc_flags
,
2900 const struct alloc_context
*ac
)
2902 struct zoneref
*z
= ac
->preferred_zoneref
;
2904 bool fair_skipped
= false;
2905 bool apply_fair
= (alloc_flags
& ALLOC_FAIR
);
2909 * Scan zonelist, looking for a zone with enough free.
2910 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2912 for_next_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
, ac
->high_zoneidx
,
2917 if (cpusets_enabled() &&
2918 (alloc_flags
& ALLOC_CPUSET
) &&
2919 !__cpuset_zone_allowed(zone
, gfp_mask
))
2922 * Distribute pages in proportion to the individual
2923 * zone size to ensure fair page aging. The zone a
2924 * page was allocated in should have no effect on the
2925 * time the page has in memory before being reclaimed.
2928 if (test_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
)) {
2929 fair_skipped
= true;
2932 if (!zone_local(ac
->preferred_zoneref
->zone
, zone
)) {
2939 * When allocating a page cache page for writing, we
2940 * want to get it from a zone that is within its dirty
2941 * limit, such that no single zone holds more than its
2942 * proportional share of globally allowed dirty pages.
2943 * The dirty limits take into account the zone's
2944 * lowmem reserves and high watermark so that kswapd
2945 * should be able to balance it without having to
2946 * write pages from its LRU list.
2948 * This may look like it could increase pressure on
2949 * lower zones by failing allocations in higher zones
2950 * before they are full. But the pages that do spill
2951 * over are limited as the lower zones are protected
2952 * by this very same mechanism. It should not become
2953 * a practical burden to them.
2955 * XXX: For now, allow allocations to potentially
2956 * exceed the per-zone dirty limit in the slowpath
2957 * (spread_dirty_pages unset) before going into reclaim,
2958 * which is important when on a NUMA setup the allowed
2959 * zones are together not big enough to reach the
2960 * global limit. The proper fix for these situations
2961 * will require awareness of zones in the
2962 * dirty-throttling and the flusher threads.
2964 if (ac
->spread_dirty_pages
&& !zone_dirty_ok(zone
))
2967 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
2968 if (!zone_watermark_fast(zone
, order
, mark
,
2969 ac_classzone_idx(ac
), alloc_flags
)) {
2972 /* Checked here to keep the fast path fast */
2973 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
2974 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
2977 if (zone_reclaim_mode
== 0 ||
2978 !zone_allows_reclaim(ac
->preferred_zoneref
->zone
, zone
))
2981 ret
= zone_reclaim(zone
, gfp_mask
, order
);
2983 case ZONE_RECLAIM_NOSCAN
:
2986 case ZONE_RECLAIM_FULL
:
2987 /* scanned but unreclaimable */
2990 /* did we reclaim enough */
2991 if (zone_watermark_ok(zone
, order
, mark
,
2992 ac_classzone_idx(ac
), alloc_flags
))
3000 page
= buffered_rmqueue(ac
->preferred_zoneref
->zone
, zone
, order
,
3001 gfp_mask
, alloc_flags
, ac
->migratetype
);
3003 prep_new_page(page
, order
, gfp_mask
, alloc_flags
);
3006 * If this is a high-order atomic allocation then check
3007 * if the pageblock should be reserved for the future
3009 if (unlikely(order
&& (alloc_flags
& ALLOC_HARDER
)))
3010 reserve_highatomic_pageblock(page
, zone
, order
);
3017 * The first pass makes sure allocations are spread fairly within the
3018 * local node. However, the local node might have free pages left
3019 * after the fairness batches are exhausted, and remote zones haven't
3020 * even been considered yet. Try once more without fairness, and
3021 * include remote zones now, before entering the slowpath and waking
3022 * kswapd: prefer spilling to a remote zone over swapping locally.
3027 fair_skipped
= false;
3028 reset_alloc_batches(ac
->preferred_zoneref
->zone
);
3036 * Large machines with many possible nodes should not always dump per-node
3037 * meminfo in irq context.
3039 static inline bool should_suppress_show_mem(void)
3044 ret
= in_interrupt();
3049 static DEFINE_RATELIMIT_STATE(nopage_rs
,
3050 DEFAULT_RATELIMIT_INTERVAL
,
3051 DEFAULT_RATELIMIT_BURST
);
3053 void warn_alloc_failed(gfp_t gfp_mask
, unsigned int order
, const char *fmt
, ...)
3055 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
3057 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
) ||
3058 debug_guardpage_minorder() > 0)
3062 * This documents exceptions given to allocations in certain
3063 * contexts that are allowed to allocate outside current's set
3066 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
3067 if (test_thread_flag(TIF_MEMDIE
) ||
3068 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
3069 filter
&= ~SHOW_MEM_FILTER_NODES
;
3070 if (in_interrupt() || !(gfp_mask
& __GFP_DIRECT_RECLAIM
))
3071 filter
&= ~SHOW_MEM_FILTER_NODES
;
3074 struct va_format vaf
;
3077 va_start(args
, fmt
);
3082 pr_warn("%pV", &vaf
);
3087 pr_warn("%s: page allocation failure: order:%u, mode:%#x(%pGg)\n",
3088 current
->comm
, order
, gfp_mask
, &gfp_mask
);
3090 if (!should_suppress_show_mem())
3094 static inline struct page
*
3095 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
3096 const struct alloc_context
*ac
, unsigned long *did_some_progress
)
3098 struct oom_control oc
= {
3099 .zonelist
= ac
->zonelist
,
3100 .nodemask
= ac
->nodemask
,
3101 .gfp_mask
= gfp_mask
,
3106 *did_some_progress
= 0;
3109 * Acquire the oom lock. If that fails, somebody else is
3110 * making progress for us.
3112 if (!mutex_trylock(&oom_lock
)) {
3113 *did_some_progress
= 1;
3114 schedule_timeout_uninterruptible(1);
3119 * Go through the zonelist yet one more time, keep very high watermark
3120 * here, this is only to catch a parallel oom killing, we must fail if
3121 * we're still under heavy pressure.
3123 page
= get_page_from_freelist(gfp_mask
| __GFP_HARDWALL
, order
,
3124 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
, ac
);
3128 if (!(gfp_mask
& __GFP_NOFAIL
)) {
3129 /* Coredumps can quickly deplete all memory reserves */
3130 if (current
->flags
& PF_DUMPCORE
)
3132 /* The OOM killer will not help higher order allocs */
3133 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
3135 /* The OOM killer does not needlessly kill tasks for lowmem */
3136 if (ac
->high_zoneidx
< ZONE_NORMAL
)
3138 if (pm_suspended_storage())
3141 * XXX: GFP_NOFS allocations should rather fail than rely on
3142 * other request to make a forward progress.
3143 * We are in an unfortunate situation where out_of_memory cannot
3144 * do much for this context but let's try it to at least get
3145 * access to memory reserved if the current task is killed (see
3146 * out_of_memory). Once filesystems are ready to handle allocation
3147 * failures more gracefully we should just bail out here.
3150 /* The OOM killer may not free memory on a specific node */
3151 if (gfp_mask
& __GFP_THISNODE
)
3154 /* Exhausted what can be done so it's blamo time */
3155 if (out_of_memory(&oc
) || WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
)) {
3156 *did_some_progress
= 1;
3158 if (gfp_mask
& __GFP_NOFAIL
) {
3159 page
= get_page_from_freelist(gfp_mask
, order
,
3160 ALLOC_NO_WATERMARKS
|ALLOC_CPUSET
, ac
);
3162 * fallback to ignore cpuset restriction if our nodes
3166 page
= get_page_from_freelist(gfp_mask
, order
,
3167 ALLOC_NO_WATERMARKS
, ac
);
3171 mutex_unlock(&oom_lock
);
3177 * Maximum number of compaction retries wit a progress before OOM
3178 * killer is consider as the only way to move forward.
3180 #define MAX_COMPACT_RETRIES 16
3182 #ifdef CONFIG_COMPACTION
3183 /* Try memory compaction for high-order allocations before reclaim */
3184 static struct page
*
3185 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
3186 unsigned int alloc_flags
, const struct alloc_context
*ac
,
3187 enum migrate_mode mode
, enum compact_result
*compact_result
)
3190 int contended_compaction
;
3195 current
->flags
|= PF_MEMALLOC
;
3196 *compact_result
= try_to_compact_pages(gfp_mask
, order
, alloc_flags
, ac
,
3197 mode
, &contended_compaction
);
3198 current
->flags
&= ~PF_MEMALLOC
;
3200 if (*compact_result
<= COMPACT_INACTIVE
)
3204 * At least in one zone compaction wasn't deferred or skipped, so let's
3205 * count a compaction stall
3207 count_vm_event(COMPACTSTALL
);
3209 page
= get_page_from_freelist(gfp_mask
, order
,
3210 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
3213 struct zone
*zone
= page_zone(page
);
3215 zone
->compact_blockskip_flush
= false;
3216 compaction_defer_reset(zone
, order
, true);
3217 count_vm_event(COMPACTSUCCESS
);
3222 * It's bad if compaction run occurs and fails. The most likely reason
3223 * is that pages exist, but not enough to satisfy watermarks.
3225 count_vm_event(COMPACTFAIL
);
3228 * In all zones where compaction was attempted (and not
3229 * deferred or skipped), lock contention has been detected.
3230 * For THP allocation we do not want to disrupt the others
3231 * so we fallback to base pages instead.
3233 if (contended_compaction
== COMPACT_CONTENDED_LOCK
)
3234 *compact_result
= COMPACT_CONTENDED
;
3237 * If compaction was aborted due to need_resched(), we do not
3238 * want to further increase allocation latency, unless it is
3239 * khugepaged trying to collapse.
3241 if (contended_compaction
== COMPACT_CONTENDED_SCHED
3242 && !(current
->flags
& PF_KTHREAD
))
3243 *compact_result
= COMPACT_CONTENDED
;
3251 should_compact_retry(struct alloc_context
*ac
, int order
, int alloc_flags
,
3252 enum compact_result compact_result
, enum migrate_mode
*migrate_mode
,
3253 int compaction_retries
)
3255 int max_retries
= MAX_COMPACT_RETRIES
;
3261 * compaction considers all the zone as desperately out of memory
3262 * so it doesn't really make much sense to retry except when the
3263 * failure could be caused by weak migration mode.
3265 if (compaction_failed(compact_result
)) {
3266 if (*migrate_mode
== MIGRATE_ASYNC
) {
3267 *migrate_mode
= MIGRATE_SYNC_LIGHT
;
3274 * make sure the compaction wasn't deferred or didn't bail out early
3275 * due to locks contention before we declare that we should give up.
3276 * But do not retry if the given zonelist is not suitable for
3279 if (compaction_withdrawn(compact_result
))
3280 return compaction_zonelist_suitable(ac
, order
, alloc_flags
);
3283 * !costly requests are much more important than __GFP_REPEAT
3284 * costly ones because they are de facto nofail and invoke OOM
3285 * killer to move on while costly can fail and users are ready
3286 * to cope with that. 1/4 retries is rather arbitrary but we
3287 * would need much more detailed feedback from compaction to
3288 * make a better decision.
3290 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
3292 if (compaction_retries
<= max_retries
)
3298 static inline struct page
*
3299 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
3300 unsigned int alloc_flags
, const struct alloc_context
*ac
,
3301 enum migrate_mode mode
, enum compact_result
*compact_result
)
3303 *compact_result
= COMPACT_SKIPPED
;
3308 should_compact_retry(struct alloc_context
*ac
, unsigned int order
, int alloc_flags
,
3309 enum compact_result compact_result
,
3310 enum migrate_mode
*migrate_mode
,
3311 int compaction_retries
)
3316 if (!order
|| order
> PAGE_ALLOC_COSTLY_ORDER
)
3320 * There are setups with compaction disabled which would prefer to loop
3321 * inside the allocator rather than hit the oom killer prematurely.
3322 * Let's give them a good hope and keep retrying while the order-0
3323 * watermarks are OK.
3325 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
, ac
->high_zoneidx
,
3327 if (zone_watermark_ok(zone
, 0, min_wmark_pages(zone
),
3328 ac_classzone_idx(ac
), alloc_flags
))
3333 #endif /* CONFIG_COMPACTION */
3335 /* Perform direct synchronous page reclaim */
3337 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
,
3338 const struct alloc_context
*ac
)
3340 struct reclaim_state reclaim_state
;
3345 /* We now go into synchronous reclaim */
3346 cpuset_memory_pressure_bump();
3347 current
->flags
|= PF_MEMALLOC
;
3348 lockdep_set_current_reclaim_state(gfp_mask
);
3349 reclaim_state
.reclaimed_slab
= 0;
3350 current
->reclaim_state
= &reclaim_state
;
3352 progress
= try_to_free_pages(ac
->zonelist
, order
, gfp_mask
,
3355 current
->reclaim_state
= NULL
;
3356 lockdep_clear_current_reclaim_state();
3357 current
->flags
&= ~PF_MEMALLOC
;
3364 /* The really slow allocator path where we enter direct reclaim */
3365 static inline struct page
*
3366 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
3367 unsigned int alloc_flags
, const struct alloc_context
*ac
,
3368 unsigned long *did_some_progress
)
3370 struct page
*page
= NULL
;
3371 bool drained
= false;
3373 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, ac
);
3374 if (unlikely(!(*did_some_progress
)))
3378 page
= get_page_from_freelist(gfp_mask
, order
,
3379 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
3382 * If an allocation failed after direct reclaim, it could be because
3383 * pages are pinned on the per-cpu lists or in high alloc reserves.
3384 * Shrink them them and try again
3386 if (!page
&& !drained
) {
3387 unreserve_highatomic_pageblock(ac
);
3388 drain_all_pages(NULL
);
3396 static void wake_all_kswapds(unsigned int order
, const struct alloc_context
*ac
)
3401 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
,
3402 ac
->high_zoneidx
, ac
->nodemask
)
3403 wakeup_kswapd(zone
, order
, ac_classzone_idx(ac
));
3406 static inline unsigned int
3407 gfp_to_alloc_flags(gfp_t gfp_mask
)
3409 unsigned int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
3411 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
3412 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
3415 * The caller may dip into page reserves a bit more if the caller
3416 * cannot run direct reclaim, or if the caller has realtime scheduling
3417 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
3418 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
3420 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
3422 if (gfp_mask
& __GFP_ATOMIC
) {
3424 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
3425 * if it can't schedule.
3427 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
3428 alloc_flags
|= ALLOC_HARDER
;
3430 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
3431 * comment for __cpuset_node_allowed().
3433 alloc_flags
&= ~ALLOC_CPUSET
;
3434 } else if (unlikely(rt_task(current
)) && !in_interrupt())
3435 alloc_flags
|= ALLOC_HARDER
;
3437 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
3438 if (gfp_mask
& __GFP_MEMALLOC
)
3439 alloc_flags
|= ALLOC_NO_WATERMARKS
;
3440 else if (in_serving_softirq() && (current
->flags
& PF_MEMALLOC
))
3441 alloc_flags
|= ALLOC_NO_WATERMARKS
;
3442 else if (!in_interrupt() &&
3443 ((current
->flags
& PF_MEMALLOC
) ||
3444 unlikely(test_thread_flag(TIF_MEMDIE
))))
3445 alloc_flags
|= ALLOC_NO_WATERMARKS
;
3448 if (gfpflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
3449 alloc_flags
|= ALLOC_CMA
;
3454 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask
)
3456 return !!(gfp_to_alloc_flags(gfp_mask
) & ALLOC_NO_WATERMARKS
);
3459 static inline bool is_thp_gfp_mask(gfp_t gfp_mask
)
3461 return (gfp_mask
& (GFP_TRANSHUGE
| __GFP_KSWAPD_RECLAIM
)) == GFP_TRANSHUGE
;
3465 * Maximum number of reclaim retries without any progress before OOM killer
3466 * is consider as the only way to move forward.
3468 #define MAX_RECLAIM_RETRIES 16
3471 * Checks whether it makes sense to retry the reclaim to make a forward progress
3472 * for the given allocation request.
3473 * The reclaim feedback represented by did_some_progress (any progress during
3474 * the last reclaim round) and no_progress_loops (number of reclaim rounds without
3475 * any progress in a row) is considered as well as the reclaimable pages on the
3476 * applicable zone list (with a backoff mechanism which is a function of
3477 * no_progress_loops).
