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>
66 #include <linux/memcontrol.h>
68 #include <asm/sections.h>
69 #include <asm/tlbflush.h>
70 #include <asm/div64.h>
73 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
74 static DEFINE_MUTEX(pcp_batch_high_lock
);
75 #define MIN_PERCPU_PAGELIST_FRACTION (8)
77 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
78 DEFINE_PER_CPU(int, numa_node
);
79 EXPORT_PER_CPU_SYMBOL(numa_node
);
82 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
84 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
85 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
86 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
87 * defined in <linux/topology.h>.
89 DEFINE_PER_CPU(int, _numa_mem_
); /* Kernel "local memory" node */
90 EXPORT_PER_CPU_SYMBOL(_numa_mem_
);
91 int _node_numa_mem_
[MAX_NUMNODES
];
95 * Array of node states.
97 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
98 [N_POSSIBLE
] = NODE_MASK_ALL
,
99 [N_ONLINE
] = { { [0] = 1UL } },
101 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
102 #ifdef CONFIG_HIGHMEM
103 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
105 #ifdef CONFIG_MOVABLE_NODE
106 [N_MEMORY
] = { { [0] = 1UL } },
108 [N_CPU
] = { { [0] = 1UL } },
111 EXPORT_SYMBOL(node_states
);
113 /* Protect totalram_pages and zone->managed_pages */
114 static DEFINE_SPINLOCK(managed_page_count_lock
);
116 unsigned long totalram_pages __read_mostly
;
117 unsigned long totalreserve_pages __read_mostly
;
118 unsigned long totalcma_pages __read_mostly
;
120 int percpu_pagelist_fraction
;
121 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
124 * A cached value of the page's pageblock's migratetype, used when the page is
125 * put on a pcplist. Used to avoid the pageblock migratetype lookup when
126 * freeing from pcplists in most cases, at the cost of possibly becoming stale.
127 * Also the migratetype set in the page does not necessarily match the pcplist
128 * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
129 * other index - this ensures that it will be put on the correct CMA freelist.
131 static inline int get_pcppage_migratetype(struct page
*page
)
136 static inline void set_pcppage_migratetype(struct page
*page
, int migratetype
)
138 page
->index
= migratetype
;
141 #ifdef CONFIG_PM_SLEEP
143 * The following functions are used by the suspend/hibernate code to temporarily
144 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
145 * while devices are suspended. To avoid races with the suspend/hibernate code,
146 * they should always be called with pm_mutex held (gfp_allowed_mask also should
147 * only be modified with pm_mutex held, unless the suspend/hibernate code is
148 * guaranteed not to run in parallel with that modification).
151 static gfp_t saved_gfp_mask
;
153 void pm_restore_gfp_mask(void)
155 WARN_ON(!mutex_is_locked(&pm_mutex
));
156 if (saved_gfp_mask
) {
157 gfp_allowed_mask
= saved_gfp_mask
;
162 void pm_restrict_gfp_mask(void)
164 WARN_ON(!mutex_is_locked(&pm_mutex
));
165 WARN_ON(saved_gfp_mask
);
166 saved_gfp_mask
= gfp_allowed_mask
;
167 gfp_allowed_mask
&= ~(__GFP_IO
| __GFP_FS
);
170 bool pm_suspended_storage(void)
172 if ((gfp_allowed_mask
& (__GFP_IO
| __GFP_FS
)) == (__GFP_IO
| __GFP_FS
))
176 #endif /* CONFIG_PM_SLEEP */
178 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
179 unsigned int pageblock_order __read_mostly
;
182 static void __free_pages_ok(struct page
*page
, unsigned int order
);
185 * results with 256, 32 in the lowmem_reserve sysctl:
186 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
187 * 1G machine -> (16M dma, 784M normal, 224M high)
188 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
189 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
190 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
192 * TBD: should special case ZONE_DMA32 machines here - in those we normally
193 * don't need any ZONE_NORMAL reservation
195 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
196 #ifdef CONFIG_ZONE_DMA
199 #ifdef CONFIG_ZONE_DMA32
202 #ifdef CONFIG_HIGHMEM
208 EXPORT_SYMBOL(totalram_pages
);
210 static char * const zone_names
[MAX_NR_ZONES
] = {
211 #ifdef CONFIG_ZONE_DMA
214 #ifdef CONFIG_ZONE_DMA32
218 #ifdef CONFIG_HIGHMEM
222 #ifdef CONFIG_ZONE_DEVICE
227 char * const migratetype_names
[MIGRATE_TYPES
] = {
235 #ifdef CONFIG_MEMORY_ISOLATION
240 compound_page_dtor
* const compound_page_dtors
[] = {
243 #ifdef CONFIG_HUGETLB_PAGE
246 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
251 int min_free_kbytes
= 1024;
252 int user_min_free_kbytes
= -1;
253 int watermark_scale_factor
= 10;
255 static unsigned long __meminitdata nr_kernel_pages
;
256 static unsigned long __meminitdata nr_all_pages
;
257 static unsigned long __meminitdata dma_reserve
;
259 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
260 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
261 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
262 static unsigned long __initdata required_kernelcore
;
263 static unsigned long __initdata required_movablecore
;
264 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
265 static bool mirrored_kernelcore
;
267 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
269 EXPORT_SYMBOL(movable_zone
);
270 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
273 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
274 int nr_online_nodes __read_mostly
= 1;
275 EXPORT_SYMBOL(nr_node_ids
);
276 EXPORT_SYMBOL(nr_online_nodes
);
279 int page_group_by_mobility_disabled __read_mostly
;
281 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
282 static inline void reset_deferred_meminit(pg_data_t
*pgdat
)
284 pgdat
->first_deferred_pfn
= ULONG_MAX
;
287 /* Returns true if the struct page for the pfn is uninitialised */
288 static inline bool __meminit
early_page_uninitialised(unsigned long pfn
)
290 int nid
= early_pfn_to_nid(pfn
);
292 if (node_online(nid
) && pfn
>= NODE_DATA(nid
)->first_deferred_pfn
)
298 static inline bool early_page_nid_uninitialised(unsigned long pfn
, int nid
)
300 if (pfn
>= NODE_DATA(nid
)->first_deferred_pfn
)
307 * Returns false when the remaining initialisation should be deferred until
308 * later in the boot cycle when it can be parallelised.
310 static inline bool update_defer_init(pg_data_t
*pgdat
,
311 unsigned long pfn
, unsigned long zone_end
,
312 unsigned long *nr_initialised
)
314 unsigned long max_initialise
;
316 /* Always populate low zones for address-contrained allocations */
317 if (zone_end
< pgdat_end_pfn(pgdat
))
320 * Initialise at least 2G of a node but also take into account that
321 * two large system hashes that can take up 1GB for 0.25TB/node.
323 max_initialise
= max(2UL << (30 - PAGE_SHIFT
),
324 (pgdat
->node_spanned_pages
>> 8));
327 if ((*nr_initialised
> max_initialise
) &&
328 (pfn
& (PAGES_PER_SECTION
- 1)) == 0) {
329 pgdat
->first_deferred_pfn
= pfn
;
336 static inline void reset_deferred_meminit(pg_data_t
*pgdat
)
340 static inline bool early_page_uninitialised(unsigned long pfn
)
345 static inline bool early_page_nid_uninitialised(unsigned long pfn
, int nid
)
350 static inline bool update_defer_init(pg_data_t
*pgdat
,
351 unsigned long pfn
, unsigned long zone_end
,
352 unsigned long *nr_initialised
)
358 /* Return a pointer to the bitmap storing bits affecting a block of pages */
359 static inline unsigned long *get_pageblock_bitmap(struct page
*page
,
362 #ifdef CONFIG_SPARSEMEM
363 return __pfn_to_section(pfn
)->pageblock_flags
;
365 return page_zone(page
)->pageblock_flags
;
366 #endif /* CONFIG_SPARSEMEM */
369 static inline int pfn_to_bitidx(struct page
*page
, unsigned long pfn
)
371 #ifdef CONFIG_SPARSEMEM
372 pfn
&= (PAGES_PER_SECTION
-1);
373 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
375 pfn
= pfn
- round_down(page_zone(page
)->zone_start_pfn
, pageblock_nr_pages
);
376 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
377 #endif /* CONFIG_SPARSEMEM */
381 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
382 * @page: The page within the block of interest
383 * @pfn: The target page frame number
384 * @end_bitidx: The last bit of interest to retrieve
385 * @mask: mask of bits that the caller is interested in
387 * Return: pageblock_bits flags
389 static __always_inline
unsigned long __get_pfnblock_flags_mask(struct page
*page
,
391 unsigned long end_bitidx
,
394 unsigned long *bitmap
;
395 unsigned long bitidx
, word_bitidx
;
398 bitmap
= get_pageblock_bitmap(page
, pfn
);
399 bitidx
= pfn_to_bitidx(page
, pfn
);
400 word_bitidx
= bitidx
/ BITS_PER_LONG
;
401 bitidx
&= (BITS_PER_LONG
-1);
403 word
= bitmap
[word_bitidx
];
404 bitidx
+= end_bitidx
;
405 return (word
>> (BITS_PER_LONG
- bitidx
- 1)) & mask
;
408 unsigned long get_pfnblock_flags_mask(struct page
*page
, unsigned long pfn
,
409 unsigned long end_bitidx
,
412 return __get_pfnblock_flags_mask(page
, pfn
, end_bitidx
, mask
);
415 static __always_inline
int get_pfnblock_migratetype(struct page
*page
, unsigned long pfn
)
417 return __get_pfnblock_flags_mask(page
, pfn
, PB_migrate_end
, MIGRATETYPE_MASK
);
421 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
422 * @page: The page within the block of interest
423 * @flags: The flags to set
424 * @pfn: The target page frame number
425 * @end_bitidx: The last bit of interest
426 * @mask: mask of bits that the caller is interested in
428 void set_pfnblock_flags_mask(struct page
*page
, unsigned long flags
,
430 unsigned long end_bitidx
,
433 unsigned long *bitmap
;
434 unsigned long bitidx
, word_bitidx
;
435 unsigned long old_word
, word
;
437 BUILD_BUG_ON(NR_PAGEBLOCK_BITS
!= 4);
439 bitmap
= get_pageblock_bitmap(page
, pfn
);
440 bitidx
= pfn_to_bitidx(page
, pfn
);
441 word_bitidx
= bitidx
/ BITS_PER_LONG
;
442 bitidx
&= (BITS_PER_LONG
-1);
444 VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page
), pfn
), page
);
446 bitidx
+= end_bitidx
;
447 mask
<<= (BITS_PER_LONG
- bitidx
- 1);
448 flags
<<= (BITS_PER_LONG
- bitidx
- 1);
450 word
= READ_ONCE(bitmap
[word_bitidx
]);
452 old_word
= cmpxchg(&bitmap
[word_bitidx
], word
, (word
& ~mask
) | flags
);
453 if (word
== old_word
)
459 void set_pageblock_migratetype(struct page
*page
, int migratetype
)
461 if (unlikely(page_group_by_mobility_disabled
&&
462 migratetype
< MIGRATE_PCPTYPES
))
463 migratetype
= MIGRATE_UNMOVABLE
;
465 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
466 PB_migrate
, PB_migrate_end
);
469 #ifdef CONFIG_DEBUG_VM
470 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
474 unsigned long pfn
= page_to_pfn(page
);
475 unsigned long sp
, start_pfn
;
478 seq
= zone_span_seqbegin(zone
);
479 start_pfn
= zone
->zone_start_pfn
;
480 sp
= zone
->spanned_pages
;
481 if (!zone_spans_pfn(zone
, pfn
))
483 } while (zone_span_seqretry(zone
, seq
));
486 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
487 pfn
, zone_to_nid(zone
), zone
->name
,
488 start_pfn
, start_pfn
+ sp
);
493 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
495 if (!pfn_valid_within(page_to_pfn(page
)))
497 if (zone
!= page_zone(page
))
503 * Temporary debugging check for pages not lying within a given zone.
505 static int bad_range(struct zone
*zone
, struct page
*page
)
507 if (page_outside_zone_boundaries(zone
, page
))
509 if (!page_is_consistent(zone
, page
))
515 static inline int bad_range(struct zone
*zone
, struct page
*page
)
521 static void bad_page(struct page
*page
, const char *reason
,
522 unsigned long bad_flags
)
524 static unsigned long resume
;
525 static unsigned long nr_shown
;
526 static unsigned long nr_unshown
;
529 * Allow a burst of 60 reports, then keep quiet for that minute;
530 * or allow a steady drip of one report per second.
532 if (nr_shown
== 60) {
533 if (time_before(jiffies
, resume
)) {
539 "BUG: Bad page state: %lu messages suppressed\n",
546 resume
= jiffies
+ 60 * HZ
;
548 pr_alert("BUG: Bad page state in process %s pfn:%05lx\n",
549 current
->comm
, page_to_pfn(page
));
550 __dump_page(page
, reason
);
551 bad_flags
&= page
->flags
;
553 pr_alert("bad because of flags: %#lx(%pGp)\n",
554 bad_flags
, &bad_flags
);
555 dump_page_owner(page
);
560 /* Leave bad fields for debug, except PageBuddy could make trouble */
561 page_mapcount_reset(page
); /* remove PageBuddy */
562 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
566 * Higher-order pages are called "compound pages". They are structured thusly:
568 * The first PAGE_SIZE page is called the "head page" and have PG_head set.
570 * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
571 * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
573 * The first tail page's ->compound_dtor holds the offset in array of compound
574 * page destructors. See compound_page_dtors.
576 * The first tail page's ->compound_order holds the order of allocation.
577 * This usage means that zero-order pages may not be compound.
580 void free_compound_page(struct page
*page
)
582 __free_pages_ok(page
, compound_order(page
));
585 void prep_compound_page(struct page
*page
, unsigned int order
)
588 int nr_pages
= 1 << order
;
590 set_compound_page_dtor(page
, COMPOUND_PAGE_DTOR
);
591 set_compound_order(page
, order
);
593 for (i
= 1; i
< nr_pages
; i
++) {
594 struct page
*p
= page
+ i
;
595 set_page_count(p
, 0);
596 p
->mapping
= TAIL_MAPPING
;
597 set_compound_head(p
, page
);
599 atomic_set(compound_mapcount_ptr(page
), -1);
602 #ifdef CONFIG_DEBUG_PAGEALLOC
603 unsigned int _debug_guardpage_minorder
;
604 bool _debug_pagealloc_enabled __read_mostly
605 = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT
);
606 EXPORT_SYMBOL(_debug_pagealloc_enabled
);
607 bool _debug_guardpage_enabled __read_mostly
;
609 static int __init
early_debug_pagealloc(char *buf
)
613 return kstrtobool(buf
, &_debug_pagealloc_enabled
);
615 early_param("debug_pagealloc", early_debug_pagealloc
);
617 static bool need_debug_guardpage(void)
619 /* If we don't use debug_pagealloc, we don't need guard page */
620 if (!debug_pagealloc_enabled())
626 static void init_debug_guardpage(void)
628 if (!debug_pagealloc_enabled())
631 _debug_guardpage_enabled
= true;
634 struct page_ext_operations debug_guardpage_ops
= {
635 .need
= need_debug_guardpage
,
636 .init
= init_debug_guardpage
,
639 static int __init
debug_guardpage_minorder_setup(char *buf
)
643 if (kstrtoul(buf
, 10, &res
) < 0 || res
> MAX_ORDER
/ 2) {
644 pr_err("Bad debug_guardpage_minorder value\n");
647 _debug_guardpage_minorder
= res
;
648 pr_info("Setting debug_guardpage_minorder to %lu\n", res
);
651 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup
);
653 static inline void set_page_guard(struct zone
*zone
, struct page
*page
,
654 unsigned int order
, int migratetype
)
656 struct page_ext
*page_ext
;
658 if (!debug_guardpage_enabled())
661 page_ext
= lookup_page_ext(page
);
662 if (unlikely(!page_ext
))
665 __set_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
667 INIT_LIST_HEAD(&page
->lru
);
668 set_page_private(page
, order
);
669 /* Guard pages are not available for any usage */
670 __mod_zone_freepage_state(zone
, -(1 << order
), migratetype
);
673 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
674 unsigned int order
, int migratetype
)
676 struct page_ext
*page_ext
;
678 if (!debug_guardpage_enabled())
681 page_ext
= lookup_page_ext(page
);
682 if (unlikely(!page_ext
))
685 __clear_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
687 set_page_private(page
, 0);
688 if (!is_migrate_isolate(migratetype
))
689 __mod_zone_freepage_state(zone
, (1 << order
), migratetype
);
692 struct page_ext_operations debug_guardpage_ops
= { NULL
, };
693 static inline void set_page_guard(struct zone
*zone
, struct page
*page
,
694 unsigned int order
, int migratetype
) {}
695 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
696 unsigned int order
, int migratetype
) {}
699 static inline void set_page_order(struct page
*page
, unsigned int order
)
701 set_page_private(page
, order
);
702 __SetPageBuddy(page
);
705 static inline void rmv_page_order(struct page
*page
)
707 __ClearPageBuddy(page
);
708 set_page_private(page
, 0);
712 * This function checks whether a page is free && is the buddy
713 * we can do coalesce a page and its buddy if
714 * (a) the buddy is not in a hole &&
715 * (b) the buddy is in the buddy system &&
716 * (c) a page and its buddy have the same order &&
717 * (d) a page and its buddy are in the same zone.
719 * For recording whether a page is in the buddy system, we set ->_mapcount
720 * PAGE_BUDDY_MAPCOUNT_VALUE.
721 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
722 * serialized by zone->lock.
724 * For recording page's order, we use page_private(page).
726 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
729 if (!pfn_valid_within(page_to_pfn(buddy
)))
732 if (page_is_guard(buddy
) && page_order(buddy
) == order
) {
733 if (page_zone_id(page
) != page_zone_id(buddy
))
736 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
741 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
743 * zone check is done late to avoid uselessly
744 * calculating zone/node ids for pages that could
747 if (page_zone_id(page
) != page_zone_id(buddy
))
750 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
758 * Freeing function for a buddy system allocator.
760 * The concept of a buddy system is to maintain direct-mapped table
761 * (containing bit values) for memory blocks of various "orders".
762 * The bottom level table contains the map for the smallest allocatable
763 * units of memory (here, pages), and each level above it describes
764 * pairs of units from the levels below, hence, "buddies".
765 * At a high level, all that happens here is marking the table entry
766 * at the bottom level available, and propagating the changes upward
767 * as necessary, plus some accounting needed to play nicely with other
768 * parts of the VM system.
769 * At each level, we keep a list of pages, which are heads of continuous
770 * free pages of length of (1 << order) and marked with _mapcount
771 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
773 * So when we are allocating or freeing one, we can derive the state of the
774 * other. That is, if we allocate a small block, and both were
775 * free, the remainder of the region must be split into blocks.
776 * If a block is freed, and its buddy is also free, then this
777 * triggers coalescing into a block of larger size.
782 static inline void __free_one_page(struct page
*page
,
784 struct zone
*zone
, unsigned int order
,
787 unsigned long page_idx
;
788 unsigned long combined_idx
;
789 unsigned long uninitialized_var(buddy_idx
);
791 unsigned int max_order
;
793 max_order
= min_t(unsigned int, MAX_ORDER
, pageblock_order
+ 1);
795 VM_BUG_ON(!zone_is_initialized(zone
));
796 VM_BUG_ON_PAGE(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
, page
);
798 VM_BUG_ON(migratetype
== -1);
799 if (likely(!is_migrate_isolate(migratetype
)))
800 __mod_zone_freepage_state(zone
, 1 << order
, migratetype
);
802 page_idx
= pfn
& ((1 << MAX_ORDER
) - 1);
804 VM_BUG_ON_PAGE(page_idx
& ((1 << order
) - 1), page
);
805 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
808 while (order
< max_order
- 1) {
809 buddy_idx
= __find_buddy_index(page_idx
, order
);
810 buddy
= page
+ (buddy_idx
- page_idx
);
811 if (!page_is_buddy(page
, buddy
, order
))
814 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
815 * merge with it and move up one order.
817 if (page_is_guard(buddy
)) {
818 clear_page_guard(zone
, buddy
, order
, migratetype
);
820 list_del(&buddy
->lru
);
821 zone
->free_area
[order
].nr_free
--;
822 rmv_page_order(buddy
);
824 combined_idx
= buddy_idx
& page_idx
;
825 page
= page
+ (combined_idx
- page_idx
);
826 page_idx
= combined_idx
;
829 if (max_order
< MAX_ORDER
) {
830 /* If we are here, it means order is >= pageblock_order.
831 * We want to prevent merge between freepages on isolate
832 * pageblock and normal pageblock. Without this, pageblock
833 * isolation could cause incorrect freepage or CMA accounting.
835 * We don't want to hit this code for the more frequent
838 if (unlikely(has_isolate_pageblock(zone
))) {
841 buddy_idx
= __find_buddy_index(page_idx
, order
);
842 buddy
= page
+ (buddy_idx
- page_idx
);
843 buddy_mt
= get_pageblock_migratetype(buddy
);
845 if (migratetype
!= buddy_mt
846 && (is_migrate_isolate(migratetype
) ||
847 is_migrate_isolate(buddy_mt
)))
851 goto continue_merging
;
855 set_page_order(page
, order
);
858 * If this is not the largest possible page, check if the buddy
859 * of the next-highest order is free. If it is, it's possible
860 * that pages are being freed that will coalesce soon. In case,
861 * that is happening, add the free page to the tail of the list
862 * so it's less likely to be used soon and more likely to be merged
863 * as a higher order page
865 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
866 struct page
*higher_page
, *higher_buddy
;
867 combined_idx
= buddy_idx
& page_idx
;
868 higher_page
= page
+ (combined_idx
- page_idx
);
869 buddy_idx
= __find_buddy_index(combined_idx
, order
+ 1);
870 higher_buddy
= higher_page
+ (buddy_idx
- combined_idx
);
871 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
872 list_add_tail(&page
->lru
,
873 &zone
->free_area
[order
].free_list
[migratetype
]);
878 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
880 zone
->free_area
[order
].nr_free
++;
884 * A bad page could be due to a number of fields. Instead of multiple branches,
885 * try and check multiple fields with one check. The caller must do a detailed
886 * check if necessary.
888 static inline bool page_expected_state(struct page
*page
,
889 unsigned long check_flags
)
891 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
894 if (unlikely((unsigned long)page
->mapping
|
895 page_ref_count(page
) |
897 (unsigned long)page
->mem_cgroup
|
899 (page
->flags
& check_flags
)))
905 static void free_pages_check_bad(struct page
*page
)
907 const char *bad_reason
;
908 unsigned long bad_flags
;
913 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
914 bad_reason
= "nonzero mapcount";
915 if (unlikely(page
->mapping
!= NULL
))
916 bad_reason
= "non-NULL mapping";
917 if (unlikely(page_ref_count(page
) != 0))
918 bad_reason
= "nonzero _refcount";
919 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
)) {
920 bad_reason
= "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
921 bad_flags
= PAGE_FLAGS_CHECK_AT_FREE
;
924 if (unlikely(page
->mem_cgroup
))
925 bad_reason
= "page still charged to cgroup";
927 bad_page(page
, bad_reason
, bad_flags
);
930 static inline int free_pages_check(struct page
*page
)
932 if (likely(page_expected_state(page
, PAGE_FLAGS_CHECK_AT_FREE
)))
935 /* Something has gone sideways, find it */
936 free_pages_check_bad(page
);
940 static int free_tail_pages_check(struct page
*head_page
, struct page
*page
)
945 * We rely page->lru.next never has bit 0 set, unless the page
946 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
948 BUILD_BUG_ON((unsigned long)LIST_POISON1
& 1);
950 if (!IS_ENABLED(CONFIG_DEBUG_VM
)) {
954 switch (page
- head_page
) {
956 /* the first tail page: ->mapping is compound_mapcount() */
957 if (unlikely(compound_mapcount(page
))) {
958 bad_page(page
, "nonzero compound_mapcount", 0);
964 * the second tail page: ->mapping is
965 * page_deferred_list().next -- ignore value.
969 if (page
->mapping
!= TAIL_MAPPING
) {
970 bad_page(page
, "corrupted mapping in tail page", 0);
975 if (unlikely(!PageTail(page
))) {
976 bad_page(page
, "PageTail not set", 0);
979 if (unlikely(compound_head(page
) != head_page
)) {
980 bad_page(page
, "compound_head not consistent", 0);
985 page
->mapping
= NULL
;
986 clear_compound_head(page
);
990 static __always_inline
bool free_pages_prepare(struct page
*page
,
991 unsigned int order
, bool check_free
)
995 VM_BUG_ON_PAGE(PageTail(page
), page
);
997 trace_mm_page_free(page
, order
);
998 kmemcheck_free_shadow(page
, order
);
1001 * Check tail pages before head page information is cleared to
1002 * avoid checking PageCompound for order-0 pages.
