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 <trace/events/oom.h>
59 #include <linux/prefetch.h>
60 #include <linux/mm_inline.h>
61 #include <linux/migrate.h>
62 #include <linux/page_ext.h>
63 #include <linux/hugetlb.h>
64 #include <linux/sched/rt.h>
65 #include <linux/page_owner.h>
66 #include <linux/kthread.h>
67 #include <linux/memcontrol.h>
69 #include <asm/sections.h>
70 #include <asm/tlbflush.h>
71 #include <asm/div64.h>
74 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
75 static DEFINE_MUTEX(pcp_batch_high_lock
);
76 #define MIN_PERCPU_PAGELIST_FRACTION (8)
78 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
79 DEFINE_PER_CPU(int, numa_node
);
80 EXPORT_PER_CPU_SYMBOL(numa_node
);
83 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
85 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
86 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
87 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
88 * defined in <linux/topology.h>.
90 DEFINE_PER_CPU(int, _numa_mem_
); /* Kernel "local memory" node */
91 EXPORT_PER_CPU_SYMBOL(_numa_mem_
);
92 int _node_numa_mem_
[MAX_NUMNODES
];
95 #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY
96 volatile unsigned long latent_entropy __latent_entropy
;
97 EXPORT_SYMBOL(latent_entropy
);
101 * Array of node states.
103 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
104 [N_POSSIBLE
] = NODE_MASK_ALL
,
105 [N_ONLINE
] = { { [0] = 1UL } },
107 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
108 #ifdef CONFIG_HIGHMEM
109 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
111 #ifdef CONFIG_MOVABLE_NODE
112 [N_MEMORY
] = { { [0] = 1UL } },
114 [N_CPU
] = { { [0] = 1UL } },
117 EXPORT_SYMBOL(node_states
);
119 /* Protect totalram_pages and zone->managed_pages */
120 static DEFINE_SPINLOCK(managed_page_count_lock
);
122 unsigned long totalram_pages __read_mostly
;
123 unsigned long totalreserve_pages __read_mostly
;
124 unsigned long totalcma_pages __read_mostly
;
126 int percpu_pagelist_fraction
;
127 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
130 * A cached value of the page's pageblock's migratetype, used when the page is
131 * put on a pcplist. Used to avoid the pageblock migratetype lookup when
132 * freeing from pcplists in most cases, at the cost of possibly becoming stale.
133 * Also the migratetype set in the page does not necessarily match the pcplist
134 * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
135 * other index - this ensures that it will be put on the correct CMA freelist.
137 static inline int get_pcppage_migratetype(struct page
*page
)
142 static inline void set_pcppage_migratetype(struct page
*page
, int migratetype
)
144 page
->index
= migratetype
;
147 #ifdef CONFIG_PM_SLEEP
149 * The following functions are used by the suspend/hibernate code to temporarily
150 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
151 * while devices are suspended. To avoid races with the suspend/hibernate code,
152 * they should always be called with pm_mutex held (gfp_allowed_mask also should
153 * only be modified with pm_mutex held, unless the suspend/hibernate code is
154 * guaranteed not to run in parallel with that modification).
157 static gfp_t saved_gfp_mask
;
159 void pm_restore_gfp_mask(void)
161 WARN_ON(!mutex_is_locked(&pm_mutex
));
162 if (saved_gfp_mask
) {
163 gfp_allowed_mask
= saved_gfp_mask
;
168 void pm_restrict_gfp_mask(void)
170 WARN_ON(!mutex_is_locked(&pm_mutex
));
171 WARN_ON(saved_gfp_mask
);
172 saved_gfp_mask
= gfp_allowed_mask
;
173 gfp_allowed_mask
&= ~(__GFP_IO
| __GFP_FS
);
176 bool pm_suspended_storage(void)
178 if ((gfp_allowed_mask
& (__GFP_IO
| __GFP_FS
)) == (__GFP_IO
| __GFP_FS
))
182 #endif /* CONFIG_PM_SLEEP */
184 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
185 unsigned int pageblock_order __read_mostly
;
188 static void __free_pages_ok(struct page
*page
, unsigned int order
);
191 * results with 256, 32 in the lowmem_reserve sysctl:
192 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
193 * 1G machine -> (16M dma, 784M normal, 224M high)
194 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
195 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
196 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
198 * TBD: should special case ZONE_DMA32 machines here - in those we normally
199 * don't need any ZONE_NORMAL reservation
201 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
202 #ifdef CONFIG_ZONE_DMA
205 #ifdef CONFIG_ZONE_DMA32
208 #ifdef CONFIG_HIGHMEM
214 EXPORT_SYMBOL(totalram_pages
);
216 static char * const zone_names
[MAX_NR_ZONES
] = {
217 #ifdef CONFIG_ZONE_DMA
220 #ifdef CONFIG_ZONE_DMA32
224 #ifdef CONFIG_HIGHMEM
228 #ifdef CONFIG_ZONE_DEVICE
233 char * const migratetype_names
[MIGRATE_TYPES
] = {
241 #ifdef CONFIG_MEMORY_ISOLATION
246 compound_page_dtor
* const compound_page_dtors
[] = {
249 #ifdef CONFIG_HUGETLB_PAGE
252 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
257 int min_free_kbytes
= 1024;
258 int user_min_free_kbytes
= -1;
259 int watermark_scale_factor
= 10;
261 static unsigned long __meminitdata nr_kernel_pages
;
262 static unsigned long __meminitdata nr_all_pages
;
263 static unsigned long __meminitdata dma_reserve
;
265 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
266 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
267 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
268 static unsigned long __initdata required_kernelcore
;
269 static unsigned long __initdata required_movablecore
;
270 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
271 static bool mirrored_kernelcore
;
273 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
275 EXPORT_SYMBOL(movable_zone
);
276 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
279 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
280 int nr_online_nodes __read_mostly
= 1;
281 EXPORT_SYMBOL(nr_node_ids
);
282 EXPORT_SYMBOL(nr_online_nodes
);
285 int page_group_by_mobility_disabled __read_mostly
;
287 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
288 static inline void reset_deferred_meminit(pg_data_t
*pgdat
)
290 pgdat
->first_deferred_pfn
= ULONG_MAX
;
293 /* Returns true if the struct page for the pfn is uninitialised */
294 static inline bool __meminit
early_page_uninitialised(unsigned long pfn
)
296 int nid
= early_pfn_to_nid(pfn
);
298 if (node_online(nid
) && pfn
>= NODE_DATA(nid
)->first_deferred_pfn
)
305 * Returns false when the remaining initialisation should be deferred until
306 * later in the boot cycle when it can be parallelised.
308 static inline bool update_defer_init(pg_data_t
*pgdat
,
309 unsigned long pfn
, unsigned long zone_end
,
310 unsigned long *nr_initialised
)
312 unsigned long max_initialise
;
314 /* Always populate low zones for address-contrained allocations */
315 if (zone_end
< pgdat_end_pfn(pgdat
))
318 * Initialise at least 2G of a node but also take into account that
319 * two large system hashes that can take up 1GB for 0.25TB/node.