3479 * Returns true if a retry is viable or false to enter the oom path.
3482 should_reclaim_retry(gfp_t gfp_mask
, unsigned order
,
3483 struct alloc_context
*ac
, int alloc_flags
,
3484 bool did_some_progress
, int no_progress_loops
)
3490 * Make sure we converge to OOM if we cannot make any progress
3491 * several times in the row.
3493 if (no_progress_loops
> MAX_RECLAIM_RETRIES
)
3497 * Keep reclaiming pages while there is a chance this will lead somewhere.
3498 * If none of the target zones can satisfy our allocation request even
3499 * if all reclaimable pages are considered then we are screwed and have
3502 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
, ac
->high_zoneidx
,
3504 unsigned long available
;
3505 unsigned long reclaimable
;
3507 available
= reclaimable
= zone_reclaimable_pages(zone
);
3508 available
-= DIV_ROUND_UP(no_progress_loops
* available
,
3509 MAX_RECLAIM_RETRIES
);
3510 available
+= zone_page_state_snapshot(zone
, NR_FREE_PAGES
);
3513 * Would the allocation succeed if we reclaimed the whole
3516 if (__zone_watermark_ok(zone
, order
, min_wmark_pages(zone
),
3517 ac_classzone_idx(ac
), alloc_flags
, available
)) {
3519 * If we didn't make any progress and have a lot of
3520 * dirty + writeback pages then we should wait for
3521 * an IO to complete to slow down the reclaim and
3522 * prevent from pre mature OOM
3524 if (!did_some_progress
) {
3525 unsigned long writeback
;
3526 unsigned long dirty
;
3528 writeback
= zone_page_state_snapshot(zone
,
3530 dirty
= zone_page_state_snapshot(zone
, NR_FILE_DIRTY
);
3532 if (2*(writeback
+ dirty
) > reclaimable
) {
3533 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3539 * Memory allocation/reclaim might be called from a WQ
3540 * context and the current implementation of the WQ
3541 * concurrency control doesn't recognize that
3542 * a particular WQ is congested if the worker thread is
3543 * looping without ever sleeping. Therefore we have to
3544 * do a short sleep here rather than calling
3547 if (current
->flags
& PF_WQ_WORKER
)
3548 schedule_timeout_uninterruptible(1);
3559 static inline struct page
*
3560 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
3561 struct alloc_context
*ac
)
3563 bool can_direct_reclaim
= gfp_mask
& __GFP_DIRECT_RECLAIM
;
3564 struct page
*page
= NULL
;
3565 unsigned int alloc_flags
;
3566 unsigned long did_some_progress
;
3567 enum migrate_mode migration_mode
= MIGRATE_ASYNC
;
3568 enum compact_result compact_result
;
3569 int compaction_retries
= 0;
3570 int no_progress_loops
= 0;
3573 * In the slowpath, we sanity check order to avoid ever trying to
3574 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3575 * be using allocators in order of preference for an area that is
3578 if (order
>= MAX_ORDER
) {
3579 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
3584 * We also sanity check to catch abuse of atomic reserves being used by
3585 * callers that are not in atomic context.
3587 if (WARN_ON_ONCE((gfp_mask
& (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)) ==
3588 (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)))
3589 gfp_mask
&= ~__GFP_ATOMIC
;
3592 if (gfp_mask
& __GFP_KSWAPD_RECLAIM
)
3593 wake_all_kswapds(order
, ac
);
3596 * OK, we're below the kswapd watermark and have kicked background
3597 * reclaim. Now things get more complex, so set up alloc_flags according
3598 * to how we want to proceed.
3600 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
3602 /* This is the last chance, in general, before the goto nopage. */
3603 page
= get_page_from_freelist(gfp_mask
, order
,
3604 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
3608 /* Allocate without watermarks if the context allows */
3609 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
3611 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3612 * the allocation is high priority and these type of
3613 * allocations are system rather than user orientated
3615 ac
->zonelist
= node_zonelist(numa_node_id(), gfp_mask
);
3616 page
= get_page_from_freelist(gfp_mask
, order
,
3617 ALLOC_NO_WATERMARKS
, ac
);
3622 /* Caller is not willing to reclaim, we can't balance anything */
3623 if (!can_direct_reclaim
) {
3625 * All existing users of the __GFP_NOFAIL are blockable, so warn
3626 * of any new users that actually allow this type of allocation
3629 WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
);
3633 /* Avoid recursion of direct reclaim */
3634 if (current
->flags
& PF_MEMALLOC
) {
3636 * __GFP_NOFAIL request from this context is rather bizarre
3637 * because we cannot reclaim anything and only can loop waiting
3638 * for somebody to do a work for us.
3640 if (WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
)) {
3647 /* Avoid allocations with no watermarks from looping endlessly */
3648 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
3652 * Try direct compaction. The first pass is asynchronous. Subsequent
3653 * attempts after direct reclaim are synchronous
3655 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
, ac
,
3661 /* Checks for THP-specific high-order allocations */
3662 if (is_thp_gfp_mask(gfp_mask
)) {
3664 * If compaction is deferred for high-order allocations, it is
3665 * because sync compaction recently failed. If this is the case
3666 * and the caller requested a THP allocation, we do not want
3667 * to heavily disrupt the system, so we fail the allocation
3668 * instead of entering direct reclaim.
3670 if (compact_result
== COMPACT_DEFERRED
)
3674 * Compaction is contended so rather back off than cause
3677 if(compact_result
== COMPACT_CONTENDED
)
3681 if (order
&& compaction_made_progress(compact_result
))
3682 compaction_retries
++;
3684 /* Try direct reclaim and then allocating */
3685 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
, alloc_flags
, ac
,
3686 &did_some_progress
);
3690 /* Do not loop if specifically requested */
3691 if (gfp_mask
& __GFP_NORETRY
)
3695 * Do not retry costly high order allocations unless they are
3698 if (order
> PAGE_ALLOC_COSTLY_ORDER
&& !(gfp_mask
& __GFP_REPEAT
))
3702 * Costly allocations might have made a progress but this doesn't mean
3703 * their order will become available due to high fragmentation so
3704 * always increment the no progress counter for them
3706 if (did_some_progress
&& order
<= PAGE_ALLOC_COSTLY_ORDER
)
3707 no_progress_loops
= 0;
3709 no_progress_loops
++;
3711 if (should_reclaim_retry(gfp_mask
, order
, ac
, alloc_flags
,
3712 did_some_progress
> 0, no_progress_loops
))
3716 * It doesn't make any sense to retry for the compaction if the order-0
3717 * reclaim is not able to make any progress because the current
3718 * implementation of the compaction depends on the sufficient amount
3719 * of free memory (see __compaction_suitable)
3721 if (did_some_progress
> 0 &&
3722 should_compact_retry(ac
, order
, alloc_flags
,
3723 compact_result
, &migration_mode
,
3724 compaction_retries
))
3727 /* Reclaim has failed us, start killing things */
3728 page
= __alloc_pages_may_oom(gfp_mask
, order
, ac
, &did_some_progress
);
3732 /* Retry as long as the OOM killer is making progress */
3733 if (did_some_progress
) {
3734 no_progress_loops
= 0;
3740 * High-order allocations do not necessarily loop after direct reclaim
3741 * and reclaim/compaction depends on compaction being called after
3742 * reclaim so call directly if necessary.
3743 * It can become very expensive to allocate transparent hugepages at
3744 * fault, so use asynchronous memory compaction for THP unless it is
3745 * khugepaged trying to collapse. All other requests should tolerate
3746 * at least light sync migration.
3748 if (is_thp_gfp_mask(gfp_mask
) && !(current
->flags
& PF_KTHREAD
))
3749 migration_mode
= MIGRATE_ASYNC
;
3751 migration_mode
= MIGRATE_SYNC_LIGHT
;
3752 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
,
3758 warn_alloc_failed(gfp_mask
, order
, NULL
);
3764 * This is the 'heart' of the zoned buddy allocator.
3767 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
3768 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
3771 unsigned int cpuset_mems_cookie
;
3772 unsigned int alloc_flags
= ALLOC_WMARK_LOW
|ALLOC_FAIR
;
3773 gfp_t alloc_mask
= gfp_mask
; /* The gfp_t that was actually used for allocation */
3774 struct alloc_context ac
= {
3775 .high_zoneidx
= gfp_zone(gfp_mask
),
3776 .zonelist
= zonelist
,
3777 .nodemask
= nodemask
,
3778 .migratetype
= gfpflags_to_migratetype(gfp_mask
),
3781 if (cpusets_enabled()) {
3782 alloc_mask
|= __GFP_HARDWALL
;
3783 alloc_flags
|= ALLOC_CPUSET
;
3785 ac
.nodemask
= &cpuset_current_mems_allowed
;
3788 gfp_mask
&= gfp_allowed_mask
;
3790 lockdep_trace_alloc(gfp_mask
);
3792 might_sleep_if(gfp_mask
& __GFP_DIRECT_RECLAIM
);
3794 if (should_fail_alloc_page(gfp_mask
, order
))
3798 * Check the zones suitable for the gfp_mask contain at least one
3799 * valid zone. It's possible to have an empty zonelist as a result
3800 * of __GFP_THISNODE and a memoryless node
3802 if (unlikely(!zonelist
->_zonerefs
->zone
))
3805 if (IS_ENABLED(CONFIG_CMA
) && ac
.migratetype
== MIGRATE_MOVABLE
)
3806 alloc_flags
|= ALLOC_CMA
;
3809 cpuset_mems_cookie
= read_mems_allowed_begin();
3811 /* Dirty zone balancing only done in the fast path */
3812 ac
.spread_dirty_pages
= (gfp_mask
& __GFP_WRITE
);
3814 /* The preferred zone is used for statistics later */
3815 ac
.preferred_zoneref
= first_zones_zonelist(ac
.zonelist
,
3816 ac
.high_zoneidx
, ac
.nodemask
);
3817 if (!ac
.preferred_zoneref
) {
3822 /* First allocation attempt */
3823 page
= get_page_from_freelist(alloc_mask
, order
, alloc_flags
, &ac
);
3828 * Runtime PM, block IO and its error handling path can deadlock
3829 * because I/O on the device might not complete.
3831 alloc_mask
= memalloc_noio_flags(gfp_mask
);
3832 ac
.spread_dirty_pages
= false;
3835 * Restore the original nodemask if it was potentially replaced with
3836 * &cpuset_current_mems_allowed to optimize the fast-path attempt.
3838 if (cpusets_enabled())
3839 ac
.nodemask
= nodemask
;
3840 page
= __alloc_pages_slowpath(alloc_mask
, order
, &ac
);
3844 * When updating a task's mems_allowed, it is possible to race with
3845 * parallel threads in such a way that an allocation can fail while
3846 * the mask is being updated. If a page allocation is about to fail,
3847 * check if the cpuset changed during allocation and if so, retry.
3849 if (unlikely(!page
&& read_mems_allowed_retry(cpuset_mems_cookie
))) {
3850 alloc_mask
= gfp_mask
;
3855 if (kmemcheck_enabled
&& page
)
3856 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
3858 trace_mm_page_alloc(page
, order
, alloc_mask
, ac
.migratetype
);
3862 EXPORT_SYMBOL(__alloc_pages_nodemask
);
3865 * Common helper functions.
3867 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
3872 * __get_free_pages() returns a 32-bit address, which cannot represent
3875 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
3877 page
= alloc_pages(gfp_mask
, order
);
3880 return (unsigned long) page_address(page
);
3882 EXPORT_SYMBOL(__get_free_pages
);
3884 unsigned long get_zeroed_page(gfp_t gfp_mask
)
3886 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
3888 EXPORT_SYMBOL(get_zeroed_page
);
3890 void __free_pages(struct page
*page
, unsigned int order
)
3892 if (put_page_testzero(page
)) {
3894 free_hot_cold_page(page
, false);
3896 __free_pages_ok(page
, order
);
3900 EXPORT_SYMBOL(__free_pages
);
3902 void free_pages(unsigned long addr
, unsigned int order
)
3905 VM_BUG_ON(!virt_addr_valid((void *)addr
));
3906 __free_pages(virt_to_page((void *)addr
), order
);
3910 EXPORT_SYMBOL(free_pages
);
3914 * An arbitrary-length arbitrary-offset area of memory which resides
3915 * within a 0 or higher order page. Multiple fragments within that page
3916 * are individually refcounted, in the page's reference counter.
3918 * The page_frag functions below provide a simple allocation framework for
3919 * page fragments. This is used by the network stack and network device
3920 * drivers to provide a backing region of memory for use as either an
3921 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3923 static struct page
*__page_frag_refill(struct page_frag_cache
*nc
,
3926 struct page
*page
= NULL
;
3927 gfp_t gfp
= gfp_mask
;
3929 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3930 gfp_mask
|= __GFP_COMP
| __GFP_NOWARN
| __GFP_NORETRY
|
3932 page
= alloc_pages_node(NUMA_NO_NODE
, gfp_mask
,
3933 PAGE_FRAG_CACHE_MAX_ORDER
);
3934 nc
->size
= page
? PAGE_FRAG_CACHE_MAX_SIZE
: PAGE_SIZE
;
3936 if (unlikely(!page
))
3937 page
= alloc_pages_node(NUMA_NO_NODE
, gfp
, 0);
3939 nc
->va
= page
? page_address(page
) : NULL
;
3944 void *__alloc_page_frag(struct page_frag_cache
*nc
,
3945 unsigned int fragsz
, gfp_t gfp_mask
)
3947 unsigned int size
= PAGE_SIZE
;
3951 if (unlikely(!nc
->va
)) {
3953 page
= __page_frag_refill(nc
, gfp_mask
);
3957 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3958 /* if size can vary use size else just use PAGE_SIZE */
3961 /* Even if we own the page, we do not use atomic_set().
3962 * This would break get_page_unless_zero() users.
3964 page_ref_add(page
, size
- 1);
3966 /* reset page count bias and offset to start of new frag */
3967 nc
->pfmemalloc
= page_is_pfmemalloc(page
);
3968 nc
->pagecnt_bias
= size
;
3972 offset
= nc
->offset
- fragsz
;
3973 if (unlikely(offset
< 0)) {
3974 page
= virt_to_page(nc
->va
);
3976 if (!page_ref_sub_and_test(page
, nc
->pagecnt_bias
))
3979 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3980 /* if size can vary use size else just use PAGE_SIZE */
3983 /* OK, page count is 0, we can safely set it */
3984 set_page_count(page
, size
);
3986 /* reset page count bias and offset to start of new frag */
3987 nc
->pagecnt_bias
= size
;
3988 offset
= size
- fragsz
;
3992 nc
->offset
= offset
;
3994 return nc
->va
+ offset
;
3996 EXPORT_SYMBOL(__alloc_page_frag
);
3999 * Frees a page fragment allocated out of either a compound or order 0 page.
4001 void __free_page_frag(void *addr
)
4003 struct page
*page
= virt_to_head_page(addr
);
4005 if (unlikely(put_page_testzero(page
)))
4006 __free_pages_ok(page
, compound_order(page
));
4008 EXPORT_SYMBOL(__free_page_frag
);
4011 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
4012 * of the current memory cgroup if __GFP_ACCOUNT is set, other than that it is
4013 * equivalent to alloc_pages.
4015 * It should be used when the caller would like to use kmalloc, but since the
4016 * allocation is large, it has to fall back to the page allocator.
