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
;
526 * Allow a burst of 60 reports, then keep quiet for that minute;
527 * or allow a steady drip of one report per second.
529 if (nr_shown
== 60) {
530 if (time_before(jiffies
, resume
)) {
536 "BUG: Bad page state: %lu messages suppressed\n",
543 resume
= jiffies
+ 60 * HZ
;
545 pr_alert("BUG: Bad page state in process %s pfn:%05lx\n",
546 current
->comm
, page_to_pfn(page
));
547 __dump_page(page
, reason
);
548 bad_flags
&= page
->flags
;
550 pr_alert("bad because of flags: %#lx(%pGp)\n",
551 bad_flags
, &bad_flags
);
552 dump_page_owner(page
);
557 /* Leave bad fields for debug, except PageBuddy could make trouble */
558 page_mapcount_reset(page
); /* remove PageBuddy */
559 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
563 * Higher-order pages are called "compound pages". They are structured thusly:
565 * The first PAGE_SIZE page is called the "head page" and have PG_head set.
567 * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
568 * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
570 * The first tail page's ->compound_dtor holds the offset in array of compound
571 * page destructors. See compound_page_dtors.
573 * The first tail page's ->compound_order holds the order of allocation.
574 * This usage means that zero-order pages may not be compound.
577 void free_compound_page(struct page
*page
)
579 __free_pages_ok(page
, compound_order(page
));
582 void prep_compound_page(struct page
*page
, unsigned int order
)
585 int nr_pages
= 1 << order
;
587 set_compound_page_dtor(page
, COMPOUND_PAGE_DTOR
);
588 set_compound_order(page
, order
);
590 for (i
= 1; i
< nr_pages
; i
++) {
591 struct page
*p
= page
+ i
;
592 set_page_count(p
, 0);
593 p
->mapping
= TAIL_MAPPING
;
594 set_compound_head(p
, page
);
596 atomic_set(compound_mapcount_ptr(page
), -1);
599 #ifdef CONFIG_DEBUG_PAGEALLOC
600 unsigned int _debug_guardpage_minorder
;
601 bool _debug_pagealloc_enabled __read_mostly
602 = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT
);
603 EXPORT_SYMBOL(_debug_pagealloc_enabled
);
604 bool _debug_guardpage_enabled __read_mostly
;
606 static int __init
early_debug_pagealloc(char *buf
)
610 return kstrtobool(buf
, &_debug_pagealloc_enabled
);
612 early_param("debug_pagealloc", early_debug_pagealloc
);
614 static bool need_debug_guardpage(void)
616 /* If we don't use debug_pagealloc, we don't need guard page */
617 if (!debug_pagealloc_enabled())
623 static void init_debug_guardpage(void)
625 if (!debug_pagealloc_enabled())
628 _debug_guardpage_enabled
= true;
631 struct page_ext_operations debug_guardpage_ops
= {
632 .need
= need_debug_guardpage
,
633 .init
= init_debug_guardpage
,
636 static int __init
debug_guardpage_minorder_setup(char *buf
)
640 if (kstrtoul(buf
, 10, &res
) < 0 || res
> MAX_ORDER
/ 2) {
641 pr_err("Bad debug_guardpage_minorder value\n");
644 _debug_guardpage_minorder
= res
;
645 pr_info("Setting debug_guardpage_minorder to %lu\n", res
);
648 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup
);
650 static inline void set_page_guard(struct zone
*zone
, struct page
*page
,
651 unsigned int order
, int migratetype
)
653 struct page_ext
*page_ext
;
655 if (!debug_guardpage_enabled())
658 page_ext
= lookup_page_ext(page
);
659 __set_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
661 INIT_LIST_HEAD(&page
->lru
);
662 set_page_private(page
, order
);
663 /* Guard pages are not available for any usage */
664 __mod_zone_freepage_state(zone
, -(1 << order
), migratetype
);
667 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
668 unsigned int order
, int migratetype
)
670 struct page_ext
*page_ext
;
672 if (!debug_guardpage_enabled())
675 page_ext
= lookup_page_ext(page
);
676 __clear_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
678 set_page_private(page
, 0);
679 if (!is_migrate_isolate(migratetype
))
680 __mod_zone_freepage_state(zone
, (1 << order
), migratetype
);
683 struct page_ext_operations debug_guardpage_ops
= { NULL
, };
684 static inline void set_page_guard(struct zone
*zone
, struct page
*page
,
685 unsigned int order
, int migratetype
) {}
686 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
687 unsigned int order
, int migratetype
) {}
690 static inline void set_page_order(struct page
*page
, unsigned int order
)
692 set_page_private(page
, order
);
693 __SetPageBuddy(page
);
696 static inline void rmv_page_order(struct page
*page
)
698 __ClearPageBuddy(page
);
699 set_page_private(page
, 0);
703 * This function checks whether a page is free && is the buddy
704 * we can do coalesce a page and its buddy if
705 * (a) the buddy is not in a hole &&
706 * (b) the buddy is in the buddy system &&
707 * (c) a page and its buddy have the same order &&
708 * (d) a page and its buddy are in the same zone.
710 * For recording whether a page is in the buddy system, we set ->_mapcount
711 * PAGE_BUDDY_MAPCOUNT_VALUE.
712 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
713 * serialized by zone->lock.
715 * For recording page's order, we use page_private(page).
717 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
720 if (!pfn_valid_within(page_to_pfn(buddy
)))
723 if (page_is_guard(buddy
) && page_order(buddy
) == order
) {
724 if (page_zone_id(page
) != page_zone_id(buddy
))
727 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
732 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
734 * zone check is done late to avoid uselessly
735 * calculating zone/node ids for pages that could
738 if (page_zone_id(page
) != page_zone_id(buddy
))
741 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
749 * Freeing function for a buddy system allocator.
751 * The concept of a buddy system is to maintain direct-mapped table
752 * (containing bit values) for memory blocks of various "orders".
753 * The bottom level table contains the map for the smallest allocatable
754 * units of memory (here, pages), and each level above it describes
755 * pairs of units from the levels below, hence, "buddies".
756 * At a high level, all that happens here is marking the table entry
757 * at the bottom level available, and propagating the changes upward
758 * as necessary, plus some accounting needed to play nicely with other
759 * parts of the VM system.
760 * At each level, we keep a list of pages, which are heads of continuous
761 * free pages of length of (1 << order) and marked with _mapcount
762 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
764 * So when we are allocating or freeing one, we can derive the state of the
765 * other. That is, if we allocate a small block, and both were
766 * free, the remainder of the region must be split into blocks.
767 * If a block is freed, and its buddy is also free, then this
768 * triggers coalescing into a block of larger size.
773 static inline void __free_one_page(struct page
*page
,
775 struct zone
*zone
, unsigned int order
,
778 unsigned long page_idx
;
779 unsigned long combined_idx
;
780 unsigned long uninitialized_var(buddy_idx
);
782 unsigned int max_order
;
784 max_order
= min_t(unsigned int, MAX_ORDER
, pageblock_order
+ 1);
786 VM_BUG_ON(!zone_is_initialized(zone
));
787 VM_BUG_ON_PAGE(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
, page
);
789 VM_BUG_ON(migratetype
== -1);
790 if (likely(!is_migrate_isolate(migratetype
)))
791 __mod_zone_freepage_state(zone
, 1 << order
, migratetype
);
793 page_idx
= pfn
& ((1 << MAX_ORDER
) - 1);
795 VM_BUG_ON_PAGE(page_idx
& ((1 << order
) - 1), page
);
796 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
799 while (order
< max_order
- 1) {
800 buddy_idx
= __find_buddy_index(page_idx
, order
);
801 buddy
= page
+ (buddy_idx
- page_idx
);
802 if (!page_is_buddy(page
, buddy
, order
))
805 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
806 * merge with it and move up one order.
808 if (page_is_guard(buddy
)) {
809 clear_page_guard(zone
, buddy
, order
, migratetype
);
811 list_del(&buddy
->lru
);
812 zone
->free_area
[order
].nr_free
--;
813 rmv_page_order(buddy
);
815 combined_idx
= buddy_idx
& page_idx
;
816 page
= page
+ (combined_idx
- page_idx
);
817 page_idx
= combined_idx
;
820 if (max_order
< MAX_ORDER
) {
821 /* If we are here, it means order is >= pageblock_order.
822 * We want to prevent merge between freepages on isolate
823 * pageblock and normal pageblock. Without this, pageblock
824 * isolation could cause incorrect freepage or CMA accounting.
826 * We don't want to hit this code for the more frequent
829 if (unlikely(has_isolate_pageblock(zone
))) {
832 buddy_idx
= __find_buddy_index(page_idx
, order
);
833 buddy
= page
+ (buddy_idx
- page_idx
);
834 buddy_mt
= get_pageblock_migratetype(buddy
);
836 if (migratetype
!= buddy_mt
837 && (is_migrate_isolate(migratetype
) ||
838 is_migrate_isolate(buddy_mt
)))
842 goto continue_merging
;
846 set_page_order(page
, order
);
849 * If this is not the largest possible page, check if the buddy
850 * of the next-highest order is free. If it is, it's possible
851 * that pages are being freed that will coalesce soon. In case,
852 * that is happening, add the free page to the tail of the list
853 * so it's less likely to be used soon and more likely to be merged
854 * as a higher order page
856 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
857 struct page
*higher_page
, *higher_buddy
;
858 combined_idx
= buddy_idx
& page_idx
;
859 higher_page
= page
+ (combined_idx
- page_idx
);
860 buddy_idx
= __find_buddy_index(combined_idx
, order
+ 1);
861 higher_buddy
= higher_page
+ (buddy_idx
- combined_idx
);
862 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
863 list_add_tail(&page
->lru
,
864 &zone
->free_area
[order
].free_list
[migratetype
]);
869 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
871 zone
->free_area
[order
].nr_free
++;
875 * A bad page could be due to a number of fields. Instead of multiple branches,
876 * try and check multiple fields with one check. The caller must do a detailed
877 * check if necessary.
879 static inline bool page_expected_state(struct page
*page
,
880 unsigned long check_flags
)
882 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
885 if (unlikely((unsigned long)page
->mapping
|
886 page_ref_count(page
) |
888 (unsigned long)page
->mem_cgroup
|
890 (page
->flags
& check_flags
)))
896 static void free_pages_check_bad(struct page
*page
)
898 const char *bad_reason
;
899 unsigned long bad_flags
;
904 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
905 bad_reason
= "nonzero mapcount";
906 if (unlikely(page
->mapping
!= NULL
))
907 bad_reason
= "non-NULL mapping";
908 if (unlikely(page_ref_count(page
) != 0))
909 bad_reason
= "nonzero _refcount";
910 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
)) {
911 bad_reason
= "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
912 bad_flags
= PAGE_FLAGS_CHECK_AT_FREE
;
915 if (unlikely(page
->mem_cgroup
))
916 bad_reason
= "page still charged to cgroup";
918 bad_page(page
, bad_reason
, bad_flags
);
921 static inline int free_pages_check(struct page
*page
)
923 if (likely(page_expected_state(page
, PAGE_FLAGS_CHECK_AT_FREE
)))
926 /* Something has gone sideways, find it */
927 free_pages_check_bad(page
);
931 static int free_tail_pages_check(struct page
*head_page
, struct page
*page
)
936 * We rely page->lru.next never has bit 0 set, unless the page
937 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
939 BUILD_BUG_ON((unsigned long)LIST_POISON1
& 1);
941 if (!IS_ENABLED(CONFIG_DEBUG_VM
)) {
945 switch (page
- head_page
) {
947 /* the first tail page: ->mapping is compound_mapcount() */
948 if (unlikely(compound_mapcount(page
))) {
949 bad_page(page
, "nonzero compound_mapcount", 0);
955 * the second tail page: ->mapping is
956 * page_deferred_list().next -- ignore value.
960 if (page
->mapping
!= TAIL_MAPPING
) {
961 bad_page(page
, "corrupted mapping in tail page", 0);
966 if (unlikely(!PageTail(page
))) {
967 bad_page(page
, "PageTail not set", 0);
970 if (unlikely(compound_head(page
) != head_page
)) {
971 bad_page(page
, "compound_head not consistent", 0);
976 page
->mapping
= NULL
;
977 clear_compound_head(page
);
981 static __always_inline
bool free_pages_prepare(struct page
*page
,
982 unsigned int order
, bool check_free
)
986 VM_BUG_ON_PAGE(PageTail(page
), page
);
988 trace_mm_page_free(page
, order
);
989 kmemcheck_free_shadow(page
, order
);
992 * Check tail pages before head page information is cleared to
993 * avoid checking PageCompound for order-0 pages.
995 if (unlikely(order
)) {
996 bool compound
= PageCompound(page
);
999 VM_BUG_ON_PAGE(compound
&& compound_order(page
) != order
, page
);
1001 for (i
= 1; i
< (1 << order
); i
++) {
1003 bad
+= free_tail_pages_check(page
, page
+ i
);
1004 if (unlikely(free_pages_check(page
+ i
))) {
1008 (page
+ i
)->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
1011 if (PageAnonHead(page
))
1012 page
->mapping
= NULL
;
1014 bad
+= free_pages_check(page
);
1018 page_cpupid_reset_last(page
);
1019 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
1020 reset_page_owner(page
, order
);
1022 if (!PageHighMem(page
)) {
1023 debug_check_no_locks_freed(page_address(page
),
1024 PAGE_SIZE
<< order
);
1025 debug_check_no_obj_freed(page_address(page
),
1026 PAGE_SIZE
<< order
);
1028 arch_free_page(page
, order
);
1029 kernel_poison_pages(page
, 1 << order
, 0);
1030 kernel_map_pages(page
, 1 << order
, 0);
1031 kasan_free_pages(page
, order
);
1036 #ifdef CONFIG_DEBUG_VM
1037 static inline bool free_pcp_prepare(struct page
*page
)
1039 return free_pages_prepare(page
, 0, true);
1042 static inline bool bulkfree_pcp_prepare(struct page
*page
)
1047 static bool free_pcp_prepare(struct page
*page
)
1049 return free_pages_prepare(page
, 0, false);
1052 static bool bulkfree_pcp_prepare(struct page
*page
)
1054 return free_pages_check(page
);
1056 #endif /* CONFIG_DEBUG_VM */
1059 * Frees a number of pages from the PCP lists
1060 * Assumes all pages on list are in same zone, and of same order.
1061 * count is the number of pages to free.
1063 * If the zone was previously in an "all pages pinned" state then look to
1064 * see if this freeing clears that state.
1066 * And clear the zone's pages_scanned counter, to hold off the "all pages are
1067 * pinned" detection logic.
1069 static void free_pcppages_bulk(struct zone
*zone
, int count
,
1070 struct per_cpu_pages
*pcp
)
1072 int migratetype
= 0;
1074 unsigned long nr_scanned
;
1075 bool isolated_pageblocks
;
1077 spin_lock(&zone
->lock
);
1078 isolated_pageblocks
= has_isolate_pageblock(zone
);
1079 nr_scanned
= zone_page_state(zone
, NR_PAGES_SCANNED
);
1081 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, -nr_scanned
);
1085 struct list_head
*list
;
1088 * Remove pages from lists in a round-robin fashion. A
1089 * batch_free count is maintained that is incremented when an
1090 * empty list is encountered. This is so more pages are freed
1091 * off fuller lists instead of spinning excessively around empty
1096 if (++migratetype
== MIGRATE_PCPTYPES
)
1098 list
= &pcp
->lists
[migratetype
];
1099 } while (list_empty(list
));
1101 /* This is the only non-empty list. Free them all. */
1102 if (batch_free
== MIGRATE_PCPTYPES
)
1106 int mt
; /* migratetype of the to-be-freed page */
1108 page
= list_last_entry(list
, struct page
, lru
);
1109 /* must delete as __free_one_page list manipulates */
1110 list_del(&page
->lru
);
1112 mt
= get_pcppage_migratetype(page
);
1113 /* MIGRATE_ISOLATE page should not go to pcplists */
1114 VM_BUG_ON_PAGE(is_migrate_isolate(mt
), page
);
1115 /* Pageblock could have been isolated meanwhile */
1116 if (unlikely(isolated_pageblocks
))
1117 mt
= get_pageblock_migratetype(page
);
1119 if (bulkfree_pcp_prepare(page
))
1122 __free_one_page(page
, page_to_pfn(page
), zone
, 0, mt
);
1123 trace_mm_page_pcpu_drain(page
, 0, mt
);
1124 } while (--count
&& --batch_free
&& !list_empty(list
));
1126 spin_unlock(&zone
->lock
);
1129 static void free_one_page(struct zone
*zone
,
1130 struct page
*page
, unsigned long pfn
,
1134 unsigned long nr_scanned
;
1135 spin_lock(&zone
->lock
);
1136 nr_scanned
= zone_page_state(zone
, NR_PAGES_SCANNED
);
1138 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, -nr_scanned
);
1140 if (unlikely(has_isolate_pageblock(zone
) ||
1141 is_migrate_isolate(migratetype
))) {
1142 migratetype
= get_pfnblock_migratetype(page
, pfn
);
1144 __free_one_page(page
, pfn
, zone
, order
, migratetype
);
1145 spin_unlock(&zone
->lock
);
1148 static void __meminit
__init_single_page(struct page
*page
, unsigned long pfn
,
1149 unsigned long zone
, int nid
)
1151 set_page_links(page
, zone
, nid
, pfn
);
1152 init_page_count(page
);
1153 page_mapcount_reset(page
);
1154 page_cpupid_reset_last(page
);
1156 INIT_LIST_HEAD(&page
->lru
);
1157 #ifdef WANT_PAGE_VIRTUAL
1158 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1159 if (!is_highmem_idx(zone
))
1160 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
1164 static void __meminit
__init_single_pfn(unsigned long pfn
, unsigned long zone
,
1167 return __init_single_page(pfn_to_page(pfn
), pfn
, zone
, nid
);
1170 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1171 static void init_reserved_page(unsigned long pfn
)
1176 if (!early_page_uninitialised(pfn
))
1179 nid
= early_pfn_to_nid(pfn
);
1180 pgdat
= NODE_DATA(nid
);
1182 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
1183 struct zone
*zone
= &pgdat
->node_zones
[zid
];
1185 if (pfn
>= zone
->zone_start_pfn
&& pfn
< zone_end_pfn(zone
))
1188 __init_single_pfn(pfn
, zid
, nid
);
1191 static inline void init_reserved_page(unsigned long pfn
)
1194 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1197 * Initialised pages do not have PageReserved set. This function is
1198 * called for each range allocated by the bootmem allocator and
1199 * marks the pages PageReserved. The remaining valid pages are later
1200 * sent to the buddy page allocator.
1202 void __meminit
reserve_bootmem_region(phys_addr_t start
, phys_addr_t end
)
1204 unsigned long start_pfn
= PFN_DOWN(start
);
1205 unsigned long end_pfn
= PFN_UP(end
);
1207 for (; start_pfn
< end_pfn
; start_pfn
++) {
1208 if (pfn_valid(start_pfn
)) {
1209 struct page
*page
= pfn_to_page(start_pfn
);
1211 init_reserved_page(start_pfn
);
1213 /* Avoid false-positive PageTail() */
1214 INIT_LIST_HEAD(&page
->lru
);
1216 SetPageReserved(page
);
1221 static void __free_pages_ok(struct page
*page
, unsigned int order
)
1223 unsigned long flags
;
1225 unsigned long pfn
= page_to_pfn(page
);
1227 if (!free_pages_prepare(page
, order
, true))
1230 migratetype
= get_pfnblock_migratetype(page
, pfn
);
1231 local_irq_save(flags
);
1232 __count_vm_events(PGFREE
, 1 << order
);
1233 free_one_page(page_zone(page
), page
, pfn
, order
, migratetype
);
1234 local_irq_restore(flags
);
1237 static void __init
__free_pages_boot_core(struct page
*page
, unsigned int order
)
1239 unsigned int nr_pages
= 1 << order
;
1240 struct page
*p
= page
;
1244 for (loop
= 0; loop
< (nr_pages
- 1); loop
++, p
++) {
1246 __ClearPageReserved(p
);
1247 set_page_count(p
, 0);
1249 __ClearPageReserved(p
);
1250 set_page_count(p
, 0);
1252 page_zone(page
)->managed_pages
+= nr_pages
;
1253 set_page_refcounted(page
);
1254 __free_pages(page
, order
);
1257 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1258 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1260 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata
;
1262 int __meminit
early_pfn_to_nid(unsigned long pfn
)
1264 static DEFINE_SPINLOCK(early_pfn_lock
);
1267 spin_lock(&early_pfn_lock
);
1268 nid
= __early_pfn_to_nid(pfn
, &early_pfnnid_cache
);
1271 spin_unlock(&early_pfn_lock
);
1277 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1278 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1279 struct mminit_pfnnid_cache
*state
)
1283 nid
= __early_pfn_to_nid(pfn
, state
);
1284 if (nid
>= 0 && nid
!= node
)
1289 /* Only safe to use early in boot when initialisation is single-threaded */
1290 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1292 return meminit_pfn_in_nid(pfn
, node
, &early_pfnnid_cache
);
1297 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1301 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1302 struct mminit_pfnnid_cache
*state
)
1309 void __init
__free_pages_bootmem(struct page
*page
, unsigned long pfn
,
1312 if (early_page_uninitialised(pfn
))
1314 return __free_pages_boot_core(page
, order
);
1318 * Check that the whole (or subset of) a pageblock given by the interval of
1319 * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1320 * with the migration of free compaction scanner. The scanners then need to
1321 * use only pfn_valid_within() check for arches that allow holes within
1324 * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1326 * It's possible on some configurations to have a setup like node0 node1 node0
1327 * i.e. it's possible that all pages within a zones range of pages do not
1328 * belong to a single zone. We assume that a border between node0 and node1
1329 * can occur within a single pageblock, but not a node0 node1 node0
1330 * interleaving within a single pageblock. It is therefore sufficient to check
1331 * the first and last page of a pageblock and avoid checking each individual
1332 * page in a pageblock.