1004 if (unlikely(order
)) {
1005 bool compound
= PageCompound(page
);
1008 VM_BUG_ON_PAGE(compound
&& compound_order(page
) != order
, page
);
1010 for (i
= 1; i
< (1 << order
); i
++) {
1012 bad
+= free_tail_pages_check(page
, page
+ i
);
1013 if (unlikely(free_pages_check(page
+ i
))) {
1017 (page
+ i
)->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
1020 if (PageMappingFlags(page
))
1021 page
->mapping
= NULL
;
1022 if (memcg_kmem_enabled() && PageKmemcg(page
)) {
1023 memcg_kmem_uncharge(page
, order
);
1024 __ClearPageKmemcg(page
);
1027 bad
+= free_pages_check(page
);
1031 page_cpupid_reset_last(page
);
1032 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
1033 reset_page_owner(page
, order
);
1035 if (!PageHighMem(page
)) {
1036 debug_check_no_locks_freed(page_address(page
),
1037 PAGE_SIZE
<< order
);
1038 debug_check_no_obj_freed(page_address(page
),
1039 PAGE_SIZE
<< order
);
1041 arch_free_page(page
, order
);
1042 kernel_poison_pages(page
, 1 << order
, 0);
1043 kernel_map_pages(page
, 1 << order
, 0);
1044 kasan_free_pages(page
, order
);
1049 #ifdef CONFIG_DEBUG_VM
1050 static inline bool free_pcp_prepare(struct page
*page
)
1052 return free_pages_prepare(page
, 0, true);
1055 static inline bool bulkfree_pcp_prepare(struct page
*page
)
1060 static bool free_pcp_prepare(struct page
*page
)
1062 return free_pages_prepare(page
, 0, false);
1065 static bool bulkfree_pcp_prepare(struct page
*page
)
1067 return free_pages_check(page
);
1069 #endif /* CONFIG_DEBUG_VM */
1072 * Frees a number of pages from the PCP lists
1073 * Assumes all pages on list are in same zone, and of same order.
1074 * count is the number of pages to free.
1076 * If the zone was previously in an "all pages pinned" state then look to
1077 * see if this freeing clears that state.
1079 * And clear the zone's pages_scanned counter, to hold off the "all pages are
1080 * pinned" detection logic.
1082 static void free_pcppages_bulk(struct zone
*zone
, int count
,
1083 struct per_cpu_pages
*pcp
)
1085 int migratetype
= 0;
1087 unsigned long nr_scanned
;
1088 bool isolated_pageblocks
;
1090 spin_lock(&zone
->lock
);
1091 isolated_pageblocks
= has_isolate_pageblock(zone
);
1092 nr_scanned
= zone_page_state(zone
, NR_PAGES_SCANNED
);
1094 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, -nr_scanned
);
1098 struct list_head
*list
;
1101 * Remove pages from lists in a round-robin fashion. A
1102 * batch_free count is maintained that is incremented when an
1103 * empty list is encountered. This is so more pages are freed
1104 * off fuller lists instead of spinning excessively around empty
1109 if (++migratetype
== MIGRATE_PCPTYPES
)
1111 list
= &pcp
->lists
[migratetype
];
1112 } while (list_empty(list
));
1114 /* This is the only non-empty list. Free them all. */
1115 if (batch_free
== MIGRATE_PCPTYPES
)
1119 int mt
; /* migratetype of the to-be-freed page */
1121 page
= list_last_entry(list
, struct page
, lru
);
1122 /* must delete as __free_one_page list manipulates */
1123 list_del(&page
->lru
);
1125 mt
= get_pcppage_migratetype(page
);
1126 /* MIGRATE_ISOLATE page should not go to pcplists */
1127 VM_BUG_ON_PAGE(is_migrate_isolate(mt
), page
);
1128 /* Pageblock could have been isolated meanwhile */
1129 if (unlikely(isolated_pageblocks
))
1130 mt
= get_pageblock_migratetype(page
);
1132 if (bulkfree_pcp_prepare(page
))
1135 __free_one_page(page
, page_to_pfn(page
), zone
, 0, mt
);
1136 trace_mm_page_pcpu_drain(page
, 0, mt
);
1137 } while (--count
&& --batch_free
&& !list_empty(list
));
1139 spin_unlock(&zone
->lock
);
1142 static void free_one_page(struct zone
*zone
,
1143 struct page
*page
, unsigned long pfn
,
1147 unsigned long nr_scanned
;
1148 spin_lock(&zone
->lock
);
1149 nr_scanned
= zone_page_state(zone
, NR_PAGES_SCANNED
);
1151 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, -nr_scanned
);
1153 if (unlikely(has_isolate_pageblock(zone
) ||
1154 is_migrate_isolate(migratetype
))) {
1155 migratetype
= get_pfnblock_migratetype(page
, pfn
);
1157 __free_one_page(page
, pfn
, zone
, order
, migratetype
);
1158 spin_unlock(&zone
->lock
);
1161 static void __meminit
__init_single_page(struct page
*page
, unsigned long pfn
,
1162 unsigned long zone
, int nid
)
1164 set_page_links(page
, zone
, nid
, pfn
);
1165 init_page_count(page
);
1166 page_mapcount_reset(page
);
1167 page_cpupid_reset_last(page
);
1169 INIT_LIST_HEAD(&page
->lru
);
1170 #ifdef WANT_PAGE_VIRTUAL
1171 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1172 if (!is_highmem_idx(zone
))
1173 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
1177 static void __meminit
__init_single_pfn(unsigned long pfn
, unsigned long zone
,
1180 return __init_single_page(pfn_to_page(pfn
), pfn
, zone
, nid
);
1183 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1184 static void init_reserved_page(unsigned long pfn
)
1189 if (!early_page_uninitialised(pfn
))
1192 nid
= early_pfn_to_nid(pfn
);
1193 pgdat
= NODE_DATA(nid
);
1195 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
1196 struct zone
*zone
= &pgdat
->node_zones
[zid
];
1198 if (pfn
>= zone
->zone_start_pfn
&& pfn
< zone_end_pfn(zone
))
1201 __init_single_pfn(pfn
, zid
, nid
);
1204 static inline void init_reserved_page(unsigned long pfn
)
1207 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1210 * Initialised pages do not have PageReserved set. This function is
1211 * called for each range allocated by the bootmem allocator and
1212 * marks the pages PageReserved. The remaining valid pages are later
1213 * sent to the buddy page allocator.
1215 void __meminit
reserve_bootmem_region(phys_addr_t start
, phys_addr_t end
)
1217 unsigned long start_pfn
= PFN_DOWN(start
);
1218 unsigned long end_pfn
= PFN_UP(end
);
1220 for (; start_pfn
< end_pfn
; start_pfn
++) {
1221 if (pfn_valid(start_pfn
)) {
1222 struct page
*page
= pfn_to_page(start_pfn
);
1224 init_reserved_page(start_pfn
);
1226 /* Avoid false-positive PageTail() */
1227 INIT_LIST_HEAD(&page
->lru
);
1229 SetPageReserved(page
);
1234 static void __free_pages_ok(struct page
*page
, unsigned int order
)
1236 unsigned long flags
;
1238 unsigned long pfn
= page_to_pfn(page
);
1240 if (!free_pages_prepare(page
, order
, true))
1243 migratetype
= get_pfnblock_migratetype(page
, pfn
);
1244 local_irq_save(flags
);
1245 __count_vm_events(PGFREE
, 1 << order
);
1246 free_one_page(page_zone(page
), page
, pfn
, order
, migratetype
);
1247 local_irq_restore(flags
);
1250 static void __init
__free_pages_boot_core(struct page
*page
, unsigned int order
)
1252 unsigned int nr_pages
= 1 << order
;
1253 struct page
*p
= page
;
1257 for (loop
= 0; loop
< (nr_pages
- 1); loop
++, p
++) {
1259 __ClearPageReserved(p
);
1260 set_page_count(p
, 0);
1262 __ClearPageReserved(p
);
1263 set_page_count(p
, 0);
1265 page_zone(page
)->managed_pages
+= nr_pages
;
1266 set_page_refcounted(page
);
1267 __free_pages(page
, order
);
1270 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1271 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1273 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata
;
1275 int __meminit
early_pfn_to_nid(unsigned long pfn
)
1277 static DEFINE_SPINLOCK(early_pfn_lock
);
1280 spin_lock(&early_pfn_lock
);
1281 nid
= __early_pfn_to_nid(pfn
, &early_pfnnid_cache
);
1283 nid
= first_online_node
;
1284 spin_unlock(&early_pfn_lock
);
1290 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1291 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1292 struct mminit_pfnnid_cache
*state
)
1296 nid
= __early_pfn_to_nid(pfn
, state
);
1297 if (nid
>= 0 && nid
!= node
)
1302 /* Only safe to use early in boot when initialisation is single-threaded */
1303 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1305 return meminit_pfn_in_nid(pfn
, node
, &early_pfnnid_cache
);
1310 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1314 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1315 struct mminit_pfnnid_cache
*state
)
1322 void __init
__free_pages_bootmem(struct page
*page
, unsigned long pfn
,
1325 if (early_page_uninitialised(pfn
))
1327 return __free_pages_boot_core(page
, order
);
1331 * Check that the whole (or subset of) a pageblock given by the interval of
1332 * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1333 * with the migration of free compaction scanner. The scanners then need to
1334 * use only pfn_valid_within() check for arches that allow holes within
1337 * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1339 * It's possible on some configurations to have a setup like node0 node1 node0
1340 * i.e. it's possible that all pages within a zones range of pages do not
1341 * belong to a single zone. We assume that a border between node0 and node1
1342 * can occur within a single pageblock, but not a node0 node1 node0
1343 * interleaving within a single pageblock. It is therefore sufficient to check
1344 * the first and last page of a pageblock and avoid checking each individual
1345 * page in a pageblock.
1347 struct page
*__pageblock_pfn_to_page(unsigned long start_pfn
,
1348 unsigned long end_pfn
, struct zone
*zone
)
1350 struct page
*start_page
;
1351 struct page
*end_page
;
1353 /* end_pfn is one past the range we are checking */
1356 if (!pfn_valid(start_pfn
) || !pfn_valid(end_pfn
))
1359 start_page
= pfn_to_page(start_pfn
);
1361 if (page_zone(start_page
) != zone
)
1364 end_page
= pfn_to_page(end_pfn
);
1366 /* This gives a shorter code than deriving page_zone(end_page) */
1367 if (page_zone_id(start_page
) != page_zone_id(end_page
))
1373 void set_zone_contiguous(struct zone
*zone
)
1375 unsigned long block_start_pfn
= zone
->zone_start_pfn
;
1376 unsigned long block_end_pfn
;
1378 block_end_pfn
= ALIGN(block_start_pfn
+ 1, pageblock_nr_pages
);
1379 for (; block_start_pfn
< zone_end_pfn(zone
);
1380 block_start_pfn
= block_end_pfn
,
1381 block_end_pfn
+= pageblock_nr_pages
) {
1383 block_end_pfn
= min(block_end_pfn
, zone_end_pfn(zone
));
1385 if (!__pageblock_pfn_to_page(block_start_pfn
,
1386 block_end_pfn
, zone
))
1390 /* We confirm that there is no hole */
1391 zone
->contiguous
= true;
1394 void clear_zone_contiguous(struct zone
*zone
)
1396 zone
->contiguous
= false;
1399 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1400 static void __init
deferred_free_range(struct page
*page
,
1401 unsigned long pfn
, int nr_pages
)
1408 /* Free a large naturally-aligned chunk if possible */
1409 if (nr_pages
== MAX_ORDER_NR_PAGES
&&
1410 (pfn
& (MAX_ORDER_NR_PAGES
-1)) == 0) {
1411 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
1412 __free_pages_boot_core(page
, MAX_ORDER
-1);
1416 for (i
= 0; i
< nr_pages
; i
++, page
++)
1417 __free_pages_boot_core(page
, 0);
1420 /* Completion tracking for deferred_init_memmap() threads */
1421 static atomic_t pgdat_init_n_undone __initdata
;
1422 static __initdata
DECLARE_COMPLETION(pgdat_init_all_done_comp
);
1424 static inline void __init
pgdat_init_report_one_done(void)
1426 if (atomic_dec_and_test(&pgdat_init_n_undone
))
1427 complete(&pgdat_init_all_done_comp
);
1430 /* Initialise remaining memory on a node */
1431 static int __init
deferred_init_memmap(void *data
)
1433 pg_data_t
*pgdat
= data
;
1434 int nid
= pgdat
->node_id
;
1435 struct mminit_pfnnid_cache nid_init_state
= { };
1436 unsigned long start
= jiffies
;
1437 unsigned long nr_pages
= 0;
1438 unsigned long walk_start
, walk_end
;
1441 unsigned long first_init_pfn
= pgdat
->first_deferred_pfn
;
1442 const struct cpumask
*cpumask
= cpumask_of_node(pgdat
->node_id
);
1444 if (first_init_pfn
== ULONG_MAX
) {
1445 pgdat_init_report_one_done();
1449 /* Bind memory initialisation thread to a local node if possible */
1450 if (!cpumask_empty(cpumask
))
1451 set_cpus_allowed_ptr(current
, cpumask
);
1453 /* Sanity check boundaries */
1454 BUG_ON(pgdat
->first_deferred_pfn
< pgdat
->node_start_pfn
);
1455 BUG_ON(pgdat
->first_deferred_pfn
> pgdat_end_pfn(pgdat
));
1456 pgdat
->first_deferred_pfn
= ULONG_MAX
;
1458 /* Only the highest zone is deferred so find it */
1459 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
1460 zone
= pgdat
->node_zones
+ zid
;
1461 if (first_init_pfn
< zone_end_pfn(zone
))
1465 for_each_mem_pfn_range(i
, nid
, &walk_start
, &walk_end
, NULL
) {
1466 unsigned long pfn
, end_pfn
;
1467 struct page
*page
= NULL
;
1468 struct page
*free_base_page
= NULL
;
1469 unsigned long free_base_pfn
= 0;
1472 end_pfn
= min(walk_end
, zone_end_pfn(zone
));
1473 pfn
= first_init_pfn
;
1474 if (pfn
< walk_start
)
1476 if (pfn
< zone
->zone_start_pfn
)
1477 pfn
= zone
->zone_start_pfn
;
1479 for (; pfn
< end_pfn
; pfn
++) {
1480 if (!pfn_valid_within(pfn
))
1484 * Ensure pfn_valid is checked every
1485 * MAX_ORDER_NR_PAGES for memory holes
1487 if ((pfn
& (MAX_ORDER_NR_PAGES
- 1)) == 0) {
1488 if (!pfn_valid(pfn
)) {
1494 if (!meminit_pfn_in_nid(pfn
, nid
, &nid_init_state
)) {
1499 /* Minimise pfn page lookups and scheduler checks */
1500 if (page
&& (pfn
& (MAX_ORDER_NR_PAGES
- 1)) != 0) {
1503 nr_pages
+= nr_to_free
;
1504 deferred_free_range(free_base_page
,
1505 free_base_pfn
, nr_to_free
);
1506 free_base_page
= NULL
;
1507 free_base_pfn
= nr_to_free
= 0;
1509 page
= pfn_to_page(pfn
);
1514 VM_BUG_ON(page_zone(page
) != zone
);
1518 __init_single_page(page
, pfn
, zid
, nid
);
1519 if (!free_base_page
) {
1520 free_base_page
= page
;
1521 free_base_pfn
= pfn
;
1526 /* Where possible, batch up pages for a single free */
1529 /* Free the current block of pages to allocator */
1530 nr_pages
+= nr_to_free
;
1531 deferred_free_range(free_base_page
, free_base_pfn
,
1533 free_base_page
= NULL
;
1534 free_base_pfn
= nr_to_free
= 0;
1537 first_init_pfn
= max(end_pfn
, first_init_pfn
);
1540 /* Sanity check that the next zone really is unpopulated */
1541 WARN_ON(++zid
< MAX_NR_ZONES
&& populated_zone(++zone
));
1543 pr_info("node %d initialised, %lu pages in %ums\n", nid
, nr_pages
,
1544 jiffies_to_msecs(jiffies
- start
));
1546 pgdat_init_report_one_done();
1549 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1551 void __init
page_alloc_init_late(void)
1555 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1558 /* There will be num_node_state(N_MEMORY) threads */
1559 atomic_set(&pgdat_init_n_undone
, num_node_state(N_MEMORY
));
1560 for_each_node_state(nid
, N_MEMORY
) {
1561 kthread_run(deferred_init_memmap
, NODE_DATA(nid
), "pgdatinit%d", nid
);
1564 /* Block until all are initialised */
1565 wait_for_completion(&pgdat_init_all_done_comp
);
1567 /* Reinit limits that are based on free pages after the kernel is up */
1568 files_maxfiles_init();
1571 for_each_populated_zone(zone
)
1572 set_zone_contiguous(zone
);
1576 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1577 void __init
init_cma_reserved_pageblock(struct page
*page
)
1579 unsigned i
= pageblock_nr_pages
;
1580 struct page
*p
= page
;
1583 __ClearPageReserved(p
);
1584 set_page_count(p
, 0);
1587 set_pageblock_migratetype(page
, MIGRATE_CMA
);
1589 if (pageblock_order
>= MAX_ORDER
) {
1590 i
= pageblock_nr_pages
;
1593 set_page_refcounted(p
);
1594 __free_pages(p
, MAX_ORDER
- 1);
1595 p
+= MAX_ORDER_NR_PAGES
;
1596 } while (i
-= MAX_ORDER_NR_PAGES
);
1598 set_page_refcounted(page
);
1599 __free_pages(page
, pageblock_order
);
1602 adjust_managed_page_count(page
, pageblock_nr_pages
);
1607 * The order of subdivision here is critical for the IO subsystem.
1608 * Please do not alter this order without good reasons and regression
1609 * testing. Specifically, as large blocks of memory are subdivided,
1610 * the order in which smaller blocks are delivered depends on the order
1611 * they're subdivided in this function. This is the primary factor
1612 * influencing the order in which pages are delivered to the IO
1613 * subsystem according to empirical testing, and this is also justified
1614 * by considering the behavior of a buddy system containing a single
1615 * large block of memory acted on by a series of small allocations.
1616 * This behavior is a critical factor in sglist merging's success.
1620 static inline void expand(struct zone
*zone
, struct page
*page
,
1621 int low
, int high
, struct free_area
*area
,
1624 unsigned long size
= 1 << high
;
1626 while (high
> low
) {
1630 VM_BUG_ON_PAGE(bad_range(zone
, &page
[size
]), &page
[size
]);
1632 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC
) &&
1633 debug_guardpage_enabled() &&
1634 high
< debug_guardpage_minorder()) {
1636 * Mark as guard pages (or page), that will allow to
1637 * merge back to allocator when buddy will be freed.
1638 * Corresponding page table entries will not be touched,
1639 * pages will stay not present in virtual address space
1641 set_page_guard(zone
, &page
[size
], high
, migratetype
);
1644 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
1646 set_page_order(&page
[size
], high
);
1650 static void check_new_page_bad(struct page
*page
)
1652 const char *bad_reason
= NULL
;
1653 unsigned long bad_flags
= 0;
1655 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
1656 bad_reason
= "nonzero mapcount";
1657 if (unlikely(page
->mapping
!= NULL
))
1658 bad_reason
= "non-NULL mapping";
1659 if (unlikely(page_ref_count(page
) != 0))
1660 bad_reason
= "nonzero _count";
1661 if (unlikely(page
->flags
& __PG_HWPOISON
)) {
1662 bad_reason
= "HWPoisoned (hardware-corrupted)";
1663 bad_flags
= __PG_HWPOISON
;
1664 /* Don't complain about hwpoisoned pages */
1665 page_mapcount_reset(page
); /* remove PageBuddy */
1668 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)) {
1669 bad_reason
= "PAGE_FLAGS_CHECK_AT_PREP flag set";
1670 bad_flags
= PAGE_FLAGS_CHECK_AT_PREP
;
1673 if (unlikely(page
->mem_cgroup
))
1674 bad_reason
= "page still charged to cgroup";
1676 bad_page(page
, bad_reason
, bad_flags
);
1680 * This page is about to be returned from the page allocator
1682 static inline int check_new_page(struct page
*page
)
1684 if (likely(page_expected_state(page
,
1685 PAGE_FLAGS_CHECK_AT_PREP
|__PG_HWPOISON
)))
1688 check_new_page_bad(page
);
1692 static inline bool free_pages_prezeroed(bool poisoned
)
1694 return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO
) &&
1695 page_poisoning_enabled() && poisoned
;
1698 #ifdef CONFIG_DEBUG_VM
1699 static bool check_pcp_refill(struct page
*page
)
1704 static bool check_new_pcp(struct page
*page
)
1706 return check_new_page(page
);
1709 static bool check_pcp_refill(struct page
*page
)
1711 return check_new_page(page
);
1713 static bool check_new_pcp(struct page
*page
)
1717 #endif /* CONFIG_DEBUG_VM */
1719 static bool check_new_pages(struct page
*page
, unsigned int order
)
1722 for (i
= 0; i
< (1 << order
); i
++) {
1723 struct page
*p
= page
+ i
;
1725 if (unlikely(check_new_page(p
)))
1732 inline void post_alloc_hook(struct page
*page
, unsigned int order
,
1735 set_page_private(page
, 0);
1736 set_page_refcounted(page
);
1738 arch_alloc_page(page
, order
);
1739 kernel_map_pages(page
, 1 << order
, 1);
1740 kernel_poison_pages(page
, 1 << order
, 1);
1741 kasan_alloc_pages(page
, order
);
1742 set_page_owner(page
, order
, gfp_flags
);
1745 static void prep_new_page(struct page
*page
, unsigned int order
, gfp_t gfp_flags
,
1746 unsigned int alloc_flags
)
1749 bool poisoned
= true;
1751 for (i
= 0; i
< (1 << order
); i
++) {
1752 struct page
*p
= page
+ i
;
1754 poisoned
&= page_is_poisoned(p
);
1757 post_alloc_hook(page
, order
, gfp_flags
);
1759 if (!free_pages_prezeroed(poisoned
) && (gfp_flags
& __GFP_ZERO
))
1760 for (i
= 0; i
< (1 << order
); i
++)
1761 clear_highpage(page
+ i
);
1763 if (order
&& (gfp_flags
& __GFP_COMP
))
1764 prep_compound_page(page
, order
);
1767 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1768 * allocate the page. The expectation is that the caller is taking
1769 * steps that will free more memory. The caller should avoid the page
1770 * being used for !PFMEMALLOC purposes.
1772 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
1773 set_page_pfmemalloc(page
);
1775 clear_page_pfmemalloc(page
);
1779 * Go through the free lists for the given migratetype and remove
1780 * the smallest available page from the freelists
1783 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
1786 unsigned int current_order
;
1787 struct free_area
*area
;
1790 /* Find a page of the appropriate size in the preferred list */
1791 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
1792 area
= &(zone
->free_area
[current_order
]);
1793 page
= list_first_entry_or_null(&area
->free_list
[migratetype
],
1797 list_del(&page
->lru
);
1798 rmv_page_order(page
);
1800 expand(zone
, page
, order
, current_order
, area
, migratetype
);
1801 set_pcppage_migratetype(page
, migratetype
);
1810 * This array describes the order lists are fallen back to when
1811 * the free lists for the desirable migrate type are depleted
1813 static int fallbacks
[MIGRATE_TYPES
][4] = {
1814 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1815 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1816 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_TYPES
},
1818 [MIGRATE_CMA
] = { MIGRATE_TYPES
}, /* Never used */
1820 #ifdef CONFIG_MEMORY_ISOLATION
1821 [MIGRATE_ISOLATE
] = { MIGRATE_TYPES
}, /* Never used */
1826 static struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1829 return __rmqueue_smallest(zone
, order
, MIGRATE_CMA
);
1832 static inline struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1833 unsigned int order
) { return NULL
; }
1837 * Move the free pages in a range to the free lists of the requested type.
1838 * Note that start_page and end_pages are not aligned on a pageblock
1839 * boundary. If alignment is required, use move_freepages_block()
1841 int move_freepages(struct zone
*zone
,
1842 struct page
*start_page
, struct page
*end_page
,
1847 int pages_moved
= 0;
1849 #ifndef CONFIG_HOLES_IN_ZONE
1851 * page_zone is not safe to call in this context when
1852 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1853 * anyway as we check zone boundaries in move_freepages_block().
1854 * Remove at a later date when no bug reports exist related to
1855 * grouping pages by mobility
1857 VM_BUG_ON(page_zone(start_page
) != page_zone(end_page
));
1860 for (page
= start_page
; page
<= end_page
;) {
1861 /* Make sure we are not inadvertently changing nodes */
1862 VM_BUG_ON_PAGE(page_to_nid(page
) != zone_to_nid(zone
), page
);
1864 if (!pfn_valid_within(page_to_pfn(page
))) {
1869 if (!PageBuddy(page
)) {
1874 order
= page_order(page
);
1875 list_move(&page
->lru
,
1876 &zone
->free_area
[order
].free_list
[migratetype
]);
1878 pages_moved
+= 1 << order
;
1884 int move_freepages_block(struct zone
*zone
, struct page
*page
,
1887 unsigned long start_pfn
, end_pfn
;
1888 struct page
*start_page
, *end_page
;
1890 start_pfn
= page_to_pfn(page
);
1891 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
1892 start_page
= pfn_to_page(start_pfn
);
1893 end_page
= start_page
+ pageblock_nr_pages
- 1;
1894 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
1896 /* Do not cross zone boundaries */
1897 if (!zone_spans_pfn(zone
, start_pfn
))
1899 if (!zone_spans_pfn(zone
, end_pfn
))
1902 return move_freepages(zone
, start_page
, end_page
, migratetype
);
1905 static void change_pageblock_range(struct page
*pageblock_page
,
1906 int start_order
, int migratetype
)
1908 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
1910 while (nr_pageblocks
--) {
1911 set_pageblock_migratetype(pageblock_page
, migratetype
);
1912 pageblock_page
+= pageblock_nr_pages
;
1917 * When we are falling back to another migratetype during allocation, try to
1918 * steal extra free pages from the same pageblocks to satisfy further
1919 * allocations, instead of polluting multiple pageblocks.