4018 struct page
*alloc_kmem_pages(gfp_t gfp_mask
, unsigned int order
)
4022 page
= alloc_pages(gfp_mask
, order
);
4023 if (page
&& memcg_kmem_charge(page
, gfp_mask
, order
) != 0) {
4024 __free_pages(page
, order
);
4030 struct page
*alloc_kmem_pages_node(int nid
, gfp_t gfp_mask
, unsigned int order
)
4034 page
= alloc_pages_node(nid
, gfp_mask
, order
);
4035 if (page
&& memcg_kmem_charge(page
, gfp_mask
, order
) != 0) {
4036 __free_pages(page
, order
);
4043 * __free_kmem_pages and free_kmem_pages will free pages allocated with
4046 void __free_kmem_pages(struct page
*page
, unsigned int order
)
4048 memcg_kmem_uncharge(page
, order
);
4049 __free_pages(page
, order
);
4052 void free_kmem_pages(unsigned long addr
, unsigned int order
)
4055 VM_BUG_ON(!virt_addr_valid((void *)addr
));
4056 __free_kmem_pages(virt_to_page((void *)addr
), order
);
4060 static void *make_alloc_exact(unsigned long addr
, unsigned int order
,
4064 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
4065 unsigned long used
= addr
+ PAGE_ALIGN(size
);
4067 split_page(virt_to_page((void *)addr
), order
);
4068 while (used
< alloc_end
) {
4073 return (void *)addr
;
4077 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
4078 * @size: the number of bytes to allocate
4079 * @gfp_mask: GFP flags for the allocation
4081 * This function is similar to alloc_pages(), except that it allocates the
4082 * minimum number of pages to satisfy the request. alloc_pages() can only
4083 * allocate memory in power-of-two pages.
4085 * This function is also limited by MAX_ORDER.
4087 * Memory allocated by this function must be released by free_pages_exact().
4089 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
4091 unsigned int order
= get_order(size
);
4094 addr
= __get_free_pages(gfp_mask
, order
);
4095 return make_alloc_exact(addr
, order
, size
);
4097 EXPORT_SYMBOL(alloc_pages_exact
);
4100 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
4102 * @nid: the preferred node ID where memory should be allocated
4103 * @size: the number of bytes to allocate
4104 * @gfp_mask: GFP flags for the allocation
4106 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
4109 void * __meminit
alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
4111 unsigned int order
= get_order(size
);
4112 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
4115 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
4119 * free_pages_exact - release memory allocated via alloc_pages_exact()
4120 * @virt: the value returned by alloc_pages_exact.
4121 * @size: size of allocation, same value as passed to alloc_pages_exact().
4123 * Release the memory allocated by a previous call to alloc_pages_exact.
4125 void free_pages_exact(void *virt
, size_t size
)
4127 unsigned long addr
= (unsigned long)virt
;
4128 unsigned long end
= addr
+ PAGE_ALIGN(size
);
4130 while (addr
< end
) {
4135 EXPORT_SYMBOL(free_pages_exact
);
4138 * nr_free_zone_pages - count number of pages beyond high watermark
4139 * @offset: The zone index of the highest zone
4141 * nr_free_zone_pages() counts the number of counts pages which are beyond the
4142 * high watermark within all zones at or below a given zone index. For each
4143 * zone, the number of pages is calculated as:
4144 * managed_pages - high_pages
4146 static unsigned long nr_free_zone_pages(int offset
)
4151 /* Just pick one node, since fallback list is circular */
4152 unsigned long sum
= 0;
4154 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
4156 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
4157 unsigned long size
= zone
->managed_pages
;
4158 unsigned long high
= high_wmark_pages(zone
);
4167 * nr_free_buffer_pages - count number of pages beyond high watermark
4169 * nr_free_buffer_pages() counts the number of pages which are beyond the high
4170 * watermark within ZONE_DMA and ZONE_NORMAL.
4172 unsigned long nr_free_buffer_pages(void)
4174 return nr_free_zone_pages(gfp_zone(GFP_USER
));
4176 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
4179 * nr_free_pagecache_pages - count number of pages beyond high watermark
4181 * nr_free_pagecache_pages() counts the number of pages which are beyond the
4182 * high watermark within all zones.
4184 unsigned long nr_free_pagecache_pages(void)
4186 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
4189 static inline void show_node(struct zone
*zone
)
4191 if (IS_ENABLED(CONFIG_NUMA
))
4192 printk("Node %d ", zone_to_nid(zone
));
4195 long si_mem_available(void)
4198 unsigned long pagecache
;
4199 unsigned long wmark_low
= 0;
4200 unsigned long pages
[NR_LRU_LISTS
];
4204 for (lru
= LRU_BASE
; lru
< NR_LRU_LISTS
; lru
++)
4205 pages
[lru
] = global_page_state(NR_LRU_BASE
+ lru
);
4208 wmark_low
+= zone
->watermark
[WMARK_LOW
];
4211 * Estimate the amount of memory available for userspace allocations,
4212 * without causing swapping.
4214 available
= global_page_state(NR_FREE_PAGES
) - totalreserve_pages
;
4217 * Not all the page cache can be freed, otherwise the system will
4218 * start swapping. Assume at least half of the page cache, or the
4219 * low watermark worth of cache, needs to stay.
4221 pagecache
= pages
[LRU_ACTIVE_FILE
] + pages
[LRU_INACTIVE_FILE
];
4222 pagecache
-= min(pagecache
/ 2, wmark_low
);
4223 available
+= pagecache
;
4226 * Part of the reclaimable slab consists of items that are in use,
4227 * and cannot be freed. Cap this estimate at the low watermark.
4229 available
+= global_page_state(NR_SLAB_RECLAIMABLE
) -
4230 min(global_page_state(NR_SLAB_RECLAIMABLE
) / 2, wmark_low
);
4236 EXPORT_SYMBOL_GPL(si_mem_available
);
4238 void si_meminfo(struct sysinfo
*val
)
4240 val
->totalram
= totalram_pages
;
4241 val
->sharedram
= global_page_state(NR_SHMEM
);
4242 val
->freeram
= global_page_state(NR_FREE_PAGES
);
4243 val
->bufferram
= nr_blockdev_pages();
4244 val
->totalhigh
= totalhigh_pages
;
4245 val
->freehigh
= nr_free_highpages();
4246 val
->mem_unit
= PAGE_SIZE
;
4249 EXPORT_SYMBOL(si_meminfo
);
4252 void si_meminfo_node(struct sysinfo
*val
, int nid
)
4254 int zone_type
; /* needs to be signed */
4255 unsigned long managed_pages
= 0;
4256 unsigned long managed_highpages
= 0;
4257 unsigned long free_highpages
= 0;
4258 pg_data_t
*pgdat
= NODE_DATA(nid
);
4260 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++)
4261 managed_pages
+= pgdat
->node_zones
[zone_type
].managed_pages
;
4262 val
->totalram
= managed_pages
;
4263 val
->sharedram
= node_page_state(nid
, NR_SHMEM
);
4264 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
4265 #ifdef CONFIG_HIGHMEM
4266 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
4267 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4269 if (is_highmem(zone
)) {
4270 managed_highpages
+= zone
->managed_pages
;
4271 free_highpages
+= zone_page_state(zone
, NR_FREE_PAGES
);
4274 val
->totalhigh
= managed_highpages
;
4275 val
->freehigh
= free_highpages
;
4277 val
->totalhigh
= managed_highpages
;
4278 val
->freehigh
= free_highpages
;
4280 val
->mem_unit
= PAGE_SIZE
;
4285 * Determine whether the node should be displayed or not, depending on whether
4286 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
4288 bool skip_free_areas_node(unsigned int flags
, int nid
)
4291 unsigned int cpuset_mems_cookie
;
4293 if (!(flags
& SHOW_MEM_FILTER_NODES
))
4297 cpuset_mems_cookie
= read_mems_allowed_begin();
4298 ret
= !node_isset(nid
, cpuset_current_mems_allowed
);
4299 } while (read_mems_allowed_retry(cpuset_mems_cookie
));
4304 #define K(x) ((x) << (PAGE_SHIFT-10))
4306 static void show_migration_types(unsigned char type
)
4308 static const char types
[MIGRATE_TYPES
] = {
4309 [MIGRATE_UNMOVABLE
] = 'U',
4310 [MIGRATE_MOVABLE
] = 'M',
4311 [MIGRATE_RECLAIMABLE
] = 'E',
4312 [MIGRATE_HIGHATOMIC
] = 'H',
4314 [MIGRATE_CMA
] = 'C',
4316 #ifdef CONFIG_MEMORY_ISOLATION
4317 [MIGRATE_ISOLATE
] = 'I',
4320 char tmp
[MIGRATE_TYPES
+ 1];
4324 for (i
= 0; i
< MIGRATE_TYPES
; i
++) {
4325 if (type
& (1 << i
))
4330 printk("(%s) ", tmp
);
4334 * Show free area list (used inside shift_scroll-lock stuff)
4335 * We also calculate the percentage fragmentation. We do this by counting the
4336 * memory on each free list with the exception of the first item on the list.
4339 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
4342 void show_free_areas(unsigned int filter
)
4344 unsigned long free_pcp
= 0;
4348 for_each_populated_zone(zone
) {
4349 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
4352 for_each_online_cpu(cpu
)
4353 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
4356 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
4357 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
4358 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
4359 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
4360 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
4361 " free:%lu free_pcp:%lu free_cma:%lu\n",
4362 global_page_state(NR_ACTIVE_ANON
),
4363 global_page_state(NR_INACTIVE_ANON
),
4364 global_page_state(NR_ISOLATED_ANON
),
4365 global_page_state(NR_ACTIVE_FILE
),
4366 global_page_state(NR_INACTIVE_FILE
),
4367 global_page_state(NR_ISOLATED_FILE
),
4368 global_page_state(NR_UNEVICTABLE
),
4369 global_page_state(NR_FILE_DIRTY
),
4370 global_page_state(NR_WRITEBACK
),
4371 global_page_state(NR_UNSTABLE_NFS
),
4372 global_page_state(NR_SLAB_RECLAIMABLE
),
4373 global_page_state(NR_SLAB_UNRECLAIMABLE
),
4374 global_page_state(NR_FILE_MAPPED
),
4375 global_page_state(NR_SHMEM
),
4376 global_page_state(NR_PAGETABLE
),
4377 global_page_state(NR_BOUNCE
),
4378 global_page_state(NR_FREE_PAGES
),
4380 global_page_state(NR_FREE_CMA_PAGES
));
4382 for_each_populated_zone(zone
) {
4385 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
4389 for_each_online_cpu(cpu
)
4390 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
4398 " active_anon:%lukB"
4399 " inactive_anon:%lukB"
4400 " active_file:%lukB"
4401 " inactive_file:%lukB"
4402 " unevictable:%lukB"
4403 " isolated(anon):%lukB"
4404 " isolated(file):%lukB"
4412 " slab_reclaimable:%lukB"
4413 " slab_unreclaimable:%lukB"
4414 " kernel_stack:%lukB"
4421 " writeback_tmp:%lukB"
4422 " pages_scanned:%lu"
4423 " all_unreclaimable? %s"
4426 K(zone_page_state(zone
, NR_FREE_PAGES
)),
4427 K(min_wmark_pages(zone
)),
4428 K(low_wmark_pages(zone
)),
4429 K(high_wmark_pages(zone
)),
4430 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
4431 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
4432 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
4433 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
4434 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
4435 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
4436 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
4437 K(zone
->present_pages
),
4438 K(zone
->managed_pages
),
4439 K(zone_page_state(zone
, NR_MLOCK
)),
4440 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
4441 K(zone_page_state(zone
, NR_WRITEBACK
)),
4442 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
4443 K(zone_page_state(zone
, NR_SHMEM
)),
4444 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
4445 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
4446 zone_page_state(zone
, NR_KERNEL_STACK
) *
4448 K(zone_page_state(zone
, NR_PAGETABLE
)),
4449 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
4450 K(zone_page_state(zone
, NR_BOUNCE
)),
4452 K(this_cpu_read(zone
->pageset
->pcp
.count
)),
4453 K(zone_page_state(zone
, NR_FREE_CMA_PAGES
)),
4454 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
4455 K(zone_page_state(zone
, NR_PAGES_SCANNED
)),
4456 (!zone_reclaimable(zone
) ? "yes" : "no")
4458 printk("lowmem_reserve[]:");
4459 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4460 printk(" %ld", zone
->lowmem_reserve
[i
]);
4464 for_each_populated_zone(zone
) {
4466 unsigned long nr
[MAX_ORDER
], flags
, total
= 0;
4467 unsigned char types
[MAX_ORDER
];
4469 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
4472 printk("%s: ", zone
->name
);
4474 spin_lock_irqsave(&zone
->lock
, flags
);
4475 for (order
= 0; order
< MAX_ORDER
; order
++) {
4476 struct free_area
*area
= &zone
->free_area
[order
];
4479 nr
[order
] = area
->nr_free
;
4480 total
+= nr
[order
] << order
;
4483 for (type
= 0; type
< MIGRATE_TYPES
; type
++) {
4484 if (!list_empty(&area
->free_list
[type
]))
4485 types
[order
] |= 1 << type
;
4488 spin_unlock_irqrestore(&zone
->lock
, flags
);
4489 for (order
= 0; order
< MAX_ORDER
; order
++) {
4490 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
4492 show_migration_types(types
[order
]);
4494 printk("= %lukB\n", K(total
));
4497 hugetlb_show_meminfo();
4499 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
4501 show_swap_cache_info();
4504 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
4506 zoneref
->zone
= zone
;
4507 zoneref
->zone_idx
= zone_idx(zone
);
4511 * Builds allocation fallback zone lists.
4513 * Add all populated zones of a node to the zonelist.
4515 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
4519 enum zone_type zone_type
= MAX_NR_ZONES
;
4523 zone
= pgdat
->node_zones
+ zone_type
;
4524 if (populated_zone(zone
)) {
4525 zoneref_set_zone(zone
,
4526 &zonelist
->_zonerefs
[nr_zones
++]);
4527 check_highest_zone(zone_type
);
4529 } while (zone_type
);
4537 * 0 = automatic detection of better ordering.
4538 * 1 = order by ([node] distance, -zonetype)
4539 * 2 = order by (-zonetype, [node] distance)
4541 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
4542 * the same zonelist. So only NUMA can configure this param.
4544 #define ZONELIST_ORDER_DEFAULT 0
4545 #define ZONELIST_ORDER_NODE 1
4546 #define ZONELIST_ORDER_ZONE 2
4548 /* zonelist order in the kernel.
4549 * set_zonelist_order() will set this to NODE or ZONE.
4551 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4552 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
4556 /* The value user specified ....changed by config */
4557 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4558 /* string for sysctl */
4559 #define NUMA_ZONELIST_ORDER_LEN 16
4560 char numa_zonelist_order
[16] = "default";
4563 * interface for configure zonelist ordering.
4564 * command line option "numa_zonelist_order"
4565 * = "[dD]efault - default, automatic configuration.
4566 * = "[nN]ode - order by node locality, then by zone within node
4567 * = "[zZ]one - order by zone, then by locality within zone
4570 static int __parse_numa_zonelist_order(char *s
)
4572 if (*s
== 'd' || *s
== 'D') {
4573 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4574 } else if (*s
== 'n' || *s
== 'N') {
4575 user_zonelist_order
= ZONELIST_ORDER_NODE
;
4576 } else if (*s
== 'z' || *s
== 'Z') {
4577 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
4579 pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s
);
4585 static __init
int setup_numa_zonelist_order(char *s
)
4592 ret
= __parse_numa_zonelist_order(s
);
4594 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
4598 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
4601 * sysctl handler for numa_zonelist_order
4603 int numa_zonelist_order_handler(struct ctl_table
*table
, int write
,
4604 void __user
*buffer
, size_t *length
,
4607 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
4609 static DEFINE_MUTEX(zl_order_mutex
);
4611 mutex_lock(&zl_order_mutex
);
4613 if (strlen((char *)table
->data
) >= NUMA_ZONELIST_ORDER_LEN
) {
4617 strcpy(saved_string
, (char *)table
->data
);
4619 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
4623 int oldval
= user_zonelist_order
;
4625 ret
= __parse_numa_zonelist_order((char *)table
->data
);
4628 * bogus value. restore saved string
4630 strncpy((char *)table
->data
, saved_string
,
4631 NUMA_ZONELIST_ORDER_LEN
);
4632 user_zonelist_order
= oldval
;
4633 } else if (oldval
!= user_zonelist_order
) {
4634 mutex_lock(&zonelists_mutex
);
4635 build_all_zonelists(NULL
, NULL
);
4636 mutex_unlock(&zonelists_mutex
);
4640 mutex_unlock(&zl_order_mutex
);
4645 #define MAX_NODE_LOAD (nr_online_nodes)
4646 static int node_load
[MAX_NUMNODES
];
4649 * find_next_best_node - find the next node that should appear in a given node's fallback list
4650 * @node: node whose fallback list we're appending
4651 * @used_node_mask: nodemask_t of already used nodes
4653 * We use a number of factors to determine which is the next node that should
4654 * appear on a given node's fallback list. The node should not have appeared
4655 * already in @node's fallback list, and it should be the next closest node
4656 * according to the distance array (which contains arbitrary distance values
4657 * from each node to each node in the system), and should also prefer nodes
4658 * with no CPUs, since presumably they'll have very little allocation pressure
4659 * on them otherwise.