1334 struct page
*__pageblock_pfn_to_page(unsigned long start_pfn
,
1335 unsigned long end_pfn
, struct zone
*zone
)
1337 struct page
*start_page
;
1338 struct page
*end_page
;
1340 /* end_pfn is one past the range we are checking */
1343 if (!pfn_valid(start_pfn
) || !pfn_valid(end_pfn
))
1346 start_page
= pfn_to_page(start_pfn
);
1348 if (page_zone(start_page
) != zone
)
1351 end_page
= pfn_to_page(end_pfn
);
1353 /* This gives a shorter code than deriving page_zone(end_page) */
1354 if (page_zone_id(start_page
) != page_zone_id(end_page
))
1360 void set_zone_contiguous(struct zone
*zone
)
1362 unsigned long block_start_pfn
= zone
->zone_start_pfn
;
1363 unsigned long block_end_pfn
;
1365 block_end_pfn
= ALIGN(block_start_pfn
+ 1, pageblock_nr_pages
);
1366 for (; block_start_pfn
< zone_end_pfn(zone
);
1367 block_start_pfn
= block_end_pfn
,
1368 block_end_pfn
+= pageblock_nr_pages
) {
1370 block_end_pfn
= min(block_end_pfn
, zone_end_pfn(zone
));
1372 if (!__pageblock_pfn_to_page(block_start_pfn
,
1373 block_end_pfn
, zone
))
1377 /* We confirm that there is no hole */
1378 zone
->contiguous
= true;
1381 void clear_zone_contiguous(struct zone
*zone
)
1383 zone
->contiguous
= false;
1386 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1387 static void __init
deferred_free_range(struct page
*page
,
1388 unsigned long pfn
, int nr_pages
)
1395 /* Free a large naturally-aligned chunk if possible */
1396 if (nr_pages
== MAX_ORDER_NR_PAGES
&&
1397 (pfn
& (MAX_ORDER_NR_PAGES
-1)) == 0) {
1398 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
1399 __free_pages_boot_core(page
, MAX_ORDER
-1);
1403 for (i
= 0; i
< nr_pages
; i
++, page
++)
1404 __free_pages_boot_core(page
, 0);
1407 /* Completion tracking for deferred_init_memmap() threads */
1408 static atomic_t pgdat_init_n_undone __initdata
;
1409 static __initdata
DECLARE_COMPLETION(pgdat_init_all_done_comp
);
1411 static inline void __init
pgdat_init_report_one_done(void)
1413 if (atomic_dec_and_test(&pgdat_init_n_undone
))
1414 complete(&pgdat_init_all_done_comp
);
1417 /* Initialise remaining memory on a node */
1418 static int __init
deferred_init_memmap(void *data
)
1420 pg_data_t
*pgdat
= data
;
1421 int nid
= pgdat
->node_id
;
1422 struct mminit_pfnnid_cache nid_init_state
= { };
1423 unsigned long start
= jiffies
;
1424 unsigned long nr_pages
= 0;
1425 unsigned long walk_start
, walk_end
;
1428 unsigned long first_init_pfn
= pgdat
->first_deferred_pfn
;
1429 const struct cpumask
*cpumask
= cpumask_of_node(pgdat
->node_id
);
1431 if (first_init_pfn
== ULONG_MAX
) {
1432 pgdat_init_report_one_done();
1436 /* Bind memory initialisation thread to a local node if possible */
1437 if (!cpumask_empty(cpumask
))
1438 set_cpus_allowed_ptr(current
, cpumask
);
1440 /* Sanity check boundaries */
1441 BUG_ON(pgdat
->first_deferred_pfn
< pgdat
->node_start_pfn
);
1442 BUG_ON(pgdat
->first_deferred_pfn
> pgdat_end_pfn(pgdat
));
1443 pgdat
->first_deferred_pfn
= ULONG_MAX
;
1445 /* Only the highest zone is deferred so find it */
1446 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
1447 zone
= pgdat
->node_zones
+ zid
;
1448 if (first_init_pfn
< zone_end_pfn(zone
))
1452 for_each_mem_pfn_range(i
, nid
, &walk_start
, &walk_end
, NULL
) {
1453 unsigned long pfn
, end_pfn
;
1454 struct page
*page
= NULL
;
1455 struct page
*free_base_page
= NULL
;
1456 unsigned long free_base_pfn
= 0;
1459 end_pfn
= min(walk_end
, zone_end_pfn(zone
));
1460 pfn
= first_init_pfn
;
1461 if (pfn
< walk_start
)
1463 if (pfn
< zone
->zone_start_pfn
)
1464 pfn
= zone
->zone_start_pfn
;
1466 for (; pfn
< end_pfn
; pfn
++) {
1467 if (!pfn_valid_within(pfn
))
1471 * Ensure pfn_valid is checked every
1472 * MAX_ORDER_NR_PAGES for memory holes
1474 if ((pfn
& (MAX_ORDER_NR_PAGES
- 1)) == 0) {
1475 if (!pfn_valid(pfn
)) {
1481 if (!meminit_pfn_in_nid(pfn
, nid
, &nid_init_state
)) {
1486 /* Minimise pfn page lookups and scheduler checks */
1487 if (page
&& (pfn
& (MAX_ORDER_NR_PAGES
- 1)) != 0) {
1490 nr_pages
+= nr_to_free
;
1491 deferred_free_range(free_base_page
,
1492 free_base_pfn
, nr_to_free
);
1493 free_base_page
= NULL
;
1494 free_base_pfn
= nr_to_free
= 0;
1496 page
= pfn_to_page(pfn
);
1501 VM_BUG_ON(page_zone(page
) != zone
);
1505 __init_single_page(page
, pfn
, zid
, nid
);
1506 if (!free_base_page
) {
1507 free_base_page
= page
;
1508 free_base_pfn
= pfn
;
1513 /* Where possible, batch up pages for a single free */
1516 /* Free the current block of pages to allocator */
1517 nr_pages
+= nr_to_free
;
1518 deferred_free_range(free_base_page
, free_base_pfn
,
1520 free_base_page
= NULL
;
1521 free_base_pfn
= nr_to_free
= 0;
1524 first_init_pfn
= max(end_pfn
, first_init_pfn
);
1527 /* Sanity check that the next zone really is unpopulated */
1528 WARN_ON(++zid
< MAX_NR_ZONES
&& populated_zone(++zone
));
1530 pr_info("node %d initialised, %lu pages in %ums\n", nid
, nr_pages
,
1531 jiffies_to_msecs(jiffies
- start
));
1533 pgdat_init_report_one_done();
1536 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1538 void __init
page_alloc_init_late(void)
1542 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1545 /* There will be num_node_state(N_MEMORY) threads */
1546 atomic_set(&pgdat_init_n_undone
, num_node_state(N_MEMORY
));
1547 for_each_node_state(nid
, N_MEMORY
) {
1548 kthread_run(deferred_init_memmap
, NODE_DATA(nid
), "pgdatinit%d", nid
);
1551 /* Block until all are initialised */
1552 wait_for_completion(&pgdat_init_all_done_comp
);
1554 /* Reinit limits that are based on free pages after the kernel is up */
1555 files_maxfiles_init();
1558 for_each_populated_zone(zone
)
1559 set_zone_contiguous(zone
);
1563 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1564 void __init
init_cma_reserved_pageblock(struct page
*page
)
1566 unsigned i
= pageblock_nr_pages
;
1567 struct page
*p
= page
;
1570 __ClearPageReserved(p
);
1571 set_page_count(p
, 0);
1574 set_pageblock_migratetype(page
, MIGRATE_CMA
);
1576 if (pageblock_order
>= MAX_ORDER
) {
1577 i
= pageblock_nr_pages
;
1580 set_page_refcounted(p
);
1581 __free_pages(p
, MAX_ORDER
- 1);
1582 p
+= MAX_ORDER_NR_PAGES
;
1583 } while (i
-= MAX_ORDER_NR_PAGES
);
1585 set_page_refcounted(page
);
1586 __free_pages(page
, pageblock_order
);
1589 adjust_managed_page_count(page
, pageblock_nr_pages
);
1594 * The order of subdivision here is critical for the IO subsystem.
1595 * Please do not alter this order without good reasons and regression
1596 * testing. Specifically, as large blocks of memory are subdivided,
1597 * the order in which smaller blocks are delivered depends on the order
1598 * they're subdivided in this function. This is the primary factor
1599 * influencing the order in which pages are delivered to the IO
1600 * subsystem according to empirical testing, and this is also justified
1601 * by considering the behavior of a buddy system containing a single
1602 * large block of memory acted on by a series of small allocations.
1603 * This behavior is a critical factor in sglist merging's success.
1607 static inline void expand(struct zone
*zone
, struct page
*page
,
1608 int low
, int high
, struct free_area
*area
,
1611 unsigned long size
= 1 << high
;
1613 while (high
> low
) {
1617 VM_BUG_ON_PAGE(bad_range(zone
, &page
[size
]), &page
[size
]);
1619 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC
) &&
1620 debug_guardpage_enabled() &&
1621 high
< debug_guardpage_minorder()) {
1623 * Mark as guard pages (or page), that will allow to
1624 * merge back to allocator when buddy will be freed.
1625 * Corresponding page table entries will not be touched,
1626 * pages will stay not present in virtual address space
1628 set_page_guard(zone
, &page
[size
], high
, migratetype
);
1631 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
1633 set_page_order(&page
[size
], high
);
1637 static void check_new_page_bad(struct page
*page
)
1639 const char *bad_reason
= NULL
;
1640 unsigned long bad_flags
= 0;
1642 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
1643 bad_reason
= "nonzero mapcount";
1644 if (unlikely(page
->mapping
!= NULL
))
1645 bad_reason
= "non-NULL mapping";
1646 if (unlikely(page_ref_count(page
) != 0))
1647 bad_reason
= "nonzero _count";
1648 if (unlikely(page
->flags
& __PG_HWPOISON
)) {
1649 bad_reason
= "HWPoisoned (hardware-corrupted)";
1650 bad_flags
= __PG_HWPOISON
;
1651 /* Don't complain about hwpoisoned pages */
1652 page_mapcount_reset(page
); /* remove PageBuddy */
1655 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)) {
1656 bad_reason
= "PAGE_FLAGS_CHECK_AT_PREP flag set";
1657 bad_flags
= PAGE_FLAGS_CHECK_AT_PREP
;
1660 if (unlikely(page
->mem_cgroup
))
1661 bad_reason
= "page still charged to cgroup";
1663 bad_page(page
, bad_reason
, bad_flags
);
1667 * This page is about to be returned from the page allocator
1669 static inline int check_new_page(struct page
*page
)
1671 if (likely(page_expected_state(page
,
1672 PAGE_FLAGS_CHECK_AT_PREP
|__PG_HWPOISON
)))
1675 check_new_page_bad(page
);
1679 static inline bool free_pages_prezeroed(bool poisoned
)
1681 return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO
) &&
1682 page_poisoning_enabled() && poisoned
;
1685 #ifdef CONFIG_DEBUG_VM
1686 static bool check_pcp_refill(struct page
*page
)
1691 static bool check_new_pcp(struct page
*page
)
1693 return check_new_page(page
);
1696 static bool check_pcp_refill(struct page
*page
)
1698 return check_new_page(page
);
1700 static bool check_new_pcp(struct page
*page
)
1704 #endif /* CONFIG_DEBUG_VM */
1706 static bool check_new_pages(struct page
*page
, unsigned int order
)
1709 for (i
= 0; i
< (1 << order
); i
++) {
1710 struct page
*p
= page
+ i
;
1712 if (unlikely(check_new_page(p
)))
1719 static void prep_new_page(struct page
*page
, unsigned int order
, gfp_t gfp_flags
,
1720 unsigned int alloc_flags
)
1723 bool poisoned
= true;
1725 for (i
= 0; i
< (1 << order
); i
++) {
1726 struct page
*p
= page
+ i
;
1728 poisoned
&= page_is_poisoned(p
);
1731 set_page_private(page
, 0);
1732 set_page_refcounted(page
);
1734 arch_alloc_page(page
, order
);
1735 kernel_map_pages(page
, 1 << order
, 1);
1736 kernel_poison_pages(page
, 1 << order
, 1);
1737 kasan_alloc_pages(page
, order
);
1739 if (!free_pages_prezeroed(poisoned
) && (gfp_flags
& __GFP_ZERO
))
1740 for (i
= 0; i
< (1 << order
); i
++)
1741 clear_highpage(page
+ i
);
1743 if (order
&& (gfp_flags
& __GFP_COMP
))
1744 prep_compound_page(page
, order
);
1746 set_page_owner(page
, order
, gfp_flags
);
1749 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1750 * allocate the page. The expectation is that the caller is taking
1751 * steps that will free more memory. The caller should avoid the page
1752 * being used for !PFMEMALLOC purposes.
1754 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
1755 set_page_pfmemalloc(page
);
1757 clear_page_pfmemalloc(page
);
1761 * Go through the free lists for the given migratetype and remove
1762 * the smallest available page from the freelists
1765 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
1768 unsigned int current_order
;
1769 struct free_area
*area
;
1772 /* Find a page of the appropriate size in the preferred list */
1773 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
1774 area
= &(zone
->free_area
[current_order
]);
1775 page
= list_first_entry_or_null(&area
->free_list
[migratetype
],
1779 list_del(&page
->lru
);
1780 rmv_page_order(page
);
1782 expand(zone
, page
, order
, current_order
, area
, migratetype
);
1783 set_pcppage_migratetype(page
, migratetype
);
1792 * This array describes the order lists are fallen back to when
1793 * the free lists for the desirable migrate type are depleted
1795 static int fallbacks
[MIGRATE_TYPES
][4] = {
1796 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1797 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1798 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_TYPES
},
1800 [MIGRATE_CMA
] = { MIGRATE_TYPES
}, /* Never used */
1802 #ifdef CONFIG_MEMORY_ISOLATION
1803 [MIGRATE_ISOLATE
] = { MIGRATE_TYPES
}, /* Never used */
1808 static struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1811 return __rmqueue_smallest(zone
, order
, MIGRATE_CMA
);
1814 static inline struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1815 unsigned int order
) { return NULL
; }
1819 * Move the free pages in a range to the free lists of the requested type.
1820 * Note that start_page and end_pages are not aligned on a pageblock
1821 * boundary. If alignment is required, use move_freepages_block()
1823 int move_freepages(struct zone
*zone
,
1824 struct page
*start_page
, struct page
*end_page
,
1829 int pages_moved
= 0;
1831 #ifndef CONFIG_HOLES_IN_ZONE
1833 * page_zone is not safe to call in this context when
1834 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1835 * anyway as we check zone boundaries in move_freepages_block().
1836 * Remove at a later date when no bug reports exist related to
1837 * grouping pages by mobility
1839 VM_BUG_ON(page_zone(start_page
) != page_zone(end_page
));
1842 for (page
= start_page
; page
<= end_page
;) {
1843 /* Make sure we are not inadvertently changing nodes */
1844 VM_BUG_ON_PAGE(page_to_nid(page
) != zone_to_nid(zone
), page
);
1846 if (!pfn_valid_within(page_to_pfn(page
))) {
1851 if (!PageBuddy(page
)) {
1856 order
= page_order(page
);
1857 list_move(&page
->lru
,
1858 &zone
->free_area
[order
].free_list
[migratetype
]);
1860 pages_moved
+= 1 << order
;
1866 int move_freepages_block(struct zone
*zone
, struct page
*page
,
1869 unsigned long start_pfn
, end_pfn
;
1870 struct page
*start_page
, *end_page
;
1872 start_pfn
= page_to_pfn(page
);
1873 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
1874 start_page
= pfn_to_page(start_pfn
);
1875 end_page
= start_page
+ pageblock_nr_pages
- 1;
1876 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
1878 /* Do not cross zone boundaries */
1879 if (!zone_spans_pfn(zone
, start_pfn
))
1881 if (!zone_spans_pfn(zone
, end_pfn
))
1884 return move_freepages(zone
, start_page
, end_page
, migratetype
);
1887 static void change_pageblock_range(struct page
*pageblock_page
,
1888 int start_order
, int migratetype
)
1890 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
1892 while (nr_pageblocks
--) {
1893 set_pageblock_migratetype(pageblock_page
, migratetype
);
1894 pageblock_page
+= pageblock_nr_pages
;
1899 * When we are falling back to another migratetype during allocation, try to
1900 * steal extra free pages from the same pageblocks to satisfy further
1901 * allocations, instead of polluting multiple pageblocks.
1903 * If we are stealing a relatively large buddy page, it is likely there will
1904 * be more free pages in the pageblock, so try to steal them all. For
1905 * reclaimable and unmovable allocations, we steal regardless of page size,
1906 * as fragmentation caused by those allocations polluting movable pageblocks
1907 * is worse than movable allocations stealing from unmovable and reclaimable
1910 static bool can_steal_fallback(unsigned int order
, int start_mt
)
1913 * Leaving this order check is intended, although there is
1914 * relaxed order check in next check. The reason is that
1915 * we can actually steal whole pageblock if this condition met,
1916 * but, below check doesn't guarantee it and that is just heuristic
1917 * so could be changed anytime.
1919 if (order
>= pageblock_order
)
1922 if (order
>= pageblock_order
/ 2 ||
1923 start_mt
== MIGRATE_RECLAIMABLE
||
1924 start_mt
== MIGRATE_UNMOVABLE
||
1925 page_group_by_mobility_disabled
)
1932 * This function implements actual steal behaviour. If order is large enough,
1933 * we can steal whole pageblock. If not, we first move freepages in this
1934 * pageblock and check whether half of pages are moved or not. If half of
1935 * pages are moved, we can change migratetype of pageblock and permanently
1936 * use it's pages as requested migratetype in the future.
1938 static void steal_suitable_fallback(struct zone
*zone
, struct page
*page
,
1941 unsigned int current_order
= page_order(page
);
1944 /* Take ownership for orders >= pageblock_order */
1945 if (current_order
>= pageblock_order
) {
1946 change_pageblock_range(page
, current_order
, start_type
);
1950 pages
= move_freepages_block(zone
, page
, start_type
);
1952 /* Claim the whole block if over half of it is free */
1953 if (pages
>= (1 << (pageblock_order
-1)) ||
1954 page_group_by_mobility_disabled
)
1955 set_pageblock_migratetype(page
, start_type
);
1959 * Check whether there is a suitable fallback freepage with requested order.
1960 * If only_stealable is true, this function returns fallback_mt only if
1961 * we can steal other freepages all together. This would help to reduce
1962 * fragmentation due to mixed migratetype pages in one pageblock.
1964 int find_suitable_fallback(struct free_area
*area
, unsigned int order
,
1965 int migratetype
, bool only_stealable
, bool *can_steal
)
1970 if (area
->nr_free
== 0)
1975 fallback_mt
= fallbacks
[migratetype
][i
];
1976 if (fallback_mt
== MIGRATE_TYPES
)
1979 if (list_empty(&area
->free_list
[fallback_mt
]))
1982 if (can_steal_fallback(order
, migratetype
))
1985 if (!only_stealable
)
1996 * Reserve a pageblock for exclusive use of high-order atomic allocations if
1997 * there are no empty page blocks that contain a page with a suitable order
1999 static void reserve_highatomic_pageblock(struct page
*page
, struct zone
*zone
,
2000 unsigned int alloc_order
)
2003 unsigned long max_managed
, flags
;
2006 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
2007 * Check is race-prone but harmless.
2009 max_managed
= (zone
->managed_pages
/ 100) + pageblock_nr_pages
;
2010 if (zone
->nr_reserved_highatomic
>= max_managed
)
2013 spin_lock_irqsave(&zone
->lock
, flags
);
2015 /* Recheck the nr_reserved_highatomic limit under the lock */
2016 if (zone
->nr_reserved_highatomic
>= max_managed
)
2020 mt
= get_pageblock_migratetype(page
);
2021 if (mt
!= MIGRATE_HIGHATOMIC
&&
2022 !is_migrate_isolate(mt
) && !is_migrate_cma(mt
)) {
2023 zone
->nr_reserved_highatomic
+= pageblock_nr_pages
;
2024 set_pageblock_migratetype(page
, MIGRATE_HIGHATOMIC
);
2025 move_freepages_block(zone
, page
, MIGRATE_HIGHATOMIC
);
2029 spin_unlock_irqrestore(&zone
->lock
, flags
);
2033 * Used when an allocation is about to fail under memory pressure. This
2034 * potentially hurts the reliability of high-order allocations when under
2035 * intense memory pressure but failed atomic allocations should be easier
2036 * to recover from than an OOM.
2038 static void unreserve_highatomic_pageblock(const struct alloc_context
*ac
)
2040 struct zonelist
*zonelist
= ac
->zonelist
;
2041 unsigned long flags
;
2047 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
, ac
->high_zoneidx
,
2049 /* Preserve at least one pageblock */
2050 if (zone
->nr_reserved_highatomic
<= pageblock_nr_pages
)
2053 spin_lock_irqsave(&zone
->lock
, flags
);
2054 for (order
= 0; order
< MAX_ORDER
; order
++) {
2055 struct free_area
*area
= &(zone
->free_area
[order
]);
2057 page
= list_first_entry_or_null(
2058 &area
->free_list
[MIGRATE_HIGHATOMIC
],
2064 * It should never happen but changes to locking could
2065 * inadvertently allow a per-cpu drain to add pages
2066 * to MIGRATE_HIGHATOMIC while unreserving so be safe
2067 * and watch for underflows.
2069 zone
->nr_reserved_highatomic
-= min(pageblock_nr_pages
,
2070 zone
->nr_reserved_highatomic
);
2073 * Convert to ac->migratetype and avoid the normal
2074 * pageblock stealing heuristics. Minimally, the caller
2075 * is doing the work and needs the pages. More
2076 * importantly, if the block was always converted to
2077 * MIGRATE_UNMOVABLE or another type then the number
2078 * of pageblocks that cannot be completely freed
2081 set_pageblock_migratetype(page
, ac
->migratetype
);
2082 move_freepages_block(zone
, page
, ac
->migratetype
);
2083 spin_unlock_irqrestore(&zone
->lock
, flags
);
2086 spin_unlock_irqrestore(&zone
->lock
, flags
);
2090 /* Remove an element from the buddy allocator from the fallback list */
2091 static inline struct page
*
2092 __rmqueue_fallback(struct zone
*zone
, unsigned int order
, int start_migratetype
)
2094 struct free_area
*area
;
2095 unsigned int current_order
;
2100 /* Find the largest possible block of pages in the other list */
2101 for (current_order
= MAX_ORDER
-1;
2102 current_order
>= order
&& current_order
<= MAX_ORDER
-1;
2104 area
= &(zone
->free_area
[current_order
]);
2105 fallback_mt
= find_suitable_fallback(area
, current_order
,
2106 start_migratetype
, false, &can_steal
);
2107 if (fallback_mt
== -1)
2110 page
= list_first_entry(&area
->free_list
[fallback_mt
],
2113 steal_suitable_fallback(zone
, page
, start_migratetype
);
2115 /* Remove the page from the freelists */
2117 list_del(&page
->lru
);
2118 rmv_page_order(page
);
2120 expand(zone
, page
, order
, current_order
, area
,
2123 * The pcppage_migratetype may differ from pageblock's
2124 * migratetype depending on the decisions in
2125 * find_suitable_fallback(). This is OK as long as it does not
2126 * differ for MIGRATE_CMA pageblocks. Those can be used as
2127 * fallback only via special __rmqueue_cma_fallback() function
2129 set_pcppage_migratetype(page
, start_migratetype
);
2131 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
2132 start_migratetype
, fallback_mt
);
2141 * Do the hard work of removing an element from the buddy allocator.
2142 * Call me with the zone->lock already held.
2144 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
2149 page
= __rmqueue_smallest(zone
, order
, migratetype
);
2150 if (unlikely(!page
)) {
2151 if (migratetype
== MIGRATE_MOVABLE
)
2152 page
= __rmqueue_cma_fallback(zone
, order
);
2155 page
= __rmqueue_fallback(zone
, order
, migratetype
);
2158 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
2163 * Obtain a specified number of elements from the buddy allocator, all under
2164 * a single hold of the lock, for efficiency. Add them to the supplied list.
2165 * Returns the number of new pages which were placed at *list.
2167 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
2168 unsigned long count
, struct list_head
*list
,
2169 int migratetype
, bool cold
)
2173 spin_lock(&zone
->lock
);
2174 for (i
= 0; i
< count
; ++i
) {
2175 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
2176 if (unlikely(page
== NULL
))
2179 if (unlikely(check_pcp_refill(page
)))
2183 * Split buddy pages returned by expand() are received here
2184 * in physical page order. The page is added to the callers and
2185 * list and the list head then moves forward. From the callers
2186 * perspective, the linked list is ordered by page number in
2187 * some conditions. This is useful for IO devices that can
2188 * merge IO requests if the physical pages are ordered
2192 list_add(&page
->lru
, list
);
2194 list_add_tail(&page
->lru
, list
);
2196 if (is_migrate_cma(get_pcppage_migratetype(page
)))
2197 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
,
2200 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
2201 spin_unlock(&zone
->lock
);
2207 * Called from the vmstat counter updater to drain pagesets of this
2208 * currently executing processor on remote nodes after they have
2211 * Note that this function must be called with the thread pinned to
2212 * a single processor.
2214 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
2216 unsigned long flags
;
2217 int to_drain
, batch
;
2219 local_irq_save(flags
);
2220 batch
= READ_ONCE(pcp
->batch
);
2221 to_drain
= min(pcp
->count
, batch
);
2223 free_pcppages_bulk(zone
, to_drain
, pcp
);
2224 pcp
->count
-= to_drain
;
2226 local_irq_restore(flags
);
2231 * Drain pcplists of the indicated processor and zone.
2233 * The processor must either be the current processor and the
2234 * thread pinned to the current processor or a processor that
2237 static void drain_pages_zone(unsigned int cpu
, struct zone
*zone
)
2239 unsigned long flags
;
2240 struct per_cpu_pageset
*pset
;
2241 struct per_cpu_pages
*pcp
;
2243 local_irq_save(flags
);
2244 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
2248 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
2251 local_irq_restore(flags
);
2255 * Drain pcplists of all zones on the indicated processor.
2257 * The processor must either be the current processor and the
2258 * thread pinned to the current processor or a processor that
2261 static void drain_pages(unsigned int cpu
)
2265 for_each_populated_zone(zone
) {
2266 drain_pages_zone(cpu
, zone
);
2271 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
2273 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2274 * the single zone's pages.