1921 * If we are stealing a relatively large buddy page, it is likely there will
1922 * be more free pages in the pageblock, so try to steal them all. For
1923 * reclaimable and unmovable allocations, we steal regardless of page size,
1924 * as fragmentation caused by those allocations polluting movable pageblocks
1925 * is worse than movable allocations stealing from unmovable and reclaimable
1928 static bool can_steal_fallback(unsigned int order
, int start_mt
)
1931 * Leaving this order check is intended, although there is
1932 * relaxed order check in next check. The reason is that
1933 * we can actually steal whole pageblock if this condition met,
1934 * but, below check doesn't guarantee it and that is just heuristic
1935 * so could be changed anytime.
1937 if (order
>= pageblock_order
)
1940 if (order
>= pageblock_order
/ 2 ||
1941 start_mt
== MIGRATE_RECLAIMABLE
||
1942 start_mt
== MIGRATE_UNMOVABLE
||
1943 page_group_by_mobility_disabled
)
1950 * This function implements actual steal behaviour. If order is large enough,
1951 * we can steal whole pageblock. If not, we first move freepages in this
1952 * pageblock and check whether half of pages are moved or not. If half of
1953 * pages are moved, we can change migratetype of pageblock and permanently
1954 * use it's pages as requested migratetype in the future.
1956 static void steal_suitable_fallback(struct zone
*zone
, struct page
*page
,
1959 unsigned int current_order
= page_order(page
);
1962 /* Take ownership for orders >= pageblock_order */
1963 if (current_order
>= pageblock_order
) {
1964 change_pageblock_range(page
, current_order
, start_type
);
1968 pages
= move_freepages_block(zone
, page
, start_type
);
1970 /* Claim the whole block if over half of it is free */
1971 if (pages
>= (1 << (pageblock_order
-1)) ||
1972 page_group_by_mobility_disabled
)
1973 set_pageblock_migratetype(page
, start_type
);
1977 * Check whether there is a suitable fallback freepage with requested order.
1978 * If only_stealable is true, this function returns fallback_mt only if
1979 * we can steal other freepages all together. This would help to reduce
1980 * fragmentation due to mixed migratetype pages in one pageblock.
1982 int find_suitable_fallback(struct free_area
*area
, unsigned int order
,
1983 int migratetype
, bool only_stealable
, bool *can_steal
)
1988 if (area
->nr_free
== 0)
1993 fallback_mt
= fallbacks
[migratetype
][i
];
1994 if (fallback_mt
== MIGRATE_TYPES
)
1997 if (list_empty(&area
->free_list
[fallback_mt
]))
2000 if (can_steal_fallback(order
, migratetype
))
2003 if (!only_stealable
)
2014 * Reserve a pageblock for exclusive use of high-order atomic allocations if
2015 * there are no empty page blocks that contain a page with a suitable order
2017 static void reserve_highatomic_pageblock(struct page
*page
, struct zone
*zone
,
2018 unsigned int alloc_order
)
2021 unsigned long max_managed
, flags
;
2024 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
2025 * Check is race-prone but harmless.
2027 max_managed
= (zone
->managed_pages
/ 100) + pageblock_nr_pages
;
2028 if (zone
->nr_reserved_highatomic
>= max_managed
)
2031 spin_lock_irqsave(&zone
->lock
, flags
);
2033 /* Recheck the nr_reserved_highatomic limit under the lock */
2034 if (zone
->nr_reserved_highatomic
>= max_managed
)
2038 mt
= get_pageblock_migratetype(page
);
2039 if (mt
!= MIGRATE_HIGHATOMIC
&&
2040 !is_migrate_isolate(mt
) && !is_migrate_cma(mt
)) {
2041 zone
->nr_reserved_highatomic
+= pageblock_nr_pages
;
2042 set_pageblock_migratetype(page
, MIGRATE_HIGHATOMIC
);
2043 move_freepages_block(zone
, page
, MIGRATE_HIGHATOMIC
);
2047 spin_unlock_irqrestore(&zone
->lock
, flags
);
2051 * Used when an allocation is about to fail under memory pressure. This
2052 * potentially hurts the reliability of high-order allocations when under
2053 * intense memory pressure but failed atomic allocations should be easier
2054 * to recover from than an OOM.
2056 static void unreserve_highatomic_pageblock(const struct alloc_context
*ac
)
2058 struct zonelist
*zonelist
= ac
->zonelist
;
2059 unsigned long flags
;
2065 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
, ac
->high_zoneidx
,
2067 /* Preserve at least one pageblock */
2068 if (zone
->nr_reserved_highatomic
<= pageblock_nr_pages
)
2071 spin_lock_irqsave(&zone
->lock
, flags
);
2072 for (order
= 0; order
< MAX_ORDER
; order
++) {
2073 struct free_area
*area
= &(zone
->free_area
[order
]);
2075 page
= list_first_entry_or_null(
2076 &area
->free_list
[MIGRATE_HIGHATOMIC
],
2082 * It should never happen but changes to locking could
2083 * inadvertently allow a per-cpu drain to add pages
2084 * to MIGRATE_HIGHATOMIC while unreserving so be safe
2085 * and watch for underflows.
2087 zone
->nr_reserved_highatomic
-= min(pageblock_nr_pages
,
2088 zone
->nr_reserved_highatomic
);
2091 * Convert to ac->migratetype and avoid the normal
2092 * pageblock stealing heuristics. Minimally, the caller
2093 * is doing the work and needs the pages. More
2094 * importantly, if the block was always converted to
2095 * MIGRATE_UNMOVABLE or another type then the number
2096 * of pageblocks that cannot be completely freed
2099 set_pageblock_migratetype(page
, ac
->migratetype
);
2100 move_freepages_block(zone
, page
, ac
->migratetype
);
2101 spin_unlock_irqrestore(&zone
->lock
, flags
);
2104 spin_unlock_irqrestore(&zone
->lock
, flags
);
2108 /* Remove an element from the buddy allocator from the fallback list */
2109 static inline struct page
*
2110 __rmqueue_fallback(struct zone
*zone
, unsigned int order
, int start_migratetype
)
2112 struct free_area
*area
;
2113 unsigned int current_order
;
2118 /* Find the largest possible block of pages in the other list */
2119 for (current_order
= MAX_ORDER
-1;
2120 current_order
>= order
&& current_order
<= MAX_ORDER
-1;
2122 area
= &(zone
->free_area
[current_order
]);
2123 fallback_mt
= find_suitable_fallback(area
, current_order
,
2124 start_migratetype
, false, &can_steal
);
2125 if (fallback_mt
== -1)
2128 page
= list_first_entry(&area
->free_list
[fallback_mt
],
2131 steal_suitable_fallback(zone
, page
, start_migratetype
);
2133 /* Remove the page from the freelists */
2135 list_del(&page
->lru
);
2136 rmv_page_order(page
);
2138 expand(zone
, page
, order
, current_order
, area
,
2141 * The pcppage_migratetype may differ from pageblock's
2142 * migratetype depending on the decisions in
2143 * find_suitable_fallback(). This is OK as long as it does not
2144 * differ for MIGRATE_CMA pageblocks. Those can be used as
2145 * fallback only via special __rmqueue_cma_fallback() function
2147 set_pcppage_migratetype(page
, start_migratetype
);
2149 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
2150 start_migratetype
, fallback_mt
);
2159 * Do the hard work of removing an element from the buddy allocator.
2160 * Call me with the zone->lock already held.
2162 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
2167 page
= __rmqueue_smallest(zone
, order
, migratetype
);
2168 if (unlikely(!page
)) {
2169 if (migratetype
== MIGRATE_MOVABLE
)
2170 page
= __rmqueue_cma_fallback(zone
, order
);
2173 page
= __rmqueue_fallback(zone
, order
, migratetype
);
2176 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
2181 * Obtain a specified number of elements from the buddy allocator, all under
2182 * a single hold of the lock, for efficiency. Add them to the supplied list.
2183 * Returns the number of new pages which were placed at *list.
2185 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
2186 unsigned long count
, struct list_head
*list
,
2187 int migratetype
, bool cold
)
2191 spin_lock(&zone
->lock
);
2192 for (i
= 0; i
< count
; ++i
) {
2193 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
2194 if (unlikely(page
== NULL
))
2197 if (unlikely(check_pcp_refill(page
)))
2201 * Split buddy pages returned by expand() are received here
2202 * in physical page order. The page is added to the callers and
2203 * list and the list head then moves forward. From the callers
2204 * perspective, the linked list is ordered by page number in
2205 * some conditions. This is useful for IO devices that can
2206 * merge IO requests if the physical pages are ordered
2210 list_add(&page
->lru
, list
);
2212 list_add_tail(&page
->lru
, list
);
2214 if (is_migrate_cma(get_pcppage_migratetype(page
)))
2215 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
,
2218 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
2219 spin_unlock(&zone
->lock
);
2225 * Called from the vmstat counter updater to drain pagesets of this
2226 * currently executing processor on remote nodes after they have
2229 * Note that this function must be called with the thread pinned to
2230 * a single processor.
2232 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
2234 unsigned long flags
;
2235 int to_drain
, batch
;
2237 local_irq_save(flags
);
2238 batch
= READ_ONCE(pcp
->batch
);
2239 to_drain
= min(pcp
->count
, batch
);
2241 free_pcppages_bulk(zone
, to_drain
, pcp
);
2242 pcp
->count
-= to_drain
;
2244 local_irq_restore(flags
);
2249 * Drain pcplists of the indicated processor and zone.
2251 * The processor must either be the current processor and the
2252 * thread pinned to the current processor or a processor that
2255 static void drain_pages_zone(unsigned int cpu
, struct zone
*zone
)
2257 unsigned long flags
;
2258 struct per_cpu_pageset
*pset
;
2259 struct per_cpu_pages
*pcp
;
2261 local_irq_save(flags
);
2262 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
2266 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
2269 local_irq_restore(flags
);
2273 * Drain pcplists of all zones on the indicated processor.
2275 * The processor must either be the current processor and the
2276 * thread pinned to the current processor or a processor that
2279 static void drain_pages(unsigned int cpu
)
2283 for_each_populated_zone(zone
) {
2284 drain_pages_zone(cpu
, zone
);
2289 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
2291 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2292 * the single zone's pages.
2294 void drain_local_pages(struct zone
*zone
)
2296 int cpu
= smp_processor_id();
2299 drain_pages_zone(cpu
, zone
);
2305 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2307 * When zone parameter is non-NULL, spill just the single zone's pages.
2309 * Note that this code is protected against sending an IPI to an offline
2310 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
2311 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
2312 * nothing keeps CPUs from showing up after we populated the cpumask and
2313 * before the call to on_each_cpu_mask().
2315 void drain_all_pages(struct zone
*zone
)
2320 * Allocate in the BSS so we wont require allocation in
2321 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2323 static cpumask_t cpus_with_pcps
;
2326 * We don't care about racing with CPU hotplug event
2327 * as offline notification will cause the notified
2328 * cpu to drain that CPU pcps and on_each_cpu_mask
2329 * disables preemption as part of its processing
2331 for_each_online_cpu(cpu
) {
2332 struct per_cpu_pageset
*pcp
;
2334 bool has_pcps
= false;
2337 pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
2341 for_each_populated_zone(z
) {
2342 pcp
= per_cpu_ptr(z
->pageset
, cpu
);
2343 if (pcp
->pcp
.count
) {
2351 cpumask_set_cpu(cpu
, &cpus_with_pcps
);
2353 cpumask_clear_cpu(cpu
, &cpus_with_pcps
);
2355 on_each_cpu_mask(&cpus_with_pcps
, (smp_call_func_t
) drain_local_pages
,
2359 #ifdef CONFIG_HIBERNATION
2361 void mark_free_pages(struct zone
*zone
)
2363 unsigned long pfn
, max_zone_pfn
;
2364 unsigned long flags
;
2365 unsigned int order
, t
;
2368 if (zone_is_empty(zone
))
2371 spin_lock_irqsave(&zone
->lock
, flags
);
2373 max_zone_pfn
= zone_end_pfn(zone
);
2374 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
2375 if (pfn_valid(pfn
)) {
2376 page
= pfn_to_page(pfn
);
2378 if (page_zone(page
) != zone
)
2381 if (!swsusp_page_is_forbidden(page
))
2382 swsusp_unset_page_free(page
);
2385 for_each_migratetype_order(order
, t
) {
2386 list_for_each_entry(page
,
2387 &zone
->free_area
[order
].free_list
[t
], lru
) {
2390 pfn
= page_to_pfn(page
);
2391 for (i
= 0; i
< (1UL << order
); i
++)
2392 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
2395 spin_unlock_irqrestore(&zone
->lock
, flags
);
2397 #endif /* CONFIG_PM */
2400 * Free a 0-order page
2401 * cold == true ? free a cold page : free a hot page
2403 void free_hot_cold_page(struct page
*page
, bool cold
)
2405 struct zone
*zone
= page_zone(page
);
2406 struct per_cpu_pages
*pcp
;
2407 unsigned long flags
;
2408 unsigned long pfn
= page_to_pfn(page
);
2411 if (!free_pcp_prepare(page
))
2414 migratetype
= get_pfnblock_migratetype(page
, pfn
);
2415 set_pcppage_migratetype(page
, migratetype
);
2416 local_irq_save(flags
);
2417 __count_vm_event(PGFREE
);
2420 * We only track unmovable, reclaimable and movable on pcp lists.
2421 * Free ISOLATE pages back to the allocator because they are being
2422 * offlined but treat RESERVE as movable pages so we can get those
2423 * areas back if necessary. Otherwise, we may have to free
2424 * excessively into the page allocator
2426 if (migratetype
>= MIGRATE_PCPTYPES
) {
2427 if (unlikely(is_migrate_isolate(migratetype
))) {
2428 free_one_page(zone
, page
, pfn
, 0, migratetype
);
2431 migratetype
= MIGRATE_MOVABLE
;
2434 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2436 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
2438 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
2440 if (pcp
->count
>= pcp
->high
) {
2441 unsigned long batch
= READ_ONCE(pcp
->batch
);
2442 free_pcppages_bulk(zone
, batch
, pcp
);
2443 pcp
->count
-= batch
;
2447 local_irq_restore(flags
);
2451 * Free a list of 0-order pages
2453 void free_hot_cold_page_list(struct list_head
*list
, bool cold
)
2455 struct page
*page
, *next
;
2457 list_for_each_entry_safe(page
, next
, list
, lru
) {
2458 trace_mm_page_free_batched(page
, cold
);
2459 free_hot_cold_page(page
, cold
);
2464 * split_page takes a non-compound higher-order page, and splits it into
2465 * n (1<<order) sub-pages: page[0..n]
2466 * Each sub-page must be freed individually.
2468 * Note: this is probably too low level an operation for use in drivers.
2469 * Please consult with lkml before using this in your driver.
2471 void split_page(struct page
*page
, unsigned int order
)
2475 VM_BUG_ON_PAGE(PageCompound(page
), page
);
2476 VM_BUG_ON_PAGE(!page_count(page
), page
);
2478 #ifdef CONFIG_KMEMCHECK
2480 * Split shadow pages too, because free(page[0]) would
2481 * otherwise free the whole shadow.
2483 if (kmemcheck_page_is_tracked(page
))
2484 split_page(virt_to_page(page
[0].shadow
), order
);
2487 for (i
= 1; i
< (1 << order
); i
++)
2488 set_page_refcounted(page
+ i
);
2489 split_page_owner(page
, order
);
2491 EXPORT_SYMBOL_GPL(split_page
);
2493 int __isolate_free_page(struct page
*page
, unsigned int order
)
2495 unsigned long watermark
;
2499 BUG_ON(!PageBuddy(page
));
2501 zone
= page_zone(page
);
2502 mt
= get_pageblock_migratetype(page
);
2504 if (!is_migrate_isolate(mt
)) {
2505 /* Obey watermarks as if the page was being allocated */
2506 watermark
= low_wmark_pages(zone
) + (1 << order
);
2507 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
2510 __mod_zone_freepage_state(zone
, -(1UL << order
), mt
);
2513 /* Remove page from free list */
2514 list_del(&page
->lru
);
2515 zone
->free_area
[order
].nr_free
--;
2516 rmv_page_order(page
);
2518 /* Set the pageblock if the isolated page is at least a pageblock */
2519 if (order
>= pageblock_order
- 1) {
2520 struct page
*endpage
= page
+ (1 << order
) - 1;
2521 for (; page
< endpage
; page
+= pageblock_nr_pages
) {
2522 int mt
= get_pageblock_migratetype(page
);
2523 if (!is_migrate_isolate(mt
) && !is_migrate_cma(mt
))
2524 set_pageblock_migratetype(page
,
2530 return 1UL << order
;
2534 * Update NUMA hit/miss statistics
2536 * Must be called with interrupts disabled.
2538 * When __GFP_OTHER_NODE is set assume the node of the preferred
2539 * zone is the local node. This is useful for daemons who allocate
2540 * memory on behalf of other processes.
2542 static inline void zone_statistics(struct zone
*preferred_zone
, struct zone
*z
,
2546 int local_nid
= numa_node_id();
2547 enum zone_stat_item local_stat
= NUMA_LOCAL
;
2549 if (unlikely(flags
& __GFP_OTHER_NODE
)) {
2550 local_stat
= NUMA_OTHER
;
2551 local_nid
= preferred_zone
->node
;
2554 if (z
->node
== local_nid
) {
2555 __inc_zone_state(z
, NUMA_HIT
);
2556 __inc_zone_state(z
, local_stat
);
2558 __inc_zone_state(z
, NUMA_MISS
);
2559 __inc_zone_state(preferred_zone
, NUMA_FOREIGN
);
2565 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2568 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
2569 struct zone
*zone
, unsigned int order
,
2570 gfp_t gfp_flags
, unsigned int alloc_flags
,
2573 unsigned long flags
;
2575 bool cold
= ((gfp_flags
& __GFP_COLD
) != 0);
2577 if (likely(order
== 0)) {
2578 struct per_cpu_pages
*pcp
;
2579 struct list_head
*list
;
2581 local_irq_save(flags
);
2583 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2584 list
= &pcp
->lists
[migratetype
];
2585 if (list_empty(list
)) {
2586 pcp
->count
+= rmqueue_bulk(zone
, 0,
2589 if (unlikely(list_empty(list
)))
2594 page
= list_last_entry(list
, struct page
, lru
);
2596 page
= list_first_entry(list
, struct page
, lru
);
2598 __dec_zone_state(zone
, NR_ALLOC_BATCH
);
2599 list_del(&page
->lru
);
2602 } while (check_new_pcp(page
));
2605 * We most definitely don't want callers attempting to
2606 * allocate greater than order-1 page units with __GFP_NOFAIL.
2608 WARN_ON_ONCE((gfp_flags
& __GFP_NOFAIL
) && (order
> 1));
2609 spin_lock_irqsave(&zone
->lock
, flags
);
2613 if (alloc_flags
& ALLOC_HARDER
) {
2614 page
= __rmqueue_smallest(zone
, order
, MIGRATE_HIGHATOMIC
);
2616 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
2619 page
= __rmqueue(zone
, order
, migratetype
);
2620 } while (page
&& check_new_pages(page
, order
));
2621 spin_unlock(&zone
->lock
);
2624 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
, -(1 << order
));
2625 __mod_zone_freepage_state(zone
, -(1 << order
),
2626 get_pcppage_migratetype(page
));
2629 if (atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]) <= 0 &&
2630 !test_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
))
2631 set_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
);
2633 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
2634 zone_statistics(preferred_zone
, zone
, gfp_flags
);
2635 local_irq_restore(flags
);
2637 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
2641 local_irq_restore(flags
);
2645 #ifdef CONFIG_FAIL_PAGE_ALLOC
2648 struct fault_attr attr
;
2650 bool ignore_gfp_highmem
;
2651 bool ignore_gfp_reclaim
;
2653 } fail_page_alloc
= {
2654 .attr
= FAULT_ATTR_INITIALIZER
,
2655 .ignore_gfp_reclaim
= true,
2656 .ignore_gfp_highmem
= true,
2660 static int __init
setup_fail_page_alloc(char *str
)
2662 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
2664 __setup("fail_page_alloc=", setup_fail_page_alloc
);
2666 static bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2668 if (order
< fail_page_alloc
.min_order
)
2670 if (gfp_mask
& __GFP_NOFAIL
)
2672 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
2674 if (fail_page_alloc
.ignore_gfp_reclaim
&&
2675 (gfp_mask
& __GFP_DIRECT_RECLAIM
))
2678 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
2681 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2683 static int __init
fail_page_alloc_debugfs(void)
2685 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
2688 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
2689 &fail_page_alloc
.attr
);
2691 return PTR_ERR(dir
);
2693 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
2694 &fail_page_alloc
.ignore_gfp_reclaim
))
2696 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
2697 &fail_page_alloc
.ignore_gfp_highmem
))
2699 if (!debugfs_create_u32("min-order", mode
, dir
,
2700 &fail_page_alloc
.min_order
))
2705 debugfs_remove_recursive(dir
);
2710 late_initcall(fail_page_alloc_debugfs
);
2712 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2714 #else /* CONFIG_FAIL_PAGE_ALLOC */
2716 static inline bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2721 #endif /* CONFIG_FAIL_PAGE_ALLOC */
2724 * Return true if free base pages are above 'mark'. For high-order checks it
2725 * will return true of the order-0 watermark is reached and there is at least
2726 * one free page of a suitable size. Checking now avoids taking the zone lock
2727 * to check in the allocation paths if no pages are free.
2729 bool __zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
2730 int classzone_idx
, unsigned int alloc_flags
,
2735 const bool alloc_harder
= (alloc_flags
& ALLOC_HARDER
);
2737 /* free_pages may go negative - that's OK */
2738 free_pages
-= (1 << order
) - 1;
2740 if (alloc_flags
& ALLOC_HIGH
)
2744 * If the caller does not have rights to ALLOC_HARDER then subtract
2745 * the high-atomic reserves. This will over-estimate the size of the
2746 * atomic reserve but it avoids a search.
2748 if (likely(!alloc_harder
))
2749 free_pages
-= z
->nr_reserved_highatomic
;
2754 /* If allocation can't use CMA areas don't use free CMA pages */
2755 if (!(alloc_flags
& ALLOC_CMA
))
2756 free_pages
-= zone_page_state(z
, NR_FREE_CMA_PAGES
);
2760 * Check watermarks for an order-0 allocation request. If these
2761 * are not met, then a high-order request also cannot go ahead
2762 * even if a suitable page happened to be free.
2764 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
2767 /* If this is an order-0 request then the watermark is fine */
2771 /* For a high-order request, check at least one suitable page is free */
2772 for (o
= order
; o
< MAX_ORDER
; o
++) {
2773 struct free_area
*area
= &z
->free_area
[o
];
2782 for (mt
= 0; mt
< MIGRATE_PCPTYPES
; mt
++) {
2783 if (!list_empty(&area
->free_list
[mt
]))
2788 if ((alloc_flags
& ALLOC_CMA
) &&
2789 !list_empty(&area
->free_list
[MIGRATE_CMA
])) {
2797 bool zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
2798 int classzone_idx
, unsigned int alloc_flags
)
2800 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
2801 zone_page_state(z
, NR_FREE_PAGES
));
2804 static inline bool zone_watermark_fast(struct zone
*z
, unsigned int order
,
2805 unsigned long mark
, int classzone_idx
, unsigned int alloc_flags
)
2807 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
2811 /* If allocation can't use CMA areas don't use free CMA pages */
2812 if (!(alloc_flags
& ALLOC_CMA
))
2813 cma_pages
= zone_page_state(z
, NR_FREE_CMA_PAGES
);
2817 * Fast check for order-0 only. If this fails then the reserves
2818 * need to be calculated. There is a corner case where the check
2819 * passes but only the high-order atomic reserve are free. If
2820 * the caller is !atomic then it'll uselessly search the free
2821 * list. That corner case is then slower but it is harmless.
2823 if (!order
&& (free_pages
- cma_pages
) > mark
+ z
->lowmem_reserve
[classzone_idx
])
2826 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
2830 bool zone_watermark_ok_safe(struct zone
*z
, unsigned int order
,
2831 unsigned long mark
, int classzone_idx
)
2833 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
2835 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
2836 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
2838 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, 0,
2843 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
2845 return local_zone
->node
== zone
->node
;
2848 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2850 return node_distance(zone_to_nid(local_zone
), zone_to_nid(zone
)) <
2853 #else /* CONFIG_NUMA */
2854 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
2859 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2863 #endif /* CONFIG_NUMA */
2865 static void reset_alloc_batches(struct zone
*preferred_zone
)
2867 struct zone
*zone
= preferred_zone
->zone_pgdat
->node_zones
;
2870 mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
2871 high_wmark_pages(zone
) - low_wmark_pages(zone
) -
2872 atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]));
2873 clear_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
);
2874 } while (zone
++ != preferred_zone
);
2878 * get_page_from_freelist goes through the zonelist trying to allocate
2881 static struct page
*
2882 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
, int alloc_flags
,
2883 const struct alloc_context
*ac
)
2885 struct zoneref
*z
= ac
->preferred_zoneref
;
2887 bool fair_skipped
= false;
2888 bool apply_fair
= (alloc_flags
& ALLOC_FAIR
);
2892 * Scan zonelist, looking for a zone with enough free.
2893 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2895 for_next_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
, ac
->high_zoneidx
,
2900 if (cpusets_enabled() &&
2901 (alloc_flags
& ALLOC_CPUSET
) &&
2902 !__cpuset_zone_allowed(zone
, gfp_mask
))
2905 * Distribute pages in proportion to the individual
2906 * zone size to ensure fair page aging. The zone a
2907 * page was allocated in should have no effect on the
2908 * time the page has in memory before being reclaimed.