4660 * It returns -1 if no node is found.
4662 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
4665 int min_val
= INT_MAX
;
4666 int best_node
= NUMA_NO_NODE
;
4667 const struct cpumask
*tmp
= cpumask_of_node(0);
4669 /* Use the local node if we haven't already */
4670 if (!node_isset(node
, *used_node_mask
)) {
4671 node_set(node
, *used_node_mask
);
4675 for_each_node_state(n
, N_MEMORY
) {
4677 /* Don't want a node to appear more than once */
4678 if (node_isset(n
, *used_node_mask
))
4681 /* Use the distance array to find the distance */
4682 val
= node_distance(node
, n
);
4684 /* Penalize nodes under us ("prefer the next node") */
4687 /* Give preference to headless and unused nodes */
4688 tmp
= cpumask_of_node(n
);
4689 if (!cpumask_empty(tmp
))
4690 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
4692 /* Slight preference for less loaded node */
4693 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
4694 val
+= node_load
[n
];
4696 if (val
< min_val
) {
4703 node_set(best_node
, *used_node_mask
);
4710 * Build zonelists ordered by node and zones within node.
4711 * This results in maximum locality--normal zone overflows into local
4712 * DMA zone, if any--but risks exhausting DMA zone.
4714 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
4717 struct zonelist
*zonelist
;
4719 zonelist
= &pgdat
->node_zonelists
[0];
4720 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
4722 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4723 zonelist
->_zonerefs
[j
].zone
= NULL
;
4724 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4728 * Build gfp_thisnode zonelists
4730 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
4733 struct zonelist
*zonelist
;
4735 zonelist
= &pgdat
->node_zonelists
[1];
4736 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4737 zonelist
->_zonerefs
[j
].zone
= NULL
;
4738 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4742 * Build zonelists ordered by zone and nodes within zones.
4743 * This results in conserving DMA zone[s] until all Normal memory is
4744 * exhausted, but results in overflowing to remote node while memory
4745 * may still exist in local DMA zone.
4747 static int node_order
[MAX_NUMNODES
];
4749 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
4752 int zone_type
; /* needs to be signed */
4754 struct zonelist
*zonelist
;
4756 zonelist
= &pgdat
->node_zonelists
[0];
4758 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
4759 for (j
= 0; j
< nr_nodes
; j
++) {
4760 node
= node_order
[j
];
4761 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
4762 if (populated_zone(z
)) {
4764 &zonelist
->_zonerefs
[pos
++]);
4765 check_highest_zone(zone_type
);
4769 zonelist
->_zonerefs
[pos
].zone
= NULL
;
4770 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
4773 #if defined(CONFIG_64BIT)
4775 * Devices that require DMA32/DMA are relatively rare and do not justify a
4776 * penalty to every machine in case the specialised case applies. Default
4777 * to Node-ordering on 64-bit NUMA machines
4779 static int default_zonelist_order(void)
4781 return ZONELIST_ORDER_NODE
;
4785 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4786 * by the kernel. If processes running on node 0 deplete the low memory zone
4787 * then reclaim will occur more frequency increasing stalls and potentially
4788 * be easier to OOM if a large percentage of the zone is under writeback or
4789 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4790 * Hence, default to zone ordering on 32-bit.
4792 static int default_zonelist_order(void)
4794 return ZONELIST_ORDER_ZONE
;
4796 #endif /* CONFIG_64BIT */
4798 static void set_zonelist_order(void)
4800 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
4801 current_zonelist_order
= default_zonelist_order();
4803 current_zonelist_order
= user_zonelist_order
;
4806 static void build_zonelists(pg_data_t
*pgdat
)
4809 nodemask_t used_mask
;
4810 int local_node
, prev_node
;
4811 struct zonelist
*zonelist
;
4812 unsigned int order
= current_zonelist_order
;
4814 /* initialize zonelists */
4815 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
4816 zonelist
= pgdat
->node_zonelists
+ i
;
4817 zonelist
->_zonerefs
[0].zone
= NULL
;
4818 zonelist
->_zonerefs
[0].zone_idx
= 0;
4821 /* NUMA-aware ordering of nodes */
4822 local_node
= pgdat
->node_id
;
4823 load
= nr_online_nodes
;
4824 prev_node
= local_node
;
4825 nodes_clear(used_mask
);
4827 memset(node_order
, 0, sizeof(node_order
));
4830 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
4832 * We don't want to pressure a particular node.
4833 * So adding penalty to the first node in same
4834 * distance group to make it round-robin.
4836 if (node_distance(local_node
, node
) !=
4837 node_distance(local_node
, prev_node
))
4838 node_load
[node
] = load
;
4842 if (order
== ZONELIST_ORDER_NODE
)
4843 build_zonelists_in_node_order(pgdat
, node
);
4845 node_order
[i
++] = node
; /* remember order */
4848 if (order
== ZONELIST_ORDER_ZONE
) {
4849 /* calculate node order -- i.e., DMA last! */
4850 build_zonelists_in_zone_order(pgdat
, i
);
4853 build_thisnode_zonelists(pgdat
);
4856 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4858 * Return node id of node used for "local" allocations.
4859 * I.e., first node id of first zone in arg node's generic zonelist.
4860 * Used for initializing percpu 'numa_mem', which is used primarily
4861 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4863 int local_memory_node(int node
)
4867 z
= first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
4868 gfp_zone(GFP_KERNEL
),
4870 return z
->zone
->node
;
4874 #else /* CONFIG_NUMA */
4876 static void set_zonelist_order(void)
4878 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
4881 static void build_zonelists(pg_data_t
*pgdat
)
4883 int node
, local_node
;
4885 struct zonelist
*zonelist
;
4887 local_node
= pgdat
->node_id
;
4889 zonelist
= &pgdat
->node_zonelists
[0];
4890 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4893 * Now we build the zonelist so that it contains the zones
4894 * of all the other nodes.
4895 * We don't want to pressure a particular node, so when
4896 * building the zones for node N, we make sure that the
4897 * zones coming right after the local ones are those from
4898 * node N+1 (modulo N)
4900 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
4901 if (!node_online(node
))
4903 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4905 for (node
= 0; node
< local_node
; node
++) {
4906 if (!node_online(node
))
4908 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4911 zonelist
->_zonerefs
[j
].zone
= NULL
;
4912 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4915 #endif /* CONFIG_NUMA */
4918 * Boot pageset table. One per cpu which is going to be used for all
4919 * zones and all nodes. The parameters will be set in such a way
4920 * that an item put on a list will immediately be handed over to
4921 * the buddy list. This is safe since pageset manipulation is done
4922 * with interrupts disabled.
4924 * The boot_pagesets must be kept even after bootup is complete for
4925 * unused processors and/or zones. They do play a role for bootstrapping
4926 * hotplugged processors.
4928 * zoneinfo_show() and maybe other functions do
4929 * not check if the processor is online before following the pageset pointer.
4930 * Other parts of the kernel may not check if the zone is available.
4932 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
4933 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
4934 static void setup_zone_pageset(struct zone
*zone
);
4937 * Global mutex to protect against size modification of zonelists
4938 * as well as to serialize pageset setup for the new populated zone.
4940 DEFINE_MUTEX(zonelists_mutex
);
4942 /* return values int ....just for stop_machine() */
4943 static int __build_all_zonelists(void *data
)
4947 pg_data_t
*self
= data
;
4950 memset(node_load
, 0, sizeof(node_load
));
4953 if (self
&& !node_online(self
->node_id
)) {
4954 build_zonelists(self
);
4957 for_each_online_node(nid
) {
4958 pg_data_t
*pgdat
= NODE_DATA(nid
);
4960 build_zonelists(pgdat
);
4964 * Initialize the boot_pagesets that are going to be used
4965 * for bootstrapping processors. The real pagesets for
4966 * each zone will be allocated later when the per cpu
4967 * allocator is available.
4969 * boot_pagesets are used also for bootstrapping offline
4970 * cpus if the system is already booted because the pagesets
4971 * are needed to initialize allocators on a specific cpu too.
4972 * F.e. the percpu allocator needs the page allocator which
4973 * needs the percpu allocator in order to allocate its pagesets
4974 * (a chicken-egg dilemma).
4976 for_each_possible_cpu(cpu
) {
4977 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
4979 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4981 * We now know the "local memory node" for each node--
4982 * i.e., the node of the first zone in the generic zonelist.
4983 * Set up numa_mem percpu variable for on-line cpus. During
4984 * boot, only the boot cpu should be on-line; we'll init the
4985 * secondary cpus' numa_mem as they come on-line. During
4986 * node/memory hotplug, we'll fixup all on-line cpus.
4988 if (cpu_online(cpu
))
4989 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
4996 static noinline
void __init
4997 build_all_zonelists_init(void)
4999 __build_all_zonelists(NULL
);
5000 mminit_verify_zonelist();
5001 cpuset_init_current_mems_allowed();
5005 * Called with zonelists_mutex held always
5006 * unless system_state == SYSTEM_BOOTING.
5008 * __ref due to (1) call of __meminit annotated setup_zone_pageset
5009 * [we're only called with non-NULL zone through __meminit paths] and
5010 * (2) call of __init annotated helper build_all_zonelists_init
5011 * [protected by SYSTEM_BOOTING].
5013 void __ref
build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
)
5015 set_zonelist_order();
5017 if (system_state
== SYSTEM_BOOTING
) {
5018 build_all_zonelists_init();
5020 #ifdef CONFIG_MEMORY_HOTPLUG
5022 setup_zone_pageset(zone
);
5024 /* we have to stop all cpus to guarantee there is no user
5026 stop_machine(__build_all_zonelists
, pgdat
, NULL
);
5027 /* cpuset refresh routine should be here */
5029 vm_total_pages
= nr_free_pagecache_pages();
5031 * Disable grouping by mobility if the number of pages in the
5032 * system is too low to allow the mechanism to work. It would be
5033 * more accurate, but expensive to check per-zone. This check is
5034 * made on memory-hotadd so a system can start with mobility
5035 * disabled and enable it later
5037 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
5038 page_group_by_mobility_disabled
= 1;
5040 page_group_by_mobility_disabled
= 0;
5042 pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n",
5044 zonelist_order_name
[current_zonelist_order
],
5045 page_group_by_mobility_disabled
? "off" : "on",
5048 pr_info("Policy zone: %s\n", zone_names
[policy_zone
]);
5053 * Helper functions to size the waitqueue hash table.
5054 * Essentially these want to choose hash table sizes sufficiently
5055 * large so that collisions trying to wait on pages are rare.
5056 * But in fact, the number of active page waitqueues on typical
5057 * systems is ridiculously low, less than 200. So this is even
5058 * conservative, even though it seems large.
5060 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
5061 * waitqueues, i.e. the size of the waitq table given the number of pages.
5063 #define PAGES_PER_WAITQUEUE 256
5065 #ifndef CONFIG_MEMORY_HOTPLUG
5066 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
5068 unsigned long size
= 1;
5070 pages
/= PAGES_PER_WAITQUEUE
;
5072 while (size
< pages
)
5076 * Once we have dozens or even hundreds of threads sleeping
5077 * on IO we've got bigger problems than wait queue collision.
5078 * Limit the size of the wait table to a reasonable size.
5080 size
= min(size
, 4096UL);
5082 return max(size
, 4UL);
5086 * A zone's size might be changed by hot-add, so it is not possible to determine
5087 * a suitable size for its wait_table. So we use the maximum size now.
5089 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
5091 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
5092 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
5093 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
5095 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
5096 * or more by the traditional way. (See above). It equals:
5098 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
5099 * ia64(16K page size) : = ( 8G + 4M)byte.
5100 * powerpc (64K page size) : = (32G +16M)byte.
5102 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
5109 * This is an integer logarithm so that shifts can be used later
5110 * to extract the more random high bits from the multiplicative
5111 * hash function before the remainder is taken.
5113 static inline unsigned long wait_table_bits(unsigned long size
)
5119 * Initially all pages are reserved - free ones are freed
5120 * up by free_all_bootmem() once the early boot process is
5121 * done. Non-atomic initialization, single-pass.
5123 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
5124 unsigned long start_pfn
, enum memmap_context context
)
5126 struct vmem_altmap
*altmap
= to_vmem_altmap(__pfn_to_phys(start_pfn
));
5127 unsigned long end_pfn
= start_pfn
+ size
;
5128 pg_data_t
*pgdat
= NODE_DATA(nid
);
5130 unsigned long nr_initialised
= 0;
5131 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5132 struct memblock_region
*r
= NULL
, *tmp
;
5135 if (highest_memmap_pfn
< end_pfn
- 1)
5136 highest_memmap_pfn
= end_pfn
- 1;
5139 * Honor reservation requested by the driver for this ZONE_DEVICE
5142 if (altmap
&& start_pfn
== altmap
->base_pfn
)
5143 start_pfn
+= altmap
->reserve
;
5145 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
5147 * There can be holes in boot-time mem_map[]s handed to this
5148 * function. They do not exist on hotplugged memory.
5150 if (context
!= MEMMAP_EARLY
)
5153 if (!early_pfn_valid(pfn
))
5155 if (!early_pfn_in_nid(pfn
, nid
))
5157 if (!update_defer_init(pgdat
, pfn
, end_pfn
, &nr_initialised
))
5160 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5162 * If not mirrored_kernelcore and ZONE_MOVABLE exists, range
5163 * from zone_movable_pfn[nid] to end of each node should be
5164 * ZONE_MOVABLE not ZONE_NORMAL. skip it.
5166 if (!mirrored_kernelcore
&& zone_movable_pfn
[nid
])
5167 if (zone
== ZONE_NORMAL
&& pfn
>= zone_movable_pfn
[nid
])
5171 * Check given memblock attribute by firmware which can affect
5172 * kernel memory layout. If zone==ZONE_MOVABLE but memory is
5173 * mirrored, it's an overlapped memmap init. skip it.
5175 if (mirrored_kernelcore
&& zone
== ZONE_MOVABLE
) {
5176 if (!r
|| pfn
>= memblock_region_memory_end_pfn(r
)) {
5177 for_each_memblock(memory
, tmp
)
5178 if (pfn
< memblock_region_memory_end_pfn(tmp
))
5182 if (pfn
>= memblock_region_memory_base_pfn(r
) &&
5183 memblock_is_mirror(r
)) {
5184 /* already initialized as NORMAL */
5185 pfn
= memblock_region_memory_end_pfn(r
);
5193 * Mark the block movable so that blocks are reserved for
5194 * movable at startup. This will force kernel allocations
5195 * to reserve their blocks rather than leaking throughout
5196 * the address space during boot when many long-lived
5197 * kernel allocations are made.
5199 * bitmap is created for zone's valid pfn range. but memmap
5200 * can be created for invalid pages (for alignment)
5201 * check here not to call set_pageblock_migratetype() against
5204 if (!(pfn
& (pageblock_nr_pages
- 1))) {
5205 struct page
*page
= pfn_to_page(pfn
);
5207 __init_single_page(page
, pfn
, zone
, nid
);
5208 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
5210 __init_single_pfn(pfn
, zone
, nid
);
5215 static void __meminit
zone_init_free_lists(struct zone
*zone
)
5217 unsigned int order
, t
;
5218 for_each_migratetype_order(order
, t
) {
5219 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
5220 zone
->free_area
[order
].nr_free
= 0;
5224 #ifndef __HAVE_ARCH_MEMMAP_INIT
5225 #define memmap_init(size, nid, zone, start_pfn) \
5226 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
5229 static int zone_batchsize(struct zone
*zone
)
5235 * The per-cpu-pages pools are set to around 1000th of the
5236 * size of the zone. But no more than 1/2 of a meg.
5238 * OK, so we don't know how big the cache is. So guess.