2276 void drain_local_pages(struct zone
*zone
)
2278 int cpu
= smp_processor_id();
2281 drain_pages_zone(cpu
, zone
);
2287 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2289 * When zone parameter is non-NULL, spill just the single zone's pages.
2291 * Note that this code is protected against sending an IPI to an offline
2292 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
2293 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
2294 * nothing keeps CPUs from showing up after we populated the cpumask and
2295 * before the call to on_each_cpu_mask().
2297 void drain_all_pages(struct zone
*zone
)
2302 * Allocate in the BSS so we wont require allocation in
2303 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2305 static cpumask_t cpus_with_pcps
;
2308 * We don't care about racing with CPU hotplug event
2309 * as offline notification will cause the notified
2310 * cpu to drain that CPU pcps and on_each_cpu_mask
2311 * disables preemption as part of its processing
2313 for_each_online_cpu(cpu
) {
2314 struct per_cpu_pageset
*pcp
;
2316 bool has_pcps
= false;
2319 pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
2323 for_each_populated_zone(z
) {
2324 pcp
= per_cpu_ptr(z
->pageset
, cpu
);
2325 if (pcp
->pcp
.count
) {
2333 cpumask_set_cpu(cpu
, &cpus_with_pcps
);
2335 cpumask_clear_cpu(cpu
, &cpus_with_pcps
);
2337 on_each_cpu_mask(&cpus_with_pcps
, (smp_call_func_t
) drain_local_pages
,
2341 #ifdef CONFIG_HIBERNATION
2343 void mark_free_pages(struct zone
*zone
)
2345 unsigned long pfn
, max_zone_pfn
;
2346 unsigned long flags
;
2347 unsigned int order
, t
;
2350 if (zone_is_empty(zone
))
2353 spin_lock_irqsave(&zone
->lock
, flags
);
2355 max_zone_pfn
= zone_end_pfn(zone
);
2356 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
2357 if (pfn_valid(pfn
)) {
2358 page
= pfn_to_page(pfn
);
2360 if (page_zone(page
) != zone
)
2363 if (!swsusp_page_is_forbidden(page
))
2364 swsusp_unset_page_free(page
);
2367 for_each_migratetype_order(order
, t
) {
2368 list_for_each_entry(page
,
2369 &zone
->free_area
[order
].free_list
[t
], lru
) {
2372 pfn
= page_to_pfn(page
);
2373 for (i
= 0; i
< (1UL << order
); i
++)
2374 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
2377 spin_unlock_irqrestore(&zone
->lock
, flags
);
2379 #endif /* CONFIG_PM */
2382 * Free a 0-order page
2383 * cold == true ? free a cold page : free a hot page
2385 void free_hot_cold_page(struct page
*page
, bool cold
)
2387 struct zone
*zone
= page_zone(page
);
2388 struct per_cpu_pages
*pcp
;
2389 unsigned long flags
;
2390 unsigned long pfn
= page_to_pfn(page
);
2393 if (!free_pcp_prepare(page
))
2396 migratetype
= get_pfnblock_migratetype(page
, pfn
);
2397 set_pcppage_migratetype(page
, migratetype
);
2398 local_irq_save(flags
);
2399 __count_vm_event(PGFREE
);
2402 * We only track unmovable, reclaimable and movable on pcp lists.
2403 * Free ISOLATE pages back to the allocator because they are being
2404 * offlined but treat RESERVE as movable pages so we can get those
2405 * areas back if necessary. Otherwise, we may have to free
2406 * excessively into the page allocator
2408 if (migratetype
>= MIGRATE_PCPTYPES
) {
2409 if (unlikely(is_migrate_isolate(migratetype
))) {
2410 free_one_page(zone
, page
, pfn
, 0, migratetype
);
2413 migratetype
= MIGRATE_MOVABLE
;
2416 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2418 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
2420 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
2422 if (pcp
->count
>= pcp
->high
) {
2423 unsigned long batch
= READ_ONCE(pcp
->batch
);
2424 free_pcppages_bulk(zone
, batch
, pcp
);
2425 pcp
->count
-= batch
;
2429 local_irq_restore(flags
);
2433 * Free a list of 0-order pages
2435 void free_hot_cold_page_list(struct list_head
*list
, bool cold
)
2437 struct page
*page
, *next
;
2439 list_for_each_entry_safe(page
, next
, list
, lru
) {
2440 trace_mm_page_free_batched(page
, cold
);
2441 free_hot_cold_page(page
, cold
);
2446 * split_page takes a non-compound higher-order page, and splits it into
2447 * n (1<<order) sub-pages: page[0..n]
2448 * Each sub-page must be freed individually.
2450 * Note: this is probably too low level an operation for use in drivers.
2451 * Please consult with lkml before using this in your driver.
2453 void split_page(struct page
*page
, unsigned int order
)
2458 VM_BUG_ON_PAGE(PageCompound(page
), page
);
2459 VM_BUG_ON_PAGE(!page_count(page
), page
);
2461 #ifdef CONFIG_KMEMCHECK
2463 * Split shadow pages too, because free(page[0]) would
2464 * otherwise free the whole shadow.
2466 if (kmemcheck_page_is_tracked(page
))
2467 split_page(virt_to_page(page
[0].shadow
), order
);
2470 gfp_mask
= get_page_owner_gfp(page
);
2471 set_page_owner(page
, 0, gfp_mask
);
2472 for (i
= 1; i
< (1 << order
); i
++) {
2473 set_page_refcounted(page
+ i
);
2474 set_page_owner(page
+ i
, 0, gfp_mask
);
2477 EXPORT_SYMBOL_GPL(split_page
);
2479 int __isolate_free_page(struct page
*page
, unsigned int order
)
2481 unsigned long watermark
;
2485 BUG_ON(!PageBuddy(page
));
2487 zone
= page_zone(page
);
2488 mt
= get_pageblock_migratetype(page
);
2490 if (!is_migrate_isolate(mt
)) {
2491 /* Obey watermarks as if the page was being allocated */
2492 watermark
= low_wmark_pages(zone
) + (1 << order
);
2493 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
2496 __mod_zone_freepage_state(zone
, -(1UL << order
), mt
);
2499 /* Remove page from free list */
2500 list_del(&page
->lru
);
2501 zone
->free_area
[order
].nr_free
--;
2502 rmv_page_order(page
);
2504 set_page_owner(page
, order
, __GFP_MOVABLE
);
2506 /* Set the pageblock if the isolated page is at least a pageblock */
2507 if (order
>= pageblock_order
- 1) {
2508 struct page
*endpage
= page
+ (1 << order
) - 1;
2509 for (; page
< endpage
; page
+= pageblock_nr_pages
) {
2510 int mt
= get_pageblock_migratetype(page
);
2511 if (!is_migrate_isolate(mt
) && !is_migrate_cma(mt
))
2512 set_pageblock_migratetype(page
,
2518 return 1UL << order
;
2522 * Similar to split_page except the page is already free. As this is only
2523 * being used for migration, the migratetype of the block also changes.
2524 * As this is called with interrupts disabled, the caller is responsible
2525 * for calling arch_alloc_page() and kernel_map_page() after interrupts
2528 * Note: this is probably too low level an operation for use in drivers.
2529 * Please consult with lkml before using this in your driver.
2531 int split_free_page(struct page
*page
)
2536 order
= page_order(page
);
2538 nr_pages
= __isolate_free_page(page
, order
);
2542 /* Split into individual pages */
2543 set_page_refcounted(page
);
2544 split_page(page
, order
);
2549 * Update NUMA hit/miss statistics
2551 * Must be called with interrupts disabled.
2553 * When __GFP_OTHER_NODE is set assume the node of the preferred
2554 * zone is the local node. This is useful for daemons who allocate
2555 * memory on behalf of other processes.
2557 static inline void zone_statistics(struct zone
*preferred_zone
, struct zone
*z
,
2561 int local_nid
= numa_node_id();
2562 enum zone_stat_item local_stat
= NUMA_LOCAL
;
2564 if (unlikely(flags
& __GFP_OTHER_NODE
)) {
2565 local_stat
= NUMA_OTHER
;
2566 local_nid
= preferred_zone
->node
;
2569 if (z
->node
== local_nid
) {
2570 __inc_zone_state(z
, NUMA_HIT
);
2571 __inc_zone_state(z
, local_stat
);
2573 __inc_zone_state(z
, NUMA_MISS
);
2574 __inc_zone_state(preferred_zone
, NUMA_FOREIGN
);
2580 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2583 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
2584 struct zone
*zone
, unsigned int order
,
2585 gfp_t gfp_flags
, unsigned int alloc_flags
,
2588 unsigned long flags
;
2590 bool cold
= ((gfp_flags
& __GFP_COLD
) != 0);
2592 if (likely(order
== 0)) {
2593 struct per_cpu_pages
*pcp
;
2594 struct list_head
*list
;
2596 local_irq_save(flags
);
2598 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2599 list
= &pcp
->lists
[migratetype
];
2600 if (list_empty(list
)) {
2601 pcp
->count
+= rmqueue_bulk(zone
, 0,
2604 if (unlikely(list_empty(list
)))
2609 page
= list_last_entry(list
, struct page
, lru
);
2611 page
= list_first_entry(list
, struct page
, lru
);
2612 } while (page
&& check_new_pcp(page
));
2614 __dec_zone_state(zone
, NR_ALLOC_BATCH
);
2615 list_del(&page
->lru
);
2619 * We most definitely don't want callers attempting to
2620 * allocate greater than order-1 page units with __GFP_NOFAIL.
2622 WARN_ON_ONCE((gfp_flags
& __GFP_NOFAIL
) && (order
> 1));
2623 spin_lock_irqsave(&zone
->lock
, flags
);
2627 if (alloc_flags
& ALLOC_HARDER
) {
2628 page
= __rmqueue_smallest(zone
, order
, MIGRATE_HIGHATOMIC
);
2630 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
2633 page
= __rmqueue(zone
, order
, migratetype
);
2634 } while (page
&& check_new_pages(page
, order
));
2635 spin_unlock(&zone
->lock
);
2638 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
, -(1 << order
));
2639 __mod_zone_freepage_state(zone
, -(1 << order
),
2640 get_pcppage_migratetype(page
));
2643 if (atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]) <= 0 &&
2644 !test_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
))
2645 set_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
);
2647 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
2648 zone_statistics(preferred_zone
, zone
, gfp_flags
);
2649 local_irq_restore(flags
);
2651 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
2655 local_irq_restore(flags
);
2659 #ifdef CONFIG_FAIL_PAGE_ALLOC
2662 struct fault_attr attr
;
2664 bool ignore_gfp_highmem
;
2665 bool ignore_gfp_reclaim
;
2667 } fail_page_alloc
= {
2668 .attr
= FAULT_ATTR_INITIALIZER
,
2669 .ignore_gfp_reclaim
= true,
2670 .ignore_gfp_highmem
= true,
2674 static int __init
setup_fail_page_alloc(char *str
)
2676 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
2678 __setup("fail_page_alloc=", setup_fail_page_alloc
);
2680 static bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2682 if (order
< fail_page_alloc
.min_order
)
2684 if (gfp_mask
& __GFP_NOFAIL
)
2686 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
2688 if (fail_page_alloc
.ignore_gfp_reclaim
&&
2689 (gfp_mask
& __GFP_DIRECT_RECLAIM
))
2692 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
2695 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2697 static int __init
fail_page_alloc_debugfs(void)
2699 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
2702 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
2703 &fail_page_alloc
.attr
);
2705 return PTR_ERR(dir
);
2707 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
2708 &fail_page_alloc
.ignore_gfp_reclaim
))
2710 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
2711 &fail_page_alloc
.ignore_gfp_highmem
))
2713 if (!debugfs_create_u32("min-order", mode
, dir
,
2714 &fail_page_alloc
.min_order
))
2719 debugfs_remove_recursive(dir
);
2724 late_initcall(fail_page_alloc_debugfs
);
2726 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2728 #else /* CONFIG_FAIL_PAGE_ALLOC */
2730 static inline bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2735 #endif /* CONFIG_FAIL_PAGE_ALLOC */
2738 * Return true if free base pages are above 'mark'. For high-order checks it
2739 * will return true of the order-0 watermark is reached and there is at least
2740 * one free page of a suitable size. Checking now avoids taking the zone lock
2741 * to check in the allocation paths if no pages are free.
2743 bool __zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
2744 int classzone_idx
, unsigned int alloc_flags
,
2749 const bool alloc_harder
= (alloc_flags
& ALLOC_HARDER
);
2751 /* free_pages may go negative - that's OK */
2752 free_pages
-= (1 << order
) - 1;
2754 if (alloc_flags
& ALLOC_HIGH
)
2758 * If the caller does not have rights to ALLOC_HARDER then subtract
2759 * the high-atomic reserves. This will over-estimate the size of the
2760 * atomic reserve but it avoids a search.
2762 if (likely(!alloc_harder
))
2763 free_pages
-= z
->nr_reserved_highatomic
;
2768 /* If allocation can't use CMA areas don't use free CMA pages */
2769 if (!(alloc_flags
& ALLOC_CMA
))
2770 free_pages
-= zone_page_state(z
, NR_FREE_CMA_PAGES
);
2774 * Check watermarks for an order-0 allocation request. If these
2775 * are not met, then a high-order request also cannot go ahead
2776 * even if a suitable page happened to be free.
2778 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
2781 /* If this is an order-0 request then the watermark is fine */
2785 /* For a high-order request, check at least one suitable page is free */
2786 for (o
= order
; o
< MAX_ORDER
; o
++) {
2787 struct free_area
*area
= &z
->free_area
[o
];
2796 for (mt
= 0; mt
< MIGRATE_PCPTYPES
; mt
++) {
2797 if (!list_empty(&area
->free_list
[mt
]))
2802 if ((alloc_flags
& ALLOC_CMA
) &&
2803 !list_empty(&area
->free_list
[MIGRATE_CMA
])) {
2811 bool zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
2812 int classzone_idx
, unsigned int alloc_flags
)
2814 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
2815 zone_page_state(z
, NR_FREE_PAGES
));
2818 static inline bool zone_watermark_fast(struct zone
*z
, unsigned int order
,
2819 unsigned long mark
, int classzone_idx
, unsigned int alloc_flags
)
2821 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
2825 /* If allocation can't use CMA areas don't use free CMA pages */
2826 if (!(alloc_flags
& ALLOC_CMA
))
2827 cma_pages
= zone_page_state(z
, NR_FREE_CMA_PAGES
);
2831 * Fast check for order-0 only. If this fails then the reserves
2832 * need to be calculated. There is a corner case where the check
2833 * passes but only the high-order atomic reserve are free. If
2834 * the caller is !atomic then it'll uselessly search the free
2835 * list. That corner case is then slower but it is harmless.
2837 if (!order
&& (free_pages
- cma_pages
) > mark
+ z
->lowmem_reserve
[classzone_idx
])
2840 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
2844 bool zone_watermark_ok_safe(struct zone
*z
, unsigned int order
,
2845 unsigned long mark
, int classzone_idx
)
2847 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
2849 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
2850 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
2852 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, 0,
2857 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
2859 return local_zone
->node
== zone
->node
;
2862 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2864 return node_distance(zone_to_nid(local_zone
), zone_to_nid(zone
)) <
2867 #else /* CONFIG_NUMA */
2868 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
2873 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2877 #endif /* CONFIG_NUMA */
2879 static void reset_alloc_batches(struct zone
*preferred_zone
)
2881 struct zone
*zone
= preferred_zone
->zone_pgdat
->node_zones
;
2884 mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
2885 high_wmark_pages(zone
) - low_wmark_pages(zone
) -
2886 atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]));
2887 clear_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
);
2888 } while (zone
++ != preferred_zone
);
2892 * get_page_from_freelist goes through the zonelist trying to allocate
2895 static struct page
*
2896 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
, int alloc_flags
,
2897 const struct alloc_context
*ac
)
2899 struct zoneref
*z
= ac
->preferred_zoneref
;
2901 bool fair_skipped
= false;
2902 bool apply_fair
= (alloc_flags
& ALLOC_FAIR
);
2906 * Scan zonelist, looking for a zone with enough free.
2907 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2909 for_next_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
, ac
->high_zoneidx
,
2914 if (cpusets_enabled() &&
2915 (alloc_flags
& ALLOC_CPUSET
) &&
2916 !__cpuset_zone_allowed(zone
, gfp_mask
))
2919 * Distribute pages in proportion to the individual
2920 * zone size to ensure fair page aging. The zone a
2921 * page was allocated in should have no effect on the
2922 * time the page has in memory before being reclaimed.
2925 if (test_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
)) {
2926 fair_skipped
= true;
2929 if (!zone_local(ac
->preferred_zoneref
->zone
, zone
)) {
2936 * When allocating a page cache page for writing, we
2937 * want to get it from a zone that is within its dirty
2938 * limit, such that no single zone holds more than its
2939 * proportional share of globally allowed dirty pages.
2940 * The dirty limits take into account the zone's
2941 * lowmem reserves and high watermark so that kswapd
2942 * should be able to balance it without having to
2943 * write pages from its LRU list.
2945 * This may look like it could increase pressure on
2946 * lower zones by failing allocations in higher zones
2947 * before they are full. But the pages that do spill
2948 * over are limited as the lower zones are protected
2949 * by this very same mechanism. It should not become
2950 * a practical burden to them.
2952 * XXX: For now, allow allocations to potentially
2953 * exceed the per-zone dirty limit in the slowpath
2954 * (spread_dirty_pages unset) before going into reclaim,
2955 * which is important when on a NUMA setup the allowed
2956 * zones are together not big enough to reach the
2957 * global limit. The proper fix for these situations
2958 * will require awareness of zones in the
2959 * dirty-throttling and the flusher threads.
2961 if (ac
->spread_dirty_pages
&& !zone_dirty_ok(zone
))
2964 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
2965 if (!zone_watermark_fast(zone
, order
, mark
,
2966 ac_classzone_idx(ac
), alloc_flags
)) {
2969 /* Checked here to keep the fast path fast */
2970 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
2971 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
2974 if (zone_reclaim_mode
== 0 ||
2975 !zone_allows_reclaim(ac
->preferred_zoneref
->zone
, zone
))
2978 ret
= zone_reclaim(zone
, gfp_mask
, order
);
2980 case ZONE_RECLAIM_NOSCAN
:
2983 case ZONE_RECLAIM_FULL
:
2984 /* scanned but unreclaimable */
2987 /* did we reclaim enough */
2988 if (zone_watermark_ok(zone
, order
, mark
,
2989 ac_classzone_idx(ac
), alloc_flags
))
2997 page
= buffered_rmqueue(ac
->preferred_zoneref
->zone
, zone
, order
,
2998 gfp_mask
, alloc_flags
, ac
->migratetype
);
3000 prep_new_page(page
, order
, gfp_mask
, alloc_flags
);
3003 * If this is a high-order atomic allocation then check
3004 * if the pageblock should be reserved for the future
3006 if (unlikely(order
&& (alloc_flags
& ALLOC_HARDER
)))
3007 reserve_highatomic_pageblock(page
, zone
, order
);
3014 * The first pass makes sure allocations are spread fairly within the
3015 * local node. However, the local node might have free pages left
3016 * after the fairness batches are exhausted, and remote zones haven't
3017 * even been considered yet. Try once more without fairness, and
3018 * include remote zones now, before entering the slowpath and waking
3019 * kswapd: prefer spilling to a remote zone over swapping locally.
3024 fair_skipped
= false;
3025 reset_alloc_batches(ac
->preferred_zoneref
->zone
);
3033 * Large machines with many possible nodes should not always dump per-node
3034 * meminfo in irq context.
3036 static inline bool should_suppress_show_mem(void)
3041 ret
= in_interrupt();
3046 static DEFINE_RATELIMIT_STATE(nopage_rs
,
3047 DEFAULT_RATELIMIT_INTERVAL
,
3048 DEFAULT_RATELIMIT_BURST
);
3050 void warn_alloc_failed(gfp_t gfp_mask
, unsigned int order
, const char *fmt
, ...)
3052 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
3054 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
) ||
3055 debug_guardpage_minorder() > 0)
3059 * This documents exceptions given to allocations in certain
3060 * contexts that are allowed to allocate outside current's set
3063 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
3064 if (test_thread_flag(TIF_MEMDIE
) ||
3065 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
3066 filter
&= ~SHOW_MEM_FILTER_NODES
;
3067 if (in_interrupt() || !(gfp_mask
& __GFP_DIRECT_RECLAIM
))
3068 filter
&= ~SHOW_MEM_FILTER_NODES
;
3071 struct va_format vaf
;
3074 va_start(args
, fmt
);
3079 pr_warn("%pV", &vaf
);
3084 pr_warn("%s: page allocation failure: order:%u, mode:%#x(%pGg)\n",
3085 current
->comm
, order
, gfp_mask
, &gfp_mask
);
3087 if (!should_suppress_show_mem())
3091 static inline struct page
*
3092 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
3093 const struct alloc_context
*ac
, unsigned long *did_some_progress
)
3095 struct oom_control oc
= {
3096 .zonelist
= ac
->zonelist
,
3097 .nodemask
= ac
->nodemask
,
3098 .gfp_mask
= gfp_mask
,
3103 *did_some_progress
= 0;
3106 * Acquire the oom lock. If that fails, somebody else is
3107 * making progress for us.
3109 if (!mutex_trylock(&oom_lock
)) {
3110 *did_some_progress
= 1;
3111 schedule_timeout_uninterruptible(1);
3116 * Go through the zonelist yet one more time, keep very high watermark
3117 * here, this is only to catch a parallel oom killing, we must fail if
3118 * we're still under heavy pressure.
3120 page
= get_page_from_freelist(gfp_mask
| __GFP_HARDWALL
, order
,
3121 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
, ac
);
3125 if (!(gfp_mask
& __GFP_NOFAIL
)) {
3126 /* Coredumps can quickly deplete all memory reserves */
3127 if (current
->flags
& PF_DUMPCORE
)
3129 /* The OOM killer will not help higher order allocs */
3130 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
3132 /* The OOM killer does not needlessly kill tasks for lowmem */
3133 if (ac
->high_zoneidx
< ZONE_NORMAL
)
3135 if (pm_suspended_storage())
3138 * XXX: GFP_NOFS allocations should rather fail than rely on
3139 * other request to make a forward progress.
3140 * We are in an unfortunate situation where out_of_memory cannot
3141 * do much for this context but let's try it to at least get
3142 * access to memory reserved if the current task is killed (see
3143 * out_of_memory). Once filesystems are ready to handle allocation
3144 * failures more gracefully we should just bail out here.
3147 /* The OOM killer may not free memory on a specific node */
3148 if (gfp_mask
& __GFP_THISNODE
)
3151 /* Exhausted what can be done so it's blamo time */
3152 if (out_of_memory(&oc
) || WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
)) {
3153 *did_some_progress
= 1;
3155 if (gfp_mask
& __GFP_NOFAIL
) {
3156 page
= get_page_from_freelist(gfp_mask
, order
,
3157 ALLOC_NO_WATERMARKS
|ALLOC_CPUSET
, ac
);
3159 * fallback to ignore cpuset restriction if our nodes
3163 page
= get_page_from_freelist(gfp_mask
, order
,
3164 ALLOC_NO_WATERMARKS
, ac
);
3168 mutex_unlock(&oom_lock
);
3174 * Maximum number of compaction retries wit a progress before OOM
3175 * killer is consider as the only way to move forward.
3177 #define MAX_COMPACT_RETRIES 16
3179 #ifdef CONFIG_COMPACTION
3180 /* Try memory compaction for high-order allocations before reclaim */
3181 static struct page
*
3182 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
3183 unsigned int alloc_flags
, const struct alloc_context
*ac
,
3184 enum migrate_mode mode
, enum compact_result
*compact_result
)
3187 int contended_compaction
;
3192 current
->flags
|= PF_MEMALLOC
;
3193 *compact_result
= try_to_compact_pages(gfp_mask
, order
, alloc_flags
, ac
,
3194 mode
, &contended_compaction
);
3195 current
->flags
&= ~PF_MEMALLOC
;
3197 if (*compact_result
<= COMPACT_INACTIVE
)
3201 * At least in one zone compaction wasn't deferred or skipped, so let's
3202 * count a compaction stall
3204 count_vm_event(COMPACTSTALL
);
3206 page
= get_page_from_freelist(gfp_mask
, order
,
3207 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
3210 struct zone
*zone
= page_zone(page
);
3212 zone
->compact_blockskip_flush
= false;
3213 compaction_defer_reset(zone
, order
, true);
3214 count_vm_event(COMPACTSUCCESS
);
3219 * It's bad if compaction run occurs and fails. The most likely reason
3220 * is that pages exist, but not enough to satisfy watermarks.