321 max_initialise
= max(2UL << (30 - PAGE_SHIFT
),
322 (pgdat
->node_spanned_pages
>> 8));
325 if ((*nr_initialised
> max_initialise
) &&
326 (pfn
& (PAGES_PER_SECTION
- 1)) == 0) {
327 pgdat
->first_deferred_pfn
= pfn
;
334 static inline void reset_deferred_meminit(pg_data_t
*pgdat
)
338 static inline bool early_page_uninitialised(unsigned long pfn
)
343 static inline bool update_defer_init(pg_data_t
*pgdat
,
344 unsigned long pfn
, unsigned long zone_end
,
345 unsigned long *nr_initialised
)
351 /* Return a pointer to the bitmap storing bits affecting a block of pages */
352 static inline unsigned long *get_pageblock_bitmap(struct page
*page
,
355 #ifdef CONFIG_SPARSEMEM
356 return __pfn_to_section(pfn
)->pageblock_flags
;
358 return page_zone(page
)->pageblock_flags
;
359 #endif /* CONFIG_SPARSEMEM */
362 static inline int pfn_to_bitidx(struct page
*page
, unsigned long pfn
)
364 #ifdef CONFIG_SPARSEMEM
365 pfn
&= (PAGES_PER_SECTION
-1);
366 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
368 pfn
= pfn
- round_down(page_zone(page
)->zone_start_pfn
, pageblock_nr_pages
);
369 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
370 #endif /* CONFIG_SPARSEMEM */
374 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
375 * @page: The page within the block of interest
376 * @pfn: The target page frame number
377 * @end_bitidx: The last bit of interest to retrieve
378 * @mask: mask of bits that the caller is interested in
380 * Return: pageblock_bits flags
382 static __always_inline
unsigned long __get_pfnblock_flags_mask(struct page
*page
,
384 unsigned long end_bitidx
,
387 unsigned long *bitmap
;
388 unsigned long bitidx
, word_bitidx
;
391 bitmap
= get_pageblock_bitmap(page
, pfn
);
392 bitidx
= pfn_to_bitidx(page
, pfn
);
393 word_bitidx
= bitidx
/ BITS_PER_LONG
;
394 bitidx
&= (BITS_PER_LONG
-1);
396 word
= bitmap
[word_bitidx
];
397 bitidx
+= end_bitidx
;
398 return (word
>> (BITS_PER_LONG
- bitidx
- 1)) & mask
;
401 unsigned long get_pfnblock_flags_mask(struct page
*page
, unsigned long pfn
,
402 unsigned long end_bitidx
,
405 return __get_pfnblock_flags_mask(page
, pfn
, end_bitidx
, mask
);
408 static __always_inline
int get_pfnblock_migratetype(struct page
*page
, unsigned long pfn
)
410 return __get_pfnblock_flags_mask(page
, pfn
, PB_migrate_end
, MIGRATETYPE_MASK
);
414 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
415 * @page: The page within the block of interest
416 * @flags: The flags to set
417 * @pfn: The target page frame number
418 * @end_bitidx: The last bit of interest
419 * @mask: mask of bits that the caller is interested in
421 void set_pfnblock_flags_mask(struct page
*page
, unsigned long flags
,
423 unsigned long end_bitidx
,
426 unsigned long *bitmap
;
427 unsigned long bitidx
, word_bitidx
;
428 unsigned long old_word
, word
;
430 BUILD_BUG_ON(NR_PAGEBLOCK_BITS
!= 4);
432 bitmap
= get_pageblock_bitmap(page
, pfn
);
433 bitidx
= pfn_to_bitidx(page
, pfn
);
434 word_bitidx
= bitidx
/ BITS_PER_LONG
;
435 bitidx
&= (BITS_PER_LONG
-1);
437 VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page
), pfn
), page
);
439 bitidx
+= end_bitidx
;
440 mask
<<= (BITS_PER_LONG
- bitidx
- 1);
441 flags
<<= (BITS_PER_LONG
- bitidx
- 1);
443 word
= READ_ONCE(bitmap
[word_bitidx
]);
445 old_word
= cmpxchg(&bitmap
[word_bitidx
], word
, (word
& ~mask
) | flags
);
446 if (word
== old_word
)
452 void set_pageblock_migratetype(struct page
*page
, int migratetype
)
454 if (unlikely(page_group_by_mobility_disabled
&&
455 migratetype
< MIGRATE_PCPTYPES
))
456 migratetype
= MIGRATE_UNMOVABLE
;
458 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
459 PB_migrate
, PB_migrate_end
);
462 #ifdef CONFIG_DEBUG_VM
463 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
467 unsigned long pfn
= page_to_pfn(page
);
468 unsigned long sp
, start_pfn
;
471 seq
= zone_span_seqbegin(zone
);
472 start_pfn
= zone
->zone_start_pfn
;
473 sp
= zone
->spanned_pages
;
474 if (!zone_spans_pfn(zone
, pfn
))
476 } while (zone_span_seqretry(zone
, seq
));
479 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
480 pfn
, zone_to_nid(zone
), zone
->name
,
481 start_pfn
, start_pfn
+ sp
);
486 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
488 if (!pfn_valid_within(page_to_pfn(page
)))
490 if (zone
!= page_zone(page
))
496 * Temporary debugging check for pages not lying within a given zone.
498 static int bad_range(struct zone
*zone
, struct page
*page
)
500 if (page_outside_zone_boundaries(zone
, page
))
502 if (!page_is_consistent(zone
, page
))
508 static inline int bad_range(struct zone
*zone
, struct page
*page
)
514 static void bad_page(struct page
*page
, const char *reason
,
515 unsigned long bad_flags
)
517 static unsigned long resume
;
518 static unsigned long nr_shown
;
519 static unsigned long nr_unshown
;
522 * Allow a burst of 60 reports, then keep quiet for that minute;
523 * or allow a steady drip of one report per second.
525 if (nr_shown
== 60) {
526 if (time_before(jiffies
, resume
)) {
532 "BUG: Bad page state: %lu messages suppressed\n",
539 resume
= jiffies
+ 60 * HZ
;
541 pr_alert("BUG: Bad page state in process %s pfn:%05lx\n",
542 current
->comm
, page_to_pfn(page
));
543 __dump_page(page
, reason
);
544 bad_flags
&= page
->flags
;
546 pr_alert("bad because of flags: %#lx(%pGp)\n",
547 bad_flags
, &bad_flags
);
548 dump_page_owner(page
);
553 /* Leave bad fields for debug, except PageBuddy could make trouble */
554 page_mapcount_reset(page
); /* remove PageBuddy */
555 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
559 * Higher-order pages are called "compound pages". They are structured thusly:
561 * The first PAGE_SIZE page is called the "head page" and have PG_head set.
563 * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
564 * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
566 * The first tail page's ->compound_dtor holds the offset in array of compound
567 * page destructors. See compound_page_dtors.
569 * The first tail page's ->compound_order holds the order of allocation.
570 * This usage means that zero-order pages may not be compound.
573 void free_compound_page(struct page
*page
)
575 __free_pages_ok(page
, compound_order(page
));
578 void prep_compound_page(struct page
*page
, unsigned int order
)
581 int nr_pages
= 1 << order
;
583 set_compound_page_dtor(page
, COMPOUND_PAGE_DTOR
);
584 set_compound_order(page
, order
);
586 for (i
= 1; i
< nr_pages
; i
++) {
587 struct page
*p
= page
+ i
;
588 set_page_count(p
, 0);
589 p
->mapping
= TAIL_MAPPING
;
590 set_compound_head(p
, page
);
592 atomic_set(compound_mapcount_ptr(page
), -1);
595 #ifdef CONFIG_DEBUG_PAGEALLOC
596 unsigned int _debug_guardpage_minorder
;
597 bool _debug_pagealloc_enabled __read_mostly
598 = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT
);
599 EXPORT_SYMBOL(_debug_pagealloc_enabled
);
600 bool _debug_guardpage_enabled __read_mostly
;
602 static int __init
early_debug_pagealloc(char *buf
)
606 return kstrtobool(buf
, &_debug_pagealloc_enabled
);
608 early_param("debug_pagealloc", early_debug_pagealloc
);
610 static bool need_debug_guardpage(void)
612 /* If we don't use debug_pagealloc, we don't need guard page */
613 if (!debug_pagealloc_enabled())
616 if (!debug_guardpage_minorder())
622 static void init_debug_guardpage(void)
624 if (!debug_pagealloc_enabled())
627 if (!debug_guardpage_minorder())
630 _debug_guardpage_enabled
= true;
633 struct page_ext_operations debug_guardpage_ops
= {
634 .need
= need_debug_guardpage
,
635 .init
= init_debug_guardpage
,
638 static int __init
debug_guardpage_minorder_setup(char *buf
)
642 if (kstrtoul(buf
, 10, &res
) < 0 || res
> MAX_ORDER
/ 2) {
643 pr_err("Bad debug_guardpage_minorder value\n");
646 _debug_guardpage_minorder
= res
;
647 pr_info("Setting debug_guardpage_minorder to %lu\n", res
);
650 early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup
);
652 static inline bool set_page_guard(struct zone
*zone
, struct page
*page
,
653 unsigned int order
, int migratetype
)
655 struct page_ext
*page_ext
;
657 if (!debug_guardpage_enabled())
660 if (order
>= debug_guardpage_minorder())
663 page_ext
= lookup_page_ext(page
);
664 if (unlikely(!page_ext
))
667 __set_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
669 INIT_LIST_HEAD(&page
->lru
);
670 set_page_private(page
, order
);
671 /* Guard pages are not available for any usage */
672 __mod_zone_freepage_state(zone
, -(1 << order
), migratetype
);
677 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
678 unsigned int order
, int migratetype
)
680 struct page_ext
*page_ext
;
682 if (!debug_guardpage_enabled())
685 page_ext
= lookup_page_ext(page
);
686 if (unlikely(!page_ext
))
689 __clear_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
691 set_page_private(page
, 0);
692 if (!is_migrate_isolate(migratetype
))
693 __mod_zone_freepage_state(zone
, (1 << order
), migratetype
);
696 struct page_ext_operations debug_guardpage_ops
;
697 static inline bool set_page_guard(struct zone
*zone
, struct page
*page
,
698 unsigned int order
, int migratetype
) { return false; }
699 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
700 unsigned int order
, int migratetype
) {}
703 static inline void set_page_order(struct page
*page
, unsigned int order
)
705 set_page_private(page
, order
);
706 __SetPageBuddy(page
);
709 static inline void rmv_page_order(struct page
*page
)
711 __ClearPageBuddy(page
);
712 set_page_private(page
, 0);
716 * This function checks whether a page is free && is the buddy
717 * we can do coalesce a page and its buddy if
718 * (a) the buddy is not in a hole (check before calling!) &&
719 * (b) the buddy is in the buddy system &&
720 * (c) a page and its buddy have the same order &&
721 * (d) a page and its buddy are in the same zone.