5240 batch
= zone
->managed_pages
/ 1024;
5241 if (batch
* PAGE_SIZE
> 512 * 1024)
5242 batch
= (512 * 1024) / PAGE_SIZE
;
5243 batch
/= 4; /* We effectively *= 4 below */
5248 * Clamp the batch to a 2^n - 1 value. Having a power
5249 * of 2 value was found to be more likely to have
5250 * suboptimal cache aliasing properties in some cases.
5252 * For example if 2 tasks are alternately allocating
5253 * batches of pages, one task can end up with a lot
5254 * of pages of one half of the possible page colors
5255 * and the other with pages of the other colors.
5257 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
5262 /* The deferral and batching of frees should be suppressed under NOMMU
5265 * The problem is that NOMMU needs to be able to allocate large chunks
5266 * of contiguous memory as there's no hardware page translation to
5267 * assemble apparent contiguous memory from discontiguous pages.
5269 * Queueing large contiguous runs of pages for batching, however,
5270 * causes the pages to actually be freed in smaller chunks. As there
5271 * can be a significant delay between the individual batches being
5272 * recycled, this leads to the once large chunks of space being
5273 * fragmented and becoming unavailable for high-order allocations.
5280 * pcp->high and pcp->batch values are related and dependent on one another:
5281 * ->batch must never be higher then ->high.
5282 * The following function updates them in a safe manner without read side
5285 * Any new users of pcp->batch and pcp->high should ensure they can cope with
5286 * those fields changing asynchronously (acording the the above rule).
5288 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
5289 * outside of boot time (or some other assurance that no concurrent updaters
5292 static void pageset_update(struct per_cpu_pages
*pcp
, unsigned long high
,
5293 unsigned long batch
)
5295 /* start with a fail safe value for batch */
5299 /* Update high, then batch, in order */
5306 /* a companion to pageset_set_high() */
5307 static void pageset_set_batch(struct per_cpu_pageset
*p
, unsigned long batch
)
5309 pageset_update(&p
->pcp
, 6 * batch
, max(1UL, 1 * batch
));
5312 static void pageset_init(struct per_cpu_pageset
*p
)
5314 struct per_cpu_pages
*pcp
;
5317 memset(p
, 0, sizeof(*p
));
5321 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
5322 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
5325 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
5328 pageset_set_batch(p
, batch
);
5332 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
5333 * to the value high for the pageset p.
5335 static void pageset_set_high(struct per_cpu_pageset
*p
,
5338 unsigned long batch
= max(1UL, high
/ 4);
5339 if ((high
/ 4) > (PAGE_SHIFT
* 8))
5340 batch
= PAGE_SHIFT
* 8;
5342 pageset_update(&p
->pcp
, high
, batch
);
5345 static void pageset_set_high_and_batch(struct zone
*zone
,
5346 struct per_cpu_pageset
*pcp
)
5348 if (percpu_pagelist_fraction
)
5349 pageset_set_high(pcp
,
5350 (zone
->managed_pages
/
5351 percpu_pagelist_fraction
));
5353 pageset_set_batch(pcp
, zone_batchsize(zone
));
5356 static void __meminit
zone_pageset_init(struct zone
*zone
, int cpu
)
5358 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
5361 pageset_set_high_and_batch(zone
, pcp
);
5364 static void __meminit
setup_zone_pageset(struct zone
*zone
)
5367 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
5368 for_each_possible_cpu(cpu
)
5369 zone_pageset_init(zone
, cpu
);
5373 * Allocate per cpu pagesets and initialize them.
5374 * Before this call only boot pagesets were available.
5376 void __init
setup_per_cpu_pageset(void)
5380 for_each_populated_zone(zone
)
5381 setup_zone_pageset(zone
);
5384 static noinline __init_refok
5385 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
5391 * The per-page waitqueue mechanism uses hashed waitqueues
5394 zone
->wait_table_hash_nr_entries
=
5395 wait_table_hash_nr_entries(zone_size_pages
);
5396 zone
->wait_table_bits
=
5397 wait_table_bits(zone
->wait_table_hash_nr_entries
);
5398 alloc_size
= zone
->wait_table_hash_nr_entries
5399 * sizeof(wait_queue_head_t
);
5401 if (!slab_is_available()) {
5402 zone
->wait_table
= (wait_queue_head_t
*)
5403 memblock_virt_alloc_node_nopanic(
5404 alloc_size
, zone
->zone_pgdat
->node_id
);
5407 * This case means that a zone whose size was 0 gets new memory
5408 * via memory hot-add.
5409 * But it may be the case that a new node was hot-added. In
5410 * this case vmalloc() will not be able to use this new node's
5411 * memory - this wait_table must be initialized to use this new
5412 * node itself as well.
5413 * To use this new node's memory, further consideration will be
5416 zone
->wait_table
= vmalloc(alloc_size
);
5418 if (!zone
->wait_table
)
5421 for (i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
5422 init_waitqueue_head(zone
->wait_table
+ i
);
5427 static __meminit
void zone_pcp_init(struct zone
*zone
)
5430 * per cpu subsystem is not up at this point. The following code
5431 * relies on the ability of the linker to provide the
5432 * offset of a (static) per cpu variable into the per cpu area.
5434 zone
->pageset
= &boot_pageset
;
5436 if (populated_zone(zone
))
5437 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
5438 zone
->name
, zone
->present_pages
,
5439 zone_batchsize(zone
));
5442 int __meminit
init_currently_empty_zone(struct zone
*zone
,
5443 unsigned long zone_start_pfn
,
5446 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
5448 ret
= zone_wait_table_init(zone
, size
);
5451 pgdat
->nr_zones
= zone_idx(zone
) + 1;
5453 zone
->zone_start_pfn
= zone_start_pfn
;
5455 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
5456 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
5458 (unsigned long)zone_idx(zone
),
5459 zone_start_pfn
, (zone_start_pfn
+ size
));
5461 zone_init_free_lists(zone
);
5466 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5467 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
5470 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
5472 int __meminit
__early_pfn_to_nid(unsigned long pfn
,
5473 struct mminit_pfnnid_cache
*state
)
5475 unsigned long start_pfn
, end_pfn
;
5478 if (state
->last_start
<= pfn
&& pfn
< state
->last_end
)
5479 return state
->last_nid
;
5481 nid
= memblock_search_pfn_nid(pfn
, &start_pfn
, &end_pfn
);
5483 state
->last_start
= start_pfn
;
5484 state
->last_end
= end_pfn
;
5485 state
->last_nid
= nid
;
5490 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
5493 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
5494 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
5495 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
5497 * If an architecture guarantees that all ranges registered contain no holes
5498 * and may be freed, this this function may be used instead of calling
5499 * memblock_free_early_nid() manually.
5501 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
5503 unsigned long start_pfn
, end_pfn
;
5506 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
5507 start_pfn
= min(start_pfn
, max_low_pfn
);
5508 end_pfn
= min(end_pfn
, max_low_pfn
);
5510 if (start_pfn
< end_pfn
)
5511 memblock_free_early_nid(PFN_PHYS(start_pfn
),
5512 (end_pfn
- start_pfn
) << PAGE_SHIFT
,
5518 * sparse_memory_present_with_active_regions - Call memory_present for each active range
5519 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
5521 * If an architecture guarantees that all ranges registered contain no holes and may
5522 * be freed, this function may be used instead of calling memory_present() manually.
5524 void __init
sparse_memory_present_with_active_regions(int nid
)
5526 unsigned long start_pfn
, end_pfn
;
5529 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
5530 memory_present(this_nid
, start_pfn
, end_pfn
);
5534 * get_pfn_range_for_nid - Return the start and end page frames for a node
5535 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
5536 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
5537 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
5539 * It returns the start and end page frame of a node based on information
5540 * provided by memblock_set_node(). If called for a node
5541 * with no available memory, a warning is printed and the start and end
5544 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
5545 unsigned long *start_pfn
, unsigned long *end_pfn
)
5547 unsigned long this_start_pfn
, this_end_pfn
;
5553 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
5554 *start_pfn
= min(*start_pfn
, this_start_pfn
);
5555 *end_pfn
= max(*end_pfn
, this_end_pfn
);
5558 if (*start_pfn
== -1UL)
5563 * This finds a zone that can be used for ZONE_MOVABLE pages. The
5564 * assumption is made that zones within a node are ordered in monotonic
5565 * increasing memory addresses so that the "highest" populated zone is used
5567 static void __init
find_usable_zone_for_movable(void)
5570 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
5571 if (zone_index
== ZONE_MOVABLE
)
5574 if (arch_zone_highest_possible_pfn
[zone_index
] >
5575 arch_zone_lowest_possible_pfn
[zone_index
])
5579 VM_BUG_ON(zone_index
== -1);
5580 movable_zone
= zone_index
;
5584 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
5585 * because it is sized independent of architecture. Unlike the other zones,
5586 * the starting point for ZONE_MOVABLE is not fixed. It may be different
5587 * in each node depending on the size of each node and how evenly kernelcore
5588 * is distributed. This helper function adjusts the zone ranges
5589 * provided by the architecture for a given node by using the end of the
5590 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5591 * zones within a node are in order of monotonic increases memory addresses
5593 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
5594 unsigned long zone_type
,
5595 unsigned long node_start_pfn
,
5596 unsigned long node_end_pfn
,
5597 unsigned long *zone_start_pfn
,
5598 unsigned long *zone_end_pfn
)
5600 /* Only adjust if ZONE_MOVABLE is on this node */
5601 if (zone_movable_pfn
[nid
]) {
5602 /* Size ZONE_MOVABLE */
5603 if (zone_type
== ZONE_MOVABLE
) {
5604 *zone_start_pfn
= zone_movable_pfn
[nid
];
5605 *zone_end_pfn
= min(node_end_pfn
,
5606 arch_zone_highest_possible_pfn
[movable_zone
]);
5608 /* Check if this whole range is within ZONE_MOVABLE */
5609 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
5610 *zone_start_pfn
= *zone_end_pfn
;
5615 * Return the number of pages a zone spans in a node, including holes
5616 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5618 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5619 unsigned long zone_type
,
5620 unsigned long node_start_pfn
,
5621 unsigned long node_end_pfn
,
5622 unsigned long *zone_start_pfn
,
5623 unsigned long *zone_end_pfn
,
5624 unsigned long *ignored
)
5626 /* When hotadd a new node from cpu_up(), the node should be empty */
5627 if (!node_start_pfn
&& !node_end_pfn
)
5630 /* Get the start and end of the zone */
5631 *zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
5632 *zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
5633 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5634 node_start_pfn
, node_end_pfn
,
5635 zone_start_pfn
, zone_end_pfn
);
5637 /* Check that this node has pages within the zone's required range */
5638 if (*zone_end_pfn
< node_start_pfn
|| *zone_start_pfn
> node_end_pfn
)
5641 /* Move the zone boundaries inside the node if necessary */
5642 *zone_end_pfn
= min(*zone_end_pfn
, node_end_pfn
);
5643 *zone_start_pfn
= max(*zone_start_pfn
, node_start_pfn
);
5645 /* Return the spanned pages */
5646 return *zone_end_pfn
- *zone_start_pfn
;
5650 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
5651 * then all holes in the requested range will be accounted for.
5653 unsigned long __meminit
__absent_pages_in_range(int nid
,
5654 unsigned long range_start_pfn
,
5655 unsigned long range_end_pfn
)
5657 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
5658 unsigned long start_pfn
, end_pfn
;
5661 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5662 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
5663 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
5664 nr_absent
-= end_pfn
- start_pfn
;
5670 * absent_pages_in_range - Return number of page frames in holes within a range
5671 * @start_pfn: The start PFN to start searching for holes
5672 * @end_pfn: The end PFN to stop searching for holes
5674 * It returns the number of pages frames in memory holes within a range.
5676 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
5677 unsigned long end_pfn
)
5679 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
5682 /* Return the number of page frames in holes in a zone on a node */
5683 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5684 unsigned long zone_type
,
5685 unsigned long node_start_pfn
,
5686 unsigned long node_end_pfn
,
5687 unsigned long *ignored
)
5689 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
5690 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
5691 unsigned long zone_start_pfn
, zone_end_pfn
;
5692 unsigned long nr_absent
;
5694 /* When hotadd a new node from cpu_up(), the node should be empty */
5695 if (!node_start_pfn
&& !node_end_pfn
)
5698 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
5699 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
5701 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5702 node_start_pfn
, node_end_pfn
,
5703 &zone_start_pfn
, &zone_end_pfn
);
5704 nr_absent
= __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
5707 * ZONE_MOVABLE handling.
5708 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
5711 if (zone_movable_pfn
[nid
]) {
5712 if (mirrored_kernelcore
) {
5713 unsigned long start_pfn
, end_pfn
;
5714 struct memblock_region
*r
;
5716 for_each_memblock(memory
, r
) {
5717 start_pfn
= clamp(memblock_region_memory_base_pfn(r
),
5718 zone_start_pfn
, zone_end_pfn
);
5719 end_pfn
= clamp(memblock_region_memory_end_pfn(r
),
5720 zone_start_pfn
, zone_end_pfn
);
5722 if (zone_type
== ZONE_MOVABLE
&&
5723 memblock_is_mirror(r
))
5724 nr_absent
+= end_pfn
- start_pfn
;
5726 if (zone_type
== ZONE_NORMAL
&&
5727 !memblock_is_mirror(r
))
5728 nr_absent
+= end_pfn
- start_pfn
;
5731 if (zone_type
== ZONE_NORMAL
)
5732 nr_absent
+= node_end_pfn
- zone_movable_pfn
[nid
];
5739 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5740 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5741 unsigned long zone_type
,
5742 unsigned long node_start_pfn
,
5743 unsigned long node_end_pfn
,
5744 unsigned long *zone_start_pfn
,
5745 unsigned long *zone_end_pfn
,
5746 unsigned long *zones_size
)
5750 *zone_start_pfn
= node_start_pfn
;
5751 for (zone
= 0; zone
< zone_type
; zone
++)
5752 *zone_start_pfn
+= zones_size
[zone
];
5754 *zone_end_pfn
= *zone_start_pfn
+ zones_size
[zone_type
];
5756 return zones_size
[zone_type
];
5759 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5760 unsigned long zone_type
,
5761 unsigned long node_start_pfn
,
5762 unsigned long node_end_pfn
,
5763 unsigned long *zholes_size
)
5768 return zholes_size
[zone_type
];
5771 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5773 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
5774 unsigned long node_start_pfn
,
5775 unsigned long node_end_pfn
,
5776 unsigned long *zones_size
,
5777 unsigned long *zholes_size
)
5779 unsigned long realtotalpages
= 0, totalpages
= 0;
5782 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5783 struct zone
*zone
= pgdat
->node_zones
+ i
;
5784 unsigned long zone_start_pfn
, zone_end_pfn
;
5785 unsigned long size
, real_size
;
5787 size
= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
5793 real_size
= size
- zone_absent_pages_in_node(pgdat
->node_id
, i
,
5794 node_start_pfn
, node_end_pfn
,
5797 zone
->zone_start_pfn
= zone_start_pfn
;
5799 zone
->zone_start_pfn
= 0;
5800 zone
->spanned_pages
= size
;
5801 zone
->present_pages
= real_size
;
5804 realtotalpages
+= real_size
;
5807 pgdat
->node_spanned_pages
= totalpages
;
5808 pgdat
->node_present_pages
= realtotalpages
;
5809 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
5813 #ifndef CONFIG_SPARSEMEM
5815 * Calculate the size of the zone->blockflags rounded to an unsigned long
5816 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5817 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
5818 * round what is now in bits to nearest long in bits, then return it in
5821 static unsigned long __init
usemap_size(unsigned long zone_start_pfn
, unsigned long zonesize
)
5823 unsigned long usemapsize
;
5825 zonesize
+= zone_start_pfn
& (pageblock_nr_pages
-1);
5826 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
5827 usemapsize
= usemapsize
>> pageblock_order
;
5828 usemapsize
*= NR_PAGEBLOCK_BITS
;
5829 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
5831 return usemapsize
/ 8;
5834 static void __init
setup_usemap(struct pglist_data
*pgdat
,
5836 unsigned long zone_start_pfn
,
5837 unsigned long zonesize
)
5839 unsigned long usemapsize
= usemap_size(zone_start_pfn
, zonesize
);
5840 zone
->pageblock_flags
= NULL
;
5842 zone
->pageblock_flags
=
5843 memblock_virt_alloc_node_nopanic(usemapsize
,
5847 static inline void setup_usemap(struct pglist_data
*pgdat
, struct zone
*zone
,
5848 unsigned long zone_start_pfn
, unsigned long zonesize
) {}
5849 #endif /* CONFIG_SPARSEMEM */
5851 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
5853 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
5854 void __paginginit
set_pageblock_order(void)
5858 /* Check that pageblock_nr_pages has not already been setup */
5859 if (pageblock_order
)
5862 if (HPAGE_SHIFT
> PAGE_SHIFT
)
5863 order
= HUGETLB_PAGE_ORDER
;
5865 order
= MAX_ORDER
- 1;
5868 * Assume the largest contiguous order of interest is a huge page.