3222 count_vm_event(COMPACTFAIL
);
3225 * In all zones where compaction was attempted (and not
3226 * deferred or skipped), lock contention has been detected.
3227 * For THP allocation we do not want to disrupt the others
3228 * so we fallback to base pages instead.
3230 if (contended_compaction
== COMPACT_CONTENDED_LOCK
)
3231 *compact_result
= COMPACT_CONTENDED
;
3234 * If compaction was aborted due to need_resched(), we do not
3235 * want to further increase allocation latency, unless it is
3236 * khugepaged trying to collapse.
3238 if (contended_compaction
== COMPACT_CONTENDED_SCHED
3239 && !(current
->flags
& PF_KTHREAD
))
3240 *compact_result
= COMPACT_CONTENDED
;
3248 should_compact_retry(struct alloc_context
*ac
, int order
, int alloc_flags
,
3249 enum compact_result compact_result
, enum migrate_mode
*migrate_mode
,
3250 int compaction_retries
)
3252 int max_retries
= MAX_COMPACT_RETRIES
;
3258 * compaction considers all the zone as desperately out of memory
3259 * so it doesn't really make much sense to retry except when the
3260 * failure could be caused by weak migration mode.
3262 if (compaction_failed(compact_result
)) {
3263 if (*migrate_mode
== MIGRATE_ASYNC
) {
3264 *migrate_mode
= MIGRATE_SYNC_LIGHT
;
3271 * make sure the compaction wasn't deferred or didn't bail out early
3272 * due to locks contention before we declare that we should give up.
3273 * But do not retry if the given zonelist is not suitable for
3276 if (compaction_withdrawn(compact_result
))
3277 return compaction_zonelist_suitable(ac
, order
, alloc_flags
);
3280 * !costly requests are much more important than __GFP_REPEAT
3281 * costly ones because they are de facto nofail and invoke OOM
3282 * killer to move on while costly can fail and users are ready
3283 * to cope with that. 1/4 retries is rather arbitrary but we
3284 * would need much more detailed feedback from compaction to
3285 * make a better decision.
3287 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
3289 if (compaction_retries
<= max_retries
)
3295 static inline struct page
*
3296 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
3297 unsigned int alloc_flags
, const struct alloc_context
*ac
,
3298 enum migrate_mode mode
, enum compact_result
*compact_result
)
3300 *compact_result
= COMPACT_SKIPPED
;
3305 should_compact_retry(struct alloc_context
*ac
, unsigned int order
, int alloc_flags
,
3306 enum compact_result compact_result
,
3307 enum migrate_mode
*migrate_mode
,
3308 int compaction_retries
)
3313 if (!order
|| order
> PAGE_ALLOC_COSTLY_ORDER
)
3317 * There are setups with compaction disabled which would prefer to loop
3318 * inside the allocator rather than hit the oom killer prematurely.
3319 * Let's give them a good hope and keep retrying while the order-0
3320 * watermarks are OK.
3322 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
, ac
->high_zoneidx
,
3324 if (zone_watermark_ok(zone
, 0, min_wmark_pages(zone
),
3325 ac_classzone_idx(ac
), alloc_flags
))
3330 #endif /* CONFIG_COMPACTION */
3332 /* Perform direct synchronous page reclaim */
3334 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
,
3335 const struct alloc_context
*ac
)
3337 struct reclaim_state reclaim_state
;
3342 /* We now go into synchronous reclaim */
3343 cpuset_memory_pressure_bump();
3344 current
->flags
|= PF_MEMALLOC
;
3345 lockdep_set_current_reclaim_state(gfp_mask
);
3346 reclaim_state
.reclaimed_slab
= 0;
3347 current
->reclaim_state
= &reclaim_state
;
3349 progress
= try_to_free_pages(ac
->zonelist
, order
, gfp_mask
,
3352 current
->reclaim_state
= NULL
;
3353 lockdep_clear_current_reclaim_state();
3354 current
->flags
&= ~PF_MEMALLOC
;
3361 /* The really slow allocator path where we enter direct reclaim */
3362 static inline struct page
*
3363 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
3364 unsigned int alloc_flags
, const struct alloc_context
*ac
,
3365 unsigned long *did_some_progress
)
3367 struct page
*page
= NULL
;
3368 bool drained
= false;
3370 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, ac
);
3371 if (unlikely(!(*did_some_progress
)))
3375 page
= get_page_from_freelist(gfp_mask
, order
,
3376 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
3379 * If an allocation failed after direct reclaim, it could be because
3380 * pages are pinned on the per-cpu lists or in high alloc reserves.
3381 * Shrink them them and try again
3383 if (!page
&& !drained
) {
3384 unreserve_highatomic_pageblock(ac
);
3385 drain_all_pages(NULL
);
3393 static void wake_all_kswapds(unsigned int order
, const struct alloc_context
*ac
)
3398 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
,
3399 ac
->high_zoneidx
, ac
->nodemask
)
3400 wakeup_kswapd(zone
, order
, ac_classzone_idx(ac
));
3403 static inline unsigned int
3404 gfp_to_alloc_flags(gfp_t gfp_mask
)
3406 unsigned int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
3408 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
3409 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
3412 * The caller may dip into page reserves a bit more if the caller
3413 * cannot run direct reclaim, or if the caller has realtime scheduling
3414 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
3415 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
3417 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
3419 if (gfp_mask
& __GFP_ATOMIC
) {
3421 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
3422 * if it can't schedule.
3424 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
3425 alloc_flags
|= ALLOC_HARDER
;
3427 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
3428 * comment for __cpuset_node_allowed().
3430 alloc_flags
&= ~ALLOC_CPUSET
;
3431 } else if (unlikely(rt_task(current
)) && !in_interrupt())
3432 alloc_flags
|= ALLOC_HARDER
;
3434 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
3435 if (gfp_mask
& __GFP_MEMALLOC
)
3436 alloc_flags
|= ALLOC_NO_WATERMARKS
;
3437 else if (in_serving_softirq() && (current
->flags
& PF_MEMALLOC
))
3438 alloc_flags
|= ALLOC_NO_WATERMARKS
;
3439 else if (!in_interrupt() &&
3440 ((current
->flags
& PF_MEMALLOC
) ||
3441 unlikely(test_thread_flag(TIF_MEMDIE
))))
3442 alloc_flags
|= ALLOC_NO_WATERMARKS
;
3445 if (gfpflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
3446 alloc_flags
|= ALLOC_CMA
;
3451 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask
)
3453 return !!(gfp_to_alloc_flags(gfp_mask
) & ALLOC_NO_WATERMARKS
);
3456 static inline bool is_thp_gfp_mask(gfp_t gfp_mask
)
3458 return (gfp_mask
& (GFP_TRANSHUGE
| __GFP_KSWAPD_RECLAIM
)) == GFP_TRANSHUGE
;
3462 * Maximum number of reclaim retries without any progress before OOM killer
3463 * is consider as the only way to move forward.
3465 #define MAX_RECLAIM_RETRIES 16
3468 * Checks whether it makes sense to retry the reclaim to make a forward progress
3469 * for the given allocation request.
3470 * The reclaim feedback represented by did_some_progress (any progress during
3471 * the last reclaim round) and no_progress_loops (number of reclaim rounds without
3472 * any progress in a row) is considered as well as the reclaimable pages on the
3473 * applicable zone list (with a backoff mechanism which is a function of
3474 * no_progress_loops).
3476 * Returns true if a retry is viable or false to enter the oom path.
3479 should_reclaim_retry(gfp_t gfp_mask
, unsigned order
,
3480 struct alloc_context
*ac
, int alloc_flags
,
3481 bool did_some_progress
, int no_progress_loops
)
3487 * Make sure we converge to OOM if we cannot make any progress
3488 * several times in the row.
3490 if (no_progress_loops
> MAX_RECLAIM_RETRIES
)
3494 * Keep reclaiming pages while there is a chance this will lead somewhere.
3495 * If none of the target zones can satisfy our allocation request even
3496 * if all reclaimable pages are considered then we are screwed and have
3499 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
, ac
->high_zoneidx
,
3501 unsigned long available
;
3502 unsigned long reclaimable
;
3504 available
= reclaimable
= zone_reclaimable_pages(zone
);
3505 available
-= DIV_ROUND_UP(no_progress_loops
* available
,
3506 MAX_RECLAIM_RETRIES
);
3507 available
+= zone_page_state_snapshot(zone
, NR_FREE_PAGES
);
3510 * Would the allocation succeed if we reclaimed the whole
3513 if (__zone_watermark_ok(zone
, order
, min_wmark_pages(zone
),
3514 ac_classzone_idx(ac
), alloc_flags
, available
)) {
3516 * If we didn't make any progress and have a lot of
3517 * dirty + writeback pages then we should wait for
3518 * an IO to complete to slow down the reclaim and
3519 * prevent from pre mature OOM
3521 if (!did_some_progress
) {
3522 unsigned long writeback
;
3523 unsigned long dirty
;
3525 writeback
= zone_page_state_snapshot(zone
,
3527 dirty
= zone_page_state_snapshot(zone
, NR_FILE_DIRTY
);
3529 if (2*(writeback
+ dirty
) > reclaimable
) {
3530 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3536 * Memory allocation/reclaim might be called from a WQ
3537 * context and the current implementation of the WQ
3538 * concurrency control doesn't recognize that
3539 * a particular WQ is congested if the worker thread is
3540 * looping without ever sleeping. Therefore we have to
3541 * do a short sleep here rather than calling
3544 if (current
->flags
& PF_WQ_WORKER
)
3545 schedule_timeout_uninterruptible(1);
3556 static inline struct page
*
3557 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
3558 struct alloc_context
*ac
)
3560 bool can_direct_reclaim
= gfp_mask
& __GFP_DIRECT_RECLAIM
;
3561 struct page
*page
= NULL
;
3562 unsigned int alloc_flags
;
3563 unsigned long did_some_progress
;
3564 enum migrate_mode migration_mode
= MIGRATE_ASYNC
;
3565 enum compact_result compact_result
;
3566 int compaction_retries
= 0;
3567 int no_progress_loops
= 0;
3570 * In the slowpath, we sanity check order to avoid ever trying to
3571 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3572 * be using allocators in order of preference for an area that is
3575 if (order
>= MAX_ORDER
) {
3576 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
3581 * We also sanity check to catch abuse of atomic reserves being used by
3582 * callers that are not in atomic context.
3584 if (WARN_ON_ONCE((gfp_mask
& (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)) ==
3585 (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)))
3586 gfp_mask
&= ~__GFP_ATOMIC
;
3589 if (gfp_mask
& __GFP_KSWAPD_RECLAIM
)
3590 wake_all_kswapds(order
, ac
);
3593 * OK, we're below the kswapd watermark and have kicked background
3594 * reclaim. Now things get more complex, so set up alloc_flags according
3595 * to how we want to proceed.
3597 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
3599 /* This is the last chance, in general, before the goto nopage. */
3600 page
= get_page_from_freelist(gfp_mask
, order
,
3601 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
3605 /* Allocate without watermarks if the context allows */
3606 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
3608 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3609 * the allocation is high priority and these type of
3610 * allocations are system rather than user orientated
3612 ac
->zonelist
= node_zonelist(numa_node_id(), gfp_mask
);
3613 page
= get_page_from_freelist(gfp_mask
, order
,
3614 ALLOC_NO_WATERMARKS
, ac
);
3619 /* Caller is not willing to reclaim, we can't balance anything */
3620 if (!can_direct_reclaim
) {
3622 * All existing users of the __GFP_NOFAIL are blockable, so warn
3623 * of any new users that actually allow this type of allocation
3626 WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
);
3630 /* Avoid recursion of direct reclaim */
3631 if (current
->flags
& PF_MEMALLOC
) {
3633 * __GFP_NOFAIL request from this context is rather bizarre
3634 * because we cannot reclaim anything and only can loop waiting
3635 * for somebody to do a work for us.
3637 if (WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
)) {
3644 /* Avoid allocations with no watermarks from looping endlessly */
3645 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
3649 * Try direct compaction. The first pass is asynchronous. Subsequent
3650 * attempts after direct reclaim are synchronous
3652 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
, ac
,
3658 /* Checks for THP-specific high-order allocations */
3659 if (is_thp_gfp_mask(gfp_mask
)) {
3661 * If compaction is deferred for high-order allocations, it is
3662 * because sync compaction recently failed. If this is the case
3663 * and the caller requested a THP allocation, we do not want
3664 * to heavily disrupt the system, so we fail the allocation
3665 * instead of entering direct reclaim.
3667 if (compact_result
== COMPACT_DEFERRED
)
3671 * Compaction is contended so rather back off than cause
3674 if(compact_result
== COMPACT_CONTENDED
)
3678 if (order
&& compaction_made_progress(compact_result
))
3679 compaction_retries
++;
3681 /* Try direct reclaim and then allocating */
3682 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
, alloc_flags
, ac
,
3683 &did_some_progress
);
3687 /* Do not loop if specifically requested */
3688 if (gfp_mask
& __GFP_NORETRY
)
3692 * Do not retry costly high order allocations unless they are
3695 if (order
> PAGE_ALLOC_COSTLY_ORDER
&& !(gfp_mask
& __GFP_REPEAT
))
3699 * Costly allocations might have made a progress but this doesn't mean
3700 * their order will become available due to high fragmentation so
3701 * always increment the no progress counter for them
3703 if (did_some_progress
&& order
<= PAGE_ALLOC_COSTLY_ORDER
)
3704 no_progress_loops
= 0;
3706 no_progress_loops
++;
3708 if (should_reclaim_retry(gfp_mask
, order
, ac
, alloc_flags
,
3709 did_some_progress
> 0, no_progress_loops
))
3713 * It doesn't make any sense to retry for the compaction if the order-0
3714 * reclaim is not able to make any progress because the current
3715 * implementation of the compaction depends on the sufficient amount
3716 * of free memory (see __compaction_suitable)
3718 if (did_some_progress
> 0 &&
3719 should_compact_retry(ac
, order
, alloc_flags
,
3720 compact_result
, &migration_mode
,
3721 compaction_retries
))
3724 /* Reclaim has failed us, start killing things */
3725 page
= __alloc_pages_may_oom(gfp_mask
, order
, ac
, &did_some_progress
);
3729 /* Retry as long as the OOM killer is making progress */
3730 if (did_some_progress
) {
3731 no_progress_loops
= 0;
3737 * High-order allocations do not necessarily loop after direct reclaim
3738 * and reclaim/compaction depends on compaction being called after
3739 * reclaim so call directly if necessary.
3740 * It can become very expensive to allocate transparent hugepages at
3741 * fault, so use asynchronous memory compaction for THP unless it is
3742 * khugepaged trying to collapse. All other requests should tolerate
3743 * at least light sync migration.
3745 if (is_thp_gfp_mask(gfp_mask
) && !(current
->flags
& PF_KTHREAD
))
3746 migration_mode
= MIGRATE_ASYNC
;
3748 migration_mode
= MIGRATE_SYNC_LIGHT
;
3749 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
,
3755 warn_alloc_failed(gfp_mask
, order
, NULL
);
3761 * This is the 'heart' of the zoned buddy allocator.
3764 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
3765 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
3768 unsigned int cpuset_mems_cookie
;
3769 unsigned int alloc_flags
= ALLOC_WMARK_LOW
|ALLOC_FAIR
;
3770 gfp_t alloc_mask
= gfp_mask
; /* The gfp_t that was actually used for allocation */
3771 struct alloc_context ac
= {
3772 .high_zoneidx
= gfp_zone(gfp_mask
),
3773 .zonelist
= zonelist
,
3774 .nodemask
= nodemask
,
3775 .migratetype
= gfpflags_to_migratetype(gfp_mask
),
3778 if (cpusets_enabled()) {
3779 alloc_mask
|= __GFP_HARDWALL
;
3780 alloc_flags
|= ALLOC_CPUSET
;
3782 ac
.nodemask
= &cpuset_current_mems_allowed
;
3785 gfp_mask
&= gfp_allowed_mask
;
3787 lockdep_trace_alloc(gfp_mask
);
3789 might_sleep_if(gfp_mask
& __GFP_DIRECT_RECLAIM
);
3791 if (should_fail_alloc_page(gfp_mask
, order
))
3795 * Check the zones suitable for the gfp_mask contain at least one
3796 * valid zone. It's possible to have an empty zonelist as a result
3797 * of __GFP_THISNODE and a memoryless node
3799 if (unlikely(!zonelist
->_zonerefs
->zone
))
3802 if (IS_ENABLED(CONFIG_CMA
) && ac
.migratetype
== MIGRATE_MOVABLE
)
3803 alloc_flags
|= ALLOC_CMA
;
3806 cpuset_mems_cookie
= read_mems_allowed_begin();
3808 /* Dirty zone balancing only done in the fast path */
3809 ac
.spread_dirty_pages
= (gfp_mask
& __GFP_WRITE
);
3811 /* The preferred zone is used for statistics later */
3812 ac
.preferred_zoneref
= first_zones_zonelist(ac
.zonelist
,
3813 ac
.high_zoneidx
, ac
.nodemask
);
3814 if (!ac
.preferred_zoneref
) {
3819 /* First allocation attempt */
3820 page
= get_page_from_freelist(alloc_mask
, order
, alloc_flags
, &ac
);
3825 * Runtime PM, block IO and its error handling path can deadlock
3826 * because I/O on the device might not complete.
3828 alloc_mask
= memalloc_noio_flags(gfp_mask
);
3829 ac
.spread_dirty_pages
= false;
3832 * Restore the original nodemask if it was potentially replaced with
3833 * &cpuset_current_mems_allowed to optimize the fast-path attempt.
3835 if (cpusets_enabled())
3836 ac
.nodemask
= nodemask
;
3837 page
= __alloc_pages_slowpath(alloc_mask
, order
, &ac
);
3841 * When updating a task's mems_allowed, it is possible to race with
3842 * parallel threads in such a way that an allocation can fail while
3843 * the mask is being updated. If a page allocation is about to fail,
3844 * check if the cpuset changed during allocation and if so, retry.
3846 if (unlikely(!page
&& read_mems_allowed_retry(cpuset_mems_cookie
))) {
3847 alloc_mask
= gfp_mask
;
3852 if (kmemcheck_enabled
&& page
)
3853 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
3855 trace_mm_page_alloc(page
, order
, alloc_mask
, ac
.migratetype
);
3859 EXPORT_SYMBOL(__alloc_pages_nodemask
);
3862 * Common helper functions.
3864 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
3869 * __get_free_pages() returns a 32-bit address, which cannot represent
3872 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
3874 page
= alloc_pages(gfp_mask
, order
);
3877 return (unsigned long) page_address(page
);
3879 EXPORT_SYMBOL(__get_free_pages
);
3881 unsigned long get_zeroed_page(gfp_t gfp_mask
)
3883 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
3885 EXPORT_SYMBOL(get_zeroed_page
);
3887 void __free_pages(struct page
*page
, unsigned int order
)
3889 if (put_page_testzero(page
)) {
3891 free_hot_cold_page(page
, false);
3893 __free_pages_ok(page
, order
);
3897 EXPORT_SYMBOL(__free_pages
);
3899 void free_pages(unsigned long addr
, unsigned int order
)
3902 VM_BUG_ON(!virt_addr_valid((void *)addr
));
3903 __free_pages(virt_to_page((void *)addr
), order
);
3907 EXPORT_SYMBOL(free_pages
);
3911 * An arbitrary-length arbitrary-offset area of memory which resides
3912 * within a 0 or higher order page. Multiple fragments within that page
3913 * are individually refcounted, in the page's reference counter.
3915 * The page_frag functions below provide a simple allocation framework for
3916 * page fragments. This is used by the network stack and network device
3917 * drivers to provide a backing region of memory for use as either an
3918 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3920 static struct page
*__page_frag_refill(struct page_frag_cache
*nc
,
3923 struct page
*page
= NULL
;
3924 gfp_t gfp
= gfp_mask
;
3926 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3927 gfp_mask
|= __GFP_COMP
| __GFP_NOWARN
| __GFP_NORETRY
|
3929 page
= alloc_pages_node(NUMA_NO_NODE
, gfp_mask
,
3930 PAGE_FRAG_CACHE_MAX_ORDER
);
3931 nc
->size
= page
? PAGE_FRAG_CACHE_MAX_SIZE
: PAGE_SIZE
;
3933 if (unlikely(!page
))
3934 page
= alloc_pages_node(NUMA_NO_NODE
, gfp
, 0);
3936 nc
->va
= page
? page_address(page
) : NULL
;
3941 void *__alloc_page_frag(struct page_frag_cache
*nc
,
3942 unsigned int fragsz
, gfp_t gfp_mask
)
3944 unsigned int size
= PAGE_SIZE
;
3948 if (unlikely(!nc
->va
)) {
3950 page
= __page_frag_refill(nc
, gfp_mask
);
3954 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3955 /* if size can vary use size else just use PAGE_SIZE */
3958 /* Even if we own the page, we do not use atomic_set().
3959 * This would break get_page_unless_zero() users.
3961 page_ref_add(page
, size
- 1);
3963 /* reset page count bias and offset to start of new frag */
3964 nc
->pfmemalloc
= page_is_pfmemalloc(page
);
3965 nc
->pagecnt_bias
= size
;
3969 offset
= nc
->offset
- fragsz
;
3970 if (unlikely(offset
< 0)) {
3971 page
= virt_to_page(nc
->va
);
3973 if (!page_ref_sub_and_test(page
, nc
->pagecnt_bias
))
3976 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3977 /* if size can vary use size else just use PAGE_SIZE */
3980 /* OK, page count is 0, we can safely set it */
3981 set_page_count(page
, size
);
3983 /* reset page count bias and offset to start of new frag */
3984 nc
->pagecnt_bias
= size
;
3985 offset
= size
- fragsz
;
3989 nc
->offset
= offset
;
3991 return nc
->va
+ offset
;
3993 EXPORT_SYMBOL(__alloc_page_frag
);
3996 * Frees a page fragment allocated out of either a compound or order 0 page.
3998 void __free_page_frag(void *addr
)
4000 struct page
*page
= virt_to_head_page(addr
);
4002 if (unlikely(put_page_testzero(page
)))
4003 __free_pages_ok(page
, compound_order(page
));
4005 EXPORT_SYMBOL(__free_page_frag
);
4008 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
4009 * of the current memory cgroup if __GFP_ACCOUNT is set, other than that it is
4010 * equivalent to alloc_pages.
4012 * It should be used when the caller would like to use kmalloc, but since the
4013 * allocation is large, it has to fall back to the page allocator.
4015 struct page
*alloc_kmem_pages(gfp_t gfp_mask
, unsigned int order
)
4019 page
= alloc_pages(gfp_mask
, order
);
4020 if (page
&& memcg_kmem_charge(page
, gfp_mask
, order
) != 0) {
4021 __free_pages(page
, order
);
4027 struct page
*alloc_kmem_pages_node(int nid
, gfp_t gfp_mask
, unsigned int order
)
4031 page
= alloc_pages_node(nid
, gfp_mask
, order
);
4032 if (page
&& memcg_kmem_charge(page
, gfp_mask
, order
) != 0) {
4033 __free_pages(page
, order
);
4040 * __free_kmem_pages and free_kmem_pages will free pages allocated with
4043 void __free_kmem_pages(struct page
*page
, unsigned int order
)
4045 memcg_kmem_uncharge(page
, order
);
4046 __free_pages(page
, order
);
4049 void free_kmem_pages(unsigned long addr
, unsigned int order
)
4052 VM_BUG_ON(!virt_addr_valid((void *)addr
));
4053 __free_kmem_pages(virt_to_page((void *)addr
), order
);
4057 static void *make_alloc_exact(unsigned long addr
, unsigned int order
,
4061 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
4062 unsigned long used
= addr
+ PAGE_ALIGN(size
);
4064 split_page(virt_to_page((void *)addr
), order
);
4065 while (used
< alloc_end
) {
4070 return (void *)addr
;
4074 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
4075 * @size: the number of bytes to allocate
4076 * @gfp_mask: GFP flags for the allocation
4078 * This function is similar to alloc_pages(), except that it allocates the
4079 * minimum number of pages to satisfy the request. alloc_pages() can only
4080 * allocate memory in power-of-two pages.