723 * For recording whether a page is in the buddy system, we set ->_mapcount
724 * PAGE_BUDDY_MAPCOUNT_VALUE.
725 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
726 * serialized by zone->lock.
728 * For recording page's order, we use page_private(page).
730 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
733 if (page_is_guard(buddy
) && page_order(buddy
) == order
) {
734 if (page_zone_id(page
) != page_zone_id(buddy
))
737 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
742 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
744 * zone check is done late to avoid uselessly
745 * calculating zone/node ids for pages that could
748 if (page_zone_id(page
) != page_zone_id(buddy
))
751 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
759 * Freeing function for a buddy system allocator.
761 * The concept of a buddy system is to maintain direct-mapped table
762 * (containing bit values) for memory blocks of various "orders".
763 * The bottom level table contains the map for the smallest allocatable
764 * units of memory (here, pages), and each level above it describes
765 * pairs of units from the levels below, hence, "buddies".
766 * At a high level, all that happens here is marking the table entry
767 * at the bottom level available, and propagating the changes upward
768 * as necessary, plus some accounting needed to play nicely with other
769 * parts of the VM system.
770 * At each level, we keep a list of pages, which are heads of continuous
771 * free pages of length of (1 << order) and marked with _mapcount
772 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
774 * So when we are allocating or freeing one, we can derive the state of the
775 * other. That is, if we allocate a small block, and both were
776 * free, the remainder of the region must be split into blocks.
777 * If a block is freed, and its buddy is also free, then this
778 * triggers coalescing into a block of larger size.
783 static inline void __free_one_page(struct page
*page
,
785 struct zone
*zone
, unsigned int order
,
788 unsigned long combined_pfn
;
789 unsigned long uninitialized_var(buddy_pfn
);
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 VM_BUG_ON_PAGE(pfn
& ((1 << order
) - 1), page
);
803 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
806 while (order
< max_order
- 1) {
807 buddy_pfn
= __find_buddy_pfn(pfn
, order
);
808 buddy
= page
+ (buddy_pfn
- pfn
);
810 if (!pfn_valid_within(buddy_pfn
))
812 if (!page_is_buddy(page
, buddy
, order
))
815 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
816 * merge with it and move up one order.
818 if (page_is_guard(buddy
)) {
819 clear_page_guard(zone
, buddy
, order
, migratetype
);
821 list_del(&buddy
->lru
);
822 zone
->free_area
[order
].nr_free
--;
823 rmv_page_order(buddy
);
825 combined_pfn
= buddy_pfn
& pfn
;
826 page
= page
+ (combined_pfn
- pfn
);
830 if (max_order
< MAX_ORDER
) {
831 /* If we are here, it means order is >= pageblock_order.
832 * We want to prevent merge between freepages on isolate
833 * pageblock and normal pageblock. Without this, pageblock
834 * isolation could cause incorrect freepage or CMA accounting.
836 * We don't want to hit this code for the more frequent
839 if (unlikely(has_isolate_pageblock(zone
))) {
842 buddy_pfn
= __find_buddy_pfn(pfn
, order
);
843 buddy
= page
+ (buddy_pfn
- pfn
);
844 buddy_mt
= get_pageblock_migratetype(buddy
);
846 if (migratetype
!= buddy_mt
847 && (is_migrate_isolate(migratetype
) ||
848 is_migrate_isolate(buddy_mt
)))
852 goto continue_merging
;
856 set_page_order(page
, order
);
859 * If this is not the largest possible page, check if the buddy
860 * of the next-highest order is free. If it is, it's possible
861 * that pages are being freed that will coalesce soon. In case,
862 * that is happening, add the free page to the tail of the list
863 * so it's less likely to be used soon and more likely to be merged
864 * as a higher order page
866 if ((order
< MAX_ORDER
-2) && pfn_valid_within(buddy_pfn
)) {
867 struct page
*higher_page
, *higher_buddy
;
868 combined_pfn
= buddy_pfn
& pfn
;
869 higher_page
= page
+ (combined_pfn
- pfn
);
870 buddy_pfn
= __find_buddy_pfn(combined_pfn
, order
+ 1);
871 higher_buddy
= higher_page
+ (buddy_pfn
- combined_pfn
);
872 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
873 list_add_tail(&page
->lru
,
874 &zone
->free_area
[order
].free_list
[migratetype
]);
879 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
881 zone
->free_area
[order
].nr_free
++;
885 * A bad page could be due to a number of fields. Instead of multiple branches,
886 * try and check multiple fields with one check. The caller must do a detailed
887 * check if necessary.
889 static inline bool page_expected_state(struct page
*page
,
890 unsigned long check_flags
)
892 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
895 if (unlikely((unsigned long)page
->mapping
|
896 page_ref_count(page
) |
898 (unsigned long)page
->mem_cgroup
|
900 (page
->flags
& check_flags
)))
906 static void free_pages_check_bad(struct page
*page
)
908 const char *bad_reason
;
909 unsigned long bad_flags
;
914 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
915 bad_reason
= "nonzero mapcount";
916 if (unlikely(page
->mapping
!= NULL
))
917 bad_reason
= "non-NULL mapping";
918 if (unlikely(page_ref_count(page
) != 0))
919 bad_reason
= "nonzero _refcount";
920 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
)) {
921 bad_reason
= "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
922 bad_flags
= PAGE_FLAGS_CHECK_AT_FREE
;
925 if (unlikely(page
->mem_cgroup
))
926 bad_reason
= "page still charged to cgroup";
928 bad_page(page
, bad_reason
, bad_flags
);
931 static inline int free_pages_check(struct page
*page
)
933 if (likely(page_expected_state(page
, PAGE_FLAGS_CHECK_AT_FREE
)))
936 /* Something has gone sideways, find it */
937 free_pages_check_bad(page
);
941 static int free_tail_pages_check(struct page
*head_page
, struct page
*page
)
946 * We rely page->lru.next never has bit 0 set, unless the page
947 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
949 BUILD_BUG_ON((unsigned long)LIST_POISON1
& 1);
951 if (!IS_ENABLED(CONFIG_DEBUG_VM
)) {
955 switch (page
- head_page
) {
957 /* the first tail page: ->mapping is compound_mapcount() */
958 if (unlikely(compound_mapcount(page
))) {
959 bad_page(page
, "nonzero compound_mapcount", 0);
965 * the second tail page: ->mapping is
966 * page_deferred_list().next -- ignore value.
970 if (page
->mapping
!= TAIL_MAPPING
) {
971 bad_page(page
, "corrupted mapping in tail page", 0);
976 if (unlikely(!PageTail(page
))) {
977 bad_page(page
, "PageTail not set", 0);
980 if (unlikely(compound_head(page
) != head_page
)) {
981 bad_page(page
, "compound_head not consistent", 0);
986 page
->mapping
= NULL
;
987 clear_compound_head(page
);
991 static __always_inline
bool free_pages_prepare(struct page
*page
,
992 unsigned int order
, bool check_free
)
996 VM_BUG_ON_PAGE(PageTail(page
), page
);
998 trace_mm_page_free(page
, order
);
999 kmemcheck_free_shadow(page
, order
);
1002 * Check tail pages before head page information is cleared to
1003 * avoid checking PageCompound for order-0 pages.
1005 if (unlikely(order
)) {
1006 bool compound
= PageCompound(page
);
1009 VM_BUG_ON_PAGE(compound
&& compound_order(page
) != order
, page
);
1012 ClearPageDoubleMap(page
);
1013 for (i
= 1; i
< (1 << order
); i
++) {
1015 bad
+= free_tail_pages_check(page
, page
+ i
);
1016 if (unlikely(free_pages_check(page
+ i
))) {
1020 (page
+ i
)->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
1023 if (PageMappingFlags(page
))
1024 page
->mapping
= NULL
;
1025 if (memcg_kmem_enabled() && PageKmemcg(page
))
1026 memcg_kmem_uncharge(page
, order
);
1028 bad
+= free_pages_check(page
);
1032 page_cpupid_reset_last(page
);
1033 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
1034 reset_page_owner(page
, order
);
1036 if (!PageHighMem(page
)) {
1037 debug_check_no_locks_freed(page_address(page
),
1038 PAGE_SIZE
<< order
);
1039 debug_check_no_obj_freed(page_address(page
),
1040 PAGE_SIZE
<< order
);
1042 arch_free_page(page
, order
);
1043 kernel_poison_pages(page
, 1 << order
, 0);
1044 kernel_map_pages(page
, 1 << order
, 0);
1045 kasan_free_pages(page
, order
);
1050 #ifdef CONFIG_DEBUG_VM
1051 static inline bool free_pcp_prepare(struct page
*page
)
1053 return free_pages_prepare(page
, 0, true);
1056 static inline bool bulkfree_pcp_prepare(struct page
*page
)
1061 static bool free_pcp_prepare(struct page
*page
)
1063 return free_pages_prepare(page
, 0, false);
1066 static bool bulkfree_pcp_prepare(struct page
*page
)
1068 return free_pages_check(page
);
1070 #endif /* CONFIG_DEBUG_VM */
1073 * Frees a number of pages from the PCP lists
1074 * Assumes all pages on list are in same zone, and of same order.