5869 * This value may be variable depending on boot parameters on IA64 and
5872 pageblock_order
= order
;
5874 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5877 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
5878 * is unused as pageblock_order is set at compile-time. See
5879 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5882 void __paginginit
set_pageblock_order(void)
5886 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5888 static unsigned long __paginginit
calc_memmap_size(unsigned long spanned_pages
,
5889 unsigned long present_pages
)
5891 unsigned long pages
= spanned_pages
;
5894 * Provide a more accurate estimation if there are holes within
5895 * the zone and SPARSEMEM is in use. If there are holes within the
5896 * zone, each populated memory region may cost us one or two extra
5897 * memmap pages due to alignment because memmap pages for each
5898 * populated regions may not naturally algined on page boundary.
5899 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5901 if (spanned_pages
> present_pages
+ (present_pages
>> 4) &&
5902 IS_ENABLED(CONFIG_SPARSEMEM
))
5903 pages
= present_pages
;
5905 return PAGE_ALIGN(pages
* sizeof(struct page
)) >> PAGE_SHIFT
;
5909 * Set up the zone data structures:
5910 * - mark all pages reserved
5911 * - mark all memory queues empty
5912 * - clear the memory bitmaps
5914 * NOTE: pgdat should get zeroed by caller.
5916 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
)
5919 int nid
= pgdat
->node_id
;
5922 pgdat_resize_init(pgdat
);
5923 #ifdef CONFIG_NUMA_BALANCING
5924 spin_lock_init(&pgdat
->numabalancing_migrate_lock
);
5925 pgdat
->numabalancing_migrate_nr_pages
= 0;
5926 pgdat
->numabalancing_migrate_next_window
= jiffies
;
5928 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5929 spin_lock_init(&pgdat
->split_queue_lock
);
5930 INIT_LIST_HEAD(&pgdat
->split_queue
);
5931 pgdat
->split_queue_len
= 0;
5933 init_waitqueue_head(&pgdat
->kswapd_wait
);
5934 init_waitqueue_head(&pgdat
->pfmemalloc_wait
);
5935 #ifdef CONFIG_COMPACTION
5936 init_waitqueue_head(&pgdat
->kcompactd_wait
);
5938 pgdat_page_ext_init(pgdat
);
5940 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5941 struct zone
*zone
= pgdat
->node_zones
+ j
;
5942 unsigned long size
, realsize
, freesize
, memmap_pages
;
5943 unsigned long zone_start_pfn
= zone
->zone_start_pfn
;
5945 size
= zone
->spanned_pages
;
5946 realsize
= freesize
= zone
->present_pages
;
5949 * Adjust freesize so that it accounts for how much memory
5950 * is used by this zone for memmap. This affects the watermark
5951 * and per-cpu initialisations
5953 memmap_pages
= calc_memmap_size(size
, realsize
);
5954 if (!is_highmem_idx(j
)) {
5955 if (freesize
>= memmap_pages
) {
5956 freesize
-= memmap_pages
;
5959 " %s zone: %lu pages used for memmap\n",
5960 zone_names
[j
], memmap_pages
);
5962 pr_warn(" %s zone: %lu pages exceeds freesize %lu\n",
5963 zone_names
[j
], memmap_pages
, freesize
);
5966 /* Account for reserved pages */
5967 if (j
== 0 && freesize
> dma_reserve
) {
5968 freesize
-= dma_reserve
;
5969 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
5970 zone_names
[0], dma_reserve
);
5973 if (!is_highmem_idx(j
))
5974 nr_kernel_pages
+= freesize
;
5975 /* Charge for highmem memmap if there are enough kernel pages */
5976 else if (nr_kernel_pages
> memmap_pages
* 2)
5977 nr_kernel_pages
-= memmap_pages
;
5978 nr_all_pages
+= freesize
;
5981 * Set an approximate value for lowmem here, it will be adjusted
5982 * when the bootmem allocator frees pages into the buddy system.
5983 * And all highmem pages will be managed by the buddy system.
5985 zone
->managed_pages
= is_highmem_idx(j
) ? realsize
: freesize
;
5988 zone
->min_unmapped_pages
= (freesize
*sysctl_min_unmapped_ratio
)
5990 zone
->min_slab_pages
= (freesize
* sysctl_min_slab_ratio
) / 100;
5992 zone
->name
= zone_names
[j
];
5993 spin_lock_init(&zone
->lock
);
5994 spin_lock_init(&zone
->lru_lock
);
5995 zone_seqlock_init(zone
);
5996 zone
->zone_pgdat
= pgdat
;
5997 zone_pcp_init(zone
);
5999 /* For bootup, initialized properly in watermark setup */
6000 mod_zone_page_state(zone
, NR_ALLOC_BATCH
, zone
->managed_pages
);
6002 lruvec_init(&zone
->lruvec
);
6006 set_pageblock_order();
6007 setup_usemap(pgdat
, zone
, zone_start_pfn
, size
);
6008 ret
= init_currently_empty_zone(zone
, zone_start_pfn
, size
);
6010 memmap_init(size
, nid
, j
, zone_start_pfn
);
6014 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
6016 unsigned long __maybe_unused start
= 0;
6017 unsigned long __maybe_unused offset
= 0;
6019 /* Skip empty nodes */
6020 if (!pgdat
->node_spanned_pages
)
6023 #ifdef CONFIG_FLAT_NODE_MEM_MAP
6024 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
6025 offset
= pgdat
->node_start_pfn
- start
;
6026 /* ia64 gets its own node_mem_map, before this, without bootmem */
6027 if (!pgdat
->node_mem_map
) {
6028 unsigned long size
, end
;
6032 * The zone's endpoints aren't required to be MAX_ORDER
6033 * aligned but the node_mem_map endpoints must be in order
6034 * for the buddy allocator to function correctly.
6036 end
= pgdat_end_pfn(pgdat
);
6037 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
6038 size
= (end
- start
) * sizeof(struct page
);
6039 map
= alloc_remap(pgdat
->node_id
, size
);
6041 map
= memblock_virt_alloc_node_nopanic(size
,
6043 pgdat
->node_mem_map
= map
+ offset
;
6045 #ifndef CONFIG_NEED_MULTIPLE_NODES
6047 * With no DISCONTIG, the global mem_map is just set as node 0's
6049 if (pgdat
== NODE_DATA(0)) {
6050 mem_map
= NODE_DATA(0)->node_mem_map
;
6051 #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
6052 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
6054 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6057 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
6060 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
6061 unsigned long node_start_pfn
, unsigned long *zholes_size
)
6063 pg_data_t
*pgdat
= NODE_DATA(nid
);
6064 unsigned long start_pfn
= 0;
6065 unsigned long end_pfn
= 0;
6067 /* pg_data_t should be reset to zero when it's allocated */
6068 WARN_ON(pgdat
->nr_zones
|| pgdat
->classzone_idx
);
6070 reset_deferred_meminit(pgdat
);
6071 pgdat
->node_id
= nid
;
6072 pgdat
->node_start_pfn
= node_start_pfn
;
6073 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6074 get_pfn_range_for_nid(nid
, &start_pfn
, &end_pfn
);
6075 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid
,
6076 (u64
)start_pfn
<< PAGE_SHIFT
,
6077 end_pfn
? ((u64
)end_pfn
<< PAGE_SHIFT
) - 1 : 0);
6079 start_pfn
= node_start_pfn
;
6081 calculate_node_totalpages(pgdat
, start_pfn
, end_pfn
,
6082 zones_size
, zholes_size
);
6084 alloc_node_mem_map(pgdat
);
6085 #ifdef CONFIG_FLAT_NODE_MEM_MAP
6086 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
6087 nid
, (unsigned long)pgdat
,
6088 (unsigned long)pgdat
->node_mem_map
);
6091 free_area_init_core(pgdat
);
6094 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6096 #if MAX_NUMNODES > 1
6098 * Figure out the number of possible node ids.
6100 void __init
setup_nr_node_ids(void)
6102 unsigned int highest
;
6104 highest
= find_last_bit(node_possible_map
.bits
, MAX_NUMNODES
);
6105 nr_node_ids
= highest
+ 1;
6110 * node_map_pfn_alignment - determine the maximum internode alignment
6112 * This function should be called after node map is populated and sorted.
6113 * It calculates the maximum power of two alignment which can distinguish
6116 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
6117 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
6118 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
6119 * shifted, 1GiB is enough and this function will indicate so.
6121 * This is used to test whether pfn -> nid mapping of the chosen memory
6122 * model has fine enough granularity to avoid incorrect mapping for the
6123 * populated node map.
6125 * Returns the determined alignment in pfn's. 0 if there is no alignment
6126 * requirement (single node).
6128 unsigned long __init
node_map_pfn_alignment(void)
6130 unsigned long accl_mask
= 0, last_end
= 0;
6131 unsigned long start
, end
, mask
;
6135 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
6136 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
6143 * Start with a mask granular enough to pin-point to the
6144 * start pfn and tick off bits one-by-one until it becomes
6145 * too coarse to separate the current node from the last.
6147 mask
= ~((1 << __ffs(start
)) - 1);
6148 while (mask
&& last_end
<= (start
& (mask
<< 1)))
6151 /* accumulate all internode masks */
6155 /* convert mask to number of pages */
6156 return ~accl_mask
+ 1;
6159 /* Find the lowest pfn for a node */
6160 static unsigned long __init
find_min_pfn_for_node(int nid
)
6162 unsigned long min_pfn
= ULONG_MAX
;
6163 unsigned long start_pfn
;
6166 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
6167 min_pfn
= min(min_pfn
, start_pfn
);
6169 if (min_pfn
== ULONG_MAX
) {
6170 pr_warn("Could not find start_pfn for node %d\n", nid
);
6178 * find_min_pfn_with_active_regions - Find the minimum PFN registered
6180 * It returns the minimum PFN based on information provided via
6181 * memblock_set_node().
6183 unsigned long __init
find_min_pfn_with_active_regions(void)
6185 return find_min_pfn_for_node(MAX_NUMNODES
);
6189 * early_calculate_totalpages()
6190 * Sum pages in active regions for movable zone.
6191 * Populate N_MEMORY for calculating usable_nodes.
6193 static unsigned long __init
early_calculate_totalpages(void)
6195 unsigned long totalpages
= 0;
6196 unsigned long start_pfn
, end_pfn
;
6199 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
6200 unsigned long pages
= end_pfn
- start_pfn
;
6202 totalpages
+= pages
;
6204 node_set_state(nid
, N_MEMORY
);
6210 * Find the PFN the Movable zone begins in each node. Kernel memory
6211 * is spread evenly between nodes as long as the nodes have enough
6212 * memory. When they don't, some nodes will have more kernelcore than
6215 static void __init
find_zone_movable_pfns_for_nodes(void)
6218 unsigned long usable_startpfn
;
6219 unsigned long kernelcore_node
, kernelcore_remaining
;
6220 /* save the state before borrow the nodemask */
6221 nodemask_t saved_node_state
= node_states
[N_MEMORY
];
6222 unsigned long totalpages
= early_calculate_totalpages();
6223 int usable_nodes
= nodes_weight(node_states
[N_MEMORY
]);
6224 struct memblock_region
*r
;
6226 /* Need to find movable_zone earlier when movable_node is specified. */
6227 find_usable_zone_for_movable();
6230 * If movable_node is specified, ignore kernelcore and movablecore
6233 if (movable_node_is_enabled()) {
6234 for_each_memblock(memory
, r
) {
6235 if (!memblock_is_hotpluggable(r
))
6240 usable_startpfn
= PFN_DOWN(r
->base
);
6241 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
6242 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
6250 * If kernelcore=mirror is specified, ignore movablecore option
6252 if (mirrored_kernelcore
) {
6253 bool mem_below_4gb_not_mirrored
= false;
6255 for_each_memblock(memory
, r
) {
6256 if (memblock_is_mirror(r
))
6261 usable_startpfn
= memblock_region_memory_base_pfn(r
);
6263 if (usable_startpfn
< 0x100000) {
6264 mem_below_4gb_not_mirrored
= true;
6268 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
6269 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
6273 if (mem_below_4gb_not_mirrored
)
6274 pr_warn("This configuration results in unmirrored kernel memory.");
6280 * If movablecore=nn[KMG] was specified, calculate what size of
6281 * kernelcore that corresponds so that memory usable for
6282 * any allocation type is evenly spread. If both kernelcore
6283 * and movablecore are specified, then the value of kernelcore
6284 * will be used for required_kernelcore if it's greater than
6285 * what movablecore would have allowed.
6287 if (required_movablecore
) {
6288 unsigned long corepages
;
6291 * Round-up so that ZONE_MOVABLE is at least as large as what
6292 * was requested by the user
6294 required_movablecore
=
6295 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
6296 required_movablecore
= min(totalpages
, required_movablecore
);
6297 corepages
= totalpages
- required_movablecore
;
6299 required_kernelcore
= max(required_kernelcore
, corepages
);
6303 * If kernelcore was not specified or kernelcore size is larger
6304 * than totalpages, there is no ZONE_MOVABLE.
6306 if (!required_kernelcore
|| required_kernelcore
>= totalpages
)
6309 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
6310 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
6313 /* Spread kernelcore memory as evenly as possible throughout nodes */
6314 kernelcore_node
= required_kernelcore
/ usable_nodes
;
6315 for_each_node_state(nid
, N_MEMORY
) {
6316 unsigned long start_pfn
, end_pfn
;
6319 * Recalculate kernelcore_node if the division per node
6320 * now exceeds what is necessary to satisfy the requested
6321 * amount of memory for the kernel
6323 if (required_kernelcore
< kernelcore_node
)
6324 kernelcore_node
= required_kernelcore
/ usable_nodes
;
6327 * As the map is walked, we track how much memory is usable
6328 * by the kernel using kernelcore_remaining. When it is
6329 * 0, the rest of the node is usable by ZONE_MOVABLE
6331 kernelcore_remaining
= kernelcore_node
;
6333 /* Go through each range of PFNs within this node */
6334 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
6335 unsigned long size_pages
;
6337 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
6338 if (start_pfn
>= end_pfn
)
6341 /* Account for what is only usable for kernelcore */
6342 if (start_pfn
< usable_startpfn
) {
6343 unsigned long kernel_pages
;
6344 kernel_pages
= min(end_pfn
, usable_startpfn
)
6347 kernelcore_remaining
-= min(kernel_pages
,
6348 kernelcore_remaining
);
6349 required_kernelcore
-= min(kernel_pages
,
6350 required_kernelcore
);
6352 /* Continue if range is now fully accounted */
6353 if (end_pfn
<= usable_startpfn
) {
6356 * Push zone_movable_pfn to the end so
6357 * that if we have to rebalance
6358 * kernelcore across nodes, we will
6359 * not double account here
6361 zone_movable_pfn
[nid
] = end_pfn
;
6364 start_pfn
= usable_startpfn
;
6368 * The usable PFN range for ZONE_MOVABLE is from
6369 * start_pfn->end_pfn. Calculate size_pages as the
6370 * number of pages used as kernelcore
6372 size_pages
= end_pfn
- start_pfn
;
6373 if (size_pages
> kernelcore_remaining
)
6374 size_pages
= kernelcore_remaining
;
6375 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
6378 * Some kernelcore has been met, update counts and
6379 * break if the kernelcore for this node has been
6382 required_kernelcore
-= min(required_kernelcore
,
6384 kernelcore_remaining
-= size_pages
;
6385 if (!kernelcore_remaining
)
6391 * If there is still required_kernelcore, we do another pass with one
6392 * less node in the count. This will push zone_movable_pfn[nid] further
6393 * along on the nodes that still have memory until kernelcore is
6397 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
6401 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
6402 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
6403 zone_movable_pfn
[nid
] =
6404 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
6407 /* restore the node_state */
6408 node_states
[N_MEMORY
] = saved_node_state
;
6411 /* Any regular or high memory on that node ? */
6412 static void check_for_memory(pg_data_t
*pgdat
, int nid
)
6414 enum zone_type zone_type
;
6416 if (N_MEMORY
== N_NORMAL_MEMORY
)
6419 for (zone_type
= 0; zone_type
<= ZONE_MOVABLE
- 1; zone_type
++) {
6420 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
6421 if (populated_zone(zone
)) {
6422 node_set_state(nid
, N_HIGH_MEMORY
);
6423 if (N_NORMAL_MEMORY
!= N_HIGH_MEMORY
&&
6424 zone_type
<= ZONE_NORMAL
)
6425 node_set_state(nid
, N_NORMAL_MEMORY
);
6432 * free_area_init_nodes - Initialise all pg_data_t and zone data
6433 * @max_zone_pfn: an array of max PFNs for each zone
6435 * This will call free_area_init_node() for each active node in the system.