4082 * This function is also limited by MAX_ORDER.
4084 * Memory allocated by this function must be released by free_pages_exact().
4086 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
4088 unsigned int order
= get_order(size
);
4091 addr
= __get_free_pages(gfp_mask
, order
);
4092 return make_alloc_exact(addr
, order
, size
);
4094 EXPORT_SYMBOL(alloc_pages_exact
);
4097 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
4099 * @nid: the preferred node ID where memory should be allocated
4100 * @size: the number of bytes to allocate
4101 * @gfp_mask: GFP flags for the allocation
4103 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
4106 void * __meminit
alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
4108 unsigned int order
= get_order(size
);
4109 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
4112 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
4116 * free_pages_exact - release memory allocated via alloc_pages_exact()
4117 * @virt: the value returned by alloc_pages_exact.
4118 * @size: size of allocation, same value as passed to alloc_pages_exact().
4120 * Release the memory allocated by a previous call to alloc_pages_exact.
4122 void free_pages_exact(void *virt
, size_t size
)
4124 unsigned long addr
= (unsigned long)virt
;
4125 unsigned long end
= addr
+ PAGE_ALIGN(size
);
4127 while (addr
< end
) {
4132 EXPORT_SYMBOL(free_pages_exact
);
4135 * nr_free_zone_pages - count number of pages beyond high watermark
4136 * @offset: The zone index of the highest zone
4138 * nr_free_zone_pages() counts the number of counts pages which are beyond the
4139 * high watermark within all zones at or below a given zone index. For each
4140 * zone, the number of pages is calculated as:
4141 * managed_pages - high_pages
4143 static unsigned long nr_free_zone_pages(int offset
)
4148 /* Just pick one node, since fallback list is circular */
4149 unsigned long sum
= 0;
4151 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
4153 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
4154 unsigned long size
= zone
->managed_pages
;
4155 unsigned long high
= high_wmark_pages(zone
);
4164 * nr_free_buffer_pages - count number of pages beyond high watermark
4166 * nr_free_buffer_pages() counts the number of pages which are beyond the high
4167 * watermark within ZONE_DMA and ZONE_NORMAL.
4169 unsigned long nr_free_buffer_pages(void)
4171 return nr_free_zone_pages(gfp_zone(GFP_USER
));
4173 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
4176 * nr_free_pagecache_pages - count number of pages beyond high watermark
4178 * nr_free_pagecache_pages() counts the number of pages which are beyond the
4179 * high watermark within all zones.
4181 unsigned long nr_free_pagecache_pages(void)
4183 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
4186 static inline void show_node(struct zone
*zone
)
4188 if (IS_ENABLED(CONFIG_NUMA
))
4189 printk("Node %d ", zone_to_nid(zone
));
4192 long si_mem_available(void)
4195 unsigned long pagecache
;
4196 unsigned long wmark_low
= 0;
4197 unsigned long pages
[NR_LRU_LISTS
];
4201 for (lru
= LRU_BASE
; lru
< NR_LRU_LISTS
; lru
++)
4202 pages
[lru
] = global_page_state(NR_LRU_BASE
+ lru
);
4205 wmark_low
+= zone
->watermark
[WMARK_LOW
];
4208 * Estimate the amount of memory available for userspace allocations,
4209 * without causing swapping.
4211 available
= global_page_state(NR_FREE_PAGES
) - totalreserve_pages
;
4214 * Not all the page cache can be freed, otherwise the system will
4215 * start swapping. Assume at least half of the page cache, or the
4216 * low watermark worth of cache, needs to stay.
4218 pagecache
= pages
[LRU_ACTIVE_FILE
] + pages
[LRU_INACTIVE_FILE
];
4219 pagecache
-= min(pagecache
/ 2, wmark_low
);
4220 available
+= pagecache
;
4223 * Part of the reclaimable slab consists of items that are in use,
4224 * and cannot be freed. Cap this estimate at the low watermark.
4226 available
+= global_page_state(NR_SLAB_RECLAIMABLE
) -
4227 min(global_page_state(NR_SLAB_RECLAIMABLE
) / 2, wmark_low
);
4233 EXPORT_SYMBOL_GPL(si_mem_available
);
4235 void si_meminfo(struct sysinfo
*val
)
4237 val
->totalram
= totalram_pages
;
4238 val
->sharedram
= global_page_state(NR_SHMEM
);
4239 val
->freeram
= global_page_state(NR_FREE_PAGES
);
4240 val
->bufferram
= nr_blockdev_pages();
4241 val
->totalhigh
= totalhigh_pages
;
4242 val
->freehigh
= nr_free_highpages();
4243 val
->mem_unit
= PAGE_SIZE
;
4246 EXPORT_SYMBOL(si_meminfo
);
4249 void si_meminfo_node(struct sysinfo
*val
, int nid
)
4251 int zone_type
; /* needs to be signed */
4252 unsigned long managed_pages
= 0;
4253 unsigned long managed_highpages
= 0;
4254 unsigned long free_highpages
= 0;
4255 pg_data_t
*pgdat
= NODE_DATA(nid
);
4257 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++)
4258 managed_pages
+= pgdat
->node_zones
[zone_type
].managed_pages
;
4259 val
->totalram
= managed_pages
;
4260 val
->sharedram
= node_page_state(nid
, NR_SHMEM
);
4261 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
4262 #ifdef CONFIG_HIGHMEM
4263 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
4264 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4266 if (is_highmem(zone
)) {
4267 managed_highpages
+= zone
->managed_pages
;
4268 free_highpages
+= zone_page_state(zone
, NR_FREE_PAGES
);
4271 val
->totalhigh
= managed_highpages
;
4272 val
->freehigh
= free_highpages
;
4274 val
->totalhigh
= managed_highpages
;
4275 val
->freehigh
= free_highpages
;
4277 val
->mem_unit
= PAGE_SIZE
;
4282 * Determine whether the node should be displayed or not, depending on whether
4283 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
4285 bool skip_free_areas_node(unsigned int flags
, int nid
)
4288 unsigned int cpuset_mems_cookie
;
4290 if (!(flags
& SHOW_MEM_FILTER_NODES
))
4294 cpuset_mems_cookie
= read_mems_allowed_begin();
4295 ret
= !node_isset(nid
, cpuset_current_mems_allowed
);
4296 } while (read_mems_allowed_retry(cpuset_mems_cookie
));
4301 #define K(x) ((x) << (PAGE_SHIFT-10))
4303 static void show_migration_types(unsigned char type
)
4305 static const char types
[MIGRATE_TYPES
] = {
4306 [MIGRATE_UNMOVABLE
] = 'U',
4307 [MIGRATE_MOVABLE
] = 'M',
4308 [MIGRATE_RECLAIMABLE
] = 'E',
4309 [MIGRATE_HIGHATOMIC
] = 'H',
4311 [MIGRATE_CMA
] = 'C',
4313 #ifdef CONFIG_MEMORY_ISOLATION
4314 [MIGRATE_ISOLATE
] = 'I',
4317 char tmp
[MIGRATE_TYPES
+ 1];
4321 for (i
= 0; i
< MIGRATE_TYPES
; i
++) {
4322 if (type
& (1 << i
))
4327 printk("(%s) ", tmp
);
4331 * Show free area list (used inside shift_scroll-lock stuff)
4332 * We also calculate the percentage fragmentation. We do this by counting the
4333 * memory on each free list with the exception of the first item on the list.
4336 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
4339 void show_free_areas(unsigned int filter
)
4341 unsigned long free_pcp
= 0;
4345 for_each_populated_zone(zone
) {
4346 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
4349 for_each_online_cpu(cpu
)
4350 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
4353 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
4354 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
4355 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
4356 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
4357 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
4358 " free:%lu free_pcp:%lu free_cma:%lu\n",
4359 global_page_state(NR_ACTIVE_ANON
),
4360 global_page_state(NR_INACTIVE_ANON
),
4361 global_page_state(NR_ISOLATED_ANON
),
4362 global_page_state(NR_ACTIVE_FILE
),
4363 global_page_state(NR_INACTIVE_FILE
),
4364 global_page_state(NR_ISOLATED_FILE
),
4365 global_page_state(NR_UNEVICTABLE
),
4366 global_page_state(NR_FILE_DIRTY
),
4367 global_page_state(NR_WRITEBACK
),
4368 global_page_state(NR_UNSTABLE_NFS
),
4369 global_page_state(NR_SLAB_RECLAIMABLE
),
4370 global_page_state(NR_SLAB_UNRECLAIMABLE
),
4371 global_page_state(NR_FILE_MAPPED
),
4372 global_page_state(NR_SHMEM
),
4373 global_page_state(NR_PAGETABLE
),
4374 global_page_state(NR_BOUNCE
),
4375 global_page_state(NR_FREE_PAGES
),
4377 global_page_state(NR_FREE_CMA_PAGES
));
4379 for_each_populated_zone(zone
) {
4382 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
4386 for_each_online_cpu(cpu
)
4387 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
4395 " active_anon:%lukB"
4396 " inactive_anon:%lukB"
4397 " active_file:%lukB"
4398 " inactive_file:%lukB"
4399 " unevictable:%lukB"
4400 " isolated(anon):%lukB"
4401 " isolated(file):%lukB"
4409 " slab_reclaimable:%lukB"
4410 " slab_unreclaimable:%lukB"
4411 " kernel_stack:%lukB"
4418 " writeback_tmp:%lukB"
4419 " pages_scanned:%lu"
4420 " all_unreclaimable? %s"
4423 K(zone_page_state(zone
, NR_FREE_PAGES
)),
4424 K(min_wmark_pages(zone
)),
4425 K(low_wmark_pages(zone
)),
4426 K(high_wmark_pages(zone
)),
4427 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
4428 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
4429 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
4430 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
4431 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
4432 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
4433 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
4434 K(zone
->present_pages
),
4435 K(zone
->managed_pages
),
4436 K(zone_page_state(zone
, NR_MLOCK
)),
4437 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
4438 K(zone_page_state(zone
, NR_WRITEBACK
)),
4439 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
4440 K(zone_page_state(zone
, NR_SHMEM
)),
4441 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
4442 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
4443 zone_page_state(zone
, NR_KERNEL_STACK
) *
4445 K(zone_page_state(zone
, NR_PAGETABLE
)),
4446 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
4447 K(zone_page_state(zone
, NR_BOUNCE
)),
4449 K(this_cpu_read(zone
->pageset
->pcp
.count
)),
4450 K(zone_page_state(zone
, NR_FREE_CMA_PAGES
)),
4451 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
4452 K(zone_page_state(zone
, NR_PAGES_SCANNED
)),
4453 (!zone_reclaimable(zone
) ? "yes" : "no")
4455 printk("lowmem_reserve[]:");
4456 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4457 printk(" %ld", zone
->lowmem_reserve
[i
]);
4461 for_each_populated_zone(zone
) {
4463 unsigned long nr
[MAX_ORDER
], flags
, total
= 0;
4464 unsigned char types
[MAX_ORDER
];
4466 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
4469 printk("%s: ", zone
->name
);
4471 spin_lock_irqsave(&zone
->lock
, flags
);
4472 for (order
= 0; order
< MAX_ORDER
; order
++) {
4473 struct free_area
*area
= &zone
->free_area
[order
];
4476 nr
[order
] = area
->nr_free
;
4477 total
+= nr
[order
] << order
;
4480 for (type
= 0; type
< MIGRATE_TYPES
; type
++) {
4481 if (!list_empty(&area
->free_list
[type
]))
4482 types
[order
] |= 1 << type
;
4485 spin_unlock_irqrestore(&zone
->lock
, flags
);
4486 for (order
= 0; order
< MAX_ORDER
; order
++) {
4487 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
4489 show_migration_types(types
[order
]);
4491 printk("= %lukB\n", K(total
));
4494 hugetlb_show_meminfo();
4496 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
4498 show_swap_cache_info();
4501 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
4503 zoneref
->zone
= zone
;
4504 zoneref
->zone_idx
= zone_idx(zone
);
4508 * Builds allocation fallback zone lists.
4510 * Add all populated zones of a node to the zonelist.
4512 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
4516 enum zone_type zone_type
= MAX_NR_ZONES
;
4520 zone
= pgdat
->node_zones
+ zone_type
;
4521 if (populated_zone(zone
)) {
4522 zoneref_set_zone(zone
,
4523 &zonelist
->_zonerefs
[nr_zones
++]);
4524 check_highest_zone(zone_type
);
4526 } while (zone_type
);
4534 * 0 = automatic detection of better ordering.
4535 * 1 = order by ([node] distance, -zonetype)
4536 * 2 = order by (-zonetype, [node] distance)
4538 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
4539 * the same zonelist. So only NUMA can configure this param.
4541 #define ZONELIST_ORDER_DEFAULT 0
4542 #define ZONELIST_ORDER_NODE 1
4543 #define ZONELIST_ORDER_ZONE 2
4545 /* zonelist order in the kernel.
4546 * set_zonelist_order() will set this to NODE or ZONE.
4548 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4549 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
4553 /* The value user specified ....changed by config */
4554 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4555 /* string for sysctl */
4556 #define NUMA_ZONELIST_ORDER_LEN 16
4557 char numa_zonelist_order
[16] = "default";
4560 * interface for configure zonelist ordering.
4561 * command line option "numa_zonelist_order"
4562 * = "[dD]efault - default, automatic configuration.
4563 * = "[nN]ode - order by node locality, then by zone within node
4564 * = "[zZ]one - order by zone, then by locality within zone
4567 static int __parse_numa_zonelist_order(char *s
)
4569 if (*s
== 'd' || *s
== 'D') {
4570 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4571 } else if (*s
== 'n' || *s
== 'N') {
4572 user_zonelist_order
= ZONELIST_ORDER_NODE
;
4573 } else if (*s
== 'z' || *s
== 'Z') {
4574 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
4576 pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s
);
4582 static __init
int setup_numa_zonelist_order(char *s
)
4589 ret
= __parse_numa_zonelist_order(s
);
4591 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
4595 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
4598 * sysctl handler for numa_zonelist_order
4600 int numa_zonelist_order_handler(struct ctl_table
*table
, int write
,
4601 void __user
*buffer
, size_t *length
,
4604 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
4606 static DEFINE_MUTEX(zl_order_mutex
);
4608 mutex_lock(&zl_order_mutex
);
4610 if (strlen((char *)table
->data
) >= NUMA_ZONELIST_ORDER_LEN
) {
4614 strcpy(saved_string
, (char *)table
->data
);
4616 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
4620 int oldval
= user_zonelist_order
;
4622 ret
= __parse_numa_zonelist_order((char *)table
->data
);
4625 * bogus value. restore saved string
4627 strncpy((char *)table
->data
, saved_string
,
4628 NUMA_ZONELIST_ORDER_LEN
);
4629 user_zonelist_order
= oldval
;
4630 } else if (oldval
!= user_zonelist_order
) {
4631 mutex_lock(&zonelists_mutex
);
4632 build_all_zonelists(NULL
, NULL
);
4633 mutex_unlock(&zonelists_mutex
);
4637 mutex_unlock(&zl_order_mutex
);
4642 #define MAX_NODE_LOAD (nr_online_nodes)
4643 static int node_load
[MAX_NUMNODES
];
4646 * find_next_best_node - find the next node that should appear in a given node's fallback list
4647 * @node: node whose fallback list we're appending
4648 * @used_node_mask: nodemask_t of already used nodes
4650 * We use a number of factors to determine which is the next node that should
4651 * appear on a given node's fallback list. The node should not have appeared
4652 * already in @node's fallback list, and it should be the next closest node
4653 * according to the distance array (which contains arbitrary distance values
4654 * from each node to each node in the system), and should also prefer nodes
4655 * with no CPUs, since presumably they'll have very little allocation pressure
4656 * on them otherwise.
4657 * It returns -1 if no node is found.
4659 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
4662 int min_val
= INT_MAX
;
4663 int best_node
= NUMA_NO_NODE
;
4664 const struct cpumask
*tmp
= cpumask_of_node(0);
4666 /* Use the local node if we haven't already */
4667 if (!node_isset(node
, *used_node_mask
)) {
4668 node_set(node
, *used_node_mask
);
4672 for_each_node_state(n
, N_MEMORY
) {
4674 /* Don't want a node to appear more than once */
4675 if (node_isset(n
, *used_node_mask
))
4678 /* Use the distance array to find the distance */
4679 val
= node_distance(node
, n
);
4681 /* Penalize nodes under us ("prefer the next node") */
4684 /* Give preference to headless and unused nodes */
4685 tmp
= cpumask_of_node(n
);
4686 if (!cpumask_empty(tmp
))
4687 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
4689 /* Slight preference for less loaded node */
4690 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
4691 val
+= node_load
[n
];
4693 if (val
< min_val
) {
4700 node_set(best_node
, *used_node_mask
);
4707 * Build zonelists ordered by node and zones within node.
4708 * This results in maximum locality--normal zone overflows into local
4709 * DMA zone, if any--but risks exhausting DMA zone.
4711 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
4714 struct zonelist
*zonelist
;
4716 zonelist
= &pgdat
->node_zonelists
[0];
4717 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
4719 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4720 zonelist
->_zonerefs
[j
].zone
= NULL
;
4721 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4725 * Build gfp_thisnode zonelists
4727 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
4730 struct zonelist
*zonelist
;
4732 zonelist
= &pgdat
->node_zonelists
[1];
4733 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4734 zonelist
->_zonerefs
[j
].zone
= NULL
;
4735 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4739 * Build zonelists ordered by zone and nodes within zones.
4740 * This results in conserving DMA zone[s] until all Normal memory is
4741 * exhausted, but results in overflowing to remote node while memory
4742 * may still exist in local DMA zone.
4744 static int node_order
[MAX_NUMNODES
];
4746 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
4749 int zone_type
; /* needs to be signed */
4751 struct zonelist
*zonelist
;
4753 zonelist
= &pgdat
->node_zonelists
[0];
4755 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
4756 for (j
= 0; j
< nr_nodes
; j
++) {
4757 node
= node_order
[j
];
4758 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
4759 if (populated_zone(z
)) {
4761 &zonelist
->_zonerefs
[pos
++]);
4762 check_highest_zone(zone_type
);
4766 zonelist
->_zonerefs
[pos
].zone
= NULL
;
4767 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
4770 #if defined(CONFIG_64BIT)
4772 * Devices that require DMA32/DMA are relatively rare and do not justify a
4773 * penalty to every machine in case the specialised case applies. Default
4774 * to Node-ordering on 64-bit NUMA machines
4776 static int default_zonelist_order(void)
4778 return ZONELIST_ORDER_NODE
;
4782 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4783 * by the kernel. If processes running on node 0 deplete the low memory zone
4784 * then reclaim will occur more frequency increasing stalls and potentially
4785 * be easier to OOM if a large percentage of the zone is under writeback or
4786 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4787 * Hence, default to zone ordering on 32-bit.
4789 static int default_zonelist_order(void)
4791 return ZONELIST_ORDER_ZONE
;
4793 #endif /* CONFIG_64BIT */
4795 static void set_zonelist_order(void)
4797 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
4798 current_zonelist_order
= default_zonelist_order();
4800 current_zonelist_order
= user_zonelist_order
;
4803 static void build_zonelists(pg_data_t
*pgdat
)
4806 nodemask_t used_mask
;
4807 int local_node
, prev_node
;
4808 struct zonelist
*zonelist
;
4809 unsigned int order
= current_zonelist_order
;
4811 /* initialize zonelists */
4812 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
4813 zonelist
= pgdat
->node_zonelists
+ i
;
4814 zonelist
->_zonerefs
[0].zone
= NULL
;
4815 zonelist
->_zonerefs
[0].zone_idx
= 0;
4818 /* NUMA-aware ordering of nodes */
4819 local_node
= pgdat
->node_id
;
4820 load
= nr_online_nodes
;
4821 prev_node
= local_node
;
4822 nodes_clear(used_mask
);
4824 memset(node_order
, 0, sizeof(node_order
));
4827 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
4829 * We don't want to pressure a particular node.
4830 * So adding penalty to the first node in same
4831 * distance group to make it round-robin.
4833 if (node_distance(local_node
, node
) !=
4834 node_distance(local_node
, prev_node
))
4835 node_load
[node
] = load
;
4839 if (order
== ZONELIST_ORDER_NODE
)
4840 build_zonelists_in_node_order(pgdat
, node
);
4842 node_order
[i
++] = node
; /* remember order */
4845 if (order
== ZONELIST_ORDER_ZONE
) {
4846 /* calculate node order -- i.e., DMA last! */
4847 build_zonelists_in_zone_order(pgdat
, i
);
4850 build_thisnode_zonelists(pgdat
);
4853 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4855 * Return node id of node used for "local" allocations.
4856 * I.e., first node id of first zone in arg node's generic zonelist.
4857 * Used for initializing percpu 'numa_mem', which is used primarily
4858 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4860 int local_memory_node(int node
)
4864 z
= first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
4865 gfp_zone(GFP_KERNEL
),
4867 return z
->zone
->node
;
4871 #else /* CONFIG_NUMA */
4873 static void set_zonelist_order(void)
4875 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
4878 static void build_zonelists(pg_data_t
*pgdat
)
4880 int node
, local_node
;
4882 struct zonelist
*zonelist
;
4884 local_node
= pgdat
->node_id
;
4886 zonelist
= &pgdat
->node_zonelists
[0];
4887 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4890 * Now we build the zonelist so that it contains the zones
4891 * of all the other nodes.
4892 * We don't want to pressure a particular node, so when
4893 * building the zones for node N, we make sure that the
4894 * zones coming right after the local ones are those from
4895 * node N+1 (modulo N)
4897 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
4898 if (!node_online(node
))
4900 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4902 for (node
= 0; node
< local_node
; node
++) {
4903 if (!node_online(node
))
4905 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4908 zonelist
->_zonerefs
[j
].zone
= NULL
;
4909 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4912 #endif /* CONFIG_NUMA */
4915 * Boot pageset table. One per cpu which is going to be used for all
4916 * zones and all nodes. The parameters will be set in such a way
4917 * that an item put on a list will immediately be handed over to
4918 * the buddy list. This is safe since pageset manipulation is done
4919 * with interrupts disabled.
4921 * The boot_pagesets must be kept even after bootup is complete for
4922 * unused processors and/or zones. They do play a role for bootstrapping
4923 * hotplugged processors.
4925 * zoneinfo_show() and maybe other functions do
4926 * not check if the processor is online before following the pageset pointer.
4927 * Other parts of the kernel may not check if the zone is available.
4929 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
4930 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
4931 static void setup_zone_pageset(struct zone
*zone
);
4934 * Global mutex to protect against size modification of zonelists
4935 * as well as to serialize pageset setup for the new populated zone.
4937 DEFINE_MUTEX(zonelists_mutex
);
4939 /* return values int ....just for stop_machine() */
4940 static int __build_all_zonelists(void *data
)
4944 pg_data_t
*self
= data
;
4947 memset(node_load
, 0, sizeof(node_load
));
4950 if (self
&& !node_online(self
->node_id
)) {
4951 build_zonelists(self
);
4954 for_each_online_node(nid
) {
4955 pg_data_t
*pgdat
= NODE_DATA(nid
);
4957 build_zonelists(pgdat
);
4961 * Initialize the boot_pagesets that are going to be used
4962 * for bootstrapping processors. The real pagesets for
4963 * each zone will be allocated later when the per cpu
4964 * allocator is available.
4966 * boot_pagesets are used also for bootstrapping offline
4967 * cpus if the system is already booted because the pagesets
4968 * are needed to initialize allocators on a specific cpu too.