2911 if (test_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
)) {
2912 fair_skipped
= true;
2915 if (!zone_local(ac
->preferred_zoneref
->zone
, zone
)) {
2922 * When allocating a page cache page for writing, we
2923 * want to get it from a zone that is within its dirty
2924 * limit, such that no single zone holds more than its
2925 * proportional share of globally allowed dirty pages.
2926 * The dirty limits take into account the zone's
2927 * lowmem reserves and high watermark so that kswapd
2928 * should be able to balance it without having to
2929 * write pages from its LRU list.
2931 * This may look like it could increase pressure on
2932 * lower zones by failing allocations in higher zones
2933 * before they are full. But the pages that do spill
2934 * over are limited as the lower zones are protected
2935 * by this very same mechanism. It should not become
2936 * a practical burden to them.
2938 * XXX: For now, allow allocations to potentially
2939 * exceed the per-zone dirty limit in the slowpath
2940 * (spread_dirty_pages unset) before going into reclaim,
2941 * which is important when on a NUMA setup the allowed
2942 * zones are together not big enough to reach the
2943 * global limit. The proper fix for these situations
2944 * will require awareness of zones in the
2945 * dirty-throttling and the flusher threads.
2947 if (ac
->spread_dirty_pages
&& !zone_dirty_ok(zone
))
2950 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
2951 if (!zone_watermark_fast(zone
, order
, mark
,
2952 ac_classzone_idx(ac
), alloc_flags
)) {
2955 /* Checked here to keep the fast path fast */
2956 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
2957 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
2960 if (zone_reclaim_mode
== 0 ||
2961 !zone_allows_reclaim(ac
->preferred_zoneref
->zone
, zone
))
2964 ret
= zone_reclaim(zone
, gfp_mask
, order
);
2966 case ZONE_RECLAIM_NOSCAN
:
2969 case ZONE_RECLAIM_FULL
:
2970 /* scanned but unreclaimable */
2973 /* did we reclaim enough */
2974 if (zone_watermark_ok(zone
, order
, mark
,
2975 ac_classzone_idx(ac
), alloc_flags
))
2983 page
= buffered_rmqueue(ac
->preferred_zoneref
->zone
, zone
, order
,
2984 gfp_mask
, alloc_flags
, ac
->migratetype
);
2986 prep_new_page(page
, order
, gfp_mask
, alloc_flags
);
2989 * If this is a high-order atomic allocation then check
2990 * if the pageblock should be reserved for the future
2992 if (unlikely(order
&& (alloc_flags
& ALLOC_HARDER
)))
2993 reserve_highatomic_pageblock(page
, zone
, order
);
3000 * The first pass makes sure allocations are spread fairly within the
3001 * local node. However, the local node might have free pages left
3002 * after the fairness batches are exhausted, and remote zones haven't
3003 * even been considered yet. Try once more without fairness, and
3004 * include remote zones now, before entering the slowpath and waking
3005 * kswapd: prefer spilling to a remote zone over swapping locally.
3010 fair_skipped
= false;
3011 reset_alloc_batches(ac
->preferred_zoneref
->zone
);
3012 z
= ac
->preferred_zoneref
;
3020 * Large machines with many possible nodes should not always dump per-node
3021 * meminfo in irq context.
3023 static inline bool should_suppress_show_mem(void)
3028 ret
= in_interrupt();
3033 static DEFINE_RATELIMIT_STATE(nopage_rs
,
3034 DEFAULT_RATELIMIT_INTERVAL
,
3035 DEFAULT_RATELIMIT_BURST
);
3037 void warn_alloc_failed(gfp_t gfp_mask
, unsigned int order
, const char *fmt
, ...)
3039 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
3041 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
) ||
3042 debug_guardpage_minorder() > 0)
3046 * This documents exceptions given to allocations in certain
3047 * contexts that are allowed to allocate outside current's set
3050 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
3051 if (test_thread_flag(TIF_MEMDIE
) ||
3052 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
3053 filter
&= ~SHOW_MEM_FILTER_NODES
;
3054 if (in_interrupt() || !(gfp_mask
& __GFP_DIRECT_RECLAIM
))
3055 filter
&= ~SHOW_MEM_FILTER_NODES
;
3058 struct va_format vaf
;
3061 va_start(args
, fmt
);
3066 pr_warn("%pV", &vaf
);
3071 pr_warn("%s: page allocation failure: order:%u, mode:%#x(%pGg)\n",
3072 current
->comm
, order
, gfp_mask
, &gfp_mask
);
3074 if (!should_suppress_show_mem())
3078 static inline struct page
*
3079 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
3080 const struct alloc_context
*ac
, unsigned long *did_some_progress
)
3082 struct oom_control oc
= {
3083 .zonelist
= ac
->zonelist
,
3084 .nodemask
= ac
->nodemask
,
3086 .gfp_mask
= gfp_mask
,
3091 *did_some_progress
= 0;
3094 * Acquire the oom lock. If that fails, somebody else is
3095 * making progress for us.
3097 if (!mutex_trylock(&oom_lock
)) {
3098 *did_some_progress
= 1;
3099 schedule_timeout_uninterruptible(1);
3104 * Go through the zonelist yet one more time, keep very high watermark
3105 * here, this is only to catch a parallel oom killing, we must fail if
3106 * we're still under heavy pressure.
3108 page
= get_page_from_freelist(gfp_mask
| __GFP_HARDWALL
, order
,
3109 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
, ac
);
3113 if (!(gfp_mask
& __GFP_NOFAIL
)) {
3114 /* Coredumps can quickly deplete all memory reserves */
3115 if (current
->flags
& PF_DUMPCORE
)
3117 /* The OOM killer will not help higher order allocs */
3118 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
3120 /* The OOM killer does not needlessly kill tasks for lowmem */
3121 if (ac
->high_zoneidx
< ZONE_NORMAL
)
3123 if (pm_suspended_storage())
3126 * XXX: GFP_NOFS allocations should rather fail than rely on
3127 * other request to make a forward progress.
3128 * We are in an unfortunate situation where out_of_memory cannot
3129 * do much for this context but let's try it to at least get
3130 * access to memory reserved if the current task is killed (see
3131 * out_of_memory). Once filesystems are ready to handle allocation
3132 * failures more gracefully we should just bail out here.
3135 /* The OOM killer may not free memory on a specific node */
3136 if (gfp_mask
& __GFP_THISNODE
)
3139 /* Exhausted what can be done so it's blamo time */
3140 if (out_of_memory(&oc
) || WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
)) {
3141 *did_some_progress
= 1;
3143 if (gfp_mask
& __GFP_NOFAIL
) {
3144 page
= get_page_from_freelist(gfp_mask
, order
,
3145 ALLOC_NO_WATERMARKS
|ALLOC_CPUSET
, ac
);
3147 * fallback to ignore cpuset restriction if our nodes
3151 page
= get_page_from_freelist(gfp_mask
, order
,
3152 ALLOC_NO_WATERMARKS
, ac
);
3156 mutex_unlock(&oom_lock
);
3162 * Maximum number of compaction retries wit a progress before OOM
3163 * killer is consider as the only way to move forward.
3165 #define MAX_COMPACT_RETRIES 16
3167 #ifdef CONFIG_COMPACTION
3168 /* Try memory compaction for high-order allocations before reclaim */
3169 static struct page
*
3170 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
3171 unsigned int alloc_flags
, const struct alloc_context
*ac
,
3172 enum migrate_mode mode
, enum compact_result
*compact_result
)
3175 int contended_compaction
;
3180 current
->flags
|= PF_MEMALLOC
;
3181 *compact_result
= try_to_compact_pages(gfp_mask
, order
, alloc_flags
, ac
,
3182 mode
, &contended_compaction
);
3183 current
->flags
&= ~PF_MEMALLOC
;
3185 if (*compact_result
<= COMPACT_INACTIVE
)
3189 * At least in one zone compaction wasn't deferred or skipped, so let's
3190 * count a compaction stall
3192 count_vm_event(COMPACTSTALL
);
3194 page
= get_page_from_freelist(gfp_mask
, order
,
3195 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
3198 struct zone
*zone
= page_zone(page
);
3200 zone
->compact_blockskip_flush
= false;
3201 compaction_defer_reset(zone
, order
, true);
3202 count_vm_event(COMPACTSUCCESS
);
3207 * It's bad if compaction run occurs and fails. The most likely reason
3208 * is that pages exist, but not enough to satisfy watermarks.
3210 count_vm_event(COMPACTFAIL
);
3213 * In all zones where compaction was attempted (and not
3214 * deferred or skipped), lock contention has been detected.
3215 * For THP allocation we do not want to disrupt the others
3216 * so we fallback to base pages instead.
3218 if (contended_compaction
== COMPACT_CONTENDED_LOCK
)
3219 *compact_result
= COMPACT_CONTENDED
;
3222 * If compaction was aborted due to need_resched(), we do not
3223 * want to further increase allocation latency, unless it is
3224 * khugepaged trying to collapse.
3226 if (contended_compaction
== COMPACT_CONTENDED_SCHED
3227 && !(current
->flags
& PF_KTHREAD
))
3228 *compact_result
= COMPACT_CONTENDED
;
3236 should_compact_retry(struct alloc_context
*ac
, int order
, int alloc_flags
,
3237 enum compact_result compact_result
, enum migrate_mode
*migrate_mode
,
3238 int compaction_retries
)
3240 int max_retries
= MAX_COMPACT_RETRIES
;
3246 * compaction considers all the zone as desperately out of memory
3247 * so it doesn't really make much sense to retry except when the
3248 * failure could be caused by weak migration mode.
3250 if (compaction_failed(compact_result
)) {
3251 if (*migrate_mode
== MIGRATE_ASYNC
) {
3252 *migrate_mode
= MIGRATE_SYNC_LIGHT
;
3259 * make sure the compaction wasn't deferred or didn't bail out early
3260 * due to locks contention before we declare that we should give up.
3261 * But do not retry if the given zonelist is not suitable for
3264 if (compaction_withdrawn(compact_result
))
3265 return compaction_zonelist_suitable(ac
, order
, alloc_flags
);
3268 * !costly requests are much more important than __GFP_REPEAT
3269 * costly ones because they are de facto nofail and invoke OOM
3270 * killer to move on while costly can fail and users are ready
3271 * to cope with that. 1/4 retries is rather arbitrary but we
3272 * would need much more detailed feedback from compaction to
3273 * make a better decision.
3275 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
3277 if (compaction_retries
<= max_retries
)
3283 static inline struct page
*
3284 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
3285 unsigned int alloc_flags
, const struct alloc_context
*ac
,
3286 enum migrate_mode mode
, enum compact_result
*compact_result
)
3288 *compact_result
= COMPACT_SKIPPED
;
3293 should_compact_retry(struct alloc_context
*ac
, unsigned int order
, int alloc_flags
,
3294 enum compact_result compact_result
,
3295 enum migrate_mode
*migrate_mode
,
3296 int compaction_retries
)
3301 if (!order
|| order
> PAGE_ALLOC_COSTLY_ORDER
)
3305 * There are setups with compaction disabled which would prefer to loop
3306 * inside the allocator rather than hit the oom killer prematurely.
3307 * Let's give them a good hope and keep retrying while the order-0
3308 * watermarks are OK.
3310 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
, ac
->high_zoneidx
,
3312 if (zone_watermark_ok(zone
, 0, min_wmark_pages(zone
),
3313 ac_classzone_idx(ac
), alloc_flags
))
3318 #endif /* CONFIG_COMPACTION */
3320 /* Perform direct synchronous page reclaim */
3322 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
,
3323 const struct alloc_context
*ac
)
3325 struct reclaim_state reclaim_state
;
3330 /* We now go into synchronous reclaim */
3331 cpuset_memory_pressure_bump();
3332 current
->flags
|= PF_MEMALLOC
;
3333 lockdep_set_current_reclaim_state(gfp_mask
);
3334 reclaim_state
.reclaimed_slab
= 0;
3335 current
->reclaim_state
= &reclaim_state
;
3337 progress
= try_to_free_pages(ac
->zonelist
, order
, gfp_mask
,
3340 current
->reclaim_state
= NULL
;
3341 lockdep_clear_current_reclaim_state();
3342 current
->flags
&= ~PF_MEMALLOC
;
3349 /* The really slow allocator path where we enter direct reclaim */
3350 static inline struct page
*
3351 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
3352 unsigned int alloc_flags
, const struct alloc_context
*ac
,
3353 unsigned long *did_some_progress
)
3355 struct page
*page
= NULL
;
3356 bool drained
= false;
3358 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, ac
);
3359 if (unlikely(!(*did_some_progress
)))
3363 page
= get_page_from_freelist(gfp_mask
, order
,
3364 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
3367 * If an allocation failed after direct reclaim, it could be because
3368 * pages are pinned on the per-cpu lists or in high alloc reserves.
3369 * Shrink them them and try again
3371 if (!page
&& !drained
) {
3372 unreserve_highatomic_pageblock(ac
);
3373 drain_all_pages(NULL
);
3381 static void wake_all_kswapds(unsigned int order
, const struct alloc_context
*ac
)
3386 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
,
3387 ac
->high_zoneidx
, ac
->nodemask
)
3388 wakeup_kswapd(zone
, order
, ac_classzone_idx(ac
));
3391 static inline unsigned int
3392 gfp_to_alloc_flags(gfp_t gfp_mask
)
3394 unsigned int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
3396 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
3397 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
3400 * The caller may dip into page reserves a bit more if the caller
3401 * cannot run direct reclaim, or if the caller has realtime scheduling
3402 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
3403 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
3405 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
3407 if (gfp_mask
& __GFP_ATOMIC
) {
3409 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
3410 * if it can't schedule.
3412 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
3413 alloc_flags
|= ALLOC_HARDER
;
3415 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
3416 * comment for __cpuset_node_allowed().
3418 alloc_flags
&= ~ALLOC_CPUSET
;
3419 } else if (unlikely(rt_task(current
)) && !in_interrupt())
3420 alloc_flags
|= ALLOC_HARDER
;
3422 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
3423 if (gfp_mask
& __GFP_MEMALLOC
)
3424 alloc_flags
|= ALLOC_NO_WATERMARKS
;
3425 else if (in_serving_softirq() && (current
->flags
& PF_MEMALLOC
))
3426 alloc_flags
|= ALLOC_NO_WATERMARKS
;
3427 else if (!in_interrupt() &&
3428 ((current
->flags
& PF_MEMALLOC
) ||
3429 unlikely(test_thread_flag(TIF_MEMDIE
))))
3430 alloc_flags
|= ALLOC_NO_WATERMARKS
;
3433 if (gfpflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
3434 alloc_flags
|= ALLOC_CMA
;
3439 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask
)
3441 return !!(gfp_to_alloc_flags(gfp_mask
) & ALLOC_NO_WATERMARKS
);
3444 static inline bool is_thp_gfp_mask(gfp_t gfp_mask
)
3446 return (gfp_mask
& (GFP_TRANSHUGE
| __GFP_KSWAPD_RECLAIM
)) == GFP_TRANSHUGE
;
3450 * Maximum number of reclaim retries without any progress before OOM killer
3451 * is consider as the only way to move forward.
3453 #define MAX_RECLAIM_RETRIES 16
3456 * Checks whether it makes sense to retry the reclaim to make a forward progress
3457 * for the given allocation request.
3458 * The reclaim feedback represented by did_some_progress (any progress during
3459 * the last reclaim round) and no_progress_loops (number of reclaim rounds without
3460 * any progress in a row) is considered as well as the reclaimable pages on the
3461 * applicable zone list (with a backoff mechanism which is a function of
3462 * no_progress_loops).
3464 * Returns true if a retry is viable or false to enter the oom path.
3467 should_reclaim_retry(gfp_t gfp_mask
, unsigned order
,
3468 struct alloc_context
*ac
, int alloc_flags
,
3469 bool did_some_progress
, int no_progress_loops
)
3475 * Make sure we converge to OOM if we cannot make any progress
3476 * several times in the row.
3478 if (no_progress_loops
> MAX_RECLAIM_RETRIES
)
3482 * Keep reclaiming pages while there is a chance this will lead somewhere.
3483 * If none of the target zones can satisfy our allocation request even
3484 * if all reclaimable pages are considered then we are screwed and have
3487 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
, ac
->high_zoneidx
,
3489 unsigned long available
;
3490 unsigned long reclaimable
;
3492 available
= reclaimable
= zone_reclaimable_pages(zone
);
3493 available
-= DIV_ROUND_UP(no_progress_loops
* available
,
3494 MAX_RECLAIM_RETRIES
);
3495 available
+= zone_page_state_snapshot(zone
, NR_FREE_PAGES
);
3498 * Would the allocation succeed if we reclaimed the whole
3501 if (__zone_watermark_ok(zone
, order
, min_wmark_pages(zone
),
3502 ac_classzone_idx(ac
), alloc_flags
, available
)) {
3504 * If we didn't make any progress and have a lot of
3505 * dirty + writeback pages then we should wait for
3506 * an IO to complete to slow down the reclaim and
3507 * prevent from pre mature OOM
3509 if (!did_some_progress
) {
3510 unsigned long writeback
;
3511 unsigned long dirty
;
3513 writeback
= zone_page_state_snapshot(zone
,
3515 dirty
= zone_page_state_snapshot(zone
, NR_FILE_DIRTY
);
3517 if (2*(writeback
+ dirty
) > reclaimable
) {
3518 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3524 * Memory allocation/reclaim might be called from a WQ
3525 * context and the current implementation of the WQ
3526 * concurrency control doesn't recognize that
3527 * a particular WQ is congested if the worker thread is
3528 * looping without ever sleeping. Therefore we have to
3529 * do a short sleep here rather than calling
3532 if (current
->flags
& PF_WQ_WORKER
)
3533 schedule_timeout_uninterruptible(1);
3544 static inline struct page
*
3545 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
3546 struct alloc_context
*ac
)
3548 bool can_direct_reclaim
= gfp_mask
& __GFP_DIRECT_RECLAIM
;
3549 struct page
*page
= NULL
;
3550 unsigned int alloc_flags
;
3551 unsigned long did_some_progress
;
3552 enum migrate_mode migration_mode
= MIGRATE_ASYNC
;
3553 enum compact_result compact_result
;
3554 int compaction_retries
= 0;
3555 int no_progress_loops
= 0;
3558 * In the slowpath, we sanity check order to avoid ever trying to
3559 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3560 * be using allocators in order of preference for an area that is
3563 if (order
>= MAX_ORDER
) {
3564 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
3569 * We also sanity check to catch abuse of atomic reserves being used by
3570 * callers that are not in atomic context.
3572 if (WARN_ON_ONCE((gfp_mask
& (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)) ==
3573 (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)))
3574 gfp_mask
&= ~__GFP_ATOMIC
;
3577 if (gfp_mask
& __GFP_KSWAPD_RECLAIM
)
3578 wake_all_kswapds(order
, ac
);
3581 * OK, we're below the kswapd watermark and have kicked background
3582 * reclaim. Now things get more complex, so set up alloc_flags according
3583 * to how we want to proceed.
3585 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
3588 * Reset the zonelist iterators if memory policies can be ignored.
3589 * These allocations are high priority and system rather than user
3592 if ((alloc_flags
& ALLOC_NO_WATERMARKS
) || !(alloc_flags
& ALLOC_CPUSET
)) {
3593 ac
->zonelist
= node_zonelist(numa_node_id(), gfp_mask
);
3594 ac
->preferred_zoneref
= first_zones_zonelist(ac
->zonelist
,
3595 ac
->high_zoneidx
, ac
->nodemask
);
3598 /* This is the last chance, in general, before the goto nopage. */
3599 page
= get_page_from_freelist(gfp_mask
, order
,
3600 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
3604 /* Allocate without watermarks if the context allows */
3605 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
3606 page
= get_page_from_freelist(gfp_mask
, order
,
3607 ALLOC_NO_WATERMARKS
, ac
);
3612 /* Caller is not willing to reclaim, we can't balance anything */
3613 if (!can_direct_reclaim
) {
3615 * All existing users of the __GFP_NOFAIL are blockable, so warn
3616 * of any new users that actually allow this type of allocation
3619 WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
);
3623 /* Avoid recursion of direct reclaim */
3624 if (current
->flags
& PF_MEMALLOC
) {
3626 * __GFP_NOFAIL request from this context is rather bizarre
3627 * because we cannot reclaim anything and only can loop waiting
3628 * for somebody to do a work for us.
3630 if (WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
)) {
3637 /* Avoid allocations with no watermarks from looping endlessly */
3638 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
3642 * Try direct compaction. The first pass is asynchronous. Subsequent
3643 * attempts after direct reclaim are synchronous
3645 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
, ac
,
3651 /* Checks for THP-specific high-order allocations */
3652 if (is_thp_gfp_mask(gfp_mask
)) {
3654 * If compaction is deferred for high-order allocations, it is
3655 * because sync compaction recently failed. If this is the case
3656 * and the caller requested a THP allocation, we do not want
3657 * to heavily disrupt the system, so we fail the allocation
3658 * instead of entering direct reclaim.
3660 if (compact_result
== COMPACT_DEFERRED
)
3664 * Compaction is contended so rather back off than cause
3667 if(compact_result
== COMPACT_CONTENDED
)
3671 if (order
&& compaction_made_progress(compact_result
))
3672 compaction_retries
++;
3674 /* Try direct reclaim and then allocating */
3675 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
, alloc_flags
, ac
,
3676 &did_some_progress
);
3680 /* Do not loop if specifically requested */
3681 if (gfp_mask
& __GFP_NORETRY
)
3685 * Do not retry costly high order allocations unless they are
3688 if (order
> PAGE_ALLOC_COSTLY_ORDER
&& !(gfp_mask
& __GFP_REPEAT
))
3692 * Costly allocations might have made a progress but this doesn't mean
3693 * their order will become available due to high fragmentation so
3694 * always increment the no progress counter for them
3696 if (did_some_progress
&& order
<= PAGE_ALLOC_COSTLY_ORDER
)
3697 no_progress_loops
= 0;
3699 no_progress_loops
++;
3701 if (should_reclaim_retry(gfp_mask
, order
, ac
, alloc_flags
,
3702 did_some_progress
> 0, no_progress_loops
))
3706 * It doesn't make any sense to retry for the compaction if the order-0
3707 * reclaim is not able to make any progress because the current
3708 * implementation of the compaction depends on the sufficient amount
3709 * of free memory (see __compaction_suitable)
3711 if (did_some_progress
> 0 &&
3712 should_compact_retry(ac
, order
, alloc_flags
,
3713 compact_result
, &migration_mode
,
3714 compaction_retries
))
3717 /* Reclaim has failed us, start killing things */
3718 page
= __alloc_pages_may_oom(gfp_mask
, order
, ac
, &did_some_progress
);
3722 /* Retry as long as the OOM killer is making progress */
3723 if (did_some_progress
) {
3724 no_progress_loops
= 0;
3730 * High-order allocations do not necessarily loop after direct reclaim
3731 * and reclaim/compaction depends on compaction being called after
3732 * reclaim so call directly if necessary.
3733 * It can become very expensive to allocate transparent hugepages at
3734 * fault, so use asynchronous memory compaction for THP unless it is
3735 * khugepaged trying to collapse. All other requests should tolerate
3736 * at least light sync migration.
3738 if (is_thp_gfp_mask(gfp_mask
) && !(current
->flags
& PF_KTHREAD
))
3739 migration_mode
= MIGRATE_ASYNC
;
3741 migration_mode
= MIGRATE_SYNC_LIGHT
;
3742 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
,
3748 warn_alloc_failed(gfp_mask
, order
, NULL
);
3754 * This is the 'heart' of the zoned buddy allocator.
3757 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
3758 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
3761 unsigned int cpuset_mems_cookie
;
3762 unsigned int alloc_flags
= ALLOC_WMARK_LOW
|ALLOC_FAIR
;
3763 gfp_t alloc_mask
= gfp_mask
; /* The gfp_t that was actually used for allocation */
3764 struct alloc_context ac
= {
3765 .high_zoneidx
= gfp_zone(gfp_mask
),
3766 .zonelist
= zonelist
,
3767 .nodemask
= nodemask
,
3768 .migratetype
= gfpflags_to_migratetype(gfp_mask
),
3771 if (cpusets_enabled()) {
3772 alloc_mask
|= __GFP_HARDWALL
;
3773 alloc_flags
|= ALLOC_CPUSET
;
3775 ac
.nodemask
= &cpuset_current_mems_allowed
;
3778 gfp_mask
&= gfp_allowed_mask
;
3780 lockdep_trace_alloc(gfp_mask
);
3782 might_sleep_if(gfp_mask
& __GFP_DIRECT_RECLAIM
);
3784 if (should_fail_alloc_page(gfp_mask
, order
))
3788 * Check the zones suitable for the gfp_mask contain at least one
3789 * valid zone. It's possible to have an empty zonelist as a result
3790 * of __GFP_THISNODE and a memoryless node
3792 if (unlikely(!zonelist
->_zonerefs
->zone
))
3795 if (IS_ENABLED(CONFIG_CMA
) && ac
.migratetype
== MIGRATE_MOVABLE
)
3796 alloc_flags
|= ALLOC_CMA
;
3799 cpuset_mems_cookie
= read_mems_allowed_begin();
3801 /* Dirty zone balancing only done in the fast path */
3802 ac
.spread_dirty_pages
= (gfp_mask
& __GFP_WRITE
);
3805 * The preferred zone is used for statistics but crucially it is
3806 * also used as the starting point for the zonelist iterator. It
3807 * may get reset for allocations that ignore memory policies.