1075 * count is the number of pages to free.
1077 * If the zone was previously in an "all pages pinned" state then look to
1078 * see if this freeing clears that state.
1080 * And clear the zone's pages_scanned counter, to hold off the "all pages are
1081 * pinned" detection logic.
1083 static void free_pcppages_bulk(struct zone
*zone
, int count
,
1084 struct per_cpu_pages
*pcp
)
1086 int migratetype
= 0;
1088 unsigned long nr_scanned
, flags
;
1089 bool isolated_pageblocks
;
1091 spin_lock_irqsave(&zone
->lock
, flags
);
1092 isolated_pageblocks
= has_isolate_pageblock(zone
);
1093 nr_scanned
= node_page_state(zone
->zone_pgdat
, NR_PAGES_SCANNED
);
1095 __mod_node_page_state(zone
->zone_pgdat
, NR_PAGES_SCANNED
, -nr_scanned
);
1099 struct list_head
*list
;
1102 * Remove pages from lists in a round-robin fashion. A
1103 * batch_free count is maintained that is incremented when an
1104 * empty list is encountered. This is so more pages are freed
1105 * off fuller lists instead of spinning excessively around empty
1110 if (++migratetype
== MIGRATE_PCPTYPES
)
1112 list
= &pcp
->lists
[migratetype
];
1113 } while (list_empty(list
));
1115 /* This is the only non-empty list. Free them all. */
1116 if (batch_free
== MIGRATE_PCPTYPES
)
1120 int mt
; /* migratetype of the to-be-freed page */
1122 page
= list_last_entry(list
, struct page
, lru
);
1123 /* must delete as __free_one_page list manipulates */
1124 list_del(&page
->lru
);
1126 mt
= get_pcppage_migratetype(page
);
1127 /* MIGRATE_ISOLATE page should not go to pcplists */
1128 VM_BUG_ON_PAGE(is_migrate_isolate(mt
), page
);
1129 /* Pageblock could have been isolated meanwhile */
1130 if (unlikely(isolated_pageblocks
))
1131 mt
= get_pageblock_migratetype(page
);
1133 if (bulkfree_pcp_prepare(page
))
1136 __free_one_page(page
, page_to_pfn(page
), zone
, 0, mt
);
1137 trace_mm_page_pcpu_drain(page
, 0, mt
);
1138 } while (--count
&& --batch_free
&& !list_empty(list
));
1140 spin_unlock_irqrestore(&zone
->lock
, flags
);
1143 static void free_one_page(struct zone
*zone
,
1144 struct page
*page
, unsigned long pfn
,
1148 unsigned long nr_scanned
, flags
;
1149 spin_lock_irqsave(&zone
->lock
, flags
);
1150 __count_vm_events(PGFREE
, 1 << order
);
1151 nr_scanned
= node_page_state(zone
->zone_pgdat
, NR_PAGES_SCANNED
);
1153 __mod_node_page_state(zone
->zone_pgdat
, NR_PAGES_SCANNED
, -nr_scanned
);
1155 if (unlikely(has_isolate_pageblock(zone
) ||
1156 is_migrate_isolate(migratetype
))) {
1157 migratetype
= get_pfnblock_migratetype(page
, pfn
);
1159 __free_one_page(page
, pfn
, zone
, order
, migratetype
);
1160 spin_unlock_irqrestore(&zone
->lock
, flags
);
1163 static void __meminit
__init_single_page(struct page
*page
, unsigned long pfn
,
1164 unsigned long zone
, int nid
)
1166 set_page_links(page
, zone
, nid
, pfn
);
1167 init_page_count(page
);
1168 page_mapcount_reset(page
);
1169 page_cpupid_reset_last(page
);
1171 INIT_LIST_HEAD(&page
->lru
);
1172 #ifdef WANT_PAGE_VIRTUAL
1173 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1174 if (!is_highmem_idx(zone
))
1175 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
1179 static void __meminit
__init_single_pfn(unsigned long pfn
, unsigned long zone
,
1182 return __init_single_page(pfn_to_page(pfn
), pfn
, zone
, nid
);
1185 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1186 static void init_reserved_page(unsigned long pfn
)
1191 if (!early_page_uninitialised(pfn
))
1194 nid
= early_pfn_to_nid(pfn
);
1195 pgdat
= NODE_DATA(nid
);
1197 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
1198 struct zone
*zone
= &pgdat
->node_zones
[zid
];
1200 if (pfn
>= zone
->zone_start_pfn
&& pfn
< zone_end_pfn(zone
))
1203 __init_single_pfn(pfn
, zid
, nid
);
1206 static inline void init_reserved_page(unsigned long pfn
)
1209 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1212 * Initialised pages do not have PageReserved set. This function is
1213 * called for each range allocated by the bootmem allocator and
1214 * marks the pages PageReserved. The remaining valid pages are later
1215 * sent to the buddy page allocator.
1217 void __meminit
reserve_bootmem_region(phys_addr_t start
, phys_addr_t end
)
1219 unsigned long start_pfn
= PFN_DOWN(start
);
1220 unsigned long end_pfn
= PFN_UP(end
);
1222 for (; start_pfn
< end_pfn
; start_pfn
++) {
1223 if (pfn_valid(start_pfn
)) {
1224 struct page
*page
= pfn_to_page(start_pfn
);
1226 init_reserved_page(start_pfn
);
1228 /* Avoid false-positive PageTail() */
1229 INIT_LIST_HEAD(&page
->lru
);
1231 SetPageReserved(page
);
1236 static void __free_pages_ok(struct page
*page
, unsigned int order
)
1239 unsigned long pfn
= page_to_pfn(page
);
1241 if (!free_pages_prepare(page
, order
, true))
1244 migratetype
= get_pfnblock_migratetype(page
, pfn
);
1245 free_one_page(page_zone(page
), page
, pfn
, order
, migratetype
);
1248 static void __init
__free_pages_boot_core(struct page
*page
, unsigned int order
)
1250 unsigned int nr_pages
= 1 << order
;
1251 struct page
*p
= page
;
1255 for (loop
= 0; loop
< (nr_pages
- 1); loop
++, p
++) {
1257 __ClearPageReserved(p
);
1258 set_page_count(p
, 0);
1260 __ClearPageReserved(p
);
1261 set_page_count(p
, 0);
1263 page_zone(page
)->managed_pages
+= nr_pages
;
1264 set_page_refcounted(page
);
1265 __free_pages(page
, order
);
1268 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1269 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1271 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata
;
1273 int __meminit
early_pfn_to_nid(unsigned long pfn
)
1275 static DEFINE_SPINLOCK(early_pfn_lock
);
1278 spin_lock(&early_pfn_lock
);
1279 nid
= __early_pfn_to_nid(pfn
, &early_pfnnid_cache
);
1281 nid
= first_online_node
;
1282 spin_unlock(&early_pfn_lock
);
1288 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1289 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1290 struct mminit_pfnnid_cache
*state
)
1294 nid
= __early_pfn_to_nid(pfn
, state
);
1295 if (nid
>= 0 && nid
!= node
)
1300 /* Only safe to use early in boot when initialisation is single-threaded */
1301 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1303 return meminit_pfn_in_nid(pfn
, node
, &early_pfnnid_cache
);
1308 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1312 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1313 struct mminit_pfnnid_cache
*state
)
1320 void __init
__free_pages_bootmem(struct page
*page
, unsigned long pfn
,
1323 if (early_page_uninitialised(pfn
))
1325 return __free_pages_boot_core(page
, order
);
1329 * Check that the whole (or subset of) a pageblock given by the interval of
1330 * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1331 * with the migration of free compaction scanner. The scanners then need to
1332 * use only pfn_valid_within() check for arches that allow holes within
1335 * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1337 * It's possible on some configurations to have a setup like node0 node1 node0
1338 * i.e. it's possible that all pages within a zones range of pages do not
1339 * belong to a single zone. We assume that a border between node0 and node1
1340 * can occur within a single pageblock, but not a node0 node1 node0
1341 * interleaving within a single pageblock. It is therefore sufficient to check
1342 * the first and last page of a pageblock and avoid checking each individual
1343 * page in a pageblock.