4969 * F.e. the percpu allocator needs the page allocator which
4970 * needs the percpu allocator in order to allocate its pagesets
4971 * (a chicken-egg dilemma).
4973 for_each_possible_cpu(cpu
) {
4974 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
4976 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4978 * We now know the "local memory node" for each node--
4979 * i.e., the node of the first zone in the generic zonelist.
4980 * Set up numa_mem percpu variable for on-line cpus. During
4981 * boot, only the boot cpu should be on-line; we'll init the
4982 * secondary cpus' numa_mem as they come on-line. During
4983 * node/memory hotplug, we'll fixup all on-line cpus.
4985 if (cpu_online(cpu
))
4986 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
4993 static noinline
void __init
4994 build_all_zonelists_init(void)
4996 __build_all_zonelists(NULL
);
4997 mminit_verify_zonelist();
4998 cpuset_init_current_mems_allowed();
5002 * Called with zonelists_mutex held always
5003 * unless system_state == SYSTEM_BOOTING.
5005 * __ref due to (1) call of __meminit annotated setup_zone_pageset
5006 * [we're only called with non-NULL zone through __meminit paths] and
5007 * (2) call of __init annotated helper build_all_zonelists_init
5008 * [protected by SYSTEM_BOOTING].
5010 void __ref
build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
)
5012 set_zonelist_order();
5014 if (system_state
== SYSTEM_BOOTING
) {
5015 build_all_zonelists_init();
5017 #ifdef CONFIG_MEMORY_HOTPLUG
5019 setup_zone_pageset(zone
);
5021 /* we have to stop all cpus to guarantee there is no user
5023 stop_machine(__build_all_zonelists
, pgdat
, NULL
);
5024 /* cpuset refresh routine should be here */
5026 vm_total_pages
= nr_free_pagecache_pages();
5028 * Disable grouping by mobility if the number of pages in the
5029 * system is too low to allow the mechanism to work. It would be
5030 * more accurate, but expensive to check per-zone. This check is
5031 * made on memory-hotadd so a system can start with mobility
5032 * disabled and enable it later
5034 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
5035 page_group_by_mobility_disabled
= 1;
5037 page_group_by_mobility_disabled
= 0;
5039 pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n",
5041 zonelist_order_name
[current_zonelist_order
],
5042 page_group_by_mobility_disabled
? "off" : "on",
5045 pr_info("Policy zone: %s\n", zone_names
[policy_zone
]);
5050 * Helper functions to size the waitqueue hash table.
5051 * Essentially these want to choose hash table sizes sufficiently
5052 * large so that collisions trying to wait on pages are rare.
5053 * But in fact, the number of active page waitqueues on typical
5054 * systems is ridiculously low, less than 200. So this is even
5055 * conservative, even though it seems large.
5057 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
5058 * waitqueues, i.e. the size of the waitq table given the number of pages.
5060 #define PAGES_PER_WAITQUEUE 256
5062 #ifndef CONFIG_MEMORY_HOTPLUG
5063 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
5065 unsigned long size
= 1;
5067 pages
/= PAGES_PER_WAITQUEUE
;
5069 while (size
< pages
)
5073 * Once we have dozens or even hundreds of threads sleeping
5074 * on IO we've got bigger problems than wait queue collision.
5075 * Limit the size of the wait table to a reasonable size.
5077 size
= min(size
, 4096UL);
5079 return max(size
, 4UL);
5083 * A zone's size might be changed by hot-add, so it is not possible to determine
5084 * a suitable size for its wait_table. So we use the maximum size now.
5086 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
5088 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
5089 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
5090 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
5092 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
5093 * or more by the traditional way. (See above). It equals:
5095 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
5096 * ia64(16K page size) : = ( 8G + 4M)byte.
5097 * powerpc (64K page size) : = (32G +16M)byte.
5099 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
5106 * This is an integer logarithm so that shifts can be used later
5107 * to extract the more random high bits from the multiplicative
5108 * hash function before the remainder is taken.
5110 static inline unsigned long wait_table_bits(unsigned long size
)
5116 * Initially all pages are reserved - free ones are freed
5117 * up by free_all_bootmem() once the early boot process is
5118 * done. Non-atomic initialization, single-pass.
5120 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
5121 unsigned long start_pfn
, enum memmap_context context
)
5123 struct vmem_altmap
*altmap
= to_vmem_altmap(__pfn_to_phys(start_pfn
));
5124 unsigned long end_pfn
= start_pfn
+ size
;
5125 pg_data_t
*pgdat
= NODE_DATA(nid
);
5127 unsigned long nr_initialised
= 0;
5128 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5129 struct memblock_region
*r
= NULL
, *tmp
;
5132 if (highest_memmap_pfn
< end_pfn
- 1)
5133 highest_memmap_pfn
= end_pfn
- 1;
5136 * Honor reservation requested by the driver for this ZONE_DEVICE
5139 if (altmap
&& start_pfn
== altmap
->base_pfn
)
5140 start_pfn
+= altmap
->reserve
;
5142 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
5144 * There can be holes in boot-time mem_map[]s handed to this
5145 * function. They do not exist on hotplugged memory.
5147 if (context
!= MEMMAP_EARLY
)
5150 if (!early_pfn_valid(pfn
))
5152 if (!early_pfn_in_nid(pfn
, nid
))
5154 if (!update_defer_init(pgdat
, pfn
, end_pfn
, &nr_initialised
))
5157 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5159 * If not mirrored_kernelcore and ZONE_MOVABLE exists, range
5160 * from zone_movable_pfn[nid] to end of each node should be
5161 * ZONE_MOVABLE not ZONE_NORMAL. skip it.
5163 if (!mirrored_kernelcore
&& zone_movable_pfn
[nid
])
5164 if (zone
== ZONE_NORMAL
&& pfn
>= zone_movable_pfn
[nid
])
5168 * Check given memblock attribute by firmware which can affect
5169 * kernel memory layout. If zone==ZONE_MOVABLE but memory is
5170 * mirrored, it's an overlapped memmap init. skip it.
5172 if (mirrored_kernelcore
&& zone
== ZONE_MOVABLE
) {
5173 if (!r
|| pfn
>= memblock_region_memory_end_pfn(r
)) {
5174 for_each_memblock(memory
, tmp
)
5175 if (pfn
< memblock_region_memory_end_pfn(tmp
))
5179 if (pfn
>= memblock_region_memory_base_pfn(r
) &&
5180 memblock_is_mirror(r
)) {
5181 /* already initialized as NORMAL */
5182 pfn
= memblock_region_memory_end_pfn(r
);
5190 * Mark the block movable so that blocks are reserved for
5191 * movable at startup. This will force kernel allocations
5192 * to reserve their blocks rather than leaking throughout
5193 * the address space during boot when many long-lived
5194 * kernel allocations are made.
5196 * bitmap is created for zone's valid pfn range. but memmap
5197 * can be created for invalid pages (for alignment)
5198 * check here not to call set_pageblock_migratetype() against
5201 if (!(pfn
& (pageblock_nr_pages
- 1))) {
5202 struct page
*page
= pfn_to_page(pfn
);
5204 __init_single_page(page
, pfn
, zone
, nid
);
5205 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
5207 __init_single_pfn(pfn
, zone
, nid
);
5212 static void __meminit
zone_init_free_lists(struct zone
*zone
)
5214 unsigned int order
, t
;
5215 for_each_migratetype_order(order
, t
) {
5216 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
5217 zone
->free_area
[order
].nr_free
= 0;
5221 #ifndef __HAVE_ARCH_MEMMAP_INIT
5222 #define memmap_init(size, nid, zone, start_pfn) \
5223 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
5226 static int zone_batchsize(struct zone
*zone
)
5232 * The per-cpu-pages pools are set to around 1000th of the
5233 * size of the zone. But no more than 1/2 of a meg.
5235 * OK, so we don't know how big the cache is. So guess.
5237 batch
= zone
->managed_pages
/ 1024;
5238 if (batch
* PAGE_SIZE
> 512 * 1024)
5239 batch
= (512 * 1024) / PAGE_SIZE
;
5240 batch
/= 4; /* We effectively *= 4 below */
5245 * Clamp the batch to a 2^n - 1 value. Having a power
5246 * of 2 value was found to be more likely to have
5247 * suboptimal cache aliasing properties in some cases.
5249 * For example if 2 tasks are alternately allocating
5250 * batches of pages, one task can end up with a lot
5251 * of pages of one half of the possible page colors
5252 * and the other with pages of the other colors.
5254 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
5259 /* The deferral and batching of frees should be suppressed under NOMMU
5262 * The problem is that NOMMU needs to be able to allocate large chunks
5263 * of contiguous memory as there's no hardware page translation to
5264 * assemble apparent contiguous memory from discontiguous pages.
5266 * Queueing large contiguous runs of pages for batching, however,
5267 * causes the pages to actually be freed in smaller chunks. As there
5268 * can be a significant delay between the individual batches being
5269 * recycled, this leads to the once large chunks of space being
5270 * fragmented and becoming unavailable for high-order allocations.
5277 * pcp->high and pcp->batch values are related and dependent on one another:
5278 * ->batch must never be higher then ->high.
5279 * The following function updates them in a safe manner without read side
5282 * Any new users of pcp->batch and pcp->high should ensure they can cope with
5283 * those fields changing asynchronously (acording the the above rule).
5285 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
5286 * outside of boot time (or some other assurance that no concurrent updaters
5289 static void pageset_update(struct per_cpu_pages
*pcp
, unsigned long high
,
5290 unsigned long batch
)
5292 /* start with a fail safe value for batch */
5296 /* Update high, then batch, in order */
5303 /* a companion to pageset_set_high() */
5304 static void pageset_set_batch(struct per_cpu_pageset
*p
, unsigned long batch
)
5306 pageset_update(&p
->pcp
, 6 * batch
, max(1UL, 1 * batch
));
5309 static void pageset_init(struct per_cpu_pageset
*p
)
5311 struct per_cpu_pages
*pcp
;
5314 memset(p
, 0, sizeof(*p
));
5318 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
5319 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
5322 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
5325 pageset_set_batch(p
, batch
);
5329 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
5330 * to the value high for the pageset p.
5332 static void pageset_set_high(struct per_cpu_pageset
*p
,
5335 unsigned long batch
= max(1UL, high
/ 4);
5336 if ((high
/ 4) > (PAGE_SHIFT
* 8))
5337 batch
= PAGE_SHIFT
* 8;
5339 pageset_update(&p
->pcp
, high
, batch
);
5342 static void pageset_set_high_and_batch(struct zone
*zone
,
5343 struct per_cpu_pageset
*pcp
)
5345 if (percpu_pagelist_fraction
)
5346 pageset_set_high(pcp
,
5347 (zone
->managed_pages
/
5348 percpu_pagelist_fraction
));
5350 pageset_set_batch(pcp
, zone_batchsize(zone
));
5353 static void __meminit
zone_pageset_init(struct zone
*zone
, int cpu
)
5355 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
5358 pageset_set_high_and_batch(zone
, pcp
);
5361 static void __meminit
setup_zone_pageset(struct zone
*zone
)
5364 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
5365 for_each_possible_cpu(cpu
)
5366 zone_pageset_init(zone
, cpu
);
5370 * Allocate per cpu pagesets and initialize them.
5371 * Before this call only boot pagesets were available.
5373 void __init
setup_per_cpu_pageset(void)
5377 for_each_populated_zone(zone
)
5378 setup_zone_pageset(zone
);
5381 static noinline __init_refok
5382 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
5388 * The per-page waitqueue mechanism uses hashed waitqueues
5391 zone
->wait_table_hash_nr_entries
=
5392 wait_table_hash_nr_entries(zone_size_pages
);
5393 zone
->wait_table_bits
=
5394 wait_table_bits(zone
->wait_table_hash_nr_entries
);
5395 alloc_size
= zone
->wait_table_hash_nr_entries
5396 * sizeof(wait_queue_head_t
);
5398 if (!slab_is_available()) {
5399 zone
->wait_table
= (wait_queue_head_t
*)
5400 memblock_virt_alloc_node_nopanic(
5401 alloc_size
, zone
->zone_pgdat
->node_id
);
5404 * This case means that a zone whose size was 0 gets new memory
5405 * via memory hot-add.
5406 * But it may be the case that a new node was hot-added. In
5407 * this case vmalloc() will not be able to use this new node's
5408 * memory - this wait_table must be initialized to use this new
5409 * node itself as well.
5410 * To use this new node's memory, further consideration will be
5413 zone
->wait_table
= vmalloc(alloc_size
);
5415 if (!zone
->wait_table
)
5418 for (i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
5419 init_waitqueue_head(zone
->wait_table
+ i
);
5424 static __meminit
void zone_pcp_init(struct zone
*zone
)
5427 * per cpu subsystem is not up at this point. The following code
5428 * relies on the ability of the linker to provide the
5429 * offset of a (static) per cpu variable into the per cpu area.
5431 zone
->pageset
= &boot_pageset
;
5433 if (populated_zone(zone
))
5434 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
5435 zone
->name
, zone
->present_pages
,
5436 zone_batchsize(zone
));
5439 int __meminit
init_currently_empty_zone(struct zone
*zone
,
5440 unsigned long zone_start_pfn
,
5443 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
5445 ret
= zone_wait_table_init(zone
, size
);
5448 pgdat
->nr_zones
= zone_idx(zone
) + 1;
5450 zone
->zone_start_pfn
= zone_start_pfn
;
5452 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
5453 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
5455 (unsigned long)zone_idx(zone
),
5456 zone_start_pfn
, (zone_start_pfn
+ size
));
5458 zone_init_free_lists(zone
);
5463 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5464 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
5467 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
5469 int __meminit
__early_pfn_to_nid(unsigned long pfn
,
5470 struct mminit_pfnnid_cache
*state
)
5472 unsigned long start_pfn
, end_pfn
;
5475 if (state
->last_start
<= pfn
&& pfn
< state
->last_end
)
5476 return state
->last_nid
;
5478 nid
= memblock_search_pfn_nid(pfn
, &start_pfn
, &end_pfn
);
5480 state
->last_start
= start_pfn
;
5481 state
->last_end
= end_pfn
;
5482 state
->last_nid
= nid
;
5487 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
5490 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
5491 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
5492 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
5494 * If an architecture guarantees that all ranges registered contain no holes
5495 * and may be freed, this this function may be used instead of calling
5496 * memblock_free_early_nid() manually.
5498 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
5500 unsigned long start_pfn
, end_pfn
;
5503 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
5504 start_pfn
= min(start_pfn
, max_low_pfn
);
5505 end_pfn
= min(end_pfn
, max_low_pfn
);
5507 if (start_pfn
< end_pfn
)
5508 memblock_free_early_nid(PFN_PHYS(start_pfn
),
5509 (end_pfn
- start_pfn
) << PAGE_SHIFT
,
5515 * sparse_memory_present_with_active_regions - Call memory_present for each active range
5516 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
5518 * If an architecture guarantees that all ranges registered contain no holes and may
5519 * be freed, this function may be used instead of calling memory_present() manually.
5521 void __init
sparse_memory_present_with_active_regions(int nid
)
5523 unsigned long start_pfn
, end_pfn
;
5526 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
5527 memory_present(this_nid
, start_pfn
, end_pfn
);
5531 * get_pfn_range_for_nid - Return the start and end page frames for a node
5532 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
5533 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
5534 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
5536 * It returns the start and end page frame of a node based on information
5537 * provided by memblock_set_node(). If called for a node
5538 * with no available memory, a warning is printed and the start and end
5541 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
5542 unsigned long *start_pfn
, unsigned long *end_pfn
)
5544 unsigned long this_start_pfn
, this_end_pfn
;
5550 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
5551 *start_pfn
= min(*start_pfn
, this_start_pfn
);
5552 *end_pfn
= max(*end_pfn
, this_end_pfn
);
5555 if (*start_pfn
== -1UL)
5560 * This finds a zone that can be used for ZONE_MOVABLE pages. The
5561 * assumption is made that zones within a node are ordered in monotonic
5562 * increasing memory addresses so that the "highest" populated zone is used
5564 static void __init
find_usable_zone_for_movable(void)
5567 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
5568 if (zone_index
== ZONE_MOVABLE
)
5571 if (arch_zone_highest_possible_pfn
[zone_index
] >
5572 arch_zone_lowest_possible_pfn
[zone_index
])
5576 VM_BUG_ON(zone_index
== -1);
5577 movable_zone
= zone_index
;
5581 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
5582 * because it is sized independent of architecture. Unlike the other zones,
5583 * the starting point for ZONE_MOVABLE is not fixed. It may be different
5584 * in each node depending on the size of each node and how evenly kernelcore
5585 * is distributed. This helper function adjusts the zone ranges
5586 * provided by the architecture for a given node by using the end of the
5587 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5588 * zones within a node are in order of monotonic increases memory addresses
5590 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
5591 unsigned long zone_type
,
5592 unsigned long node_start_pfn
,
5593 unsigned long node_end_pfn
,
5594 unsigned long *zone_start_pfn
,
5595 unsigned long *zone_end_pfn
)
5597 /* Only adjust if ZONE_MOVABLE is on this node */
5598 if (zone_movable_pfn
[nid
]) {
5599 /* Size ZONE_MOVABLE */
5600 if (zone_type
== ZONE_MOVABLE
) {
5601 *zone_start_pfn
= zone_movable_pfn
[nid
];
5602 *zone_end_pfn
= min(node_end_pfn
,
5603 arch_zone_highest_possible_pfn
[movable_zone
]);
5605 /* Check if this whole range is within ZONE_MOVABLE */
5606 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
5607 *zone_start_pfn
= *zone_end_pfn
;
5612 * Return the number of pages a zone spans in a node, including holes
5613 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5615 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5616 unsigned long zone_type
,
5617 unsigned long node_start_pfn
,
5618 unsigned long node_end_pfn
,
5619 unsigned long *zone_start_pfn
,
5620 unsigned long *zone_end_pfn
,
5621 unsigned long *ignored
)
5623 /* When hotadd a new node from cpu_up(), the node should be empty */
5624 if (!node_start_pfn
&& !node_end_pfn
)
5627 /* Get the start and end of the zone */
5628 *zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
5629 *zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
5630 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5631 node_start_pfn
, node_end_pfn
,
5632 zone_start_pfn
, zone_end_pfn
);
5634 /* Check that this node has pages within the zone's required range */
5635 if (*zone_end_pfn
< node_start_pfn
|| *zone_start_pfn
> node_end_pfn
)
5638 /* Move the zone boundaries inside the node if necessary */
5639 *zone_end_pfn
= min(*zone_end_pfn
, node_end_pfn
);
5640 *zone_start_pfn
= max(*zone_start_pfn
, node_start_pfn
);
5642 /* Return the spanned pages */
5643 return *zone_end_pfn
- *zone_start_pfn
;
5647 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
5648 * then all holes in the requested range will be accounted for.
5650 unsigned long __meminit
__absent_pages_in_range(int nid
,
5651 unsigned long range_start_pfn
,
5652 unsigned long range_end_pfn
)
5654 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
5655 unsigned long start_pfn
, end_pfn
;
5658 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5659 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
5660 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
5661 nr_absent
-= end_pfn
- start_pfn
;
5667 * absent_pages_in_range - Return number of page frames in holes within a range
5668 * @start_pfn: The start PFN to start searching for holes
5669 * @end_pfn: The end PFN to stop searching for holes
5671 * It returns the number of pages frames in memory holes within a range.
5673 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
5674 unsigned long end_pfn
)
5676 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
5679 /* Return the number of page frames in holes in a zone on a node */
5680 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5681 unsigned long zone_type
,
5682 unsigned long node_start_pfn
,
5683 unsigned long node_end_pfn
,
5684 unsigned long *ignored
)
5686 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
5687 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
5688 unsigned long zone_start_pfn
, zone_end_pfn
;
5689 unsigned long nr_absent
;
5691 /* When hotadd a new node from cpu_up(), the node should be empty */
5692 if (!node_start_pfn
&& !node_end_pfn
)
5695 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
5696 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
5698 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5699 node_start_pfn
, node_end_pfn
,
5700 &zone_start_pfn
, &zone_end_pfn
);
5701 nr_absent
= __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
5704 * ZONE_MOVABLE handling.
5705 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
5708 if (zone_movable_pfn
[nid
]) {
5709 if (mirrored_kernelcore
) {
5710 unsigned long start_pfn
, end_pfn
;
5711 struct memblock_region
*r
;
5713 for_each_memblock(memory
, r
) {
5714 start_pfn
= clamp(memblock_region_memory_base_pfn(r
),
5715 zone_start_pfn
, zone_end_pfn
);
5716 end_pfn
= clamp(memblock_region_memory_end_pfn(r
),
5717 zone_start_pfn
, zone_end_pfn
);
5719 if (zone_type
== ZONE_MOVABLE
&&
5720 memblock_is_mirror(r
))
5721 nr_absent
+= end_pfn
- start_pfn
;
5723 if (zone_type
== ZONE_NORMAL
&&
5724 !memblock_is_mirror(r
))
5725 nr_absent
+= end_pfn
- start_pfn
;
5728 if (zone_type
== ZONE_NORMAL
)
5729 nr_absent
+= node_end_pfn
- zone_movable_pfn
[nid
];
5736 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5737 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5738 unsigned long zone_type
,
5739 unsigned long node_start_pfn
,
5740 unsigned long node_end_pfn
,
5741 unsigned long *zone_start_pfn
,
5742 unsigned long *zone_end_pfn
,
5743 unsigned long *zones_size
)
5747 *zone_start_pfn
= node_start_pfn
;
5748 for (zone
= 0; zone
< zone_type
; zone
++)
5749 *zone_start_pfn
+= zones_size
[zone
];
5751 *zone_end_pfn
= *zone_start_pfn
+ zones_size
[zone_type
];
5753 return zones_size
[zone_type
];
5756 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5757 unsigned long zone_type
,
5758 unsigned long node_start_pfn
,
5759 unsigned long node_end_pfn
,
5760 unsigned long *zholes_size
)
5765 return zholes_size
[zone_type
];
5768 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5770 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
5771 unsigned long node_start_pfn
,
5772 unsigned long node_end_pfn
,
5773 unsigned long *zones_size
,
5774 unsigned long *zholes_size
)
5776 unsigned long realtotalpages
= 0, totalpages
= 0;
5779 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5780 struct zone
*zone
= pgdat
->node_zones
+ i
;
5781 unsigned long zone_start_pfn
, zone_end_pfn
;
5782 unsigned long size
, real_size
;
5784 size
= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
5790 real_size
= size
- zone_absent_pages_in_node(pgdat
->node_id
, i
,
5791 node_start_pfn
, node_end_pfn
,
5794 zone
->zone_start_pfn
= zone_start_pfn
;
5796 zone
->zone_start_pfn
= 0;
5797 zone
->spanned_pages
= size
;
5798 zone
->present_pages
= real_size
;
5801 realtotalpages
+= real_size
;
5804 pgdat
->node_spanned_pages
= totalpages
;
5805 pgdat
->node_present_pages
= realtotalpages
;
5806 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
5810 #ifndef CONFIG_SPARSEMEM
5812 * Calculate the size of the zone->blockflags rounded to an unsigned long
5813 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5814 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
5815 * round what is now in bits to nearest long in bits, then return it in
5818 static unsigned long __init
usemap_size(unsigned long zone_start_pfn
, unsigned long zonesize
)
5820 unsigned long usemapsize
;
5822 zonesize
+= zone_start_pfn
& (pageblock_nr_pages
-1);
5823 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
5824 usemapsize
= usemapsize
>> pageblock_order
;
5825 usemapsize
*= NR_PAGEBLOCK_BITS
;
5826 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
5828 return usemapsize
/ 8;
5831 static void __init
setup_usemap(struct pglist_data
*pgdat
,
5833 unsigned long zone_start_pfn
,
5834 unsigned long zonesize
)
5836 unsigned long usemapsize
= usemap_size(zone_start_pfn
, zonesize
);
5837 zone
->pageblock_flags
= NULL
;
5839 zone
->pageblock_flags
=
5840 memblock_virt_alloc_node_nopanic(usemapsize
,
5844 static inline void setup_usemap(struct pglist_data
*pgdat
, struct zone
*zone
,
5845 unsigned long zone_start_pfn
, unsigned long zonesize
) {}
5846 #endif /* CONFIG_SPARSEMEM */
5848 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
5850 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
5851 void __paginginit
set_pageblock_order(void)
5855 /* Check that pageblock_nr_pages has not already been setup */
5856 if (pageblock_order
)
5859 if (HPAGE_SHIFT
> PAGE_SHIFT
)
5860 order
= HUGETLB_PAGE_ORDER
;
5862 order
= MAX_ORDER
- 1;
5865 * Assume the largest contiguous order of interest is a huge page.