3809 ac
.preferred_zoneref
= first_zones_zonelist(ac
.zonelist
,
3810 ac
.high_zoneidx
, ac
.nodemask
);
3811 if (!ac
.preferred_zoneref
) {
3816 /* First allocation attempt */
3817 page
= get_page_from_freelist(alloc_mask
, order
, alloc_flags
, &ac
);
3822 * Runtime PM, block IO and its error handling path can deadlock
3823 * because I/O on the device might not complete.
3825 alloc_mask
= memalloc_noio_flags(gfp_mask
);
3826 ac
.spread_dirty_pages
= false;
3829 * Restore the original nodemask if it was potentially replaced with
3830 * &cpuset_current_mems_allowed to optimize the fast-path attempt.
3832 if (cpusets_enabled())
3833 ac
.nodemask
= nodemask
;
3834 page
= __alloc_pages_slowpath(alloc_mask
, order
, &ac
);
3838 * When updating a task's mems_allowed, it is possible to race with
3839 * parallel threads in such a way that an allocation can fail while
3840 * the mask is being updated. If a page allocation is about to fail,
3841 * check if the cpuset changed during allocation and if so, retry.
3843 if (unlikely(!page
&& read_mems_allowed_retry(cpuset_mems_cookie
))) {
3844 alloc_mask
= gfp_mask
;
3849 if (memcg_kmem_enabled() && (gfp_mask
& __GFP_ACCOUNT
) && page
) {
3850 if (unlikely(memcg_kmem_charge(page
, gfp_mask
, order
))) {
3851 __free_pages(page
, order
);
3854 __SetPageKmemcg(page
);
3857 if (kmemcheck_enabled
&& page
)
3858 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
3860 trace_mm_page_alloc(page
, order
, alloc_mask
, ac
.migratetype
);
3864 EXPORT_SYMBOL(__alloc_pages_nodemask
);
3867 * Common helper functions.
3869 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
3874 * __get_free_pages() returns a 32-bit address, which cannot represent
3877 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
3879 page
= alloc_pages(gfp_mask
, order
);
3882 return (unsigned long) page_address(page
);
3884 EXPORT_SYMBOL(__get_free_pages
);
3886 unsigned long get_zeroed_page(gfp_t gfp_mask
)
3888 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
3890 EXPORT_SYMBOL(get_zeroed_page
);
3892 void __free_pages(struct page
*page
, unsigned int order
)
3894 if (put_page_testzero(page
)) {
3896 free_hot_cold_page(page
, false);
3898 __free_pages_ok(page
, order
);
3902 EXPORT_SYMBOL(__free_pages
);
3904 void free_pages(unsigned long addr
, unsigned int order
)
3907 VM_BUG_ON(!virt_addr_valid((void *)addr
));
3908 __free_pages(virt_to_page((void *)addr
), order
);
3912 EXPORT_SYMBOL(free_pages
);
3916 * An arbitrary-length arbitrary-offset area of memory which resides
3917 * within a 0 or higher order page. Multiple fragments within that page
3918 * are individually refcounted, in the page's reference counter.
3920 * The page_frag functions below provide a simple allocation framework for
3921 * page fragments. This is used by the network stack and network device
3922 * drivers to provide a backing region of memory for use as either an
3923 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3925 static struct page
*__page_frag_refill(struct page_frag_cache
*nc
,
3928 struct page
*page
= NULL
;
3929 gfp_t gfp
= gfp_mask
;
3931 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3932 gfp_mask
|= __GFP_COMP
| __GFP_NOWARN
| __GFP_NORETRY
|
3934 page
= alloc_pages_node(NUMA_NO_NODE
, gfp_mask
,
3935 PAGE_FRAG_CACHE_MAX_ORDER
);
3936 nc
->size
= page
? PAGE_FRAG_CACHE_MAX_SIZE
: PAGE_SIZE
;
3938 if (unlikely(!page
))
3939 page
= alloc_pages_node(NUMA_NO_NODE
, gfp
, 0);
3941 nc
->va
= page
? page_address(page
) : NULL
;
3946 void *__alloc_page_frag(struct page_frag_cache
*nc
,
3947 unsigned int fragsz
, gfp_t gfp_mask
)
3949 unsigned int size
= PAGE_SIZE
;
3953 if (unlikely(!nc
->va
)) {
3955 page
= __page_frag_refill(nc
, gfp_mask
);
3959 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3960 /* if size can vary use size else just use PAGE_SIZE */
3963 /* Even if we own the page, we do not use atomic_set().
3964 * This would break get_page_unless_zero() users.
3966 page_ref_add(page
, size
- 1);
3968 /* reset page count bias and offset to start of new frag */
3969 nc
->pfmemalloc
= page_is_pfmemalloc(page
);
3970 nc
->pagecnt_bias
= size
;
3974 offset
= nc
->offset
- fragsz
;
3975 if (unlikely(offset
< 0)) {
3976 page
= virt_to_page(nc
->va
);
3978 if (!page_ref_sub_and_test(page
, nc
->pagecnt_bias
))
3981 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3982 /* if size can vary use size else just use PAGE_SIZE */
3985 /* OK, page count is 0, we can safely set it */
3986 set_page_count(page
, size
);
3988 /* reset page count bias and offset to start of new frag */
3989 nc
->pagecnt_bias
= size
;
3990 offset
= size
- fragsz
;
3994 nc
->offset
= offset
;
3996 return nc
->va
+ offset
;
3998 EXPORT_SYMBOL(__alloc_page_frag
);
4001 * Frees a page fragment allocated out of either a compound or order 0 page.
4003 void __free_page_frag(void *addr
)
4005 struct page
*page
= virt_to_head_page(addr
);
4007 if (unlikely(put_page_testzero(page
)))
4008 __free_pages_ok(page
, compound_order(page
));
4010 EXPORT_SYMBOL(__free_page_frag
);
4012 static void *make_alloc_exact(unsigned long addr
, unsigned int order
,
4016 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
4017 unsigned long used
= addr
+ PAGE_ALIGN(size
);
4019 split_page(virt_to_page((void *)addr
), order
);
4020 while (used
< alloc_end
) {
4025 return (void *)addr
;
4029 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
4030 * @size: the number of bytes to allocate
4031 * @gfp_mask: GFP flags for the allocation
4033 * This function is similar to alloc_pages(), except that it allocates the
4034 * minimum number of pages to satisfy the request. alloc_pages() can only
4035 * allocate memory in power-of-two pages.
4037 * This function is also limited by MAX_ORDER.
4039 * Memory allocated by this function must be released by free_pages_exact().
4041 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
4043 unsigned int order
= get_order(size
);
4046 addr
= __get_free_pages(gfp_mask
, order
);
4047 return make_alloc_exact(addr
, order
, size
);
4049 EXPORT_SYMBOL(alloc_pages_exact
);
4052 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
4054 * @nid: the preferred node ID where memory should be allocated
4055 * @size: the number of bytes to allocate
4056 * @gfp_mask: GFP flags for the allocation
4058 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
4061 void * __meminit
alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
4063 unsigned int order
= get_order(size
);
4064 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
4067 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
4071 * free_pages_exact - release memory allocated via alloc_pages_exact()
4072 * @virt: the value returned by alloc_pages_exact.
4073 * @size: size of allocation, same value as passed to alloc_pages_exact().
4075 * Release the memory allocated by a previous call to alloc_pages_exact.
4077 void free_pages_exact(void *virt
, size_t size
)
4079 unsigned long addr
= (unsigned long)virt
;
4080 unsigned long end
= addr
+ PAGE_ALIGN(size
);
4082 while (addr
< end
) {
4087 EXPORT_SYMBOL(free_pages_exact
);
4090 * nr_free_zone_pages - count number of pages beyond high watermark
4091 * @offset: The zone index of the highest zone
4093 * nr_free_zone_pages() counts the number of counts pages which are beyond the
4094 * high watermark within all zones at or below a given zone index. For each
4095 * zone, the number of pages is calculated as:
4096 * managed_pages - high_pages
4098 static unsigned long nr_free_zone_pages(int offset
)
4103 /* Just pick one node, since fallback list is circular */
4104 unsigned long sum
= 0;
4106 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
4108 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
4109 unsigned long size
= zone
->managed_pages
;
4110 unsigned long high
= high_wmark_pages(zone
);
4119 * nr_free_buffer_pages - count number of pages beyond high watermark
4121 * nr_free_buffer_pages() counts the number of pages which are beyond the high
4122 * watermark within ZONE_DMA and ZONE_NORMAL.
4124 unsigned long nr_free_buffer_pages(void)
4126 return nr_free_zone_pages(gfp_zone(GFP_USER
));
4128 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
4131 * nr_free_pagecache_pages - count number of pages beyond high watermark
4133 * nr_free_pagecache_pages() counts the number of pages which are beyond the
4134 * high watermark within all zones.
4136 unsigned long nr_free_pagecache_pages(void)
4138 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
4141 static inline void show_node(struct zone
*zone
)
4143 if (IS_ENABLED(CONFIG_NUMA
))
4144 printk("Node %d ", zone_to_nid(zone
));
4147 long si_mem_available(void)
4150 unsigned long pagecache
;
4151 unsigned long wmark_low
= 0;
4152 unsigned long pages
[NR_LRU_LISTS
];
4156 for (lru
= LRU_BASE
; lru
< NR_LRU_LISTS
; lru
++)
4157 pages
[lru
] = global_page_state(NR_LRU_BASE
+ lru
);
4160 wmark_low
+= zone
->watermark
[WMARK_LOW
];
4163 * Estimate the amount of memory available for userspace allocations,
4164 * without causing swapping.
4166 available
= global_page_state(NR_FREE_PAGES
) - totalreserve_pages
;
4169 * Not all the page cache can be freed, otherwise the system will
4170 * start swapping. Assume at least half of the page cache, or the
4171 * low watermark worth of cache, needs to stay.
4173 pagecache
= pages
[LRU_ACTIVE_FILE
] + pages
[LRU_INACTIVE_FILE
];
4174 pagecache
-= min(pagecache
/ 2, wmark_low
);
4175 available
+= pagecache
;
4178 * Part of the reclaimable slab consists of items that are in use,
4179 * and cannot be freed. Cap this estimate at the low watermark.
4181 available
+= global_page_state(NR_SLAB_RECLAIMABLE
) -
4182 min(global_page_state(NR_SLAB_RECLAIMABLE
) / 2, wmark_low
);
4188 EXPORT_SYMBOL_GPL(si_mem_available
);
4190 void si_meminfo(struct sysinfo
*val
)
4192 val
->totalram
= totalram_pages
;
4193 val
->sharedram
= global_page_state(NR_SHMEM
);
4194 val
->freeram
= global_page_state(NR_FREE_PAGES
);
4195 val
->bufferram
= nr_blockdev_pages();
4196 val
->totalhigh
= totalhigh_pages
;
4197 val
->freehigh
= nr_free_highpages();
4198 val
->mem_unit
= PAGE_SIZE
;
4201 EXPORT_SYMBOL(si_meminfo
);
4204 void si_meminfo_node(struct sysinfo
*val
, int nid
)
4206 int zone_type
; /* needs to be signed */
4207 unsigned long managed_pages
= 0;
4208 unsigned long managed_highpages
= 0;
4209 unsigned long free_highpages
= 0;
4210 pg_data_t
*pgdat
= NODE_DATA(nid
);
4212 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++)
4213 managed_pages
+= pgdat
->node_zones
[zone_type
].managed_pages
;
4214 val
->totalram
= managed_pages
;
4215 val
->sharedram
= node_page_state(nid
, NR_SHMEM
);
4216 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
4217 #ifdef CONFIG_HIGHMEM
4218 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
4219 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4221 if (is_highmem(zone
)) {
4222 managed_highpages
+= zone
->managed_pages
;
4223 free_highpages
+= zone_page_state(zone
, NR_FREE_PAGES
);
4226 val
->totalhigh
= managed_highpages
;
4227 val
->freehigh
= free_highpages
;
4229 val
->totalhigh
= managed_highpages
;
4230 val
->freehigh
= free_highpages
;
4232 val
->mem_unit
= PAGE_SIZE
;
4237 * Determine whether the node should be displayed or not, depending on whether
4238 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
4240 bool skip_free_areas_node(unsigned int flags
, int nid
)
4243 unsigned int cpuset_mems_cookie
;
4245 if (!(flags
& SHOW_MEM_FILTER_NODES
))
4249 cpuset_mems_cookie
= read_mems_allowed_begin();
4250 ret
= !node_isset(nid
, cpuset_current_mems_allowed
);
4251 } while (read_mems_allowed_retry(cpuset_mems_cookie
));
4256 #define K(x) ((x) << (PAGE_SHIFT-10))
4258 static void show_migration_types(unsigned char type
)
4260 static const char types
[MIGRATE_TYPES
] = {
4261 [MIGRATE_UNMOVABLE
] = 'U',
4262 [MIGRATE_MOVABLE
] = 'M',
4263 [MIGRATE_RECLAIMABLE
] = 'E',
4264 [MIGRATE_HIGHATOMIC
] = 'H',
4266 [MIGRATE_CMA
] = 'C',
4268 #ifdef CONFIG_MEMORY_ISOLATION
4269 [MIGRATE_ISOLATE
] = 'I',
4272 char tmp
[MIGRATE_TYPES
+ 1];
4276 for (i
= 0; i
< MIGRATE_TYPES
; i
++) {
4277 if (type
& (1 << i
))
4282 printk("(%s) ", tmp
);
4286 * Show free area list (used inside shift_scroll-lock stuff)
4287 * We also calculate the percentage fragmentation. We do this by counting the
4288 * memory on each free list with the exception of the first item on the list.
4291 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
4294 void show_free_areas(unsigned int filter
)
4296 unsigned long free_pcp
= 0;
4300 for_each_populated_zone(zone
) {
4301 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
4304 for_each_online_cpu(cpu
)
4305 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
4308 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
4309 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
4310 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
4311 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
4312 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
4313 " free:%lu free_pcp:%lu free_cma:%lu\n",
4314 global_page_state(NR_ACTIVE_ANON
),
4315 global_page_state(NR_INACTIVE_ANON
),
4316 global_page_state(NR_ISOLATED_ANON
),
4317 global_page_state(NR_ACTIVE_FILE
),
4318 global_page_state(NR_INACTIVE_FILE
),
4319 global_page_state(NR_ISOLATED_FILE
),
4320 global_page_state(NR_UNEVICTABLE
),
4321 global_page_state(NR_FILE_DIRTY
),
4322 global_page_state(NR_WRITEBACK
),
4323 global_page_state(NR_UNSTABLE_NFS
),
4324 global_page_state(NR_SLAB_RECLAIMABLE
),
4325 global_page_state(NR_SLAB_UNRECLAIMABLE
),
4326 global_page_state(NR_FILE_MAPPED
),
4327 global_page_state(NR_SHMEM
),
4328 global_page_state(NR_PAGETABLE
),
4329 global_page_state(NR_BOUNCE
),
4330 global_page_state(NR_FREE_PAGES
),
4332 global_page_state(NR_FREE_CMA_PAGES
));
4334 for_each_populated_zone(zone
) {
4337 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
4341 for_each_online_cpu(cpu
)
4342 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
4350 " active_anon:%lukB"
4351 " inactive_anon:%lukB"
4352 " active_file:%lukB"
4353 " inactive_file:%lukB"
4354 " unevictable:%lukB"
4355 " isolated(anon):%lukB"
4356 " isolated(file):%lukB"
4364 " slab_reclaimable:%lukB"
4365 " slab_unreclaimable:%lukB"
4366 " kernel_stack:%lukB"
4373 " writeback_tmp:%lukB"
4374 " pages_scanned:%lu"
4375 " all_unreclaimable? %s"
4378 K(zone_page_state(zone
, NR_FREE_PAGES
)),
4379 K(min_wmark_pages(zone
)),
4380 K(low_wmark_pages(zone
)),
4381 K(high_wmark_pages(zone
)),
4382 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
4383 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
4384 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
4385 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
4386 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
4387 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
4388 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
4389 K(zone
->present_pages
),
4390 K(zone
->managed_pages
),
4391 K(zone_page_state(zone
, NR_MLOCK
)),
4392 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
4393 K(zone_page_state(zone
, NR_WRITEBACK
)),
4394 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
4395 K(zone_page_state(zone
, NR_SHMEM
)),
4396 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
4397 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
4398 zone_page_state(zone
, NR_KERNEL_STACK
) *
4400 K(zone_page_state(zone
, NR_PAGETABLE
)),
4401 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
4402 K(zone_page_state(zone
, NR_BOUNCE
)),
4404 K(this_cpu_read(zone
->pageset
->pcp
.count
)),
4405 K(zone_page_state(zone
, NR_FREE_CMA_PAGES
)),
4406 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
4407 K(zone_page_state(zone
, NR_PAGES_SCANNED
)),
4408 (!zone_reclaimable(zone
) ? "yes" : "no")
4410 printk("lowmem_reserve[]:");
4411 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4412 printk(" %ld", zone
->lowmem_reserve
[i
]);
4416 for_each_populated_zone(zone
) {
4418 unsigned long nr
[MAX_ORDER
], flags
, total
= 0;
4419 unsigned char types
[MAX_ORDER
];
4421 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
4424 printk("%s: ", zone
->name
);
4426 spin_lock_irqsave(&zone
->lock
, flags
);
4427 for (order
= 0; order
< MAX_ORDER
; order
++) {
4428 struct free_area
*area
= &zone
->free_area
[order
];
4431 nr
[order
] = area
->nr_free
;
4432 total
+= nr
[order
] << order
;
4435 for (type
= 0; type
< MIGRATE_TYPES
; type
++) {
4436 if (!list_empty(&area
->free_list
[type
]))
4437 types
[order
] |= 1 << type
;
4440 spin_unlock_irqrestore(&zone
->lock
, flags
);
4441 for (order
= 0; order
< MAX_ORDER
; order
++) {
4442 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
4444 show_migration_types(types
[order
]);
4446 printk("= %lukB\n", K(total
));
4449 hugetlb_show_meminfo();
4451 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
4453 show_swap_cache_info();
4456 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
4458 zoneref
->zone
= zone
;
4459 zoneref
->zone_idx
= zone_idx(zone
);
4463 * Builds allocation fallback zone lists.
4465 * Add all populated zones of a node to the zonelist.
4467 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
4471 enum zone_type zone_type
= MAX_NR_ZONES
;
4475 zone
= pgdat
->node_zones
+ zone_type
;
4476 if (populated_zone(zone
)) {
4477 zoneref_set_zone(zone
,
4478 &zonelist
->_zonerefs
[nr_zones
++]);
4479 check_highest_zone(zone_type
);
4481 } while (zone_type
);
4489 * 0 = automatic detection of better ordering.
4490 * 1 = order by ([node] distance, -zonetype)
4491 * 2 = order by (-zonetype, [node] distance)
4493 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
4494 * the same zonelist. So only NUMA can configure this param.
4496 #define ZONELIST_ORDER_DEFAULT 0
4497 #define ZONELIST_ORDER_NODE 1
4498 #define ZONELIST_ORDER_ZONE 2
4500 /* zonelist order in the kernel.
4501 * set_zonelist_order() will set this to NODE or ZONE.
4503 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4504 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
4508 /* The value user specified ....changed by config */
4509 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4510 /* string for sysctl */
4511 #define NUMA_ZONELIST_ORDER_LEN 16
4512 char numa_zonelist_order
[16] = "default";
4515 * interface for configure zonelist ordering.
4516 * command line option "numa_zonelist_order"
4517 * = "[dD]efault - default, automatic configuration.
4518 * = "[nN]ode - order by node locality, then by zone within node
4519 * = "[zZ]one - order by zone, then by locality within zone
4522 static int __parse_numa_zonelist_order(char *s
)
4524 if (*s
== 'd' || *s
== 'D') {
4525 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4526 } else if (*s
== 'n' || *s
== 'N') {
4527 user_zonelist_order
= ZONELIST_ORDER_NODE
;
4528 } else if (*s
== 'z' || *s
== 'Z') {
4529 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
4531 pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s
);
4537 static __init
int setup_numa_zonelist_order(char *s
)
4544 ret
= __parse_numa_zonelist_order(s
);
4546 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
4550 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
4553 * sysctl handler for numa_zonelist_order
4555 int numa_zonelist_order_handler(struct ctl_table
*table
, int write
,
4556 void __user
*buffer
, size_t *length
,
4559 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
4561 static DEFINE_MUTEX(zl_order_mutex
);
4563 mutex_lock(&zl_order_mutex
);
4565 if (strlen((char *)table
->data
) >= NUMA_ZONELIST_ORDER_LEN
) {
4569 strcpy(saved_string
, (char *)table
->data
);
4571 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
4575 int oldval
= user_zonelist_order
;
4577 ret
= __parse_numa_zonelist_order((char *)table
->data
);
4580 * bogus value. restore saved string
4582 strncpy((char *)table
->data
, saved_string
,
4583 NUMA_ZONELIST_ORDER_LEN
);
4584 user_zonelist_order
= oldval
;
4585 } else if (oldval
!= user_zonelist_order
) {
4586 mutex_lock(&zonelists_mutex
);
4587 build_all_zonelists(NULL
, NULL
);
4588 mutex_unlock(&zonelists_mutex
);
4592 mutex_unlock(&zl_order_mutex
);
4597 #define MAX_NODE_LOAD (nr_online_nodes)
4598 static int node_load
[MAX_NUMNODES
];
4601 * find_next_best_node - find the next node that should appear in a given node's fallback list
4602 * @node: node whose fallback list we're appending
4603 * @used_node_mask: nodemask_t of already used nodes
4605 * We use a number of factors to determine which is the next node that should
4606 * appear on a given node's fallback list. The node should not have appeared
4607 * already in @node's fallback list, and it should be the next closest node
4608 * according to the distance array (which contains arbitrary distance values
4609 * from each node to each node in the system), and should also prefer nodes
4610 * with no CPUs, since presumably they'll have very little allocation pressure
4611 * on them otherwise.
4612 * It returns -1 if no node is found.
4614 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
4617 int min_val
= INT_MAX
;
4618 int best_node
= NUMA_NO_NODE
;
4619 const struct cpumask
*tmp
= cpumask_of_node(0);
4621 /* Use the local node if we haven't already */
4622 if (!node_isset(node
, *used_node_mask
)) {
4623 node_set(node
, *used_node_mask
);
4627 for_each_node_state(n
, N_MEMORY
) {
4629 /* Don't want a node to appear more than once */
4630 if (node_isset(n
, *used_node_mask
))
4633 /* Use the distance array to find the distance */
4634 val
= node_distance(node
, n
);
4636 /* Penalize nodes under us ("prefer the next node") */
4639 /* Give preference to headless and unused nodes */
4640 tmp
= cpumask_of_node(n
);
4641 if (!cpumask_empty(tmp
))
4642 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
4644 /* Slight preference for less loaded node */
4645 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
4646 val
+= node_load
[n
];
4648 if (val
< min_val
) {
4655 node_set(best_node
, *used_node_mask
);
4662 * Build zonelists ordered by node and zones within node.
4663 * This results in maximum locality--normal zone overflows into local
4664 * DMA zone, if any--but risks exhausting DMA zone.
4666 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
4669 struct zonelist
*zonelist
;
4671 zonelist
= &pgdat
->node_zonelists
[0];
4672 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
4674 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4675 zonelist
->_zonerefs
[j
].zone
= NULL
;
4676 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4680 * Build gfp_thisnode zonelists
4682 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
4685 struct zonelist
*zonelist
;
4687 zonelist
= &pgdat
->node_zonelists
[1];
4688 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4689 zonelist
->_zonerefs
[j
].zone
= NULL
;
4690 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4694 * Build zonelists ordered by zone and nodes within zones.
4695 * This results in conserving DMA zone[s] until all Normal memory is
4696 * exhausted, but results in overflowing to remote node while memory
4697 * may still exist in local DMA zone.
4699 static int node_order
[MAX_NUMNODES
];
4701 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
4704 int zone_type
; /* needs to be signed */
4706 struct zonelist
*zonelist
;
4708 zonelist
= &pgdat
->node_zonelists
[0];
4710 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
4711 for (j
= 0; j
< nr_nodes
; j
++) {
4712 node
= node_order
[j
];
4713 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
4714 if (populated_zone(z
)) {
4716 &zonelist
->_zonerefs
[pos
++]);
4717 check_highest_zone(zone_type
);
4721 zonelist
->_zonerefs
[pos
].zone
= NULL
;
4722 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
4725 #if defined(CONFIG_64BIT)
4727 * Devices that require DMA32/DMA are relatively rare and do not justify a
4728 * penalty to every machine in case the specialised case applies. Default
4729 * to Node-ordering on 64-bit NUMA machines
4731 static int default_zonelist_order(void)
4733 return ZONELIST_ORDER_NODE
;
4737 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4738 * by the kernel. If processes running on node 0 deplete the low memory zone
4739 * then reclaim will occur more frequency increasing stalls and potentially
4740 * be easier to OOM if a large percentage of the zone is under writeback or
4741 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4742 * Hence, default to zone ordering on 32-bit.