1345 struct page
*__pageblock_pfn_to_page(unsigned long start_pfn
,
1346 unsigned long end_pfn
, struct zone
*zone
)
1348 struct page
*start_page
;
1349 struct page
*end_page
;
1351 /* end_pfn is one past the range we are checking */
1354 if (!pfn_valid(start_pfn
) || !pfn_valid(end_pfn
))
1357 start_page
= pfn_to_page(start_pfn
);
1359 if (page_zone(start_page
) != zone
)
1362 end_page
= pfn_to_page(end_pfn
);
1364 /* This gives a shorter code than deriving page_zone(end_page) */
1365 if (page_zone_id(start_page
) != page_zone_id(end_page
))
1371 void set_zone_contiguous(struct zone
*zone
)
1373 unsigned long block_start_pfn
= zone
->zone_start_pfn
;
1374 unsigned long block_end_pfn
;
1376 block_end_pfn
= ALIGN(block_start_pfn
+ 1, pageblock_nr_pages
);
1377 for (; block_start_pfn
< zone_end_pfn(zone
);
1378 block_start_pfn
= block_end_pfn
,
1379 block_end_pfn
+= pageblock_nr_pages
) {
1381 block_end_pfn
= min(block_end_pfn
, zone_end_pfn(zone
));
1383 if (!__pageblock_pfn_to_page(block_start_pfn
,
1384 block_end_pfn
, zone
))
1388 /* We confirm that there is no hole */
1389 zone
->contiguous
= true;
1392 void clear_zone_contiguous(struct zone
*zone
)
1394 zone
->contiguous
= false;
1397 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1398 static void __init
deferred_free_range(struct page
*page
,
1399 unsigned long pfn
, int nr_pages
)
1406 /* Free a large naturally-aligned chunk if possible */
1407 if (nr_pages
== pageblock_nr_pages
&&
1408 (pfn
& (pageblock_nr_pages
- 1)) == 0) {
1409 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
1410 __free_pages_boot_core(page
, pageblock_order
);
1414 for (i
= 0; i
< nr_pages
; i
++, page
++, pfn
++) {
1415 if ((pfn
& (pageblock_nr_pages
- 1)) == 0)
1416 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
1417 __free_pages_boot_core(page
, 0);
1421 /* Completion tracking for deferred_init_memmap() threads */
1422 static atomic_t pgdat_init_n_undone __initdata
;
1423 static __initdata
DECLARE_COMPLETION(pgdat_init_all_done_comp
);
1425 static inline void __init
pgdat_init_report_one_done(void)
1427 if (atomic_dec_and_test(&pgdat_init_n_undone
))
1428 complete(&pgdat_init_all_done_comp
);
1431 /* Initialise remaining memory on a node */
1432 static int __init
deferred_init_memmap(void *data
)
1434 pg_data_t
*pgdat
= data
;
1435 int nid
= pgdat
->node_id
;
1436 struct mminit_pfnnid_cache nid_init_state
= { };
1437 unsigned long start
= jiffies
;
1438 unsigned long nr_pages
= 0;
1439 unsigned long walk_start
, walk_end
;
1442 unsigned long first_init_pfn
= pgdat
->first_deferred_pfn
;
1443 const struct cpumask
*cpumask
= cpumask_of_node(pgdat
->node_id
);
1445 if (first_init_pfn
== ULONG_MAX
) {
1446 pgdat_init_report_one_done();
1450 /* Bind memory initialisation thread to a local node if possible */
1451 if (!cpumask_empty(cpumask
))
1452 set_cpus_allowed_ptr(current
, cpumask
);
1454 /* Sanity check boundaries */
1455 BUG_ON(pgdat
->first_deferred_pfn
< pgdat
->node_start_pfn
);
1456 BUG_ON(pgdat
->first_deferred_pfn
> pgdat_end_pfn(pgdat
));
1457 pgdat
->first_deferred_pfn
= ULONG_MAX
;
1459 /* Only the highest zone is deferred so find it */
1460 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
1461 zone
= pgdat
->node_zones
+ zid
;
1462 if (first_init_pfn
< zone_end_pfn(zone
))
1466 for_each_mem_pfn_range(i
, nid
, &walk_start
, &walk_end
, NULL
) {
1467 unsigned long pfn
, end_pfn
;
1468 struct page
*page
= NULL
;
1469 struct page
*free_base_page
= NULL
;
1470 unsigned long free_base_pfn
= 0;
1473 end_pfn
= min(walk_end
, zone_end_pfn(zone
));
1474 pfn
= first_init_pfn
;
1475 if (pfn
< walk_start
)
1477 if (pfn
< zone
->zone_start_pfn
)
1478 pfn
= zone
->zone_start_pfn
;
1480 for (; pfn
< end_pfn
; pfn
++) {
1481 if (!pfn_valid_within(pfn
))
1485 * Ensure pfn_valid is checked every
1486 * pageblock_nr_pages for memory holes
1488 if ((pfn
& (pageblock_nr_pages
- 1)) == 0) {
1489 if (!pfn_valid(pfn
)) {
1495 if (!meminit_pfn_in_nid(pfn
, nid
, &nid_init_state
)) {
1500 /* Minimise pfn page lookups and scheduler checks */
1501 if (page
&& (pfn
& (pageblock_nr_pages
- 1)) != 0) {
1504 nr_pages
+= nr_to_free
;
1505 deferred_free_range(free_base_page
,
1506 free_base_pfn
, nr_to_free
);
1507 free_base_page
= NULL
;
1508 free_base_pfn
= nr_to_free
= 0;
1510 page
= pfn_to_page(pfn
);
1515 VM_BUG_ON(page_zone(page
) != zone
);
1519 __init_single_page(page
, pfn
, zid
, nid
);
1520 if (!free_base_page
) {
1521 free_base_page
= page
;
1522 free_base_pfn
= pfn
;
1527 /* Where possible, batch up pages for a single free */
1530 /* Free the current block of pages to allocator */
1531 nr_pages
+= nr_to_free
;
1532 deferred_free_range(free_base_page
, free_base_pfn
,
1534 free_base_page
= NULL
;
1535 free_base_pfn
= nr_to_free
= 0;
1537 /* Free the last block of pages to allocator */
1538 nr_pages
+= nr_to_free
;
1539 deferred_free_range(free_base_page
, free_base_pfn
, nr_to_free
);
1541 first_init_pfn
= max(end_pfn
, first_init_pfn
);
1544 /* Sanity check that the next zone really is unpopulated */
1545 WARN_ON(++zid
< MAX_NR_ZONES
&& populated_zone(++zone
));
1547 pr_info("node %d initialised, %lu pages in %ums\n", nid
, nr_pages
,
1548 jiffies_to_msecs(jiffies
- start
));
1550 pgdat_init_report_one_done();
1553 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1555 void __init
page_alloc_init_late(void)
1559 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1562 /* There will be num_node_state(N_MEMORY) threads */
1563 atomic_set(&pgdat_init_n_undone
, num_node_state(N_MEMORY
));
1564 for_each_node_state(nid
, N_MEMORY
) {
1565 kthread_run(deferred_init_memmap
, NODE_DATA(nid
), "pgdatinit%d", nid
);
1568 /* Block until all are initialised */
1569 wait_for_completion(&pgdat_init_all_done_comp
);
1571 /* Reinit limits that are based on free pages after the kernel is up */
1572 files_maxfiles_init();
1575 for_each_populated_zone(zone
)
1576 set_zone_contiguous(zone
);
1580 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1581 void __init
init_cma_reserved_pageblock(struct page
*page
)
1583 unsigned i
= pageblock_nr_pages
;
1584 struct page
*p
= page
;
1587 __ClearPageReserved(p
);
1588 set_page_count(p
, 0);
1591 set_pageblock_migratetype(page
, MIGRATE_CMA
);
1593 if (pageblock_order
>= MAX_ORDER
) {
1594 i
= pageblock_nr_pages
;
1597 set_page_refcounted(p
);
1598 __free_pages(p
, MAX_ORDER
- 1);
1599 p
+= MAX_ORDER_NR_PAGES
;
1600 } while (i
-= MAX_ORDER_NR_PAGES
);
1602 set_page_refcounted(page
);
1603 __free_pages(page
, pageblock_order
);
1606 adjust_managed_page_count(page
, pageblock_nr_pages
);
1611 * The order of subdivision here is critical for the IO subsystem.
1612 * Please do not alter this order without good reasons and regression
1613 * testing. Specifically, as large blocks of memory are subdivided,
1614 * the order in which smaller blocks are delivered depends on the order
1615 * they're subdivided in this function. This is the primary factor
1616 * influencing the order in which pages are delivered to the IO
1617 * subsystem according to empirical testing, and this is also justified
1618 * by considering the behavior of a buddy system containing a single
1619 * large block of memory acted on by a series of small allocations.
1620 * This behavior is a critical factor in sglist merging's success.
1624 static inline void expand(struct zone
*zone
, struct page
*page
,
1625 int low
, int high
, struct free_area
*area
,
1628 unsigned long size
= 1 << high
;
1630 while (high
> low
) {
1634 VM_BUG_ON_PAGE(bad_range(zone
, &page
[size
]), &page
[size
]);
1637 * Mark as guard pages (or page), that will allow to
1638 * merge back to allocator when buddy will be freed.