5866 * This value may be variable depending on boot parameters on IA64 and
5869 pageblock_order
= order
;
5871 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5874 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
5875 * is unused as pageblock_order is set at compile-time. See
5876 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5879 void __paginginit
set_pageblock_order(void)
5883 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5885 static unsigned long __paginginit
calc_memmap_size(unsigned long spanned_pages
,
5886 unsigned long present_pages
)
5888 unsigned long pages
= spanned_pages
;
5891 * Provide a more accurate estimation if there are holes within
5892 * the zone and SPARSEMEM is in use. If there are holes within the
5893 * zone, each populated memory region may cost us one or two extra
5894 * memmap pages due to alignment because memmap pages for each
5895 * populated regions may not naturally algined on page boundary.
5896 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5898 if (spanned_pages
> present_pages
+ (present_pages
>> 4) &&
5899 IS_ENABLED(CONFIG_SPARSEMEM
))
5900 pages
= present_pages
;
5902 return PAGE_ALIGN(pages
* sizeof(struct page
)) >> PAGE_SHIFT
;
5906 * Set up the zone data structures:
5907 * - mark all pages reserved
5908 * - mark all memory queues empty
5909 * - clear the memory bitmaps
5911 * NOTE: pgdat should get zeroed by caller.
5913 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
)
5916 int nid
= pgdat
->node_id
;
5919 pgdat_resize_init(pgdat
);
5920 #ifdef CONFIG_NUMA_BALANCING
5921 spin_lock_init(&pgdat
->numabalancing_migrate_lock
);
5922 pgdat
->numabalancing_migrate_nr_pages
= 0;
5923 pgdat
->numabalancing_migrate_next_window
= jiffies
;
5925 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5926 spin_lock_init(&pgdat
->split_queue_lock
);
5927 INIT_LIST_HEAD(&pgdat
->split_queue
);
5928 pgdat
->split_queue_len
= 0;
5930 init_waitqueue_head(&pgdat
->kswapd_wait
);
5931 init_waitqueue_head(&pgdat
->pfmemalloc_wait
);
5932 #ifdef CONFIG_COMPACTION
5933 init_waitqueue_head(&pgdat
->kcompactd_wait
);
5935 pgdat_page_ext_init(pgdat
);
5937 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5938 struct zone
*zone
= pgdat
->node_zones
+ j
;
5939 unsigned long size
, realsize
, freesize
, memmap_pages
;
5940 unsigned long zone_start_pfn
= zone
->zone_start_pfn
;
5942 size
= zone
->spanned_pages
;
5943 realsize
= freesize
= zone
->present_pages
;
5946 * Adjust freesize so that it accounts for how much memory
5947 * is used by this zone for memmap. This affects the watermark
5948 * and per-cpu initialisations
5950 memmap_pages
= calc_memmap_size(size
, realsize
);
5951 if (!is_highmem_idx(j
)) {
5952 if (freesize
>= memmap_pages
) {
5953 freesize
-= memmap_pages
;
5956 " %s zone: %lu pages used for memmap\n",
5957 zone_names
[j
], memmap_pages
);
5959 pr_warn(" %s zone: %lu pages exceeds freesize %lu\n",
5960 zone_names
[j
], memmap_pages
, freesize
);
5963 /* Account for reserved pages */
5964 if (j
== 0 && freesize
> dma_reserve
) {
5965 freesize
-= dma_reserve
;
5966 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
5967 zone_names
[0], dma_reserve
);
5970 if (!is_highmem_idx(j
))
5971 nr_kernel_pages
+= freesize
;
5972 /* Charge for highmem memmap if there are enough kernel pages */
5973 else if (nr_kernel_pages
> memmap_pages
* 2)
5974 nr_kernel_pages
-= memmap_pages
;
5975 nr_all_pages
+= freesize
;
5978 * Set an approximate value for lowmem here, it will be adjusted
5979 * when the bootmem allocator frees pages into the buddy system.
5980 * And all highmem pages will be managed by the buddy system.
5982 zone
->managed_pages
= is_highmem_idx(j
) ? realsize
: freesize
;
5985 zone
->min_unmapped_pages
= (freesize
*sysctl_min_unmapped_ratio
)
5987 zone
->min_slab_pages
= (freesize
* sysctl_min_slab_ratio
) / 100;
5989 zone
->name
= zone_names
[j
];
5990 spin_lock_init(&zone
->lock
);
5991 spin_lock_init(&zone
->lru_lock
);
5992 zone_seqlock_init(zone
);
5993 zone
->zone_pgdat
= pgdat
;
5994 zone_pcp_init(zone
);
5996 /* For bootup, initialized properly in watermark setup */
5997 mod_zone_page_state(zone
, NR_ALLOC_BATCH
, zone
->managed_pages
);
5999 lruvec_init(&zone
->lruvec
);
6003 set_pageblock_order();
6004 setup_usemap(pgdat
, zone
, zone_start_pfn
, size
);
6005 ret
= init_currently_empty_zone(zone
, zone_start_pfn
, size
);
6007 memmap_init(size
, nid
, j
, zone_start_pfn
);
6011 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
6013 unsigned long __maybe_unused start
= 0;
6014 unsigned long __maybe_unused offset
= 0;
6016 /* Skip empty nodes */
6017 if (!pgdat
->node_spanned_pages
)
6020 #ifdef CONFIG_FLAT_NODE_MEM_MAP
6021 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
6022 offset
= pgdat
->node_start_pfn
- start
;
6023 /* ia64 gets its own node_mem_map, before this, without bootmem */
6024 if (!pgdat
->node_mem_map
) {
6025 unsigned long size
, end
;
6029 * The zone's endpoints aren't required to be MAX_ORDER
6030 * aligned but the node_mem_map endpoints must be in order
6031 * for the buddy allocator to function correctly.
6033 end
= pgdat_end_pfn(pgdat
);
6034 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
6035 size
= (end
- start
) * sizeof(struct page
);
6036 map
= alloc_remap(pgdat
->node_id
, size
);
6038 map
= memblock_virt_alloc_node_nopanic(size
,
6040 pgdat
->node_mem_map
= map
+ offset
;
6042 #ifndef CONFIG_NEED_MULTIPLE_NODES
6044 * With no DISCONTIG, the global mem_map is just set as node 0's
6046 if (pgdat
== NODE_DATA(0)) {
6047 mem_map
= NODE_DATA(0)->node_mem_map
;
6048 #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
6049 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
6051 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6054 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
6057 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
6058 unsigned long node_start_pfn
, unsigned long *zholes_size
)
6060 pg_data_t
*pgdat
= NODE_DATA(nid
);
6061 unsigned long start_pfn
= 0;
6062 unsigned long end_pfn
= 0;
6064 /* pg_data_t should be reset to zero when it's allocated */
6065 WARN_ON(pgdat
->nr_zones
|| pgdat
->classzone_idx
);
6067 reset_deferred_meminit(pgdat
);
6068 pgdat
->node_id
= nid
;
6069 pgdat
->node_start_pfn
= node_start_pfn
;
6070 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6071 get_pfn_range_for_nid(nid
, &start_pfn
, &end_pfn
);
6072 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid
,
6073 (u64
)start_pfn
<< PAGE_SHIFT
,
6074 end_pfn
? ((u64
)end_pfn
<< PAGE_SHIFT
) - 1 : 0);
6076 start_pfn
= node_start_pfn
;
6078 calculate_node_totalpages(pgdat
, start_pfn
, end_pfn
,
6079 zones_size
, zholes_size
);
6081 alloc_node_mem_map(pgdat
);
6082 #ifdef CONFIG_FLAT_NODE_MEM_MAP
6083 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
6084 nid
, (unsigned long)pgdat
,
6085 (unsigned long)pgdat
->node_mem_map
);
6088 free_area_init_core(pgdat
);
6091 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6093 #if MAX_NUMNODES > 1
6095 * Figure out the number of possible node ids.
6097 void __init
setup_nr_node_ids(void)
6099 unsigned int highest
;
6101 highest
= find_last_bit(node_possible_map
.bits
, MAX_NUMNODES
);
6102 nr_node_ids
= highest
+ 1;
6107 * node_map_pfn_alignment - determine the maximum internode alignment
6109 * This function should be called after node map is populated and sorted.
6110 * It calculates the maximum power of two alignment which can distinguish
6113 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
6114 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
6115 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
6116 * shifted, 1GiB is enough and this function will indicate so.
6118 * This is used to test whether pfn -> nid mapping of the chosen memory
6119 * model has fine enough granularity to avoid incorrect mapping for the
6120 * populated node map.
6122 * Returns the determined alignment in pfn's. 0 if there is no alignment
6123 * requirement (single node).
6125 unsigned long __init
node_map_pfn_alignment(void)
6127 unsigned long accl_mask
= 0, last_end
= 0;
6128 unsigned long start
, end
, mask
;
6132 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
6133 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
6140 * Start with a mask granular enough to pin-point to the
6141 * start pfn and tick off bits one-by-one until it becomes
6142 * too coarse to separate the current node from the last.
6144 mask
= ~((1 << __ffs(start
)) - 1);
6145 while (mask
&& last_end
<= (start
& (mask
<< 1)))
6148 /* accumulate all internode masks */
6152 /* convert mask to number of pages */
6153 return ~accl_mask
+ 1;
6156 /* Find the lowest pfn for a node */
6157 static unsigned long __init
find_min_pfn_for_node(int nid
)
6159 unsigned long min_pfn
= ULONG_MAX
;
6160 unsigned long start_pfn
;
6163 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
6164 min_pfn
= min(min_pfn
, start_pfn
);
6166 if (min_pfn
== ULONG_MAX
) {
6167 pr_warn("Could not find start_pfn for node %d\n", nid
);
6175 * find_min_pfn_with_active_regions - Find the minimum PFN registered
6177 * It returns the minimum PFN based on information provided via
6178 * memblock_set_node().
6180 unsigned long __init
find_min_pfn_with_active_regions(void)
6182 return find_min_pfn_for_node(MAX_NUMNODES
);
6186 * early_calculate_totalpages()
6187 * Sum pages in active regions for movable zone.
6188 * Populate N_MEMORY for calculating usable_nodes.
6190 static unsigned long __init
early_calculate_totalpages(void)
6192 unsigned long totalpages
= 0;
6193 unsigned long start_pfn
, end_pfn
;
6196 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
6197 unsigned long pages
= end_pfn
- start_pfn
;
6199 totalpages
+= pages
;
6201 node_set_state(nid
, N_MEMORY
);
6207 * Find the PFN the Movable zone begins in each node. Kernel memory
6208 * is spread evenly between nodes as long as the nodes have enough
6209 * memory. When they don't, some nodes will have more kernelcore than
6212 static void __init
find_zone_movable_pfns_for_nodes(void)
6215 unsigned long usable_startpfn
;
6216 unsigned long kernelcore_node
, kernelcore_remaining
;
6217 /* save the state before borrow the nodemask */
6218 nodemask_t saved_node_state
= node_states
[N_MEMORY
];
6219 unsigned long totalpages
= early_calculate_totalpages();
6220 int usable_nodes
= nodes_weight(node_states
[N_MEMORY
]);
6221 struct memblock_region
*r
;
6223 /* Need to find movable_zone earlier when movable_node is specified. */
6224 find_usable_zone_for_movable();
6227 * If movable_node is specified, ignore kernelcore and movablecore
6230 if (movable_node_is_enabled()) {
6231 for_each_memblock(memory
, r
) {
6232 if (!memblock_is_hotpluggable(r
))
6237 usable_startpfn
= PFN_DOWN(r
->base
);
6238 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
6239 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
6247 * If kernelcore=mirror is specified, ignore movablecore option
6249 if (mirrored_kernelcore
) {
6250 bool mem_below_4gb_not_mirrored
= false;
6252 for_each_memblock(memory
, r
) {
6253 if (memblock_is_mirror(r
))
6258 usable_startpfn
= memblock_region_memory_base_pfn(r
);
6260 if (usable_startpfn
< 0x100000) {
6261 mem_below_4gb_not_mirrored
= true;
6265 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
6266 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
6270 if (mem_below_4gb_not_mirrored
)
6271 pr_warn("This configuration results in unmirrored kernel memory.");
6277 * If movablecore=nn[KMG] was specified, calculate what size of
6278 * kernelcore that corresponds so that memory usable for
6279 * any allocation type is evenly spread. If both kernelcore
6280 * and movablecore are specified, then the value of kernelcore
6281 * will be used for required_kernelcore if it's greater than
6282 * what movablecore would have allowed.
6284 if (required_movablecore
) {
6285 unsigned long corepages
;
6288 * Round-up so that ZONE_MOVABLE is at least as large as what
6289 * was requested by the user
6291 required_movablecore
=
6292 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
6293 required_movablecore
= min(totalpages
, required_movablecore
);
6294 corepages
= totalpages
- required_movablecore
;
6296 required_kernelcore
= max(required_kernelcore
, corepages
);
6300 * If kernelcore was not specified or kernelcore size is larger
6301 * than totalpages, there is no ZONE_MOVABLE.
6303 if (!required_kernelcore
|| required_kernelcore
>= totalpages
)
6306 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
6307 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
6310 /* Spread kernelcore memory as evenly as possible throughout nodes */
6311 kernelcore_node
= required_kernelcore
/ usable_nodes
;
6312 for_each_node_state(nid
, N_MEMORY
) {
6313 unsigned long start_pfn
, end_pfn
;
6316 * Recalculate kernelcore_node if the division per node
6317 * now exceeds what is necessary to satisfy the requested
6318 * amount of memory for the kernel
6320 if (required_kernelcore
< kernelcore_node
)
6321 kernelcore_node
= required_kernelcore
/ usable_nodes
;
6324 * As the map is walked, we track how much memory is usable
6325 * by the kernel using kernelcore_remaining. When it is
6326 * 0, the rest of the node is usable by ZONE_MOVABLE
6328 kernelcore_remaining
= kernelcore_node
;
6330 /* Go through each range of PFNs within this node */
6331 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
6332 unsigned long size_pages
;
6334 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
6335 if (start_pfn
>= end_pfn
)
6338 /* Account for what is only usable for kernelcore */
6339 if (start_pfn
< usable_startpfn
) {
6340 unsigned long kernel_pages
;
6341 kernel_pages
= min(end_pfn
, usable_startpfn
)
6344 kernelcore_remaining
-= min(kernel_pages
,
6345 kernelcore_remaining
);
6346 required_kernelcore
-= min(kernel_pages
,
6347 required_kernelcore
);
6349 /* Continue if range is now fully accounted */
6350 if (end_pfn
<= usable_startpfn
) {
6353 * Push zone_movable_pfn to the end so
6354 * that if we have to rebalance
6355 * kernelcore across nodes, we will
6356 * not double account here
6358 zone_movable_pfn
[nid
] = end_pfn
;
6361 start_pfn
= usable_startpfn
;
6365 * The usable PFN range for ZONE_MOVABLE is from
6366 * start_pfn->end_pfn. Calculate size_pages as the
6367 * number of pages used as kernelcore
6369 size_pages
= end_pfn
- start_pfn
;
6370 if (size_pages
> kernelcore_remaining
)
6371 size_pages
= kernelcore_remaining
;
6372 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
6375 * Some kernelcore has been met, update counts and
6376 * break if the kernelcore for this node has been
6379 required_kernelcore
-= min(required_kernelcore
,
6381 kernelcore_remaining
-= size_pages
;
6382 if (!kernelcore_remaining
)
6388 * If there is still required_kernelcore, we do another pass with one
6389 * less node in the count. This will push zone_movable_pfn[nid] further
6390 * along on the nodes that still have memory until kernelcore is
6394 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
6398 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
6399 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
6400 zone_movable_pfn
[nid
] =
6401 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
6404 /* restore the node_state */
6405 node_states
[N_MEMORY
] = saved_node_state
;
6408 /* Any regular or high memory on that node ? */
6409 static void check_for_memory(pg_data_t
*pgdat
, int nid
)
6411 enum zone_type zone_type
;
6413 if (N_MEMORY
== N_NORMAL_MEMORY
)
6416 for (zone_type
= 0; zone_type
<= ZONE_MOVABLE
- 1; zone_type
++) {
6417 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
6418 if (populated_zone(zone
)) {
6419 node_set_state(nid
, N_HIGH_MEMORY
);
6420 if (N_NORMAL_MEMORY
!= N_HIGH_MEMORY
&&
6421 zone_type
<= ZONE_NORMAL
)
6422 node_set_state(nid
, N_NORMAL_MEMORY
);
6429 * free_area_init_nodes - Initialise all pg_data_t and zone data
6430 * @max_zone_pfn: an array of max PFNs for each zone
6432 * This will call free_area_init_node() for each active node in the system.
6433 * Using the page ranges provided by memblock_set_node(), the size of each
6434 * zone in each node and their holes is calculated. If the maximum PFN
6435 * between two adjacent zones match, it is assumed that the zone is empty.
6436 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
6437 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
6438 * starts where the previous one ended. For example, ZONE_DMA32 starts
6439 * at arch_max_dma_pfn.
6441 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
6443 unsigned long start_pfn
, end_pfn
;
6446 /* Record where the zone boundaries are */
6447 memset(arch_zone_lowest_possible_pfn
, 0,
6448 sizeof(arch_zone_lowest_possible_pfn
));
6449 memset(arch_zone_highest_possible_pfn
, 0,
6450 sizeof(arch_zone_highest_possible_pfn
));
6451 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
6452 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
6453 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
6454 if (i
== ZONE_MOVABLE
)
6456 arch_zone_lowest_possible_pfn
[i
] =
6457 arch_zone_highest_possible_pfn
[i
-1];
6458 arch_zone_highest_possible_pfn
[i
] =
6459 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
6461 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
6462 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
6464 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
6465 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
6466 find_zone_movable_pfns_for_nodes();
6468 /* Print out the zone ranges */
6469 pr_info("Zone ranges:\n");
6470 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
6471 if (i
== ZONE_MOVABLE
)
6473 pr_info(" %-8s ", zone_names
[i
]);
6474 if (arch_zone_lowest_possible_pfn
[i
] ==
6475 arch_zone_highest_possible_pfn
[i
])
6478 pr_cont("[mem %#018Lx-%#018Lx]\n",
6479 (u64
)arch_zone_lowest_possible_pfn
[i
]
6481 ((u64
)arch_zone_highest_possible_pfn
[i
]
6482 << PAGE_SHIFT
) - 1);
6485 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
6486 pr_info("Movable zone start for each node\n");
6487 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
6488 if (zone_movable_pfn
[i
])
6489 pr_info(" Node %d: %#018Lx\n", i
,
6490 (u64
)zone_movable_pfn
[i
] << PAGE_SHIFT
);
6493 /* Print out the early node map */
6494 pr_info("Early memory node ranges\n");
6495 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
6496 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid
,
6497 (u64
)start_pfn
<< PAGE_SHIFT
,
6498 ((u64
)end_pfn
<< PAGE_SHIFT
) - 1);
6500 /* Initialise every node */
6501 mminit_verify_pageflags_layout();
6502 setup_nr_node_ids();
6503 for_each_online_node(nid
) {
6504 pg_data_t
*pgdat
= NODE_DATA(nid
);
6505 free_area_init_node(nid
, NULL
,
6506 find_min_pfn_for_node(nid
), NULL
);
6508 /* Any memory on that node */
6509 if (pgdat
->node_present_pages
)
6510 node_set_state(nid
, N_MEMORY
);
6511 check_for_memory(pgdat
, nid
);
6515 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
6517 unsigned long long coremem
;
6521 coremem
= memparse(p
, &p
);
6522 *core
= coremem
>> PAGE_SHIFT
;
6524 /* Paranoid check that UL is enough for the coremem value */
6525 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
6531 * kernelcore=size sets the amount of memory for use for allocations that
6532 * cannot be reclaimed or migrated.
6534 static int __init
cmdline_parse_kernelcore(char *p
)
6536 /* parse kernelcore=mirror */
6537 if (parse_option_str(p
, "mirror")) {
6538 mirrored_kernelcore
= true;
6542 return cmdline_parse_core(p
, &required_kernelcore
);
6546 * movablecore=size sets the amount of memory for use for allocations that
6547 * can be reclaimed or migrated.
6549 static int __init
cmdline_parse_movablecore(char *p
)
6551 return cmdline_parse_core(p
, &required_movablecore
);
6554 early_param("kernelcore", cmdline_parse_kernelcore
);
6555 early_param("movablecore", cmdline_parse_movablecore
);
6557 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6559 void adjust_managed_page_count(struct page
*page
, long count
)
6561 spin_lock(&managed_page_count_lock
);
6562 page_zone(page
)->managed_pages
+= count
;
6563 totalram_pages
+= count
;
6564 #ifdef CONFIG_HIGHMEM
6565 if (PageHighMem(page
))
6566 totalhigh_pages
+= count
;
6568 spin_unlock(&managed_page_count_lock
);
6570 EXPORT_SYMBOL(adjust_managed_page_count
);
6572 unsigned long free_reserved_area(void *start
, void *end
, int poison
, char *s
)
6575 unsigned long pages
= 0;
6577 start
= (void *)PAGE_ALIGN((unsigned long)start
);
6578 end
= (void *)((unsigned long)end
& PAGE_MASK
);
6579 for (pos
= start
; pos
< end
; pos
+= PAGE_SIZE
, pages
++) {
6580 if ((unsigned int)poison
<= 0xFF)
6581 memset(pos
, poison
, PAGE_SIZE
);
6582 free_reserved_page(virt_to_page(pos
));
6586 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
6587 s
, pages
<< (PAGE_SHIFT
- 10), start
, end
);
6591 EXPORT_SYMBOL(free_reserved_area
);
6593 #ifdef CONFIG_HIGHMEM
6594 void free_highmem_page(struct page
*page
)
6596 __free_reserved_page(page
);
6598 page_zone(page
)->managed_pages
++;
6604 void __init
mem_init_print_info(const char *str
)
6606 unsigned long physpages
, codesize
, datasize
, rosize
, bss_size
;
6607 unsigned long init_code_size
, init_data_size
;
6609 physpages
= get_num_physpages();
6610 codesize
= _etext
- _stext
;
6611 datasize
= _edata
- _sdata
;
6612 rosize
= __end_rodata
- __start_rodata
;
6613 bss_size
= __bss_stop
- __bss_start
;
6614 init_data_size
= __init_end
- __init_begin
;
6615 init_code_size
= _einittext
- _sinittext
;
6618 * Detect special cases and adjust section sizes accordingly:
6619 * 1) .init.* may be embedded into .data sections
6620 * 2) .init.text.* may be out of [__init_begin, __init_end],
6621 * please refer to arch/tile/kernel/vmlinux.lds.S.
6622 * 3) .rodata.* may be embedded into .text or .data sections.
6624 #define adj_init_size(start, end, size, pos, adj) \
6626 if (start <= pos && pos < end && size > adj) \
6630 adj_init_size(__init_begin
, __init_end
, init_data_size
,
6631 _sinittext
, init_code_size
);
6632 adj_init_size(_stext
, _etext
, codesize
, _sinittext
, init_code_size
);
6633 adj_init_size(_sdata
, _edata
, datasize
, __init_begin
, init_data_size
);
6634 adj_init_size(_stext
, _etext
, codesize
, __start_rodata
, rosize
);
6635 adj_init_size(_sdata
, _edata
, datasize
, __start_rodata
, rosize
);
6637 #undef adj_init_size
6639 pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
6640 #ifdef CONFIG_HIGHMEM
6644 nr_free_pages() << (PAGE_SHIFT
- 10),
6645 physpages
<< (PAGE_SHIFT
- 10),
6646 codesize
>> 10, datasize
>> 10, rosize
>> 10,
6647 (init_data_size
+ init_code_size
) >> 10, bss_size
>> 10,
6648 (physpages
- totalram_pages
- totalcma_pages
) << (PAGE_SHIFT
- 10),
6649 totalcma_pages
<< (PAGE_SHIFT
- 10),
6650 #ifdef CONFIG_HIGHMEM
6651 totalhigh_pages
<< (PAGE_SHIFT
- 10),
6653 str
? ", " : "", str
? str
: "");
6657 * set_dma_reserve - set the specified number of pages reserved in the first zone
6658 * @new_dma_reserve: The number of pages to mark reserved
6660 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
6661 * In the DMA zone, a significant percentage may be consumed by kernel image
6662 * and other unfreeable allocations which can skew the watermarks badly. This
6663 * function may optionally be used to account for unfreeable pages in the
6664 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6665 * smaller per-cpu batchsize.