6436 * Using the page ranges provided by memblock_set_node(), the size of each
6437 * zone in each node and their holes is calculated. If the maximum PFN
6438 * between two adjacent zones match, it is assumed that the zone is empty.
6439 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
6440 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
6441 * starts where the previous one ended. For example, ZONE_DMA32 starts
6442 * at arch_max_dma_pfn.
6444 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
6446 unsigned long start_pfn
, end_pfn
;
6449 /* Record where the zone boundaries are */
6450 memset(arch_zone_lowest_possible_pfn
, 0,
6451 sizeof(arch_zone_lowest_possible_pfn
));
6452 memset(arch_zone_highest_possible_pfn
, 0,
6453 sizeof(arch_zone_highest_possible_pfn
));
6454 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
6455 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
6456 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
6457 if (i
== ZONE_MOVABLE
)
6459 arch_zone_lowest_possible_pfn
[i
] =
6460 arch_zone_highest_possible_pfn
[i
-1];
6461 arch_zone_highest_possible_pfn
[i
] =
6462 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
6464 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
6465 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
6467 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
6468 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
6469 find_zone_movable_pfns_for_nodes();
6471 /* Print out the zone ranges */
6472 pr_info("Zone ranges:\n");
6473 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
6474 if (i
== ZONE_MOVABLE
)
6476 pr_info(" %-8s ", zone_names
[i
]);
6477 if (arch_zone_lowest_possible_pfn
[i
] ==
6478 arch_zone_highest_possible_pfn
[i
])
6481 pr_cont("[mem %#018Lx-%#018Lx]\n",
6482 (u64
)arch_zone_lowest_possible_pfn
[i
]
6484 ((u64
)arch_zone_highest_possible_pfn
[i
]
6485 << PAGE_SHIFT
) - 1);
6488 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
6489 pr_info("Movable zone start for each node\n");
6490 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
6491 if (zone_movable_pfn
[i
])
6492 pr_info(" Node %d: %#018Lx\n", i
,
6493 (u64
)zone_movable_pfn
[i
] << PAGE_SHIFT
);
6496 /* Print out the early node map */
6497 pr_info("Early memory node ranges\n");
6498 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
6499 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid
,
6500 (u64
)start_pfn
<< PAGE_SHIFT
,
6501 ((u64
)end_pfn
<< PAGE_SHIFT
) - 1);
6503 /* Initialise every node */
6504 mminit_verify_pageflags_layout();
6505 setup_nr_node_ids();
6506 for_each_online_node(nid
) {
6507 pg_data_t
*pgdat
= NODE_DATA(nid
);
6508 free_area_init_node(nid
, NULL
,
6509 find_min_pfn_for_node(nid
), NULL
);
6511 /* Any memory on that node */
6512 if (pgdat
->node_present_pages
)
6513 node_set_state(nid
, N_MEMORY
);
6514 check_for_memory(pgdat
, nid
);
6518 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
6520 unsigned long long coremem
;
6524 coremem
= memparse(p
, &p
);
6525 *core
= coremem
>> PAGE_SHIFT
;
6527 /* Paranoid check that UL is enough for the coremem value */
6528 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
6534 * kernelcore=size sets the amount of memory for use for allocations that
6535 * cannot be reclaimed or migrated.
6537 static int __init
cmdline_parse_kernelcore(char *p
)
6539 /* parse kernelcore=mirror */
6540 if (parse_option_str(p
, "mirror")) {
6541 mirrored_kernelcore
= true;
6545 return cmdline_parse_core(p
, &required_kernelcore
);
6549 * movablecore=size sets the amount of memory for use for allocations that
6550 * can be reclaimed or migrated.
6552 static int __init
cmdline_parse_movablecore(char *p
)
6554 return cmdline_parse_core(p
, &required_movablecore
);
6557 early_param("kernelcore", cmdline_parse_kernelcore
);
6558 early_param("movablecore", cmdline_parse_movablecore
);
6560 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6562 void adjust_managed_page_count(struct page
*page
, long count
)
6564 spin_lock(&managed_page_count_lock
);
6565 page_zone(page
)->managed_pages
+= count
;
6566 totalram_pages
+= count
;
6567 #ifdef CONFIG_HIGHMEM
6568 if (PageHighMem(page
))
6569 totalhigh_pages
+= count
;
6571 spin_unlock(&managed_page_count_lock
);
6573 EXPORT_SYMBOL(adjust_managed_page_count
);
6575 unsigned long free_reserved_area(void *start
, void *end
, int poison
, char *s
)
6578 unsigned long pages
= 0;
6580 start
= (void *)PAGE_ALIGN((unsigned long)start
);
6581 end
= (void *)((unsigned long)end
& PAGE_MASK
);
6582 for (pos
= start
; pos
< end
; pos
+= PAGE_SIZE
, pages
++) {
6583 if ((unsigned int)poison
<= 0xFF)
6584 memset(pos
, poison
, PAGE_SIZE
);
6585 free_reserved_page(virt_to_page(pos
));
6589 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
6590 s
, pages
<< (PAGE_SHIFT
- 10), start
, end
);
6594 EXPORT_SYMBOL(free_reserved_area
);
6596 #ifdef CONFIG_HIGHMEM
6597 void free_highmem_page(struct page
*page
)
6599 __free_reserved_page(page
);
6601 page_zone(page
)->managed_pages
++;
6607 void __init
mem_init_print_info(const char *str
)
6609 unsigned long physpages
, codesize
, datasize
, rosize
, bss_size
;
6610 unsigned long init_code_size
, init_data_size
;
6612 physpages
= get_num_physpages();
6613 codesize
= _etext
- _stext
;
6614 datasize
= _edata
- _sdata
;
6615 rosize
= __end_rodata
- __start_rodata
;
6616 bss_size
= __bss_stop
- __bss_start
;
6617 init_data_size
= __init_end
- __init_begin
;
6618 init_code_size
= _einittext
- _sinittext
;
6621 * Detect special cases and adjust section sizes accordingly:
6622 * 1) .init.* may be embedded into .data sections
6623 * 2) .init.text.* may be out of [__init_begin, __init_end],
6624 * please refer to arch/tile/kernel/vmlinux.lds.S.
6625 * 3) .rodata.* may be embedded into .text or .data sections.
6627 #define adj_init_size(start, end, size, pos, adj) \
6629 if (start <= pos && pos < end && size > adj) \
6633 adj_init_size(__init_begin
, __init_end
, init_data_size
,
6634 _sinittext
, init_code_size
);
6635 adj_init_size(_stext
, _etext
, codesize
, _sinittext
, init_code_size
);
6636 adj_init_size(_sdata
, _edata
, datasize
, __init_begin
, init_data_size
);
6637 adj_init_size(_stext
, _etext
, codesize
, __start_rodata
, rosize
);
6638 adj_init_size(_sdata
, _edata
, datasize
, __start_rodata
, rosize
);
6640 #undef adj_init_size
6642 pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
6643 #ifdef CONFIG_HIGHMEM
6647 nr_free_pages() << (PAGE_SHIFT
- 10),
6648 physpages
<< (PAGE_SHIFT
- 10),
6649 codesize
>> 10, datasize
>> 10, rosize
>> 10,
6650 (init_data_size
+ init_code_size
) >> 10, bss_size
>> 10,
6651 (physpages
- totalram_pages
- totalcma_pages
) << (PAGE_SHIFT
- 10),
6652 totalcma_pages
<< (PAGE_SHIFT
- 10),
6653 #ifdef CONFIG_HIGHMEM
6654 totalhigh_pages
<< (PAGE_SHIFT
- 10),
6656 str
? ", " : "", str
? str
: "");
6660 * set_dma_reserve - set the specified number of pages reserved in the first zone
6661 * @new_dma_reserve: The number of pages to mark reserved
6663 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
6664 * In the DMA zone, a significant percentage may be consumed by kernel image
6665 * and other unfreeable allocations which can skew the watermarks badly. This
6666 * function may optionally be used to account for unfreeable pages in the
6667 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6668 * smaller per-cpu batchsize.
6670 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
6672 dma_reserve
= new_dma_reserve
;
6675 void __init
free_area_init(unsigned long *zones_size
)
6677 free_area_init_node(0, zones_size
,
6678 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
6681 static int page_alloc_cpu_notify(struct notifier_block
*self
,
6682 unsigned long action
, void *hcpu
)
6684 int cpu
= (unsigned long)hcpu
;
6686 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
6687 lru_add_drain_cpu(cpu
);
6691 * Spill the event counters of the dead processor
6692 * into the current processors event counters.
6693 * This artificially elevates the count of the current
6696 vm_events_fold_cpu(cpu
);
6699 * Zero the differential counters of the dead processor
6700 * so that the vm statistics are consistent.
6702 * This is only okay since the processor is dead and cannot
6703 * race with what we are doing.
6705 cpu_vm_stats_fold(cpu
);
6710 void __init
page_alloc_init(void)
6712 hotcpu_notifier(page_alloc_cpu_notify
, 0);
6716 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
6717 * or min_free_kbytes changes.
6719 static void calculate_totalreserve_pages(void)
6721 struct pglist_data
*pgdat
;
6722 unsigned long reserve_pages
= 0;
6723 enum zone_type i
, j
;
6725 for_each_online_pgdat(pgdat
) {
6726 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
6727 struct zone
*zone
= pgdat
->node_zones
+ i
;
6730 /* Find valid and maximum lowmem_reserve in the zone */
6731 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
6732 if (zone
->lowmem_reserve
[j
] > max
)
6733 max
= zone
->lowmem_reserve
[j
];
6736 /* we treat the high watermark as reserved pages. */
6737 max
+= high_wmark_pages(zone
);
6739 if (max
> zone
->managed_pages
)
6740 max
= zone
->managed_pages
;
6742 zone
->totalreserve_pages
= max
;
6744 reserve_pages
+= max
;
6747 totalreserve_pages
= reserve_pages
;
6751 * setup_per_zone_lowmem_reserve - called whenever
6752 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
6753 * has a correct pages reserved value, so an adequate number of
6754 * pages are left in the zone after a successful __alloc_pages().
6756 static void setup_per_zone_lowmem_reserve(void)
6758 struct pglist_data
*pgdat
;
6759 enum zone_type j
, idx
;
6761 for_each_online_pgdat(pgdat
) {
6762 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
6763 struct zone
*zone
= pgdat
->node_zones
+ j
;
6764 unsigned long managed_pages
= zone
->managed_pages
;
6766 zone
->lowmem_reserve
[j
] = 0;
6770 struct zone
*lower_zone
;
6774 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
6775 sysctl_lowmem_reserve_ratio
[idx
] = 1;
6777 lower_zone
= pgdat
->node_zones
+ idx
;
6778 lower_zone
->lowmem_reserve
[j
] = managed_pages
/
6779 sysctl_lowmem_reserve_ratio
[idx
];
6780 managed_pages
+= lower_zone
->managed_pages
;
6785 /* update totalreserve_pages */
6786 calculate_totalreserve_pages();
6789 static void __setup_per_zone_wmarks(void)
6791 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
6792 unsigned long lowmem_pages
= 0;
6794 unsigned long flags
;
6796 /* Calculate total number of !ZONE_HIGHMEM pages */
6797 for_each_zone(zone
) {
6798 if (!is_highmem(zone
))
6799 lowmem_pages
+= zone
->managed_pages
;
6802 for_each_zone(zone
) {
6805 spin_lock_irqsave(&zone
->lock
, flags
);
6806 tmp
= (u64
)pages_min
* zone
->managed_pages
;
6807 do_div(tmp
, lowmem_pages
);
6808 if (is_highmem(zone
)) {
6810 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6811 * need highmem pages, so cap pages_min to a small
6814 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
6815 * deltas control asynch page reclaim, and so should
6816 * not be capped for highmem.
6818 unsigned long min_pages
;
6820 min_pages
= zone
->managed_pages
/ 1024;
6821 min_pages
= clamp(min_pages
, SWAP_CLUSTER_MAX
, 128UL);
6822 zone
->watermark
[WMARK_MIN
] = min_pages
;
6825 * If it's a lowmem zone, reserve a number of pages
6826 * proportionate to the zone's size.
6828 zone
->watermark
[WMARK_MIN
] = tmp
;
6832 * Set the kswapd watermarks distance according to the
6833 * scale factor in proportion to available memory, but
6834 * ensure a minimum size on small systems.
6836 tmp
= max_t(u64
, tmp
>> 2,
6837 mult_frac(zone
->managed_pages
,
6838 watermark_scale_factor
, 10000));
6840 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + tmp
;
6841 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + tmp
* 2;
6843 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
6844 high_wmark_pages(zone
) - low_wmark_pages(zone
) -
6845 atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]));
6847 spin_unlock_irqrestore(&zone
->lock
, flags
);
6850 /* update totalreserve_pages */
6851 calculate_totalreserve_pages();
6855 * setup_per_zone_wmarks - called when min_free_kbytes changes
6856 * or when memory is hot-{added|removed}
6858 * Ensures that the watermark[min,low,high] values for each zone are set
6859 * correctly with respect to min_free_kbytes.
6861 void setup_per_zone_wmarks(void)
6863 mutex_lock(&zonelists_mutex
);
6864 __setup_per_zone_wmarks();
6865 mutex_unlock(&zonelists_mutex
);
6869 * Initialise min_free_kbytes.
6871 * For small machines we want it small (128k min). For large machines
6872 * we want it large (64MB max). But it is not linear, because network
6873 * bandwidth does not increase linearly with machine size. We use
6875 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
6876 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6892 int __meminit
init_per_zone_wmark_min(void)
6894 unsigned long lowmem_kbytes
;
6895 int new_min_free_kbytes
;
6897 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
6898 new_min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
6900 if (new_min_free_kbytes
> user_min_free_kbytes
) {
6901 min_free_kbytes
= new_min_free_kbytes
;
6902 if (min_free_kbytes
< 128)
6903 min_free_kbytes
= 128;
6904 if (min_free_kbytes
> 65536)
6905 min_free_kbytes
= 65536;
6907 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6908 new_min_free_kbytes
, user_min_free_kbytes
);
6910 setup_per_zone_wmarks();
6911 refresh_zone_stat_thresholds();
6912 setup_per_zone_lowmem_reserve();
6915 core_initcall(init_per_zone_wmark_min
)
6918 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
6919 * that we can call two helper functions whenever min_free_kbytes
6922 int min_free_kbytes_sysctl_handler(struct ctl_table
*table
, int write
,
6923 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6927 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6932 user_min_free_kbytes
= min_free_kbytes
;
6933 setup_per_zone_wmarks();
6938 int watermark_scale_factor_sysctl_handler(struct ctl_table
*table
, int write
,
6939 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6943 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6948 setup_per_zone_wmarks();
6954 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6955 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6960 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6965 zone
->min_unmapped_pages
= (zone
->managed_pages
*
6966 sysctl_min_unmapped_ratio
) / 100;
6970 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6971 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6976 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6981 zone
->min_slab_pages
= (zone
->managed_pages
*
6982 sysctl_min_slab_ratio
) / 100;
6988 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6989 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6990 * whenever sysctl_lowmem_reserve_ratio changes.