1639 * Corresponding page table entries will not be touched,
1640 * pages will stay not present in virtual address space
1642 if (set_page_guard(zone
, &page
[size
], high
, migratetype
))
1645 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
1647 set_page_order(&page
[size
], high
);
1651 static void check_new_page_bad(struct page
*page
)
1653 const char *bad_reason
= NULL
;
1654 unsigned long bad_flags
= 0;
1656 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
1657 bad_reason
= "nonzero mapcount";
1658 if (unlikely(page
->mapping
!= NULL
))
1659 bad_reason
= "non-NULL mapping";
1660 if (unlikely(page_ref_count(page
) != 0))
1661 bad_reason
= "nonzero _count";
1662 if (unlikely(page
->flags
& __PG_HWPOISON
)) {
1663 bad_reason
= "HWPoisoned (hardware-corrupted)";
1664 bad_flags
= __PG_HWPOISON
;
1665 /* Don't complain about hwpoisoned pages */
1666 page_mapcount_reset(page
); /* remove PageBuddy */
1669 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)) {
1670 bad_reason
= "PAGE_FLAGS_CHECK_AT_PREP flag set";
1671 bad_flags
= PAGE_FLAGS_CHECK_AT_PREP
;
1674 if (unlikely(page
->mem_cgroup
))
1675 bad_reason
= "page still charged to cgroup";
1677 bad_page(page
, bad_reason
, bad_flags
);
1681 * This page is about to be returned from the page allocator
1683 static inline int check_new_page(struct page
*page
)
1685 if (likely(page_expected_state(page
,
1686 PAGE_FLAGS_CHECK_AT_PREP
|__PG_HWPOISON
)))
1689 check_new_page_bad(page
);
1693 static inline bool free_pages_prezeroed(bool poisoned
)
1695 return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO
) &&
1696 page_poisoning_enabled() && poisoned
;
1699 #ifdef CONFIG_DEBUG_VM
1700 static bool check_pcp_refill(struct page
*page
)
1705 static bool check_new_pcp(struct page
*page
)
1707 return check_new_page(page
);
1710 static bool check_pcp_refill(struct page
*page
)
1712 return check_new_page(page
);
1714 static bool check_new_pcp(struct page
*page
)
1718 #endif /* CONFIG_DEBUG_VM */
1720 static bool check_new_pages(struct page
*page
, unsigned int order
)
1723 for (i
= 0; i
< (1 << order
); i
++) {
1724 struct page
*p
= page
+ i
;
1726 if (unlikely(check_new_page(p
)))
1733 inline void post_alloc_hook(struct page
*page
, unsigned int order
,
1736 set_page_private(page
, 0);
1737 set_page_refcounted(page
);
1739 arch_alloc_page(page
, order
);
1740 kernel_map_pages(page
, 1 << order
, 1);
1741 kernel_poison_pages(page
, 1 << order
, 1);
1742 kasan_alloc_pages(page
, order
);
1743 set_page_owner(page
, order
, gfp_flags
);
1746 static void prep_new_page(struct page
*page
, unsigned int order
, gfp_t gfp_flags
,
1747 unsigned int alloc_flags
)
1750 bool poisoned
= true;
1752 for (i
= 0; i
< (1 << order
); i
++) {
1753 struct page
*p
= page
+ i
;
1755 poisoned
&= page_is_poisoned(p
);
1758 post_alloc_hook(page
, order
, gfp_flags
);
1760 if (!free_pages_prezeroed(poisoned
) && (gfp_flags
& __GFP_ZERO
))
1761 for (i
= 0; i
< (1 << order
); i
++)
1762 clear_highpage(page
+ i
);
1764 if (order
&& (gfp_flags
& __GFP_COMP
))
1765 prep_compound_page(page
, order
);
1768 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1769 * allocate the page. The expectation is that the caller is taking
1770 * steps that will free more memory. The caller should avoid the page
1771 * being used for !PFMEMALLOC purposes.
1773 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
1774 set_page_pfmemalloc(page
);
1776 clear_page_pfmemalloc(page
);
1780 * Go through the free lists for the given migratetype and remove
1781 * the smallest available page from the freelists
1784 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
1787 unsigned int current_order
;
1788 struct free_area
*area
;
1791 /* Find a page of the appropriate size in the preferred list */
1792 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
1793 area
= &(zone
->free_area
[current_order
]);
1794 page
= list_first_entry_or_null(&area
->free_list
[migratetype
],
1798 list_del(&page
->lru
);
1799 rmv_page_order(page
);
1801 expand(zone
, page
, order
, current_order
, area
, migratetype
);
1802 set_pcppage_migratetype(page
, migratetype
);
1811 * This array describes the order lists are fallen back to when
1812 * the free lists for the desirable migrate type are depleted
1814 static int fallbacks
[MIGRATE_TYPES
][4] = {
1815 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1816 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1817 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_TYPES
},
1819 [MIGRATE_CMA
] = { MIGRATE_TYPES
}, /* Never used */
1821 #ifdef CONFIG_MEMORY_ISOLATION
1822 [MIGRATE_ISOLATE
] = { MIGRATE_TYPES
}, /* Never used */
1827 static struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1830 return __rmqueue_smallest(zone
, order
, MIGRATE_CMA
);
1833 static inline struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1834 unsigned int order
) { return NULL
; }
1838 * Move the free pages in a range to the free lists of the requested type.
1839 * Note that start_page and end_pages are not aligned on a pageblock
1840 * boundary. If alignment is required, use move_freepages_block()
1842 int move_freepages(struct zone
*zone
,
1843 struct page
*start_page
, struct page
*end_page
,
1848 int pages_moved
= 0;
1850 #ifndef CONFIG_HOLES_IN_ZONE
1852 * page_zone is not safe to call in this context when
1853 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1854 * anyway as we check zone boundaries in move_freepages_block().
1855 * Remove at a later date when no bug reports exist related to
1856 * grouping pages by mobility
1858 VM_BUG_ON(page_zone(start_page
) != page_zone(end_page
));
1861 for (page
= start_page
; page
<= end_page
;) {
1862 if (!pfn_valid_within(page_to_pfn(page
))) {
1867 /* Make sure we are not inadvertently changing nodes */
1868 VM_BUG_ON_PAGE(page_to_nid(page
) != zone_to_nid(zone
), page
);
1870 if (!PageBuddy(page
)) {
1875 order
= page_order(page
);
1876 list_move(&page
->lru
,
1877 &zone
->free_area
[order
].free_list
[migratetype
]);
1879 pages_moved
+= 1 << order
;
1885 int move_freepages_block(struct zone
*zone
, struct page
*page
,
1888 unsigned long start_pfn
, end_pfn
;
1889 struct page
*start_page
, *end_page
;
1891 start_pfn
= page_to_pfn(page
);
1892 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
1893 start_page
= pfn_to_page(start_pfn
);
1894 end_page
= start_page
+ pageblock_nr_pages
- 1;
1895 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
1897 /* Do not cross zone boundaries */
1898 if (!zone_spans_pfn(zone
, start_pfn
))
1900 if (!zone_spans_pfn(zone
, end_pfn
))
1903 return move_freepages(zone
, start_page
, end_page
, migratetype
);
1906 static void change_pageblock_range(struct page
*pageblock_page
,
1907 int start_order
, int migratetype
)
1909 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
1911 while (nr_pageblocks
--) {
1912 set_pageblock_migratetype(pageblock_page
, migratetype
);
1913 pageblock_page
+= pageblock_nr_pages
;
1918 * When we are falling back to another migratetype during allocation, try to
1919 * steal extra free pages from the same pageblocks to satisfy further
1920 * allocations, instead of polluting multiple pageblocks.
1922 * If we are stealing a relatively large buddy page, it is likely there will
1923 * be more free pages in the pageblock, so try to steal them all. For
1924 * reclaimable and unmovable allocations, we steal regardless of page size,
1925 * as fragmentation caused by those allocations polluting movable pageblocks
1926 * is worse than movable allocations stealing from unmovable and reclaimable
1929 static bool can_steal_fallback(unsigned int order
, int start_mt
)
1932 * Leaving this order check is intended, although there is
1933 * relaxed order check in next check. The reason is that
1934 * we can actually steal whole pageblock if this condition met,
1935 * but, below check doesn't guarantee it and that is just heuristic
1936 * so could be changed anytime.
1938 if (order
>= pageblock_order
)
1941 if (order
>= pageblock_order
/ 2 ||
1942 start_mt
== MIGRATE_RECLAIMABLE
||
1943 start_mt
== MIGRATE_UNMOVABLE
||
1944 page_group_by_mobility_disabled
)
1951 * This function implements actual steal behaviour. If order is large enough,
1952 * we can steal whole pageblock. If not, we first move freepages in this
1953 * pageblock and check whether half of pages are moved or not. If half of
1954 * pages are moved, we can change migratetype of pageblock and permanently
1955 * use it's pages as requested migratetype in the future.
1957 static void steal_suitable_fallback(struct zone
*zone
, struct page
*page
,
1960 unsigned int current_order
= page_order(page
);
1963 /* Take ownership for orders >= pageblock_order */
1964 if (current_order
>= pageblock_order
) {
1965 change_pageblock_range(page
, current_order
, start_type
);
1969 pages
= move_freepages_block(zone
, page
, start_type
);
1971 /* Claim the whole block if over half of it is free */
1972 if (pages
>= (1 << (pageblock_order
-1)) ||
1973 page_group_by_mobility_disabled
)
1974 set_pageblock_migratetype(page
, start_type
);
1978 * Check whether there is a suitable fallback freepage with requested order.