4744 static int default_zonelist_order(void)
4746 return ZONELIST_ORDER_ZONE
;
4748 #endif /* CONFIG_64BIT */
4750 static void set_zonelist_order(void)
4752 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
4753 current_zonelist_order
= default_zonelist_order();
4755 current_zonelist_order
= user_zonelist_order
;
4758 static void build_zonelists(pg_data_t
*pgdat
)
4761 nodemask_t used_mask
;
4762 int local_node
, prev_node
;
4763 struct zonelist
*zonelist
;
4764 unsigned int order
= current_zonelist_order
;
4766 /* initialize zonelists */
4767 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
4768 zonelist
= pgdat
->node_zonelists
+ i
;
4769 zonelist
->_zonerefs
[0].zone
= NULL
;
4770 zonelist
->_zonerefs
[0].zone_idx
= 0;
4773 /* NUMA-aware ordering of nodes */
4774 local_node
= pgdat
->node_id
;
4775 load
= nr_online_nodes
;
4776 prev_node
= local_node
;
4777 nodes_clear(used_mask
);
4779 memset(node_order
, 0, sizeof(node_order
));
4782 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
4784 * We don't want to pressure a particular node.
4785 * So adding penalty to the first node in same
4786 * distance group to make it round-robin.
4788 if (node_distance(local_node
, node
) !=
4789 node_distance(local_node
, prev_node
))
4790 node_load
[node
] = load
;
4794 if (order
== ZONELIST_ORDER_NODE
)
4795 build_zonelists_in_node_order(pgdat
, node
);
4797 node_order
[i
++] = node
; /* remember order */
4800 if (order
== ZONELIST_ORDER_ZONE
) {
4801 /* calculate node order -- i.e., DMA last! */
4802 build_zonelists_in_zone_order(pgdat
, i
);
4805 build_thisnode_zonelists(pgdat
);
4808 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4810 * Return node id of node used for "local" allocations.
4811 * I.e., first node id of first zone in arg node's generic zonelist.
4812 * Used for initializing percpu 'numa_mem', which is used primarily
4813 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4815 int local_memory_node(int node
)
4819 z
= first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
4820 gfp_zone(GFP_KERNEL
),
4822 return z
->zone
->node
;
4826 #else /* CONFIG_NUMA */
4828 static void set_zonelist_order(void)
4830 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
4833 static void build_zonelists(pg_data_t
*pgdat
)
4835 int node
, local_node
;
4837 struct zonelist
*zonelist
;
4839 local_node
= pgdat
->node_id
;
4841 zonelist
= &pgdat
->node_zonelists
[0];
4842 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4845 * Now we build the zonelist so that it contains the zones
4846 * of all the other nodes.
4847 * We don't want to pressure a particular node, so when
4848 * building the zones for node N, we make sure that the
4849 * zones coming right after the local ones are those from
4850 * node N+1 (modulo N)
4852 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
4853 if (!node_online(node
))
4855 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4857 for (node
= 0; node
< local_node
; node
++) {
4858 if (!node_online(node
))
4860 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4863 zonelist
->_zonerefs
[j
].zone
= NULL
;
4864 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4867 #endif /* CONFIG_NUMA */
4870 * Boot pageset table. One per cpu which is going to be used for all
4871 * zones and all nodes. The parameters will be set in such a way
4872 * that an item put on a list will immediately be handed over to
4873 * the buddy list. This is safe since pageset manipulation is done
4874 * with interrupts disabled.
4876 * The boot_pagesets must be kept even after bootup is complete for
4877 * unused processors and/or zones. They do play a role for bootstrapping
4878 * hotplugged processors.
4880 * zoneinfo_show() and maybe other functions do
4881 * not check if the processor is online before following the pageset pointer.
4882 * Other parts of the kernel may not check if the zone is available.
4884 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
4885 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
4886 static void setup_zone_pageset(struct zone
*zone
);
4889 * Global mutex to protect against size modification of zonelists
4890 * as well as to serialize pageset setup for the new populated zone.
4892 DEFINE_MUTEX(zonelists_mutex
);
4894 /* return values int ....just for stop_machine() */
4895 static int __build_all_zonelists(void *data
)
4899 pg_data_t
*self
= data
;
4902 memset(node_load
, 0, sizeof(node_load
));
4905 if (self
&& !node_online(self
->node_id
)) {
4906 build_zonelists(self
);
4909 for_each_online_node(nid
) {
4910 pg_data_t
*pgdat
= NODE_DATA(nid
);
4912 build_zonelists(pgdat
);
4916 * Initialize the boot_pagesets that are going to be used
4917 * for bootstrapping processors. The real pagesets for
4918 * each zone will be allocated later when the per cpu
4919 * allocator is available.
4921 * boot_pagesets are used also for bootstrapping offline
4922 * cpus if the system is already booted because the pagesets
4923 * are needed to initialize allocators on a specific cpu too.
4924 * F.e. the percpu allocator needs the page allocator which
4925 * needs the percpu allocator in order to allocate its pagesets
4926 * (a chicken-egg dilemma).
4928 for_each_possible_cpu(cpu
) {
4929 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
4931 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4933 * We now know the "local memory node" for each node--
4934 * i.e., the node of the first zone in the generic zonelist.
4935 * Set up numa_mem percpu variable for on-line cpus. During
4936 * boot, only the boot cpu should be on-line; we'll init the
4937 * secondary cpus' numa_mem as they come on-line. During
4938 * node/memory hotplug, we'll fixup all on-line cpus.
4940 if (cpu_online(cpu
))
4941 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
4948 static noinline
void __init
4949 build_all_zonelists_init(void)
4951 __build_all_zonelists(NULL
);
4952 mminit_verify_zonelist();
4953 cpuset_init_current_mems_allowed();
4957 * Called with zonelists_mutex held always
4958 * unless system_state == SYSTEM_BOOTING.
4960 * __ref due to (1) call of __meminit annotated setup_zone_pageset
4961 * [we're only called with non-NULL zone through __meminit paths] and
4962 * (2) call of __init annotated helper build_all_zonelists_init
4963 * [protected by SYSTEM_BOOTING].
4965 void __ref
build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
)
4967 set_zonelist_order();
4969 if (system_state
== SYSTEM_BOOTING
) {
4970 build_all_zonelists_init();
4972 #ifdef CONFIG_MEMORY_HOTPLUG
4974 setup_zone_pageset(zone
);
4976 /* we have to stop all cpus to guarantee there is no user
4978 stop_machine(__build_all_zonelists
, pgdat
, NULL
);
4979 /* cpuset refresh routine should be here */
4981 vm_total_pages
= nr_free_pagecache_pages();
4983 * Disable grouping by mobility if the number of pages in the
4984 * system is too low to allow the mechanism to work. It would be
4985 * more accurate, but expensive to check per-zone. This check is
4986 * made on memory-hotadd so a system can start with mobility
4987 * disabled and enable it later
4989 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
4990 page_group_by_mobility_disabled
= 1;
4992 page_group_by_mobility_disabled
= 0;
4994 pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n",
4996 zonelist_order_name
[current_zonelist_order
],
4997 page_group_by_mobility_disabled
? "off" : "on",
5000 pr_info("Policy zone: %s\n", zone_names
[policy_zone
]);
5005 * Helper functions to size the waitqueue hash table.
5006 * Essentially these want to choose hash table sizes sufficiently
5007 * large so that collisions trying to wait on pages are rare.
5008 * But in fact, the number of active page waitqueues on typical
5009 * systems is ridiculously low, less than 200. So this is even
5010 * conservative, even though it seems large.
5012 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
5013 * waitqueues, i.e. the size of the waitq table given the number of pages.
5015 #define PAGES_PER_WAITQUEUE 256
5017 #ifndef CONFIG_MEMORY_HOTPLUG
5018 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
5020 unsigned long size
= 1;
5022 pages
/= PAGES_PER_WAITQUEUE
;
5024 while (size
< pages
)
5028 * Once we have dozens or even hundreds of threads sleeping
5029 * on IO we've got bigger problems than wait queue collision.
5030 * Limit the size of the wait table to a reasonable size.
5032 size
= min(size
, 4096UL);
5034 return max(size
, 4UL);
5038 * A zone's size might be changed by hot-add, so it is not possible to determine
5039 * a suitable size for its wait_table. So we use the maximum size now.
5041 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
5043 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
5044 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
5045 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
5047 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
5048 * or more by the traditional way. (See above). It equals:
5050 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
5051 * ia64(16K page size) : = ( 8G + 4M)byte.
5052 * powerpc (64K page size) : = (32G +16M)byte.
5054 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
5061 * This is an integer logarithm so that shifts can be used later
5062 * to extract the more random high bits from the multiplicative
5063 * hash function before the remainder is taken.
5065 static inline unsigned long wait_table_bits(unsigned long size
)
5071 * Initially all pages are reserved - free ones are freed
5072 * up by free_all_bootmem() once the early boot process is
5073 * done. Non-atomic initialization, single-pass.
5075 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
5076 unsigned long start_pfn
, enum memmap_context context
)
5078 struct vmem_altmap
*altmap
= to_vmem_altmap(__pfn_to_phys(start_pfn
));
5079 unsigned long end_pfn
= start_pfn
+ size
;
5080 pg_data_t
*pgdat
= NODE_DATA(nid
);
5082 unsigned long nr_initialised
= 0;
5083 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5084 struct memblock_region
*r
= NULL
, *tmp
;
5087 if (highest_memmap_pfn
< end_pfn
- 1)
5088 highest_memmap_pfn
= end_pfn
- 1;
5091 * Honor reservation requested by the driver for this ZONE_DEVICE
5094 if (altmap
&& start_pfn
== altmap
->base_pfn
)
5095 start_pfn
+= altmap
->reserve
;
5097 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
5099 * There can be holes in boot-time mem_map[]s handed to this
5100 * function. They do not exist on hotplugged memory.
5102 if (context
!= MEMMAP_EARLY
)
5105 if (!early_pfn_valid(pfn
))
5107 if (!early_pfn_in_nid(pfn
, nid
))
5109 if (!update_defer_init(pgdat
, pfn
, end_pfn
, &nr_initialised
))
5112 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5114 * If not mirrored_kernelcore and ZONE_MOVABLE exists, range
5115 * from zone_movable_pfn[nid] to end of each node should be
5116 * ZONE_MOVABLE not ZONE_NORMAL. skip it.
5118 if (!mirrored_kernelcore
&& zone_movable_pfn
[nid
])
5119 if (zone
== ZONE_NORMAL
&& pfn
>= zone_movable_pfn
[nid
])
5123 * Check given memblock attribute by firmware which can affect
5124 * kernel memory layout. If zone==ZONE_MOVABLE but memory is
5125 * mirrored, it's an overlapped memmap init. skip it.
5127 if (mirrored_kernelcore
&& zone
== ZONE_MOVABLE
) {
5128 if (!r
|| pfn
>= memblock_region_memory_end_pfn(r
)) {
5129 for_each_memblock(memory
, tmp
)
5130 if (pfn
< memblock_region_memory_end_pfn(tmp
))
5134 if (pfn
>= memblock_region_memory_base_pfn(r
) &&
5135 memblock_is_mirror(r
)) {
5136 /* already initialized as NORMAL */
5137 pfn
= memblock_region_memory_end_pfn(r
);
5145 * Mark the block movable so that blocks are reserved for
5146 * movable at startup. This will force kernel allocations
5147 * to reserve their blocks rather than leaking throughout
5148 * the address space during boot when many long-lived
5149 * kernel allocations are made.
5151 * bitmap is created for zone's valid pfn range. but memmap
5152 * can be created for invalid pages (for alignment)
5153 * check here not to call set_pageblock_migratetype() against
5156 if (!(pfn
& (pageblock_nr_pages
- 1))) {
5157 struct page
*page
= pfn_to_page(pfn
);
5159 __init_single_page(page
, pfn
, zone
, nid
);
5160 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
5162 __init_single_pfn(pfn
, zone
, nid
);
5167 static void __meminit
zone_init_free_lists(struct zone
*zone
)
5169 unsigned int order
, t
;
5170 for_each_migratetype_order(order
, t
) {
5171 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
5172 zone
->free_area
[order
].nr_free
= 0;
5176 #ifndef __HAVE_ARCH_MEMMAP_INIT
5177 #define memmap_init(size, nid, zone, start_pfn) \
5178 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
5181 static int zone_batchsize(struct zone
*zone
)
5187 * The per-cpu-pages pools are set to around 1000th of the
5188 * size of the zone. But no more than 1/2 of a meg.
5190 * OK, so we don't know how big the cache is. So guess.
5192 batch
= zone
->managed_pages
/ 1024;
5193 if (batch
* PAGE_SIZE
> 512 * 1024)
5194 batch
= (512 * 1024) / PAGE_SIZE
;
5195 batch
/= 4; /* We effectively *= 4 below */
5200 * Clamp the batch to a 2^n - 1 value. Having a power
5201 * of 2 value was found to be more likely to have
5202 * suboptimal cache aliasing properties in some cases.
5204 * For example if 2 tasks are alternately allocating
5205 * batches of pages, one task can end up with a lot
5206 * of pages of one half of the possible page colors
5207 * and the other with pages of the other colors.
5209 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
5214 /* The deferral and batching of frees should be suppressed under NOMMU
5217 * The problem is that NOMMU needs to be able to allocate large chunks
5218 * of contiguous memory as there's no hardware page translation to
5219 * assemble apparent contiguous memory from discontiguous pages.
5221 * Queueing large contiguous runs of pages for batching, however,
5222 * causes the pages to actually be freed in smaller chunks. As there
5223 * can be a significant delay between the individual batches being
5224 * recycled, this leads to the once large chunks of space being
5225 * fragmented and becoming unavailable for high-order allocations.
5232 * pcp->high and pcp->batch values are related and dependent on one another:
5233 * ->batch must never be higher then ->high.
5234 * The following function updates them in a safe manner without read side
5237 * Any new users of pcp->batch and pcp->high should ensure they can cope with
5238 * those fields changing asynchronously (acording the the above rule).
5240 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
5241 * outside of boot time (or some other assurance that no concurrent updaters
5244 static void pageset_update(struct per_cpu_pages
*pcp
, unsigned long high
,
5245 unsigned long batch
)
5247 /* start with a fail safe value for batch */
5251 /* Update high, then batch, in order */
5258 /* a companion to pageset_set_high() */
5259 static void pageset_set_batch(struct per_cpu_pageset
*p
, unsigned long batch
)
5261 pageset_update(&p
->pcp
, 6 * batch
, max(1UL, 1 * batch
));
5264 static void pageset_init(struct per_cpu_pageset
*p
)
5266 struct per_cpu_pages
*pcp
;
5269 memset(p
, 0, sizeof(*p
));
5273 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
5274 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
5277 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
5280 pageset_set_batch(p
, batch
);
5284 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
5285 * to the value high for the pageset p.
5287 static void pageset_set_high(struct per_cpu_pageset
*p
,
5290 unsigned long batch
= max(1UL, high
/ 4);
5291 if ((high
/ 4) > (PAGE_SHIFT
* 8))
5292 batch
= PAGE_SHIFT
* 8;
5294 pageset_update(&p
->pcp
, high
, batch
);
5297 static void pageset_set_high_and_batch(struct zone
*zone
,
5298 struct per_cpu_pageset
*pcp
)
5300 if (percpu_pagelist_fraction
)
5301 pageset_set_high(pcp
,
5302 (zone
->managed_pages
/
5303 percpu_pagelist_fraction
));
5305 pageset_set_batch(pcp
, zone_batchsize(zone
));
5308 static void __meminit
zone_pageset_init(struct zone
*zone
, int cpu
)
5310 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
5313 pageset_set_high_and_batch(zone
, pcp
);
5316 static void __meminit
setup_zone_pageset(struct zone
*zone
)
5319 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
5320 for_each_possible_cpu(cpu
)
5321 zone_pageset_init(zone
, cpu
);
5325 * Allocate per cpu pagesets and initialize them.
5326 * Before this call only boot pagesets were available.
5328 void __init
setup_per_cpu_pageset(void)
5332 for_each_populated_zone(zone
)
5333 setup_zone_pageset(zone
);
5336 static noinline __init_refok
5337 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
5343 * The per-page waitqueue mechanism uses hashed waitqueues
5346 zone
->wait_table_hash_nr_entries
=
5347 wait_table_hash_nr_entries(zone_size_pages
);
5348 zone
->wait_table_bits
=
5349 wait_table_bits(zone
->wait_table_hash_nr_entries
);
5350 alloc_size
= zone
->wait_table_hash_nr_entries
5351 * sizeof(wait_queue_head_t
);
5353 if (!slab_is_available()) {
5354 zone
->wait_table
= (wait_queue_head_t
*)
5355 memblock_virt_alloc_node_nopanic(
5356 alloc_size
, zone
->zone_pgdat
->node_id
);
5359 * This case means that a zone whose size was 0 gets new memory
5360 * via memory hot-add.
5361 * But it may be the case that a new node was hot-added. In
5362 * this case vmalloc() will not be able to use this new node's
5363 * memory - this wait_table must be initialized to use this new
5364 * node itself as well.
5365 * To use this new node's memory, further consideration will be
5368 zone
->wait_table
= vmalloc(alloc_size
);
5370 if (!zone
->wait_table
)
5373 for (i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
5374 init_waitqueue_head(zone
->wait_table
+ i
);
5379 static __meminit
void zone_pcp_init(struct zone
*zone
)
5382 * per cpu subsystem is not up at this point. The following code
5383 * relies on the ability of the linker to provide the
5384 * offset of a (static) per cpu variable into the per cpu area.
5386 zone
->pageset
= &boot_pageset
;
5388 if (populated_zone(zone
))
5389 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
5390 zone
->name
, zone
->present_pages
,
5391 zone_batchsize(zone
));
5394 int __meminit
init_currently_empty_zone(struct zone
*zone
,
5395 unsigned long zone_start_pfn
,
5398 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
5400 ret
= zone_wait_table_init(zone
, size
);
5403 pgdat
->nr_zones
= zone_idx(zone
) + 1;
5405 zone
->zone_start_pfn
= zone_start_pfn
;
5407 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
5408 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
5410 (unsigned long)zone_idx(zone
),
5411 zone_start_pfn
, (zone_start_pfn
+ size
));
5413 zone_init_free_lists(zone
);
5418 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5419 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
5422 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
5424 int __meminit
__early_pfn_to_nid(unsigned long pfn
,
5425 struct mminit_pfnnid_cache
*state
)
5427 unsigned long start_pfn
, end_pfn
;
5430 if (state
->last_start
<= pfn
&& pfn
< state
->last_end
)
5431 return state
->last_nid
;
5433 nid
= memblock_search_pfn_nid(pfn
, &start_pfn
, &end_pfn
);
5435 state
->last_start
= start_pfn
;
5436 state
->last_end
= end_pfn
;
5437 state
->last_nid
= nid
;
5442 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
5445 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
5446 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
5447 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
5449 * If an architecture guarantees that all ranges registered contain no holes
5450 * and may be freed, this this function may be used instead of calling
5451 * memblock_free_early_nid() manually.
5453 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
5455 unsigned long start_pfn
, end_pfn
;
5458 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
5459 start_pfn
= min(start_pfn
, max_low_pfn
);
5460 end_pfn
= min(end_pfn
, max_low_pfn
);
5462 if (start_pfn
< end_pfn
)
5463 memblock_free_early_nid(PFN_PHYS(start_pfn
),
5464 (end_pfn
- start_pfn
) << PAGE_SHIFT
,
5470 * sparse_memory_present_with_active_regions - Call memory_present for each active range
5471 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
5473 * If an architecture guarantees that all ranges registered contain no holes and may
5474 * be freed, this function may be used instead of calling memory_present() manually.
5476 void __init
sparse_memory_present_with_active_regions(int nid
)
5478 unsigned long start_pfn
, end_pfn
;
5481 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
5482 memory_present(this_nid
, start_pfn
, end_pfn
);
5486 * get_pfn_range_for_nid - Return the start and end page frames for a node
5487 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
5488 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
5489 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
5491 * It returns the start and end page frame of a node based on information
5492 * provided by memblock_set_node(). If called for a node
5493 * with no available memory, a warning is printed and the start and end
5496 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
5497 unsigned long *start_pfn
, unsigned long *end_pfn
)
5499 unsigned long this_start_pfn
, this_end_pfn
;
5505 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
5506 *start_pfn
= min(*start_pfn
, this_start_pfn
);
5507 *end_pfn
= max(*end_pfn
, this_end_pfn
);
5510 if (*start_pfn
== -1UL)
5515 * This finds a zone that can be used for ZONE_MOVABLE pages. The
5516 * assumption is made that zones within a node are ordered in monotonic
5517 * increasing memory addresses so that the "highest" populated zone is used
5519 static void __init
find_usable_zone_for_movable(void)
5522 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
5523 if (zone_index
== ZONE_MOVABLE
)
5526 if (arch_zone_highest_possible_pfn
[zone_index
] >
5527 arch_zone_lowest_possible_pfn
[zone_index
])
5531 VM_BUG_ON(zone_index
== -1);
5532 movable_zone
= zone_index
;
5536 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
5537 * because it is sized independent of architecture. Unlike the other zones,
5538 * the starting point for ZONE_MOVABLE is not fixed. It may be different
5539 * in each node depending on the size of each node and how evenly kernelcore
5540 * is distributed. This helper function adjusts the zone ranges
5541 * provided by the architecture for a given node by using the end of the
5542 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5543 * zones within a node are in order of monotonic increases memory addresses
5545 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
5546 unsigned long zone_type
,
5547 unsigned long node_start_pfn
,
5548 unsigned long node_end_pfn
,
5549 unsigned long *zone_start_pfn
,
5550 unsigned long *zone_end_pfn
)
5552 /* Only adjust if ZONE_MOVABLE is on this node */
5553 if (zone_movable_pfn
[nid
]) {
5554 /* Size ZONE_MOVABLE */
5555 if (zone_type
== ZONE_MOVABLE
) {
5556 *zone_start_pfn
= zone_movable_pfn
[nid
];
5557 *zone_end_pfn
= min(node_end_pfn
,
5558 arch_zone_highest_possible_pfn
[movable_zone
]);
5560 /* Check if this whole range is within ZONE_MOVABLE */
5561 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
5562 *zone_start_pfn
= *zone_end_pfn
;
5567 * Return the number of pages a zone spans in a node, including holes
5568 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5570 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5571 unsigned long zone_type
,
5572 unsigned long node_start_pfn
,
5573 unsigned long node_end_pfn
,
5574 unsigned long *zone_start_pfn
,
5575 unsigned long *zone_end_pfn
,
5576 unsigned long *ignored
)
5578 /* When hotadd a new node from cpu_up(), the node should be empty */
5579 if (!node_start_pfn
&& !node_end_pfn
)
5582 /* Get the start and end of the zone */
5583 *zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
5584 *zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
5585 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5586 node_start_pfn
, node_end_pfn
,
5587 zone_start_pfn
, zone_end_pfn
);
5589 /* Check that this node has pages within the zone's required range */
5590 if (*zone_end_pfn
< node_start_pfn
|| *zone_start_pfn
> node_end_pfn
)
5593 /* Move the zone boundaries inside the node if necessary */
5594 *zone_end_pfn
= min(*zone_end_pfn
, node_end_pfn
);
5595 *zone_start_pfn
= max(*zone_start_pfn
, node_start_pfn
);
5597 /* Return the spanned pages */
5598 return *zone_end_pfn
- *zone_start_pfn
;
5602 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
5603 * then all holes in the requested range will be accounted for.
5605 unsigned long __meminit
__absent_pages_in_range(int nid
,
5606 unsigned long range_start_pfn
,
5607 unsigned long range_end_pfn
)
5609 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
5610 unsigned long start_pfn
, end_pfn
;
5613 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5614 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
5615 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
5616 nr_absent
-= end_pfn
- start_pfn
;
5622 * absent_pages_in_range - Return number of page frames in holes within a range
5623 * @start_pfn: The start PFN to start searching for holes
5624 * @end_pfn: The end PFN to stop searching for holes
5626 * It returns the number of pages frames in memory holes within a range.
5628 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
5629 unsigned long end_pfn
)
5631 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
5634 /* Return the number of page frames in holes in a zone on a node */
5635 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5636 unsigned long zone_type
,
5637 unsigned long node_start_pfn
,
5638 unsigned long node_end_pfn
,
5639 unsigned long *ignored
)
5641 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
5642 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
5643 unsigned long zone_start_pfn
, zone_end_pfn
;
5644 unsigned long nr_absent
;
5646 /* When hotadd a new node from cpu_up(), the node should be empty */
5647 if (!node_start_pfn
&& !node_end_pfn
)
5650 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
5651 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
5653 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5654 node_start_pfn
, node_end_pfn
,
5655 &zone_start_pfn
, &zone_end_pfn
);
5656 nr_absent
= __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
5659 * ZONE_MOVABLE handling.