6992 * The reserve ratio obviously has absolutely no relation with the
6993 * minimum watermarks. The lowmem reserve ratio can only make sense
6994 * if in function of the boot time zone sizes.
6996 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6997 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6999 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
7000 setup_per_zone_lowmem_reserve();
7005 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
7006 * cpu. It is the fraction of total pages in each zone that a hot per cpu
7007 * pagelist can have before it gets flushed back to buddy allocator.
7009 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table
*table
, int write
,
7010 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
7013 int old_percpu_pagelist_fraction
;
7016 mutex_lock(&pcp_batch_high_lock
);
7017 old_percpu_pagelist_fraction
= percpu_pagelist_fraction
;
7019 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
7020 if (!write
|| ret
< 0)
7023 /* Sanity checking to avoid pcp imbalance */
7024 if (percpu_pagelist_fraction
&&
7025 percpu_pagelist_fraction
< MIN_PERCPU_PAGELIST_FRACTION
) {
7026 percpu_pagelist_fraction
= old_percpu_pagelist_fraction
;
7032 if (percpu_pagelist_fraction
== old_percpu_pagelist_fraction
)
7035 for_each_populated_zone(zone
) {
7038 for_each_possible_cpu(cpu
)
7039 pageset_set_high_and_batch(zone
,
7040 per_cpu_ptr(zone
->pageset
, cpu
));
7043 mutex_unlock(&pcp_batch_high_lock
);
7048 int hashdist
= HASHDIST_DEFAULT
;
7050 static int __init
set_hashdist(char *str
)
7054 hashdist
= simple_strtoul(str
, &str
, 0);
7057 __setup("hashdist=", set_hashdist
);
7061 * allocate a large system hash table from bootmem
7062 * - it is assumed that the hash table must contain an exact power-of-2
7063 * quantity of entries
7064 * - limit is the number of hash buckets, not the total allocation size
7066 void *__init
alloc_large_system_hash(const char *tablename
,
7067 unsigned long bucketsize
,
7068 unsigned long numentries
,
7071 unsigned int *_hash_shift
,
7072 unsigned int *_hash_mask
,
7073 unsigned long low_limit
,
7074 unsigned long high_limit
)
7076 unsigned long long max
= high_limit
;
7077 unsigned long log2qty
, size
;
7080 /* allow the kernel cmdline to have a say */
7082 /* round applicable memory size up to nearest megabyte */
7083 numentries
= nr_kernel_pages
;
7085 /* It isn't necessary when PAGE_SIZE >= 1MB */
7086 if (PAGE_SHIFT
< 20)
7087 numentries
= round_up(numentries
, (1<<20)/PAGE_SIZE
);
7089 /* limit to 1 bucket per 2^scale bytes of low memory */
7090 if (scale
> PAGE_SHIFT
)
7091 numentries
>>= (scale
- PAGE_SHIFT
);
7093 numentries
<<= (PAGE_SHIFT
- scale
);
7095 /* Make sure we've got at least a 0-order allocation.. */
7096 if (unlikely(flags
& HASH_SMALL
)) {
7097 /* Makes no sense without HASH_EARLY */
7098 WARN_ON(!(flags
& HASH_EARLY
));
7099 if (!(numentries
>> *_hash_shift
)) {
7100 numentries
= 1UL << *_hash_shift
;
7101 BUG_ON(!numentries
);
7103 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
7104 numentries
= PAGE_SIZE
/ bucketsize
;
7106 numentries
= roundup_pow_of_two(numentries
);
7108 /* limit allocation size to 1/16 total memory by default */
7110 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
7111 do_div(max
, bucketsize
);
7113 max
= min(max
, 0x80000000ULL
);
7115 if (numentries
< low_limit
)
7116 numentries
= low_limit
;
7117 if (numentries
> max
)
7120 log2qty
= ilog2(numentries
);
7123 size
= bucketsize
<< log2qty
;
7124 if (flags
& HASH_EARLY
)
7125 table
= memblock_virt_alloc_nopanic(size
, 0);
7127 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
7130 * If bucketsize is not a power-of-two, we may free
7131 * some pages at the end of hash table which
7132 * alloc_pages_exact() automatically does
7134 if (get_order(size
) < MAX_ORDER
) {
7135 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
7136 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
7139 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
7142 panic("Failed to allocate %s hash table\n", tablename
);
7144 pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n",
7145 tablename
, 1UL << log2qty
, ilog2(size
) - PAGE_SHIFT
, size
);
7148 *_hash_shift
= log2qty
;
7150 *_hash_mask
= (1 << log2qty
) - 1;
7156 * This function checks whether pageblock includes unmovable pages or not.
7157 * If @count is not zero, it is okay to include less @count unmovable pages
7159 * PageLRU check without isolation or lru_lock could race so that
7160 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
7161 * expect this function should be exact.
7163 bool has_unmovable_pages(struct zone
*zone
, struct page
*page
, int count
,
7164 bool skip_hwpoisoned_pages
)
7166 unsigned long pfn
, iter
, found
;
7170 * For avoiding noise data, lru_add_drain_all() should be called
7171 * If ZONE_MOVABLE, the zone never contains unmovable pages
7173 if (zone_idx(zone
) == ZONE_MOVABLE
)
7175 mt
= get_pageblock_migratetype(page
);
7176 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
7179 pfn
= page_to_pfn(page
);
7180 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
7181 unsigned long check
= pfn
+ iter
;
7183 if (!pfn_valid_within(check
))
7186 page
= pfn_to_page(check
);
7189 * Hugepages are not in LRU lists, but they're movable.
7190 * We need not scan over tail pages bacause we don't
7191 * handle each tail page individually in migration.
7193 if (PageHuge(page
)) {
7194 iter
= round_up(iter
+ 1, 1<<compound_order(page
)) - 1;
7199 * We can't use page_count without pin a page
7200 * because another CPU can free compound page.
7201 * This check already skips compound tails of THP
7202 * because their page->_refcount is zero at all time.
7204 if (!page_ref_count(page
)) {
7205 if (PageBuddy(page
))
7206 iter
+= (1 << page_order(page
)) - 1;
7211 * The HWPoisoned page may be not in buddy system, and
7212 * page_count() is not 0.
7214 if (skip_hwpoisoned_pages
&& PageHWPoison(page
))
7220 * If there are RECLAIMABLE pages, we need to check
7221 * it. But now, memory offline itself doesn't call
7222 * shrink_node_slabs() and it still to be fixed.
7225 * If the page is not RAM, page_count()should be 0.
7226 * we don't need more check. This is an _used_ not-movable page.
7228 * The problematic thing here is PG_reserved pages. PG_reserved
7229 * is set to both of a memory hole page and a _used_ kernel
7238 bool is_pageblock_removable_nolock(struct page
*page
)
7244 * We have to be careful here because we are iterating over memory
7245 * sections which are not zone aware so we might end up outside of
7246 * the zone but still within the section.
7247 * We have to take care about the node as well. If the node is offline
7248 * its NODE_DATA will be NULL - see page_zone.
7250 if (!node_online(page_to_nid(page
)))
7253 zone
= page_zone(page
);
7254 pfn
= page_to_pfn(page
);
7255 if (!zone_spans_pfn(zone
, pfn
))
7258 return !has_unmovable_pages(zone
, page
, 0, true);
7261 #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
7263 static unsigned long pfn_max_align_down(unsigned long pfn
)
7265 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
7266 pageblock_nr_pages
) - 1);
7269 static unsigned long pfn_max_align_up(unsigned long pfn
)
7271 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
7272 pageblock_nr_pages
));
7275 /* [start, end) must belong to a single zone. */
7276 static int __alloc_contig_migrate_range(struct compact_control
*cc
,
7277 unsigned long start
, unsigned long end
)
7279 /* This function is based on compact_zone() from compaction.c. */
7280 unsigned long nr_reclaimed
;
7281 unsigned long pfn
= start
;
7282 unsigned int tries
= 0;
7287 while (pfn
< end
|| !list_empty(&cc
->migratepages
)) {
7288 if (fatal_signal_pending(current
)) {
7293 if (list_empty(&cc
->migratepages
)) {
7294 cc
->nr_migratepages
= 0;
7295 pfn
= isolate_migratepages_range(cc
, pfn
, end
);
7301 } else if (++tries
== 5) {
7302 ret
= ret
< 0 ? ret
: -EBUSY
;
7306 nr_reclaimed
= reclaim_clean_pages_from_list(cc
->zone
,
7308 cc
->nr_migratepages
-= nr_reclaimed
;
7310 ret
= migrate_pages(&cc
->migratepages
, alloc_migrate_target
,
7311 NULL
, 0, cc
->mode
, MR_CMA
);
7314 putback_movable_pages(&cc
->migratepages
);
7321 * alloc_contig_range() -- tries to allocate given range of pages
7322 * @start: start PFN to allocate
7323 * @end: one-past-the-last PFN to allocate
7324 * @migratetype: migratetype of the underlaying pageblocks (either
7325 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
7326 * in range must have the same migratetype and it must
7327 * be either of the two.
7329 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
7330 * aligned, however it's the caller's responsibility to guarantee that
7331 * we are the only thread that changes migrate type of pageblocks the
7334 * The PFN range must belong to a single zone.
7336 * Returns zero on success or negative error code. On success all
7337 * pages which PFN is in [start, end) are allocated for the caller and
7338 * need to be freed with free_contig_range().
7340 int alloc_contig_range(unsigned long start
, unsigned long end
,
7341 unsigned migratetype
)
7343 unsigned long outer_start
, outer_end
;
7347 struct compact_control cc
= {
7348 .nr_migratepages
= 0,
7350 .zone
= page_zone(pfn_to_page(start
)),
7351 .mode
= MIGRATE_SYNC
,
7352 .ignore_skip_hint
= true,
7354 INIT_LIST_HEAD(&cc
.migratepages
);
7357 * What we do here is we mark all pageblocks in range as
7358 * MIGRATE_ISOLATE. Because pageblock and max order pages may
7359 * have different sizes, and due to the way page allocator
7360 * work, we align the range to biggest of the two pages so
7361 * that page allocator won't try to merge buddies from
7362 * different pageblocks and change MIGRATE_ISOLATE to some
7363 * other migration type.
7365 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
7366 * migrate the pages from an unaligned range (ie. pages that
7367 * we are interested in). This will put all the pages in
7368 * range back to page allocator as MIGRATE_ISOLATE.
7370 * When this is done, we take the pages in range from page
7371 * allocator removing them from the buddy system. This way
7372 * page allocator will never consider using them.
7374 * This lets us mark the pageblocks back as
7375 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
7376 * aligned range but not in the unaligned, original range are
7377 * put back to page allocator so that buddy can use them.
7380 ret
= start_isolate_page_range(pfn_max_align_down(start
),
7381 pfn_max_align_up(end
), migratetype
,
7387 * In case of -EBUSY, we'd like to know which page causes problem.
7388 * So, just fall through. We will check it in test_pages_isolated().
7390 ret
= __alloc_contig_migrate_range(&cc
, start
, end
);
7391 if (ret
&& ret
!= -EBUSY
)
7395 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
7396 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
7397 * more, all pages in [start, end) are free in page allocator.
7398 * What we are going to do is to allocate all pages from
7399 * [start, end) (that is remove them from page allocator).
7401 * The only problem is that pages at the beginning and at the
7402 * end of interesting range may be not aligned with pages that
7403 * page allocator holds, ie. they can be part of higher order
7404 * pages. Because of this, we reserve the bigger range and
7405 * once this is done free the pages we are not interested in.
7407 * We don't have to hold zone->lock here because the pages are
7408 * isolated thus they won't get removed from buddy.
7411 lru_add_drain_all();
7412 drain_all_pages(cc
.zone
);
7415 outer_start
= start
;
7416 while (!PageBuddy(pfn_to_page(outer_start
))) {
7417 if (++order
>= MAX_ORDER
) {
7418 outer_start
= start
;
7421 outer_start
&= ~0UL << order
;
7424 if (outer_start
!= start
) {
7425 order
= page_order(pfn_to_page(outer_start
));
7428 * outer_start page could be small order buddy page and
7429 * it doesn't include start page. Adjust outer_start
7430 * in this case to report failed page properly
7431 * on tracepoint in test_pages_isolated()
7433 if (outer_start
+ (1UL << order
) <= start
)
7434 outer_start
= start
;
7437 /* Make sure the range is really isolated. */
7438 if (test_pages_isolated(outer_start
, end
, false)) {
7439 pr_info("%s: [%lx, %lx) PFNs busy\n",
7440 __func__
, outer_start
, end
);
7445 /* Grab isolated pages from freelists. */
7446 outer_end
= isolate_freepages_range(&cc
, outer_start
, end
);
7452 /* Free head and tail (if any) */
7453 if (start
!= outer_start
)
7454 free_contig_range(outer_start
, start
- outer_start
);
7455 if (end
!= outer_end
)
7456 free_contig_range(end
, outer_end
- end
);
7459 undo_isolate_page_range(pfn_max_align_down(start
),
7460 pfn_max_align_up(end
), migratetype
);
7464 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
7466 unsigned int count
= 0;
7468 for (; nr_pages
--; pfn
++) {
7469 struct page
*page
= pfn_to_page(pfn
);
7471 count
+= page_count(page
) != 1;
7474 WARN(count
!= 0, "%d pages are still in use!\n", count
);
7478 #ifdef CONFIG_MEMORY_HOTPLUG
7480 * The zone indicated has a new number of managed_pages; batch sizes and percpu
7481 * page high values need to be recalulated.
7483 void __meminit
zone_pcp_update(struct zone
*zone
)
7486 mutex_lock(&pcp_batch_high_lock
);
7487 for_each_possible_cpu(cpu
)
7488 pageset_set_high_and_batch(zone
,
7489 per_cpu_ptr(zone
->pageset
, cpu
));
7490 mutex_unlock(&pcp_batch_high_lock
);
7494 void zone_pcp_reset(struct zone
*zone
)
7496 unsigned long flags
;
7498 struct per_cpu_pageset
*pset
;
7500 /* avoid races with drain_pages() */
7501 local_irq_save(flags
);
7502 if (zone
->pageset
!= &boot_pageset
) {
7503 for_each_online_cpu(cpu
) {
7504 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
7505 drain_zonestat(zone
, pset
);
7507 free_percpu(zone
->pageset
);
7508 zone
->pageset
= &boot_pageset
;
7510 local_irq_restore(flags
);
7513 #ifdef CONFIG_MEMORY_HOTREMOVE
7515 * All pages in the range must be in a single zone and isolated
7516 * before calling this.
7519 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
7523 unsigned int order
, i
;
7525 unsigned long flags
;
7526 /* find the first valid pfn */
7527 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
7532 zone
= page_zone(pfn_to_page(pfn
));
7533 spin_lock_irqsave(&zone
->lock
, flags
);
7535 while (pfn
< end_pfn
) {
7536 if (!pfn_valid(pfn
)) {
7540 page
= pfn_to_page(pfn
);
7542 * The HWPoisoned page may be not in buddy system, and
7543 * page_count() is not 0.
7545 if (unlikely(!PageBuddy(page
) && PageHWPoison(page
))) {
7547 SetPageReserved(page
);
7551 BUG_ON(page_count(page
));
7552 BUG_ON(!PageBuddy(page
));
7553 order
= page_order(page
);
7554 #ifdef CONFIG_DEBUG_VM
7555 pr_info("remove from free list %lx %d %lx\n",
7556 pfn
, 1 << order
, end_pfn
);
7558 list_del(&page
->lru
);
7559 rmv_page_order(page
);
7560 zone
->free_area
[order
].nr_free
--;
7561 for (i
= 0; i
< (1 << order
); i
++)
7562 SetPageReserved((page
+i
));
7563 pfn
+= (1 << order
);
7565 spin_unlock_irqrestore(&zone
->lock
, flags
);
7569 bool is_free_buddy_page(struct page
*page
)
7571 struct zone
*zone
= page_zone(page
);
7572 unsigned long pfn
= page_to_pfn(page
);
7573 unsigned long flags
;
7576 spin_lock_irqsave(&zone
->lock
, flags
);
7577 for (order
= 0; order
< MAX_ORDER
; order
++) {
7578 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
7580 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
7583 spin_unlock_irqrestore(&zone
->lock
, flags
);
7585 return order
< MAX_ORDER
;