6667 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
6669 dma_reserve
= new_dma_reserve
;
6672 void __init
free_area_init(unsigned long *zones_size
)
6674 free_area_init_node(0, zones_size
,
6675 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
6678 static int page_alloc_cpu_notify(struct notifier_block
*self
,
6679 unsigned long action
, void *hcpu
)
6681 int cpu
= (unsigned long)hcpu
;
6683 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
6684 lru_add_drain_cpu(cpu
);
6688 * Spill the event counters of the dead processor
6689 * into the current processors event counters.
6690 * This artificially elevates the count of the current
6693 vm_events_fold_cpu(cpu
);
6696 * Zero the differential counters of the dead processor
6697 * so that the vm statistics are consistent.
6699 * This is only okay since the processor is dead and cannot
6700 * race with what we are doing.
6702 cpu_vm_stats_fold(cpu
);
6707 void __init
page_alloc_init(void)
6709 hotcpu_notifier(page_alloc_cpu_notify
, 0);
6713 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
6714 * or min_free_kbytes changes.
6716 static void calculate_totalreserve_pages(void)
6718 struct pglist_data
*pgdat
;
6719 unsigned long reserve_pages
= 0;
6720 enum zone_type i
, j
;
6722 for_each_online_pgdat(pgdat
) {
6723 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
6724 struct zone
*zone
= pgdat
->node_zones
+ i
;
6727 /* Find valid and maximum lowmem_reserve in the zone */
6728 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
6729 if (zone
->lowmem_reserve
[j
] > max
)
6730 max
= zone
->lowmem_reserve
[j
];
6733 /* we treat the high watermark as reserved pages. */
6734 max
+= high_wmark_pages(zone
);
6736 if (max
> zone
->managed_pages
)
6737 max
= zone
->managed_pages
;
6739 zone
->totalreserve_pages
= max
;
6741 reserve_pages
+= max
;
6744 totalreserve_pages
= reserve_pages
;
6748 * setup_per_zone_lowmem_reserve - called whenever
6749 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
6750 * has a correct pages reserved value, so an adequate number of
6751 * pages are left in the zone after a successful __alloc_pages().
6753 static void setup_per_zone_lowmem_reserve(void)
6755 struct pglist_data
*pgdat
;
6756 enum zone_type j
, idx
;
6758 for_each_online_pgdat(pgdat
) {
6759 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
6760 struct zone
*zone
= pgdat
->node_zones
+ j
;
6761 unsigned long managed_pages
= zone
->managed_pages
;
6763 zone
->lowmem_reserve
[j
] = 0;
6767 struct zone
*lower_zone
;
6771 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
6772 sysctl_lowmem_reserve_ratio
[idx
] = 1;
6774 lower_zone
= pgdat
->node_zones
+ idx
;
6775 lower_zone
->lowmem_reserve
[j
] = managed_pages
/
6776 sysctl_lowmem_reserve_ratio
[idx
];
6777 managed_pages
+= lower_zone
->managed_pages
;
6782 /* update totalreserve_pages */
6783 calculate_totalreserve_pages();
6786 static void __setup_per_zone_wmarks(void)
6788 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
6789 unsigned long lowmem_pages
= 0;
6791 unsigned long flags
;
6793 /* Calculate total number of !ZONE_HIGHMEM pages */
6794 for_each_zone(zone
) {
6795 if (!is_highmem(zone
))
6796 lowmem_pages
+= zone
->managed_pages
;
6799 for_each_zone(zone
) {
6802 spin_lock_irqsave(&zone
->lock
, flags
);
6803 tmp
= (u64
)pages_min
* zone
->managed_pages
;
6804 do_div(tmp
, lowmem_pages
);
6805 if (is_highmem(zone
)) {
6807 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6808 * need highmem pages, so cap pages_min to a small
6811 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
6812 * deltas control asynch page reclaim, and so should
6813 * not be capped for highmem.
6815 unsigned long min_pages
;
6817 min_pages
= zone
->managed_pages
/ 1024;
6818 min_pages
= clamp(min_pages
, SWAP_CLUSTER_MAX
, 128UL);
6819 zone
->watermark
[WMARK_MIN
] = min_pages
;
6822 * If it's a lowmem zone, reserve a number of pages
6823 * proportionate to the zone's size.
6825 zone
->watermark
[WMARK_MIN
] = tmp
;
6829 * Set the kswapd watermarks distance according to the
6830 * scale factor in proportion to available memory, but
6831 * ensure a minimum size on small systems.
6833 tmp
= max_t(u64
, tmp
>> 2,
6834 mult_frac(zone
->managed_pages
,
6835 watermark_scale_factor
, 10000));
6837 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + tmp
;
6838 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + tmp
* 2;
6840 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
6841 high_wmark_pages(zone
) - low_wmark_pages(zone
) -
6842 atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]));
6844 spin_unlock_irqrestore(&zone
->lock
, flags
);
6847 /* update totalreserve_pages */
6848 calculate_totalreserve_pages();
6852 * setup_per_zone_wmarks - called when min_free_kbytes changes
6853 * or when memory is hot-{added|removed}
6855 * Ensures that the watermark[min,low,high] values for each zone are set
6856 * correctly with respect to min_free_kbytes.
6858 void setup_per_zone_wmarks(void)
6860 mutex_lock(&zonelists_mutex
);
6861 __setup_per_zone_wmarks();
6862 mutex_unlock(&zonelists_mutex
);
6866 * Initialise min_free_kbytes.
6868 * For small machines we want it small (128k min). For large machines
6869 * we want it large (64MB max). But it is not linear, because network
6870 * bandwidth does not increase linearly with machine size. We use
6872 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
6873 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6889 int __meminit
init_per_zone_wmark_min(void)
6891 unsigned long lowmem_kbytes
;
6892 int new_min_free_kbytes
;
6894 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
6895 new_min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
6897 if (new_min_free_kbytes
> user_min_free_kbytes
) {
6898 min_free_kbytes
= new_min_free_kbytes
;
6899 if (min_free_kbytes
< 128)
6900 min_free_kbytes
= 128;
6901 if (min_free_kbytes
> 65536)
6902 min_free_kbytes
= 65536;
6904 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6905 new_min_free_kbytes
, user_min_free_kbytes
);
6907 setup_per_zone_wmarks();
6908 refresh_zone_stat_thresholds();
6909 setup_per_zone_lowmem_reserve();
6912 core_initcall(init_per_zone_wmark_min
)
6915 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
6916 * that we can call two helper functions whenever min_free_kbytes
6919 int min_free_kbytes_sysctl_handler(struct ctl_table
*table
, int write
,
6920 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6924 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6929 user_min_free_kbytes
= min_free_kbytes
;
6930 setup_per_zone_wmarks();
6935 int watermark_scale_factor_sysctl_handler(struct ctl_table
*table
, int write
,
6936 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6940 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6945 setup_per_zone_wmarks();
6951 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6952 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6957 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6962 zone
->min_unmapped_pages
= (zone
->managed_pages
*
6963 sysctl_min_unmapped_ratio
) / 100;
6967 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6968 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6973 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6978 zone
->min_slab_pages
= (zone
->managed_pages
*
6979 sysctl_min_slab_ratio
) / 100;
6985 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6986 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6987 * whenever sysctl_lowmem_reserve_ratio changes.
6989 * The reserve ratio obviously has absolutely no relation with the
6990 * minimum watermarks. The lowmem reserve ratio can only make sense
6991 * if in function of the boot time zone sizes.
6993 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6994 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6996 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6997 setup_per_zone_lowmem_reserve();
7002 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
7003 * cpu. It is the fraction of total pages in each zone that a hot per cpu
7004 * pagelist can have before it gets flushed back to buddy allocator.
7006 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table
*table
, int write
,
7007 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
7010 int old_percpu_pagelist_fraction
;
7013 mutex_lock(&pcp_batch_high_lock
);
7014 old_percpu_pagelist_fraction
= percpu_pagelist_fraction
;
7016 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
7017 if (!write
|| ret
< 0)
7020 /* Sanity checking to avoid pcp imbalance */
7021 if (percpu_pagelist_fraction
&&
7022 percpu_pagelist_fraction
< MIN_PERCPU_PAGELIST_FRACTION
) {
7023 percpu_pagelist_fraction
= old_percpu_pagelist_fraction
;
7029 if (percpu_pagelist_fraction
== old_percpu_pagelist_fraction
)
7032 for_each_populated_zone(zone
) {
7035 for_each_possible_cpu(cpu
)
7036 pageset_set_high_and_batch(zone
,
7037 per_cpu_ptr(zone
->pageset
, cpu
));
7040 mutex_unlock(&pcp_batch_high_lock
);
7045 int hashdist
= HASHDIST_DEFAULT
;
7047 static int __init
set_hashdist(char *str
)
7051 hashdist
= simple_strtoul(str
, &str
, 0);
7054 __setup("hashdist=", set_hashdist
);
7058 * allocate a large system hash table from bootmem
7059 * - it is assumed that the hash table must contain an exact power-of-2
7060 * quantity of entries
7061 * - limit is the number of hash buckets, not the total allocation size
7063 void *__init
alloc_large_system_hash(const char *tablename
,
7064 unsigned long bucketsize
,
7065 unsigned long numentries
,
7068 unsigned int *_hash_shift
,
7069 unsigned int *_hash_mask
,
7070 unsigned long low_limit
,
7071 unsigned long high_limit
)
7073 unsigned long long max
= high_limit
;
7074 unsigned long log2qty
, size
;
7077 /* allow the kernel cmdline to have a say */
7079 /* round applicable memory size up to nearest megabyte */
7080 numentries
= nr_kernel_pages
;
7082 /* It isn't necessary when PAGE_SIZE >= 1MB */
7083 if (PAGE_SHIFT
< 20)
7084 numentries
= round_up(numentries
, (1<<20)/PAGE_SIZE
);
7086 /* limit to 1 bucket per 2^scale bytes of low memory */
7087 if (scale
> PAGE_SHIFT
)
7088 numentries
>>= (scale
- PAGE_SHIFT
);
7090 numentries
<<= (PAGE_SHIFT
- scale
);
7092 /* Make sure we've got at least a 0-order allocation.. */
7093 if (unlikely(flags
& HASH_SMALL
)) {
7094 /* Makes no sense without HASH_EARLY */
7095 WARN_ON(!(flags
& HASH_EARLY
));
7096 if (!(numentries
>> *_hash_shift
)) {
7097 numentries
= 1UL << *_hash_shift
;
7098 BUG_ON(!numentries
);
7100 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
7101 numentries
= PAGE_SIZE
/ bucketsize
;
7103 numentries
= roundup_pow_of_two(numentries
);
7105 /* limit allocation size to 1/16 total memory by default */
7107 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
7108 do_div(max
, bucketsize
);
7110 max
= min(max
, 0x80000000ULL
);
7112 if (numentries
< low_limit
)
7113 numentries
= low_limit
;
7114 if (numentries
> max
)
7117 log2qty
= ilog2(numentries
);
7120 size
= bucketsize
<< log2qty
;
7121 if (flags
& HASH_EARLY
)
7122 table
= memblock_virt_alloc_nopanic(size
, 0);
7124 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
7127 * If bucketsize is not a power-of-two, we may free
7128 * some pages at the end of hash table which
7129 * alloc_pages_exact() automatically does
7131 if (get_order(size
) < MAX_ORDER
) {
7132 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
7133 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
7136 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
7139 panic("Failed to allocate %s hash table\n", tablename
);
7141 pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n",
7142 tablename
, 1UL << log2qty
, ilog2(size
) - PAGE_SHIFT
, size
);
7145 *_hash_shift
= log2qty
;
7147 *_hash_mask
= (1 << log2qty
) - 1;
7153 * This function checks whether pageblock includes unmovable pages or not.
7154 * If @count is not zero, it is okay to include less @count unmovable pages
7156 * PageLRU check without isolation or lru_lock could race so that
7157 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
7158 * expect this function should be exact.
7160 bool has_unmovable_pages(struct zone
*zone
, struct page
*page
, int count
,
7161 bool skip_hwpoisoned_pages
)
7163 unsigned long pfn
, iter
, found
;
7167 * For avoiding noise data, lru_add_drain_all() should be called
7168 * If ZONE_MOVABLE, the zone never contains unmovable pages
7170 if (zone_idx(zone
) == ZONE_MOVABLE
)
7172 mt
= get_pageblock_migratetype(page
);
7173 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
7176 pfn
= page_to_pfn(page
);
7177 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
7178 unsigned long check
= pfn
+ iter
;
7180 if (!pfn_valid_within(check
))
7183 page
= pfn_to_page(check
);
7186 * Hugepages are not in LRU lists, but they're movable.
7187 * We need not scan over tail pages bacause we don't
7188 * handle each tail page individually in migration.
7190 if (PageHuge(page
)) {
7191 iter
= round_up(iter
+ 1, 1<<compound_order(page
)) - 1;
7196 * We can't use page_count without pin a page
7197 * because another CPU can free compound page.
7198 * This check already skips compound tails of THP
7199 * because their page->_refcount is zero at all time.
7201 if (!page_ref_count(page
)) {
7202 if (PageBuddy(page
))
7203 iter
+= (1 << page_order(page
)) - 1;
7208 * The HWPoisoned page may be not in buddy system, and
7209 * page_count() is not 0.
7211 if (skip_hwpoisoned_pages
&& PageHWPoison(page
))
7217 * If there are RECLAIMABLE pages, we need to check
7218 * it. But now, memory offline itself doesn't call
7219 * shrink_node_slabs() and it still to be fixed.
7222 * If the page is not RAM, page_count()should be 0.
7223 * we don't need more check. This is an _used_ not-movable page.
7225 * The problematic thing here is PG_reserved pages. PG_reserved
7226 * is set to both of a memory hole page and a _used_ kernel
7235 bool is_pageblock_removable_nolock(struct page
*page
)
7241 * We have to be careful here because we are iterating over memory
7242 * sections which are not zone aware so we might end up outside of
7243 * the zone but still within the section.
7244 * We have to take care about the node as well. If the node is offline
7245 * its NODE_DATA will be NULL - see page_zone.
7247 if (!node_online(page_to_nid(page
)))
7250 zone
= page_zone(page
);
7251 pfn
= page_to_pfn(page
);
7252 if (!zone_spans_pfn(zone
, pfn
))
7255 return !has_unmovable_pages(zone
, page
, 0, true);
7258 #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
7260 static unsigned long pfn_max_align_down(unsigned long pfn
)
7262 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
7263 pageblock_nr_pages
) - 1);
7266 static unsigned long pfn_max_align_up(unsigned long pfn
)
7268 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
7269 pageblock_nr_pages
));
7272 /* [start, end) must belong to a single zone. */
7273 static int __alloc_contig_migrate_range(struct compact_control
*cc
,
7274 unsigned long start
, unsigned long end
)
7276 /* This function is based on compact_zone() from compaction.c. */
7277 unsigned long nr_reclaimed
;
7278 unsigned long pfn
= start
;
7279 unsigned int tries
= 0;
7284 while (pfn
< end
|| !list_empty(&cc
->migratepages
)) {
7285 if (fatal_signal_pending(current
)) {
7290 if (list_empty(&cc
->migratepages
)) {
7291 cc
->nr_migratepages
= 0;
7292 pfn
= isolate_migratepages_range(cc
, pfn
, end
);
7298 } else if (++tries
== 5) {
7299 ret
= ret
< 0 ? ret
: -EBUSY
;
7303 nr_reclaimed
= reclaim_clean_pages_from_list(cc
->zone
,
7305 cc
->nr_migratepages
-= nr_reclaimed
;
7307 ret
= migrate_pages(&cc
->migratepages
, alloc_migrate_target
,
7308 NULL
, 0, cc
->mode
, MR_CMA
);
7311 putback_movable_pages(&cc
->migratepages
);
7318 * alloc_contig_range() -- tries to allocate given range of pages
7319 * @start: start PFN to allocate
7320 * @end: one-past-the-last PFN to allocate
7321 * @migratetype: migratetype of the underlaying pageblocks (either
7322 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
7323 * in range must have the same migratetype and it must
7324 * be either of the two.
7326 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
7327 * aligned, however it's the caller's responsibility to guarantee that
7328 * we are the only thread that changes migrate type of pageblocks the
7331 * The PFN range must belong to a single zone.
7333 * Returns zero on success or negative error code. On success all
7334 * pages which PFN is in [start, end) are allocated for the caller and
7335 * need to be freed with free_contig_range().
7337 int alloc_contig_range(unsigned long start
, unsigned long end
,
7338 unsigned migratetype
)
7340 unsigned long outer_start
, outer_end
;
7344 struct compact_control cc
= {
7345 .nr_migratepages
= 0,
7347 .zone
= page_zone(pfn_to_page(start
)),
7348 .mode
= MIGRATE_SYNC
,
7349 .ignore_skip_hint
= true,
7351 INIT_LIST_HEAD(&cc
.migratepages
);
7354 * What we do here is we mark all pageblocks in range as
7355 * MIGRATE_ISOLATE. Because pageblock and max order pages may
7356 * have different sizes, and due to the way page allocator
7357 * work, we align the range to biggest of the two pages so
7358 * that page allocator won't try to merge buddies from
7359 * different pageblocks and change MIGRATE_ISOLATE to some
7360 * other migration type.
7362 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
7363 * migrate the pages from an unaligned range (ie. pages that
7364 * we are interested in). This will put all the pages in
7365 * range back to page allocator as MIGRATE_ISOLATE.
7367 * When this is done, we take the pages in range from page
7368 * allocator removing them from the buddy system. This way
7369 * page allocator will never consider using them.
7371 * This lets us mark the pageblocks back as
7372 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
7373 * aligned range but not in the unaligned, original range are
7374 * put back to page allocator so that buddy can use them.
7377 ret
= start_isolate_page_range(pfn_max_align_down(start
),
7378 pfn_max_align_up(end
), migratetype
,
7384 * In case of -EBUSY, we'd like to know which page causes problem.
7385 * So, just fall through. We will check it in test_pages_isolated().
7387 ret
= __alloc_contig_migrate_range(&cc
, start
, end
);
7388 if (ret
&& ret
!= -EBUSY
)
7392 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
7393 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
7394 * more, all pages in [start, end) are free in page allocator.
7395 * What we are going to do is to allocate all pages from
7396 * [start, end) (that is remove them from page allocator).
7398 * The only problem is that pages at the beginning and at the
7399 * end of interesting range may be not aligned with pages that
7400 * page allocator holds, ie. they can be part of higher order
7401 * pages. Because of this, we reserve the bigger range and
7402 * once this is done free the pages we are not interested in.
7404 * We don't have to hold zone->lock here because the pages are
7405 * isolated thus they won't get removed from buddy.
7408 lru_add_drain_all();
7409 drain_all_pages(cc
.zone
);
7412 outer_start
= start
;
7413 while (!PageBuddy(pfn_to_page(outer_start
))) {
7414 if (++order
>= MAX_ORDER
) {
7415 outer_start
= start
;
7418 outer_start
&= ~0UL << order
;
7421 if (outer_start
!= start
) {
7422 order
= page_order(pfn_to_page(outer_start
));
7425 * outer_start page could be small order buddy page and
7426 * it doesn't include start page. Adjust outer_start
7427 * in this case to report failed page properly
7428 * on tracepoint in test_pages_isolated()
7430 if (outer_start
+ (1UL << order
) <= start
)
7431 outer_start
= start
;
7434 /* Make sure the range is really isolated. */
7435 if (test_pages_isolated(outer_start
, end
, false)) {
7436 pr_info("%s: [%lx, %lx) PFNs busy\n",
7437 __func__
, outer_start
, end
);
7442 /* Grab isolated pages from freelists. */
7443 outer_end
= isolate_freepages_range(&cc
, outer_start
, end
);
7449 /* Free head and tail (if any) */
7450 if (start
!= outer_start
)
7451 free_contig_range(outer_start
, start
- outer_start
);
7452 if (end
!= outer_end
)
7453 free_contig_range(end
, outer_end
- end
);
7456 undo_isolate_page_range(pfn_max_align_down(start
),
7457 pfn_max_align_up(end
), migratetype
);
7461 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
7463 unsigned int count
= 0;
7465 for (; nr_pages
--; pfn
++) {
7466 struct page
*page
= pfn_to_page(pfn
);
7468 count
+= page_count(page
) != 1;
7471 WARN(count
!= 0, "%d pages are still in use!\n", count
);
7475 #ifdef CONFIG_MEMORY_HOTPLUG
7477 * The zone indicated has a new number of managed_pages; batch sizes and percpu
7478 * page high values need to be recalulated.
7480 void __meminit
zone_pcp_update(struct zone
*zone
)
7483 mutex_lock(&pcp_batch_high_lock
);
7484 for_each_possible_cpu(cpu
)
7485 pageset_set_high_and_batch(zone
,
7486 per_cpu_ptr(zone
->pageset
, cpu
));
7487 mutex_unlock(&pcp_batch_high_lock
);
7491 void zone_pcp_reset(struct zone
*zone
)
7493 unsigned long flags
;
7495 struct per_cpu_pageset
*pset
;
7497 /* avoid races with drain_pages() */
7498 local_irq_save(flags
);
7499 if (zone
->pageset
!= &boot_pageset
) {
7500 for_each_online_cpu(cpu
) {
7501 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
7502 drain_zonestat(zone
, pset
);
7504 free_percpu(zone
->pageset
);
7505 zone
->pageset
= &boot_pageset
;
7507 local_irq_restore(flags
);
7510 #ifdef CONFIG_MEMORY_HOTREMOVE
7512 * All pages in the range must be in a single zone and isolated
7513 * before calling this.
7516 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
7520 unsigned int order
, i
;
7522 unsigned long flags
;
7523 /* find the first valid pfn */
7524 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
7529 zone
= page_zone(pfn_to_page(pfn
));
7530 spin_lock_irqsave(&zone
->lock
, flags
);
7532 while (pfn
< end_pfn
) {
7533 if (!pfn_valid(pfn
)) {
7537 page
= pfn_to_page(pfn
);
7539 * The HWPoisoned page may be not in buddy system, and
7540 * page_count() is not 0.
7542 if (unlikely(!PageBuddy(page
) && PageHWPoison(page
))) {
7544 SetPageReserved(page
);
7548 BUG_ON(page_count(page
));
7549 BUG_ON(!PageBuddy(page
));
7550 order
= page_order(page
);
7551 #ifdef CONFIG_DEBUG_VM
7552 pr_info("remove from free list %lx %d %lx\n",
7553 pfn
, 1 << order
, end_pfn
);
7555 list_del(&page
->lru
);
7556 rmv_page_order(page
);
7557 zone
->free_area
[order
].nr_free
--;
7558 for (i
= 0; i
< (1 << order
); i
++)
7559 SetPageReserved((page
+i
));
7560 pfn
+= (1 << order
);
7562 spin_unlock_irqrestore(&zone
->lock
, flags
);
7566 bool is_free_buddy_page(struct page
*page
)
7568 struct zone
*zone
= page_zone(page
);
7569 unsigned long pfn
= page_to_pfn(page
);
7570 unsigned long flags
;
7573 spin_lock_irqsave(&zone
->lock
, flags
);
7574 for (order
= 0; order
< MAX_ORDER
; order
++) {
7575 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
7577 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
7580 spin_unlock_irqrestore(&zone
->lock
, flags
);
7582 return order
< MAX_ORDER
;