5660 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
5663 if (zone_movable_pfn
[nid
]) {
5664 if (mirrored_kernelcore
) {
5665 unsigned long start_pfn
, end_pfn
;
5666 struct memblock_region
*r
;
5668 for_each_memblock(memory
, r
) {
5669 start_pfn
= clamp(memblock_region_memory_base_pfn(r
),
5670 zone_start_pfn
, zone_end_pfn
);
5671 end_pfn
= clamp(memblock_region_memory_end_pfn(r
),
5672 zone_start_pfn
, zone_end_pfn
);
5674 if (zone_type
== ZONE_MOVABLE
&&
5675 memblock_is_mirror(r
))
5676 nr_absent
+= end_pfn
- start_pfn
;
5678 if (zone_type
== ZONE_NORMAL
&&
5679 !memblock_is_mirror(r
))
5680 nr_absent
+= end_pfn
- start_pfn
;
5683 if (zone_type
== ZONE_NORMAL
)
5684 nr_absent
+= node_end_pfn
- zone_movable_pfn
[nid
];
5691 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5692 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5693 unsigned long zone_type
,
5694 unsigned long node_start_pfn
,
5695 unsigned long node_end_pfn
,
5696 unsigned long *zone_start_pfn
,
5697 unsigned long *zone_end_pfn
,
5698 unsigned long *zones_size
)
5702 *zone_start_pfn
= node_start_pfn
;
5703 for (zone
= 0; zone
< zone_type
; zone
++)
5704 *zone_start_pfn
+= zones_size
[zone
];
5706 *zone_end_pfn
= *zone_start_pfn
+ zones_size
[zone_type
];
5708 return zones_size
[zone_type
];
5711 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5712 unsigned long zone_type
,
5713 unsigned long node_start_pfn
,
5714 unsigned long node_end_pfn
,
5715 unsigned long *zholes_size
)
5720 return zholes_size
[zone_type
];
5723 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5725 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
5726 unsigned long node_start_pfn
,
5727 unsigned long node_end_pfn
,
5728 unsigned long *zones_size
,
5729 unsigned long *zholes_size
)
5731 unsigned long realtotalpages
= 0, totalpages
= 0;
5734 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5735 struct zone
*zone
= pgdat
->node_zones
+ i
;
5736 unsigned long zone_start_pfn
, zone_end_pfn
;
5737 unsigned long size
, real_size
;
5739 size
= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
5745 real_size
= size
- zone_absent_pages_in_node(pgdat
->node_id
, i
,
5746 node_start_pfn
, node_end_pfn
,
5749 zone
->zone_start_pfn
= zone_start_pfn
;
5751 zone
->zone_start_pfn
= 0;
5752 zone
->spanned_pages
= size
;
5753 zone
->present_pages
= real_size
;
5756 realtotalpages
+= real_size
;
5759 pgdat
->node_spanned_pages
= totalpages
;
5760 pgdat
->node_present_pages
= realtotalpages
;
5761 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
5765 #ifndef CONFIG_SPARSEMEM
5767 * Calculate the size of the zone->blockflags rounded to an unsigned long
5768 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5769 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
5770 * round what is now in bits to nearest long in bits, then return it in
5773 static unsigned long __init
usemap_size(unsigned long zone_start_pfn
, unsigned long zonesize
)
5775 unsigned long usemapsize
;
5777 zonesize
+= zone_start_pfn
& (pageblock_nr_pages
-1);
5778 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
5779 usemapsize
= usemapsize
>> pageblock_order
;
5780 usemapsize
*= NR_PAGEBLOCK_BITS
;
5781 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
5783 return usemapsize
/ 8;
5786 static void __init
setup_usemap(struct pglist_data
*pgdat
,
5788 unsigned long zone_start_pfn
,
5789 unsigned long zonesize
)
5791 unsigned long usemapsize
= usemap_size(zone_start_pfn
, zonesize
);
5792 zone
->pageblock_flags
= NULL
;
5794 zone
->pageblock_flags
=
5795 memblock_virt_alloc_node_nopanic(usemapsize
,
5799 static inline void setup_usemap(struct pglist_data
*pgdat
, struct zone
*zone
,
5800 unsigned long zone_start_pfn
, unsigned long zonesize
) {}
5801 #endif /* CONFIG_SPARSEMEM */
5803 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
5805 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
5806 void __paginginit
set_pageblock_order(void)
5810 /* Check that pageblock_nr_pages has not already been setup */
5811 if (pageblock_order
)
5814 if (HPAGE_SHIFT
> PAGE_SHIFT
)
5815 order
= HUGETLB_PAGE_ORDER
;
5817 order
= MAX_ORDER
- 1;
5820 * Assume the largest contiguous order of interest is a huge page.
5821 * This value may be variable depending on boot parameters on IA64 and
5824 pageblock_order
= order
;
5826 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5829 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
5830 * is unused as pageblock_order is set at compile-time. See
5831 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5834 void __paginginit
set_pageblock_order(void)
5838 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5840 static unsigned long __paginginit
calc_memmap_size(unsigned long spanned_pages
,
5841 unsigned long present_pages
)
5843 unsigned long pages
= spanned_pages
;
5846 * Provide a more accurate estimation if there are holes within
5847 * the zone and SPARSEMEM is in use. If there are holes within the
5848 * zone, each populated memory region may cost us one or two extra
5849 * memmap pages due to alignment because memmap pages for each
5850 * populated regions may not naturally algined on page boundary.
5851 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5853 if (spanned_pages
> present_pages
+ (present_pages
>> 4) &&
5854 IS_ENABLED(CONFIG_SPARSEMEM
))
5855 pages
= present_pages
;
5857 return PAGE_ALIGN(pages
* sizeof(struct page
)) >> PAGE_SHIFT
;
5861 * Set up the zone data structures:
5862 * - mark all pages reserved
5863 * - mark all memory queues empty
5864 * - clear the memory bitmaps
5866 * NOTE: pgdat should get zeroed by caller.
5868 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
)
5871 int nid
= pgdat
->node_id
;
5874 pgdat_resize_init(pgdat
);
5875 #ifdef CONFIG_NUMA_BALANCING
5876 spin_lock_init(&pgdat
->numabalancing_migrate_lock
);
5877 pgdat
->numabalancing_migrate_nr_pages
= 0;
5878 pgdat
->numabalancing_migrate_next_window
= jiffies
;
5880 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5881 spin_lock_init(&pgdat
->split_queue_lock
);
5882 INIT_LIST_HEAD(&pgdat
->split_queue
);
5883 pgdat
->split_queue_len
= 0;
5885 init_waitqueue_head(&pgdat
->kswapd_wait
);
5886 init_waitqueue_head(&pgdat
->pfmemalloc_wait
);
5887 #ifdef CONFIG_COMPACTION
5888 init_waitqueue_head(&pgdat
->kcompactd_wait
);
5890 pgdat_page_ext_init(pgdat
);
5892 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5893 struct zone
*zone
= pgdat
->node_zones
+ j
;
5894 unsigned long size
, realsize
, freesize
, memmap_pages
;
5895 unsigned long zone_start_pfn
= zone
->zone_start_pfn
;
5897 size
= zone
->spanned_pages
;
5898 realsize
= freesize
= zone
->present_pages
;
5901 * Adjust freesize so that it accounts for how much memory
5902 * is used by this zone for memmap. This affects the watermark
5903 * and per-cpu initialisations
5905 memmap_pages
= calc_memmap_size(size
, realsize
);
5906 if (!is_highmem_idx(j
)) {
5907 if (freesize
>= memmap_pages
) {
5908 freesize
-= memmap_pages
;
5911 " %s zone: %lu pages used for memmap\n",
5912 zone_names
[j
], memmap_pages
);
5914 pr_warn(" %s zone: %lu pages exceeds freesize %lu\n",
5915 zone_names
[j
], memmap_pages
, freesize
);
5918 /* Account for reserved pages */
5919 if (j
== 0 && freesize
> dma_reserve
) {
5920 freesize
-= dma_reserve
;
5921 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
5922 zone_names
[0], dma_reserve
);
5925 if (!is_highmem_idx(j
))
5926 nr_kernel_pages
+= freesize
;
5927 /* Charge for highmem memmap if there are enough kernel pages */
5928 else if (nr_kernel_pages
> memmap_pages
* 2)
5929 nr_kernel_pages
-= memmap_pages
;
5930 nr_all_pages
+= freesize
;
5933 * Set an approximate value for lowmem here, it will be adjusted
5934 * when the bootmem allocator frees pages into the buddy system.
5935 * And all highmem pages will be managed by the buddy system.
5937 zone
->managed_pages
= is_highmem_idx(j
) ? realsize
: freesize
;
5940 zone
->min_unmapped_pages
= (freesize
*sysctl_min_unmapped_ratio
)
5942 zone
->min_slab_pages
= (freesize
* sysctl_min_slab_ratio
) / 100;
5944 zone
->name
= zone_names
[j
];
5945 spin_lock_init(&zone
->lock
);
5946 spin_lock_init(&zone
->lru_lock
);
5947 zone_seqlock_init(zone
);
5948 zone
->zone_pgdat
= pgdat
;
5949 zone_pcp_init(zone
);
5951 /* For bootup, initialized properly in watermark setup */
5952 mod_zone_page_state(zone
, NR_ALLOC_BATCH
, zone
->managed_pages
);
5954 lruvec_init(&zone
->lruvec
);
5958 set_pageblock_order();
5959 setup_usemap(pgdat
, zone
, zone_start_pfn
, size
);
5960 ret
= init_currently_empty_zone(zone
, zone_start_pfn
, size
);
5962 memmap_init(size
, nid
, j
, zone_start_pfn
);
5966 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
5968 unsigned long __maybe_unused start
= 0;
5969 unsigned long __maybe_unused offset
= 0;
5971 /* Skip empty nodes */
5972 if (!pgdat
->node_spanned_pages
)
5975 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5976 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
5977 offset
= pgdat
->node_start_pfn
- start
;
5978 /* ia64 gets its own node_mem_map, before this, without bootmem */
5979 if (!pgdat
->node_mem_map
) {
5980 unsigned long size
, end
;
5984 * The zone's endpoints aren't required to be MAX_ORDER
5985 * aligned but the node_mem_map endpoints must be in order
5986 * for the buddy allocator to function correctly.
5988 end
= pgdat_end_pfn(pgdat
);
5989 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
5990 size
= (end
- start
) * sizeof(struct page
);
5991 map
= alloc_remap(pgdat
->node_id
, size
);
5993 map
= memblock_virt_alloc_node_nopanic(size
,
5995 pgdat
->node_mem_map
= map
+ offset
;
5997 #ifndef CONFIG_NEED_MULTIPLE_NODES
5999 * With no DISCONTIG, the global mem_map is just set as node 0's
6001 if (pgdat
== NODE_DATA(0)) {
6002 mem_map
= NODE_DATA(0)->node_mem_map
;
6003 #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
6004 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
6006 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6009 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
6012 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
6013 unsigned long node_start_pfn
, unsigned long *zholes_size
)
6015 pg_data_t
*pgdat
= NODE_DATA(nid
);
6016 unsigned long start_pfn
= 0;
6017 unsigned long end_pfn
= 0;
6019 /* pg_data_t should be reset to zero when it's allocated */
6020 WARN_ON(pgdat
->nr_zones
|| pgdat
->classzone_idx
);
6022 reset_deferred_meminit(pgdat
);
6023 pgdat
->node_id
= nid
;
6024 pgdat
->node_start_pfn
= node_start_pfn
;
6025 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6026 get_pfn_range_for_nid(nid
, &start_pfn
, &end_pfn
);
6027 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid
,
6028 (u64
)start_pfn
<< PAGE_SHIFT
,
6029 end_pfn
? ((u64
)end_pfn
<< PAGE_SHIFT
) - 1 : 0);
6031 start_pfn
= node_start_pfn
;
6033 calculate_node_totalpages(pgdat
, start_pfn
, end_pfn
,
6034 zones_size
, zholes_size
);
6036 alloc_node_mem_map(pgdat
);
6037 #ifdef CONFIG_FLAT_NODE_MEM_MAP
6038 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
6039 nid
, (unsigned long)pgdat
,
6040 (unsigned long)pgdat
->node_mem_map
);
6043 free_area_init_core(pgdat
);
6046 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6048 #if MAX_NUMNODES > 1
6050 * Figure out the number of possible node ids.
6052 void __init
setup_nr_node_ids(void)
6054 unsigned int highest
;
6056 highest
= find_last_bit(node_possible_map
.bits
, MAX_NUMNODES
);
6057 nr_node_ids
= highest
+ 1;
6062 * node_map_pfn_alignment - determine the maximum internode alignment
6064 * This function should be called after node map is populated and sorted.
6065 * It calculates the maximum power of two alignment which can distinguish
6068 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
6069 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
6070 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
6071 * shifted, 1GiB is enough and this function will indicate so.
6073 * This is used to test whether pfn -> nid mapping of the chosen memory
6074 * model has fine enough granularity to avoid incorrect mapping for the
6075 * populated node map.
6077 * Returns the determined alignment in pfn's. 0 if there is no alignment
6078 * requirement (single node).
6080 unsigned long __init
node_map_pfn_alignment(void)
6082 unsigned long accl_mask
= 0, last_end
= 0;
6083 unsigned long start
, end
, mask
;
6087 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
6088 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
6095 * Start with a mask granular enough to pin-point to the
6096 * start pfn and tick off bits one-by-one until it becomes
6097 * too coarse to separate the current node from the last.
6099 mask
= ~((1 << __ffs(start
)) - 1);
6100 while (mask
&& last_end
<= (start
& (mask
<< 1)))
6103 /* accumulate all internode masks */
6107 /* convert mask to number of pages */
6108 return ~accl_mask
+ 1;
6111 /* Find the lowest pfn for a node */
6112 static unsigned long __init
find_min_pfn_for_node(int nid
)
6114 unsigned long min_pfn
= ULONG_MAX
;
6115 unsigned long start_pfn
;
6118 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
6119 min_pfn
= min(min_pfn
, start_pfn
);
6121 if (min_pfn
== ULONG_MAX
) {
6122 pr_warn("Could not find start_pfn for node %d\n", nid
);
6130 * find_min_pfn_with_active_regions - Find the minimum PFN registered
6132 * It returns the minimum PFN based on information provided via
6133 * memblock_set_node().
6135 unsigned long __init
find_min_pfn_with_active_regions(void)
6137 return find_min_pfn_for_node(MAX_NUMNODES
);
6141 * early_calculate_totalpages()
6142 * Sum pages in active regions for movable zone.
6143 * Populate N_MEMORY for calculating usable_nodes.
6145 static unsigned long __init
early_calculate_totalpages(void)
6147 unsigned long totalpages
= 0;
6148 unsigned long start_pfn
, end_pfn
;
6151 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
6152 unsigned long pages
= end_pfn
- start_pfn
;
6154 totalpages
+= pages
;
6156 node_set_state(nid
, N_MEMORY
);
6162 * Find the PFN the Movable zone begins in each node. Kernel memory
6163 * is spread evenly between nodes as long as the nodes have enough
6164 * memory. When they don't, some nodes will have more kernelcore than
6167 static void __init
find_zone_movable_pfns_for_nodes(void)
6170 unsigned long usable_startpfn
;
6171 unsigned long kernelcore_node
, kernelcore_remaining
;
6172 /* save the state before borrow the nodemask */
6173 nodemask_t saved_node_state
= node_states
[N_MEMORY
];
6174 unsigned long totalpages
= early_calculate_totalpages();
6175 int usable_nodes
= nodes_weight(node_states
[N_MEMORY
]);
6176 struct memblock_region
*r
;
6178 /* Need to find movable_zone earlier when movable_node is specified. */
6179 find_usable_zone_for_movable();
6182 * If movable_node is specified, ignore kernelcore and movablecore
6185 if (movable_node_is_enabled()) {
6186 for_each_memblock(memory
, r
) {
6187 if (!memblock_is_hotpluggable(r
))
6192 usable_startpfn
= PFN_DOWN(r
->base
);
6193 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
6194 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
6202 * If kernelcore=mirror is specified, ignore movablecore option
6204 if (mirrored_kernelcore
) {
6205 bool mem_below_4gb_not_mirrored
= false;
6207 for_each_memblock(memory
, r
) {
6208 if (memblock_is_mirror(r
))
6213 usable_startpfn
= memblock_region_memory_base_pfn(r
);
6215 if (usable_startpfn
< 0x100000) {
6216 mem_below_4gb_not_mirrored
= true;
6220 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
6221 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
6225 if (mem_below_4gb_not_mirrored
)
6226 pr_warn("This configuration results in unmirrored kernel memory.");
6232 * If movablecore=nn[KMG] was specified, calculate what size of
6233 * kernelcore that corresponds so that memory usable for
6234 * any allocation type is evenly spread. If both kernelcore
6235 * and movablecore are specified, then the value of kernelcore
6236 * will be used for required_kernelcore if it's greater than
6237 * what movablecore would have allowed.
6239 if (required_movablecore
) {
6240 unsigned long corepages
;
6243 * Round-up so that ZONE_MOVABLE is at least as large as what
6244 * was requested by the user
6246 required_movablecore
=
6247 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
6248 required_movablecore
= min(totalpages
, required_movablecore
);
6249 corepages
= totalpages
- required_movablecore
;
6251 required_kernelcore
= max(required_kernelcore
, corepages
);
6255 * If kernelcore was not specified or kernelcore size is larger
6256 * than totalpages, there is no ZONE_MOVABLE.
6258 if (!required_kernelcore
|| required_kernelcore
>= totalpages
)
6261 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
6262 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
6265 /* Spread kernelcore memory as evenly as possible throughout nodes */
6266 kernelcore_node
= required_kernelcore
/ usable_nodes
;
6267 for_each_node_state(nid
, N_MEMORY
) {
6268 unsigned long start_pfn
, end_pfn
;
6271 * Recalculate kernelcore_node if the division per node
6272 * now exceeds what is necessary to satisfy the requested
6273 * amount of memory for the kernel
6275 if (required_kernelcore
< kernelcore_node
)
6276 kernelcore_node
= required_kernelcore
/ usable_nodes
;
6279 * As the map is walked, we track how much memory is usable
6280 * by the kernel using kernelcore_remaining. When it is
6281 * 0, the rest of the node is usable by ZONE_MOVABLE
6283 kernelcore_remaining
= kernelcore_node
;
6285 /* Go through each range of PFNs within this node */
6286 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
6287 unsigned long size_pages
;
6289 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
6290 if (start_pfn
>= end_pfn
)
6293 /* Account for what is only usable for kernelcore */
6294 if (start_pfn
< usable_startpfn
) {
6295 unsigned long kernel_pages
;
6296 kernel_pages
= min(end_pfn
, usable_startpfn
)
6299 kernelcore_remaining
-= min(kernel_pages
,
6300 kernelcore_remaining
);
6301 required_kernelcore
-= min(kernel_pages
,
6302 required_kernelcore
);
6304 /* Continue if range is now fully accounted */
6305 if (end_pfn
<= usable_startpfn
) {
6308 * Push zone_movable_pfn to the end so
6309 * that if we have to rebalance
6310 * kernelcore across nodes, we will
6311 * not double account here
6313 zone_movable_pfn
[nid
] = end_pfn
;
6316 start_pfn
= usable_startpfn
;
6320 * The usable PFN range for ZONE_MOVABLE is from
6321 * start_pfn->end_pfn. Calculate size_pages as the
6322 * number of pages used as kernelcore
6324 size_pages
= end_pfn
- start_pfn
;
6325 if (size_pages
> kernelcore_remaining
)
6326 size_pages
= kernelcore_remaining
;
6327 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
6330 * Some kernelcore has been met, update counts and
6331 * break if the kernelcore for this node has been
6334 required_kernelcore
-= min(required_kernelcore
,
6336 kernelcore_remaining
-= size_pages
;
6337 if (!kernelcore_remaining
)
6343 * If there is still required_kernelcore, we do another pass with one
6344 * less node in the count. This will push zone_movable_pfn[nid] further
6345 * along on the nodes that still have memory until kernelcore is
6349 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
6353 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
6354 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
6355 zone_movable_pfn
[nid
] =
6356 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
6359 /* restore the node_state */
6360 node_states
[N_MEMORY
] = saved_node_state
;
6363 /* Any regular or high memory on that node ? */
6364 static void check_for_memory(pg_data_t
*pgdat
, int nid
)
6366 enum zone_type zone_type
;
6368 if (N_MEMORY
== N_NORMAL_MEMORY
)
6371 for (zone_type
= 0; zone_type
<= ZONE_MOVABLE
- 1; zone_type
++) {
6372 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
6373 if (populated_zone(zone
)) {
6374 node_set_state(nid
, N_HIGH_MEMORY
);
6375 if (N_NORMAL_MEMORY
!= N_HIGH_MEMORY
&&
6376 zone_type
<= ZONE_NORMAL
)
6377 node_set_state(nid
, N_NORMAL_MEMORY
);
6384 * free_area_init_nodes - Initialise all pg_data_t and zone data
6385 * @max_zone_pfn: an array of max PFNs for each zone
6387 * This will call free_area_init_node() for each active node in the system.
6388 * Using the page ranges provided by memblock_set_node(), the size of each
6389 * zone in each node and their holes is calculated. If the maximum PFN
6390 * between two adjacent zones match, it is assumed that the zone is empty.
6391 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
6392 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
6393 * starts where the previous one ended. For example, ZONE_DMA32 starts
6394 * at arch_max_dma_pfn.
6396 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
6398 unsigned long start_pfn
, end_pfn
;
6401 /* Record where the zone boundaries are */
6402 memset(arch_zone_lowest_possible_pfn
, 0,
6403 sizeof(arch_zone_lowest_possible_pfn
));
6404 memset(arch_zone_highest_possible_pfn
, 0,
6405 sizeof(arch_zone_highest_possible_pfn
));
6407 start_pfn
= find_min_pfn_with_active_regions();
6409 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
6410 if (i
== ZONE_MOVABLE
)
6413 end_pfn
= max(max_zone_pfn
[i
], start_pfn
);
6414 arch_zone_lowest_possible_pfn
[i
] = start_pfn
;
6415 arch_zone_highest_possible_pfn
[i
] = end_pfn
;
6417 start_pfn
= end_pfn
;
6419 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
6420 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
6422 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
6423 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
6424 find_zone_movable_pfns_for_nodes();
6426 /* Print out the zone ranges */
6427 pr_info("Zone ranges:\n");
6428 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
6429 if (i
== ZONE_MOVABLE
)
6431 pr_info(" %-8s ", zone_names
[i
]);
6432 if (arch_zone_lowest_possible_pfn
[i
] ==
6433 arch_zone_highest_possible_pfn
[i
])
6436 pr_cont("[mem %#018Lx-%#018Lx]\n",
6437 (u64
)arch_zone_lowest_possible_pfn
[i
]
6439 ((u64
)arch_zone_highest_possible_pfn
[i
]
6440 << PAGE_SHIFT
) - 1);
6443 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
6444 pr_info("Movable zone start for each node\n");
6445 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
6446 if (zone_movable_pfn
[i
])
6447 pr_info(" Node %d: %#018Lx\n", i
,
6448 (u64
)zone_movable_pfn
[i
] << PAGE_SHIFT
);
6451 /* Print out the early node map */
6452 pr_info("Early memory node ranges\n");
6453 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
6454 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid
,
6455 (u64
)start_pfn
<< PAGE_SHIFT
,
6456 ((u64
)end_pfn
<< PAGE_SHIFT
) - 1);
6458 /* Initialise every node */
6459 mminit_verify_pageflags_layout();
6460 setup_nr_node_ids();
6461 for_each_online_node(nid
) {
6462 pg_data_t
*pgdat
= NODE_DATA(nid
);
6463 free_area_init_node(nid
, NULL
,
6464 find_min_pfn_for_node(nid
), NULL
);
6466 /* Any memory on that node */
6467 if (pgdat
->node_present_pages
)
6468 node_set_state(nid
, N_MEMORY
);
6469 check_for_memory(pgdat
, nid
);
6473 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
6475 unsigned long long coremem
;
6479 coremem
= memparse(p
, &p
);
6480 *core
= coremem
>> PAGE_SHIFT
;
6482 /* Paranoid check that UL is enough for the coremem value */
6483 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
6489 * kernelcore=size sets the amount of memory for use for allocations that
6490 * cannot be reclaimed or migrated.
6492 static int __init
cmdline_parse_kernelcore(char *p
)
6494 /* parse kernelcore=mirror */
6495 if (parse_option_str(p
, "mirror")) {
6496 mirrored_kernelcore
= true;
6500 return cmdline_parse_core(p
, &required_kernelcore
);
6504 * movablecore=size sets the amount of memory for use for allocations that
6505 * can be reclaimed or migrated.
6507 static int __init
cmdline_parse_movablecore(char *p
)
6509 return cmdline_parse_core(p
, &required_movablecore
);
6512 early_param("kernelcore", cmdline_parse_kernelcore
);
6513 early_param("movablecore", cmdline_parse_movablecore
);
6515 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6517 void adjust_managed_page_count(struct page
*page
, long count
)
6519 spin_lock(&managed_page_count_lock
);
6520 page_zone(page
)->managed_pages
+= count
;
6521 totalram_pages
+= count
;
6522 #ifdef CONFIG_HIGHMEM
6523 if (PageHighMem(page
))
6524 totalhigh_pages
+= count
;
6526 spin_unlock(&managed_page_count_lock
);
6528 EXPORT_SYMBOL(adjust_managed_page_count
);
6530 unsigned long free_reserved_area(void *start
, void *end
, int poison
, char *s
)
6533 unsigned long pages
= 0;
6535 start
= (void *)PAGE_ALIGN((unsigned long)start
);
6536 end
= (void *)((unsigned long)end
& PAGE_MASK
);
6537 for (pos
= start
; pos
< end
; pos
+= PAGE_SIZE
, pages
++) {
6538 if ((unsigned int)poison
<= 0xFF)
6539 memset(pos
, poison
, PAGE_SIZE
);
6540 free_reserved_page(virt_to_page(pos
));
6544 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
6545 s
, pages
<< (PAGE_SHIFT
- 10), start
, end
);
6549 EXPORT_SYMBOL(free_reserved_area
);
6551 #ifdef CONFIG_HIGHMEM
6552 void free_highmem_page(struct page
*page
)
6554 __free_reserved_page(page
);
6556 page_zone(page
)->managed_pages
++;
6562 void __init
mem_init_print_info(const char *str
)
6564 unsigned long physpages
, codesize
, datasize
, rosize
, bss_size
;
6565 unsigned long init_code_size
, init_data_size
;
6567 physpages
= get_num_physpages();
6568 codesize
= _etext
- _stext
;
6569 datasize
= _edata
- _sdata
;
6570 rosize
= __end_rodata
- __start_rodata
;
6571 bss_size
= __bss_stop
- __bss_start
;
6572 init_data_size
= __init_end
- __init_begin
;
6573 init_code_size
= _einittext
- _sinittext
;
6576 * Detect special cases and adjust section sizes accordingly:
6577 * 1) .init.* may be embedded into .data sections
6578 * 2) .init.text.* may be out of [__init_begin, __init_end],
6579 * please refer to arch/tile/kernel/vmlinux.lds.S.