1979 * If only_stealable is true, this function returns fallback_mt only if
1980 * we can steal other freepages all together. This would help to reduce
1981 * fragmentation due to mixed migratetype pages in one pageblock.
1983 int find_suitable_fallback(struct free_area
*area
, unsigned int order
,
1984 int migratetype
, bool only_stealable
, bool *can_steal
)
1989 if (area
->nr_free
== 0)
1994 fallback_mt
= fallbacks
[migratetype
][i
];
1995 if (fallback_mt
== MIGRATE_TYPES
)
1998 if (list_empty(&area
->free_list
[fallback_mt
]))
2001 if (can_steal_fallback(order
, migratetype
))
2004 if (!only_stealable
)
2015 * Reserve a pageblock for exclusive use of high-order atomic allocations if
2016 * there are no empty page blocks that contain a page with a suitable order
2018 static void reserve_highatomic_pageblock(struct page
*page
, struct zone
*zone
,
2019 unsigned int alloc_order
)
2022 unsigned long max_managed
, flags
;
2025 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
2026 * Check is race-prone but harmless.
2028 max_managed
= (zone
->managed_pages
/ 100) + pageblock_nr_pages
;
2029 if (zone
->nr_reserved_highatomic
>= max_managed
)
2032 spin_lock_irqsave(&zone
->lock
, flags
);
2034 /* Recheck the nr_reserved_highatomic limit under the lock */
2035 if (zone
->nr_reserved_highatomic
>= max_managed
)
2039 mt
= get_pageblock_migratetype(page
);
2040 if (mt
!= MIGRATE_HIGHATOMIC
&&
2041 !is_migrate_isolate(mt
) && !is_migrate_cma(mt
)) {
2042 zone
->nr_reserved_highatomic
+= pageblock_nr_pages
;
2043 set_pageblock_migratetype(page
, MIGRATE_HIGHATOMIC
);
2044 move_freepages_block(zone
, page
, MIGRATE_HIGHATOMIC
);
2048 spin_unlock_irqrestore(&zone
->lock
, flags
);
2052 * Used when an allocation is about to fail under memory pressure. This
2053 * potentially hurts the reliability of high-order allocations when under
2054 * intense memory pressure but failed atomic allocations should be easier
2055 * to recover from than an OOM.
2057 * If @force is true, try to unreserve a pageblock even though highatomic
2058 * pageblock is exhausted.
2060 static bool unreserve_highatomic_pageblock(const struct alloc_context
*ac
,
2063 struct zonelist
*zonelist
= ac
->zonelist
;
2064 unsigned long flags
;
2071 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
, ac
->high_zoneidx
,
2074 * Preserve at least one pageblock unless memory pressure
2077 if (!force
&& zone
->nr_reserved_highatomic
<=
2081 spin_lock_irqsave(&zone
->lock
, flags
);
2082 for (order
= 0; order
< MAX_ORDER
; order
++) {
2083 struct free_area
*area
= &(zone
->free_area
[order
]);
2085 page
= list_first_entry_or_null(
2086 &area
->free_list
[MIGRATE_HIGHATOMIC
],
2092 * In page freeing path, migratetype change is racy so
2093 * we can counter several free pages in a pageblock
2094 * in this loop althoug we changed the pageblock type
2095 * from highatomic to ac->migratetype. So we should
2096 * adjust the count once.
2098 if (get_pageblock_migratetype(page
) ==
2099 MIGRATE_HIGHATOMIC
) {
2101 * It should never happen but changes to
2102 * locking could inadvertently allow a per-cpu
2103 * drain to add pages to MIGRATE_HIGHATOMIC
2104 * while unreserving so be safe and watch for
2107 zone
->nr_reserved_highatomic
-= min(
2109 zone
->nr_reserved_highatomic
);
2113 * Convert to ac->migratetype and avoid the normal
2114 * pageblock stealing heuristics. Minimally, the caller
2115 * is doing the work and needs the pages. More
2116 * importantly, if the block was always converted to
2117 * MIGRATE_UNMOVABLE or another type then the number
2118 * of pageblocks that cannot be completely freed
2121 set_pageblock_migratetype(page
, ac
->migratetype
);
2122 ret
= move_freepages_block(zone
, page
, ac
->migratetype
);
2124 spin_unlock_irqrestore(&zone
->lock
, flags
);
2128 spin_unlock_irqrestore(&zone
->lock
, flags
);
2134 /* Remove an element from the buddy allocator from the fallback list */
2135 static inline struct page
*
2136 __rmqueue_fallback(struct zone
*zone
, unsigned int order
, int start_migratetype
)
2138 struct free_area
*area
;
2139 unsigned int current_order
;
2144 /* Find the largest possible block of pages in the other list */
2145 for (current_order
= MAX_ORDER
-1;
2146 current_order
>= order
&& current_order
<= MAX_ORDER
-1;
2148 area
= &(zone
->free_area
[current_order
]);
2149 fallback_mt
= find_suitable_fallback(area
, current_order
,
2150 start_migratetype
, false, &can_steal
);
2151 if (fallback_mt
== -1)
2154 page
= list_first_entry(&area
->free_list
[fallback_mt
],
2157 get_pageblock_migratetype(page
) != MIGRATE_HIGHATOMIC
)
2158 steal_suitable_fallback(zone
, page
, start_migratetype
);
2160 /* Remove the page from the freelists */
2162 list_del(&page
->lru
);
2163 rmv_page_order(page
);
2165 expand(zone
, page
, order
, current_order
, area
,
2168 * The pcppage_migratetype may differ from pageblock's
2169 * migratetype depending on the decisions in
2170 * find_suitable_fallback(). This is OK as long as it does not
2171 * differ for MIGRATE_CMA pageblocks. Those can be used as
2172 * fallback only via special __rmqueue_cma_fallback() function
2174 set_pcppage_migratetype(page
, start_migratetype
);
2176 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
2177 start_migratetype
, fallback_mt
);
2186 * Do the hard work of removing an element from the buddy allocator.
2187 * Call me with the zone->lock already held.
2189 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
2194 page
= __rmqueue_smallest(zone
, order
, migratetype
);
2195 if (unlikely(!page
)) {
2196 if (migratetype
== MIGRATE_MOVABLE
)
2197 page
= __rmqueue_cma_fallback(zone
, order
);
2200 page
= __rmqueue_fallback(zone
, order
, migratetype
);
2203 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
2208 * Obtain a specified number of elements from the buddy allocator, all under
2209 * a single hold of the lock, for efficiency. Add them to the supplied list.
2210 * Returns the number of new pages which were placed at *list.
2212 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
2213 unsigned long count
, struct list_head
*list
,
2214 int migratetype
, bool cold
)
2217 unsigned long flags
;
2219 spin_lock_irqsave(&zone
->lock
, flags
);
2220 for (i
= 0; i
< count
; ++i
) {
2221 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
2222 if (unlikely(page
== NULL
))
2225 if (unlikely(check_pcp_refill(page
)))
2229 * Split buddy pages returned by expand() are received here
2230 * in physical page order. The page is added to the callers and
2231 * list and the list head then moves forward. From the callers
2232 * perspective, the linked list is ordered by page number in
2233 * some conditions. This is useful for IO devices that can
2234 * merge IO requests if the physical pages are ordered
2238 list_add(&page
->lru
, list
);
2240 list_add_tail(&page
->lru
, list
);
2243 if (is_migrate_cma(get_pcppage_migratetype(page
)))
2244 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
,
2249 * i pages were removed from the buddy list even if some leak due
2250 * to check_pcp_refill failing so adjust NR_FREE_PAGES based
2251 * on i. Do not confuse with 'alloced' which is the number of
2252 * pages added to the pcp list.
2254 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
2255 spin_unlock_irqrestore(&zone
->lock
, flags
);
2261 * Called from the vmstat counter updater to drain pagesets of this
2262 * currently executing processor on remote nodes after they have
2265 * Note that this function must be called with the thread pinned to
2266 * a single processor.
2268 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
2270 unsigned long flags
;
2271 int to_drain
, batch
;
2273 local_irq_save(flags
);
2274 batch
= READ_ONCE(pcp
->batch
);
2275 to_drain
= min(pcp
->count
, batch
);
2277 free_pcppages_bulk(zone
, to_drain
, pcp
);
2278 pcp
->count
-= to_drain
;
2280 local_irq_restore(flags
);
2285 * Drain pcplists of the indicated processor and zone.
2287 * The processor must either be the current processor and the
2288 * thread pinned to the current processor or a processor that
2291 static void drain_pages_zone(unsigned int cpu
, struct zone
*zone
)
2293 unsigned long flags
;
2294 struct per_cpu_pageset
*pset
;
2295 struct per_cpu_pages
*pcp
;
2297 local_irq_save(flags
);
2298 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
2302 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
2305 local_irq_restore(flags
);
2309 * Drain pcplists of all zones on the indicated processor.
2311 * The processor must either be the current processor and the
2312 * thread pinned to the current processor or a processor that
2315 static void drain_pages(unsigned int cpu
)
2319 for_each_populated_zone(zone
) {
2320 drain_pages_zone(cpu
, zone
);
2325 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
2327 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2328 * the single zone's pages.