6580 * 3) .rodata.* may be embedded into .text or .data sections.
6582 #define adj_init_size(start, end, size, pos, adj) \
6584 if (start <= pos && pos < end && size > adj) \
6588 adj_init_size(__init_begin
, __init_end
, init_data_size
,
6589 _sinittext
, init_code_size
);
6590 adj_init_size(_stext
, _etext
, codesize
, _sinittext
, init_code_size
);
6591 adj_init_size(_sdata
, _edata
, datasize
, __init_begin
, init_data_size
);
6592 adj_init_size(_stext
, _etext
, codesize
, __start_rodata
, rosize
);
6593 adj_init_size(_sdata
, _edata
, datasize
, __start_rodata
, rosize
);
6595 #undef adj_init_size
6597 pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
6598 #ifdef CONFIG_HIGHMEM
6602 nr_free_pages() << (PAGE_SHIFT
- 10),
6603 physpages
<< (PAGE_SHIFT
- 10),
6604 codesize
>> 10, datasize
>> 10, rosize
>> 10,
6605 (init_data_size
+ init_code_size
) >> 10, bss_size
>> 10,
6606 (physpages
- totalram_pages
- totalcma_pages
) << (PAGE_SHIFT
- 10),
6607 totalcma_pages
<< (PAGE_SHIFT
- 10),
6608 #ifdef CONFIG_HIGHMEM
6609 totalhigh_pages
<< (PAGE_SHIFT
- 10),
6611 str
? ", " : "", str
? str
: "");
6615 * set_dma_reserve - set the specified number of pages reserved in the first zone
6616 * @new_dma_reserve: The number of pages to mark reserved
6618 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
6619 * In the DMA zone, a significant percentage may be consumed by kernel image
6620 * and other unfreeable allocations which can skew the watermarks badly. This
6621 * function may optionally be used to account for unfreeable pages in the
6622 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6623 * smaller per-cpu batchsize.
6625 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
6627 dma_reserve
= new_dma_reserve
;
6630 void __init
free_area_init(unsigned long *zones_size
)
6632 free_area_init_node(0, zones_size
,
6633 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
6636 static int page_alloc_cpu_notify(struct notifier_block
*self
,
6637 unsigned long action
, void *hcpu
)
6639 int cpu
= (unsigned long)hcpu
;
6641 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
6642 lru_add_drain_cpu(cpu
);
6646 * Spill the event counters of the dead processor
6647 * into the current processors event counters.
6648 * This artificially elevates the count of the current
6651 vm_events_fold_cpu(cpu
);
6654 * Zero the differential counters of the dead processor
6655 * so that the vm statistics are consistent.
6657 * This is only okay since the processor is dead and cannot
6658 * race with what we are doing.
6660 cpu_vm_stats_fold(cpu
);
6665 void __init
page_alloc_init(void)
6667 hotcpu_notifier(page_alloc_cpu_notify
, 0);
6671 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
6672 * or min_free_kbytes changes.
6674 static void calculate_totalreserve_pages(void)
6676 struct pglist_data
*pgdat
;
6677 unsigned long reserve_pages
= 0;
6678 enum zone_type i
, j
;
6680 for_each_online_pgdat(pgdat
) {
6681 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
6682 struct zone
*zone
= pgdat
->node_zones
+ i
;
6685 /* Find valid and maximum lowmem_reserve in the zone */
6686 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
6687 if (zone
->lowmem_reserve
[j
] > max
)
6688 max
= zone
->lowmem_reserve
[j
];
6691 /* we treat the high watermark as reserved pages. */
6692 max
+= high_wmark_pages(zone
);
6694 if (max
> zone
->managed_pages
)
6695 max
= zone
->managed_pages
;
6697 zone
->totalreserve_pages
= max
;
6699 reserve_pages
+= max
;
6702 totalreserve_pages
= reserve_pages
;
6706 * setup_per_zone_lowmem_reserve - called whenever
6707 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
6708 * has a correct pages reserved value, so an adequate number of
6709 * pages are left in the zone after a successful __alloc_pages().
6711 static void setup_per_zone_lowmem_reserve(void)
6713 struct pglist_data
*pgdat
;
6714 enum zone_type j
, idx
;
6716 for_each_online_pgdat(pgdat
) {
6717 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
6718 struct zone
*zone
= pgdat
->node_zones
+ j
;
6719 unsigned long managed_pages
= zone
->managed_pages
;
6721 zone
->lowmem_reserve
[j
] = 0;
6725 struct zone
*lower_zone
;
6729 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
6730 sysctl_lowmem_reserve_ratio
[idx
] = 1;
6732 lower_zone
= pgdat
->node_zones
+ idx
;
6733 lower_zone
->lowmem_reserve
[j
] = managed_pages
/
6734 sysctl_lowmem_reserve_ratio
[idx
];
6735 managed_pages
+= lower_zone
->managed_pages
;
6740 /* update totalreserve_pages */
6741 calculate_totalreserve_pages();
6744 static void __setup_per_zone_wmarks(void)
6746 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
6747 unsigned long lowmem_pages
= 0;
6749 unsigned long flags
;
6751 /* Calculate total number of !ZONE_HIGHMEM pages */
6752 for_each_zone(zone
) {
6753 if (!is_highmem(zone
))
6754 lowmem_pages
+= zone
->managed_pages
;
6757 for_each_zone(zone
) {
6760 spin_lock_irqsave(&zone
->lock
, flags
);
6761 tmp
= (u64
)pages_min
* zone
->managed_pages
;
6762 do_div(tmp
, lowmem_pages
);
6763 if (is_highmem(zone
)) {
6765 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6766 * need highmem pages, so cap pages_min to a small
6769 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
6770 * deltas control asynch page reclaim, and so should
6771 * not be capped for highmem.
6773 unsigned long min_pages
;
6775 min_pages
= zone
->managed_pages
/ 1024;
6776 min_pages
= clamp(min_pages
, SWAP_CLUSTER_MAX
, 128UL);
6777 zone
->watermark
[WMARK_MIN
] = min_pages
;
6780 * If it's a lowmem zone, reserve a number of pages
6781 * proportionate to the zone's size.
6783 zone
->watermark
[WMARK_MIN
] = tmp
;
6787 * Set the kswapd watermarks distance according to the
6788 * scale factor in proportion to available memory, but
6789 * ensure a minimum size on small systems.
6791 tmp
= max_t(u64
, tmp
>> 2,
6792 mult_frac(zone
->managed_pages
,
6793 watermark_scale_factor
, 10000));
6795 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + tmp
;
6796 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + tmp
* 2;
6798 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
6799 high_wmark_pages(zone
) - low_wmark_pages(zone
) -
6800 atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]));
6802 spin_unlock_irqrestore(&zone
->lock
, flags
);
6805 /* update totalreserve_pages */
6806 calculate_totalreserve_pages();
6810 * setup_per_zone_wmarks - called when min_free_kbytes changes
6811 * or when memory is hot-{added|removed}
6813 * Ensures that the watermark[min,low,high] values for each zone are set
6814 * correctly with respect to min_free_kbytes.
6816 void setup_per_zone_wmarks(void)
6818 mutex_lock(&zonelists_mutex
);
6819 __setup_per_zone_wmarks();
6820 mutex_unlock(&zonelists_mutex
);
6824 * Initialise min_free_kbytes.
6826 * For small machines we want it small (128k min). For large machines
6827 * we want it large (64MB max). But it is not linear, because network
6828 * bandwidth does not increase linearly with machine size. We use
6830 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
6831 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6847 int __meminit
init_per_zone_wmark_min(void)
6849 unsigned long lowmem_kbytes
;
6850 int new_min_free_kbytes
;
6852 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
6853 new_min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
6855 if (new_min_free_kbytes
> user_min_free_kbytes
) {
6856 min_free_kbytes
= new_min_free_kbytes
;
6857 if (min_free_kbytes
< 128)
6858 min_free_kbytes
= 128;
6859 if (min_free_kbytes
> 65536)
6860 min_free_kbytes
= 65536;
6862 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6863 new_min_free_kbytes
, user_min_free_kbytes
);
6865 setup_per_zone_wmarks();
6866 refresh_zone_stat_thresholds();
6867 setup_per_zone_lowmem_reserve();
6870 core_initcall(init_per_zone_wmark_min
)
6873 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
6874 * that we can call two helper functions whenever min_free_kbytes
6877 int min_free_kbytes_sysctl_handler(struct ctl_table
*table
, int write
,
6878 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6882 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6887 user_min_free_kbytes
= min_free_kbytes
;
6888 setup_per_zone_wmarks();
6893 int watermark_scale_factor_sysctl_handler(struct ctl_table
*table
, int write
,
6894 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6898 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6903 setup_per_zone_wmarks();
6909 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6910 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6915 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6920 zone
->min_unmapped_pages
= (zone
->managed_pages
*
6921 sysctl_min_unmapped_ratio
) / 100;
6925 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6926 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6931 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6936 zone
->min_slab_pages
= (zone
->managed_pages
*
6937 sysctl_min_slab_ratio
) / 100;
6943 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6944 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6945 * whenever sysctl_lowmem_reserve_ratio changes.
6947 * The reserve ratio obviously has absolutely no relation with the
6948 * minimum watermarks. The lowmem reserve ratio can only make sense
6949 * if in function of the boot time zone sizes.
6951 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6952 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6954 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6955 setup_per_zone_lowmem_reserve();
6960 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
6961 * cpu. It is the fraction of total pages in each zone that a hot per cpu
6962 * pagelist can have before it gets flushed back to buddy allocator.
6964 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table
*table
, int write
,
6965 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6968 int old_percpu_pagelist_fraction
;
6971 mutex_lock(&pcp_batch_high_lock
);
6972 old_percpu_pagelist_fraction
= percpu_pagelist_fraction
;
6974 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6975 if (!write
|| ret
< 0)
6978 /* Sanity checking to avoid pcp imbalance */
6979 if (percpu_pagelist_fraction
&&
6980 percpu_pagelist_fraction
< MIN_PERCPU_PAGELIST_FRACTION
) {
6981 percpu_pagelist_fraction
= old_percpu_pagelist_fraction
;
6987 if (percpu_pagelist_fraction
== old_percpu_pagelist_fraction
)
6990 for_each_populated_zone(zone
) {
6993 for_each_possible_cpu(cpu
)
6994 pageset_set_high_and_batch(zone
,
6995 per_cpu_ptr(zone
->pageset
, cpu
));
6998 mutex_unlock(&pcp_batch_high_lock
);
7003 int hashdist
= HASHDIST_DEFAULT
;
7005 static int __init
set_hashdist(char *str
)
7009 hashdist
= simple_strtoul(str
, &str
, 0);
7012 __setup("hashdist=", set_hashdist
);
7016 * allocate a large system hash table from bootmem
7017 * - it is assumed that the hash table must contain an exact power-of-2
7018 * quantity of entries
7019 * - limit is the number of hash buckets, not the total allocation size
7021 void *__init
alloc_large_system_hash(const char *tablename
,
7022 unsigned long bucketsize
,
7023 unsigned long numentries
,
7026 unsigned int *_hash_shift
,
7027 unsigned int *_hash_mask
,
7028 unsigned long low_limit
,
7029 unsigned long high_limit
)
7031 unsigned long long max
= high_limit
;
7032 unsigned long log2qty
, size
;
7035 /* allow the kernel cmdline to have a say */
7037 /* round applicable memory size up to nearest megabyte */
7038 numentries
= nr_kernel_pages
;
7040 /* It isn't necessary when PAGE_SIZE >= 1MB */
7041 if (PAGE_SHIFT
< 20)
7042 numentries
= round_up(numentries
, (1<<20)/PAGE_SIZE
);
7044 /* limit to 1 bucket per 2^scale bytes of low memory */
7045 if (scale
> PAGE_SHIFT
)
7046 numentries
>>= (scale
- PAGE_SHIFT
);
7048 numentries
<<= (PAGE_SHIFT
- scale
);
7050 /* Make sure we've got at least a 0-order allocation.. */
7051 if (unlikely(flags
& HASH_SMALL
)) {
7052 /* Makes no sense without HASH_EARLY */
7053 WARN_ON(!(flags
& HASH_EARLY
));
7054 if (!(numentries
>> *_hash_shift
)) {
7055 numentries
= 1UL << *_hash_shift
;
7056 BUG_ON(!numentries
);
7058 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
7059 numentries
= PAGE_SIZE
/ bucketsize
;
7061 numentries
= roundup_pow_of_two(numentries
);
7063 /* limit allocation size to 1/16 total memory by default */
7065 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
7066 do_div(max
, bucketsize
);
7068 max
= min(max
, 0x80000000ULL
);
7070 if (numentries
< low_limit
)
7071 numentries
= low_limit
;
7072 if (numentries
> max
)
7075 log2qty
= ilog2(numentries
);
7078 size
= bucketsize
<< log2qty
;
7079 if (flags
& HASH_EARLY
)
7080 table
= memblock_virt_alloc_nopanic(size
, 0);
7082 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
7085 * If bucketsize is not a power-of-two, we may free
7086 * some pages at the end of hash table which
7087 * alloc_pages_exact() automatically does
7089 if (get_order(size
) < MAX_ORDER
) {
7090 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
7091 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
7094 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
7097 panic("Failed to allocate %s hash table\n", tablename
);
7099 pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n",
7100 tablename
, 1UL << log2qty
, ilog2(size
) - PAGE_SHIFT
, size
);
7103 *_hash_shift
= log2qty
;
7105 *_hash_mask
= (1 << log2qty
) - 1;
7111 * This function checks whether pageblock includes unmovable pages or not.
7112 * If @count is not zero, it is okay to include less @count unmovable pages
7114 * PageLRU check without isolation or lru_lock could race so that
7115 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
7116 * expect this function should be exact.
7118 bool has_unmovable_pages(struct zone
*zone
, struct page
*page
, int count
,
7119 bool skip_hwpoisoned_pages
)
7121 unsigned long pfn
, iter
, found
;
7125 * For avoiding noise data, lru_add_drain_all() should be called
7126 * If ZONE_MOVABLE, the zone never contains unmovable pages
7128 if (zone_idx(zone
) == ZONE_MOVABLE
)
7130 mt
= get_pageblock_migratetype(page
);
7131 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
7134 pfn
= page_to_pfn(page
);
7135 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
7136 unsigned long check
= pfn
+ iter
;
7138 if (!pfn_valid_within(check
))
7141 page
= pfn_to_page(check
);
7144 * Hugepages are not in LRU lists, but they're movable.
7145 * We need not scan over tail pages bacause we don't
7146 * handle each tail page individually in migration.
7148 if (PageHuge(page
)) {
7149 iter
= round_up(iter
+ 1, 1<<compound_order(page
)) - 1;
7154 * We can't use page_count without pin a page
7155 * because another CPU can free compound page.
7156 * This check already skips compound tails of THP
7157 * because their page->_refcount is zero at all time.
7159 if (!page_ref_count(page
)) {
7160 if (PageBuddy(page
))
7161 iter
+= (1 << page_order(page
)) - 1;
7166 * The HWPoisoned page may be not in buddy system, and
7167 * page_count() is not 0.
7169 if (skip_hwpoisoned_pages
&& PageHWPoison(page
))
7175 * If there are RECLAIMABLE pages, we need to check
7176 * it. But now, memory offline itself doesn't call
7177 * shrink_node_slabs() and it still to be fixed.
7180 * If the page is not RAM, page_count()should be 0.
7181 * we don't need more check. This is an _used_ not-movable page.
7183 * The problematic thing here is PG_reserved pages. PG_reserved
7184 * is set to both of a memory hole page and a _used_ kernel
7193 bool is_pageblock_removable_nolock(struct page
*page
)
7199 * We have to be careful here because we are iterating over memory
7200 * sections which are not zone aware so we might end up outside of
7201 * the zone but still within the section.
7202 * We have to take care about the node as well. If the node is offline
7203 * its NODE_DATA will be NULL - see page_zone.
7205 if (!node_online(page_to_nid(page
)))
7208 zone
= page_zone(page
);
7209 pfn
= page_to_pfn(page
);
7210 if (!zone_spans_pfn(zone
, pfn
))
7213 return !has_unmovable_pages(zone
, page
, 0, true);
7216 #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
7218 static unsigned long pfn_max_align_down(unsigned long pfn
)
7220 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
7221 pageblock_nr_pages
) - 1);
7224 static unsigned long pfn_max_align_up(unsigned long pfn
)
7226 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
7227 pageblock_nr_pages
));
7230 /* [start, end) must belong to a single zone. */
7231 static int __alloc_contig_migrate_range(struct compact_control
*cc
,
7232 unsigned long start
, unsigned long end
)
7234 /* This function is based on compact_zone() from compaction.c. */
7235 unsigned long nr_reclaimed
;
7236 unsigned long pfn
= start
;
7237 unsigned int tries
= 0;
7242 while (pfn
< end
|| !list_empty(&cc
->migratepages
)) {
7243 if (fatal_signal_pending(current
)) {
7248 if (list_empty(&cc
->migratepages
)) {
7249 cc
->nr_migratepages
= 0;
7250 pfn
= isolate_migratepages_range(cc
, pfn
, end
);
7256 } else if (++tries
== 5) {
7257 ret
= ret
< 0 ? ret
: -EBUSY
;
7261 nr_reclaimed
= reclaim_clean_pages_from_list(cc
->zone
,
7263 cc
->nr_migratepages
-= nr_reclaimed
;
7265 ret
= migrate_pages(&cc
->migratepages
, alloc_migrate_target
,
7266 NULL
, 0, cc
->mode
, MR_CMA
);
7269 putback_movable_pages(&cc
->migratepages
);
7276 * alloc_contig_range() -- tries to allocate given range of pages
7277 * @start: start PFN to allocate
7278 * @end: one-past-the-last PFN to allocate
7279 * @migratetype: migratetype of the underlaying pageblocks (either
7280 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
7281 * in range must have the same migratetype and it must
7282 * be either of the two.
7284 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
7285 * aligned, however it's the caller's responsibility to guarantee that
7286 * we are the only thread that changes migrate type of pageblocks the
7289 * The PFN range must belong to a single zone.
7291 * Returns zero on success or negative error code. On success all
7292 * pages which PFN is in [start, end) are allocated for the caller and
7293 * need to be freed with free_contig_range().
7295 int alloc_contig_range(unsigned long start
, unsigned long end
,
7296 unsigned migratetype
)
7298 unsigned long outer_start
, outer_end
;
7302 struct compact_control cc
= {
7303 .nr_migratepages
= 0,
7305 .zone
= page_zone(pfn_to_page(start
)),
7306 .mode
= MIGRATE_SYNC
,
7307 .ignore_skip_hint
= true,
7309 INIT_LIST_HEAD(&cc
.migratepages
);
7312 * What we do here is we mark all pageblocks in range as
7313 * MIGRATE_ISOLATE. Because pageblock and max order pages may
7314 * have different sizes, and due to the way page allocator
7315 * work, we align the range to biggest of the two pages so
7316 * that page allocator won't try to merge buddies from
7317 * different pageblocks and change MIGRATE_ISOLATE to some
7318 * other migration type.
7320 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
7321 * migrate the pages from an unaligned range (ie. pages that
7322 * we are interested in). This will put all the pages in
7323 * range back to page allocator as MIGRATE_ISOLATE.
7325 * When this is done, we take the pages in range from page
7326 * allocator removing them from the buddy system. This way
7327 * page allocator will never consider using them.
7329 * This lets us mark the pageblocks back as
7330 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
7331 * aligned range but not in the unaligned, original range are
7332 * put back to page allocator so that buddy can use them.
7335 ret
= start_isolate_page_range(pfn_max_align_down(start
),
7336 pfn_max_align_up(end
), migratetype
,
7342 * In case of -EBUSY, we'd like to know which page causes problem.
7343 * So, just fall through. We will check it in test_pages_isolated().
7345 ret
= __alloc_contig_migrate_range(&cc
, start
, end
);
7346 if (ret
&& ret
!= -EBUSY
)
7350 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
7351 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
7352 * more, all pages in [start, end) are free in page allocator.
7353 * What we are going to do is to allocate all pages from
7354 * [start, end) (that is remove them from page allocator).
7356 * The only problem is that pages at the beginning and at the
7357 * end of interesting range may be not aligned with pages that
7358 * page allocator holds, ie. they can be part of higher order
7359 * pages. Because of this, we reserve the bigger range and
7360 * once this is done free the pages we are not interested in.
7362 * We don't have to hold zone->lock here because the pages are
7363 * isolated thus they won't get removed from buddy.
7366 lru_add_drain_all();
7367 drain_all_pages(cc
.zone
);
7370 outer_start
= start
;
7371 while (!PageBuddy(pfn_to_page(outer_start
))) {
7372 if (++order
>= MAX_ORDER
) {
7373 outer_start
= start
;
7376 outer_start
&= ~0UL << order
;
7379 if (outer_start
!= start
) {
7380 order
= page_order(pfn_to_page(outer_start
));
7383 * outer_start page could be small order buddy page and
7384 * it doesn't include start page. Adjust outer_start
7385 * in this case to report failed page properly
7386 * on tracepoint in test_pages_isolated()
7388 if (outer_start
+ (1UL << order
) <= start
)
7389 outer_start
= start
;
7392 /* Make sure the range is really isolated. */
7393 if (test_pages_isolated(outer_start
, end
, false)) {
7394 pr_info("%s: [%lx, %lx) PFNs busy\n",
7395 __func__
, outer_start
, end
);
7400 /* Grab isolated pages from freelists. */
7401 outer_end
= isolate_freepages_range(&cc
, outer_start
, end
);
7407 /* Free head and tail (if any) */
7408 if (start
!= outer_start
)
7409 free_contig_range(outer_start
, start
- outer_start
);
7410 if (end
!= outer_end
)
7411 free_contig_range(end
, outer_end
- end
);
7414 undo_isolate_page_range(pfn_max_align_down(start
),
7415 pfn_max_align_up(end
), migratetype
);
7419 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
7421 unsigned int count
= 0;
7423 for (; nr_pages
--; pfn
++) {
7424 struct page
*page
= pfn_to_page(pfn
);
7426 count
+= page_count(page
) != 1;
7429 WARN(count
!= 0, "%d pages are still in use!\n", count
);
7433 #ifdef CONFIG_MEMORY_HOTPLUG
7435 * The zone indicated has a new number of managed_pages; batch sizes and percpu
7436 * page high values need to be recalulated.
7438 void __meminit
zone_pcp_update(struct zone
*zone
)
7441 mutex_lock(&pcp_batch_high_lock
);
7442 for_each_possible_cpu(cpu
)
7443 pageset_set_high_and_batch(zone
,
7444 per_cpu_ptr(zone
->pageset
, cpu
));
7445 mutex_unlock(&pcp_batch_high_lock
);
7449 void zone_pcp_reset(struct zone
*zone
)
7451 unsigned long flags
;
7453 struct per_cpu_pageset
*pset
;
7455 /* avoid races with drain_pages() */
7456 local_irq_save(flags
);
7457 if (zone
->pageset
!= &boot_pageset
) {
7458 for_each_online_cpu(cpu
) {
7459 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
7460 drain_zonestat(zone
, pset
);
7462 free_percpu(zone
->pageset
);
7463 zone
->pageset
= &boot_pageset
;
7465 local_irq_restore(flags
);
7468 #ifdef CONFIG_MEMORY_HOTREMOVE
7470 * All pages in the range must be in a single zone and isolated
7471 * before calling this.
7474 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
7478 unsigned int order
, i
;
7480 unsigned long flags
;
7481 /* find the first valid pfn */
7482 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
7487 zone
= page_zone(pfn_to_page(pfn
));
7488 spin_lock_irqsave(&zone
->lock
, flags
);
7490 while (pfn
< end_pfn
) {
7491 if (!pfn_valid(pfn
)) {
7495 page
= pfn_to_page(pfn
);
7497 * The HWPoisoned page may be not in buddy system, and
7498 * page_count() is not 0.
7500 if (unlikely(!PageBuddy(page
) && PageHWPoison(page
))) {
7502 SetPageReserved(page
);
7506 BUG_ON(page_count(page
));
7507 BUG_ON(!PageBuddy(page
));
7508 order
= page_order(page
);
7509 #ifdef CONFIG_DEBUG_VM
7510 pr_info("remove from free list %lx %d %lx\n",
7511 pfn
, 1 << order
, end_pfn
);
7513 list_del(&page
->lru
);
7514 rmv_page_order(page
);
7515 zone
->free_area
[order
].nr_free
--;
7516 for (i
= 0; i
< (1 << order
); i
++)
7517 SetPageReserved((page
+i
));
7518 pfn
+= (1 << order
);
7520 spin_unlock_irqrestore(&zone
->lock
, flags
);
7524 bool is_free_buddy_page(struct page
*page
)
7526 struct zone
*zone
= page_zone(page
);
7527 unsigned long pfn
= page_to_pfn(page
);
7528 unsigned long flags
;
7531 spin_lock_irqsave(&zone
->lock
, flags
);
7532 for (order
= 0; order
< MAX_ORDER
; order
++) {
7533 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
7535 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
7538 spin_unlock_irqrestore(&zone
->lock
, flags
);
7540 return order
< MAX_ORDER
;