2330 void drain_local_pages(struct zone
*zone
)
2332 int cpu
= smp_processor_id();
2335 drain_pages_zone(cpu
, zone
);
2340 static void drain_local_pages_wq(struct work_struct
*work
)
2343 * drain_all_pages doesn't use proper cpu hotplug protection so
2344 * we can race with cpu offline when the WQ can move this from
2345 * a cpu pinned worker to an unbound one. We can operate on a different
2346 * cpu which is allright but we also have to make sure to not move to
2350 drain_local_pages(NULL
);
2355 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2357 * When zone parameter is non-NULL, spill just the single zone's pages.
2359 * Note that this can be extremely slow as the draining happens in a workqueue.
2361 void drain_all_pages(struct zone
*zone
)
2363 struct work_struct __percpu
*works
;
2367 * Allocate in the BSS so we wont require allocation in
2368 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2370 static cpumask_t cpus_with_pcps
;
2372 /* Workqueues cannot recurse */
2373 if (current
->flags
& PF_WQ_WORKER
)
2376 works
= alloc_percpu_gfp(struct work_struct
, GFP_ATOMIC
);
2379 * We don't care about racing with CPU hotplug event
2380 * as offline notification will cause the notified
2381 * cpu to drain that CPU pcps and on_each_cpu_mask
2382 * disables preemption as part of its processing
2384 for_each_online_cpu(cpu
) {
2385 struct per_cpu_pageset
*pcp
;
2387 bool has_pcps
= false;
2390 pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
2394 for_each_populated_zone(z
) {
2395 pcp
= per_cpu_ptr(z
->pageset
, cpu
);
2396 if (pcp
->pcp
.count
) {
2404 cpumask_set_cpu(cpu
, &cpus_with_pcps
);
2406 cpumask_clear_cpu(cpu
, &cpus_with_pcps
);
2410 for_each_cpu(cpu
, &cpus_with_pcps
) {
2411 struct work_struct
*work
= per_cpu_ptr(works
, cpu
);
2412 INIT_WORK(work
, drain_local_pages_wq
);
2413 schedule_work_on(cpu
, work
);
2415 for_each_cpu(cpu
, &cpus_with_pcps
)
2416 flush_work(per_cpu_ptr(works
, cpu
));
2418 for_each_cpu(cpu
, &cpus_with_pcps
) {
2419 struct work_struct work
;
2421 INIT_WORK(&work
, drain_local_pages_wq
);
2422 schedule_work_on(cpu
, &work
);
2428 #ifdef CONFIG_HIBERNATION
2430 void mark_free_pages(struct zone
*zone
)
2432 unsigned long pfn
, max_zone_pfn
;
2433 unsigned long flags
;
2434 unsigned int order
, t
;
2437 if (zone_is_empty(zone
))
2440 spin_lock_irqsave(&zone
->lock
, flags
);
2442 max_zone_pfn
= zone_end_pfn(zone
);
2443 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
2444 if (pfn_valid(pfn
)) {
2445 page
= pfn_to_page(pfn
);
2447 if (page_zone(page
) != zone
)
2450 if (!swsusp_page_is_forbidden(page
))
2451 swsusp_unset_page_free(page
);
2454 for_each_migratetype_order(order
, t
) {
2455 list_for_each_entry(page
,
2456 &zone
->free_area
[order
].free_list
[t
], lru
) {
2459 pfn
= page_to_pfn(page
);
2460 for (i
= 0; i
< (1UL << order
); i
++)
2461 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
2464 spin_unlock_irqrestore(&zone
->lock
, flags
);
2466 #endif /* CONFIG_PM */
2469 * Free a 0-order page
2470 * cold == true ? free a cold page : free a hot page
2472 void free_hot_cold_page(struct page
*page
, bool cold
)
2474 struct zone
*zone
= page_zone(page
);
2475 struct per_cpu_pages
*pcp
;
2476 unsigned long pfn
= page_to_pfn(page
);
2479 if (in_interrupt()) {
2480 __free_pages_ok(page
, 0);
2484 if (!free_pcp_prepare(page
))
2487 migratetype
= get_pfnblock_migratetype(page
, pfn
);
2488 set_pcppage_migratetype(page
, migratetype
);
2492 * We only track unmovable, reclaimable and movable on pcp lists.
2493 * Free ISOLATE pages back to the allocator because they are being
2494 * offlined but treat RESERVE as movable pages so we can get those
2495 * areas back if necessary. Otherwise, we may have to free
2496 * excessively into the page allocator
2498 if (migratetype
>= MIGRATE_PCPTYPES
) {
2499 if (unlikely(is_migrate_isolate(migratetype
))) {
2500 free_one_page(zone
, page
, pfn
, 0, migratetype
);
2503 migratetype
= MIGRATE_MOVABLE
;
2506 __count_vm_event(PGFREE
);
2507 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2509 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
2511 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
2513 if (pcp
->count
>= pcp
->high
) {
2514 unsigned long batch
= READ_ONCE(pcp
->batch
);
2515 free_pcppages_bulk(zone
, batch
, pcp
);
2516 pcp
->count
-= batch
;
2524 * Free a list of 0-order pages
2526 void free_hot_cold_page_list(struct list_head
*list
, bool cold
)
2528 struct page
*page
, *next
;
2530 list_for_each_entry_safe(page
, next
, list
, lru
) {
2531 trace_mm_page_free_batched(page
, cold
);
2532 free_hot_cold_page(page
, cold
);
2537 * split_page takes a non-compound higher-order page, and splits it into
2538 * n (1<<order) sub-pages: page[0..n]
2539 * Each sub-page must be freed individually.
2541 * Note: this is probably too low level an operation for use in drivers.
2542 * Please consult with lkml before using this in your driver.
2544 void split_page(struct page
*page
, unsigned int order
)
2548 VM_BUG_ON_PAGE(PageCompound(page
), page
);
2549 VM_BUG_ON_PAGE(!page_count(page
), page
);
2551 #ifdef CONFIG_KMEMCHECK
2553 * Split shadow pages too, because free(page[0]) would
2554 * otherwise free the whole shadow.
2556 if (kmemcheck_page_is_tracked(page
))
2557 split_page(virt_to_page(page
[0].shadow
), order
);
2560 for (i
= 1; i
< (1 << order
); i
++)
2561 set_page_refcounted(page
+ i
);
2562 split_page_owner(page
, order
);
2564 EXPORT_SYMBOL_GPL(split_page
);
2566 int __isolate_free_page(struct page
*page
, unsigned int order
)
2568 unsigned long watermark
;
2572 BUG_ON(!PageBuddy(page
));
2574 zone
= page_zone(page
);
2575 mt
= get_pageblock_migratetype(page
);
2577 if (!is_migrate_isolate(mt
)) {
2579 * Obey watermarks as if the page was being allocated. We can
2580 * emulate a high-order watermark check with a raised order-0
2581 * watermark, because we already know our high-order page
2584 watermark
= min_wmark_pages(zone
) + (1UL << order
);
2585 if (!zone_watermark_ok(zone
, 0, watermark
, 0, ALLOC_CMA
))
2588 __mod_zone_freepage_state(zone
, -(1UL << order
), mt
);
2591 /* Remove page from free list */
2592 list_del(&page
->lru
);
2593 zone
->free_area
[order
].nr_free
--;
2594 rmv_page_order(page
);
2597 * Set the pageblock if the isolated page is at least half of a
2600 if (order
>= pageblock_order
- 1) {
2601 struct page
*endpage
= page
+ (1 << order
) - 1;
2602 for (; page
< endpage
; page
+= pageblock_nr_pages
) {
2603 int mt
= get_pageblock_migratetype(page
);
2604 if (!is_migrate_isolate(mt
) && !is_migrate_cma(mt
)
2605 && mt
!= MIGRATE_HIGHATOMIC
)
2606 set_pageblock_migratetype(page
,
2612 return 1UL << order
;
2616 * Update NUMA hit/miss statistics
2618 * Must be called with interrupts disabled.
2620 static inline void zone_statistics(struct zone
*preferred_zone
, struct zone
*z
)
2623 enum zone_stat_item local_stat
= NUMA_LOCAL
;
2625 if (z
->node
!= numa_node_id())
2626 local_stat
= NUMA_OTHER
;
2628 if (z
->node
== preferred_zone
->node
)
2629 __inc_zone_state(z
, NUMA_HIT
);
2631 __inc_zone_state(z
, NUMA_MISS
);
2632 __inc_zone_state(preferred_zone
, NUMA_FOREIGN
);
2634 __inc_zone_state(z
, local_stat
);
2638 /* Remove page from the per-cpu list, caller must protect the list */
2639 static struct page
*__rmqueue_pcplist(struct zone
*zone
, int migratetype
,
2640 bool cold
, struct per_cpu_pages
*pcp
,
2641 struct list_head
*list
)
2645 VM_BUG_ON(in_interrupt());
2648 if (list_empty(list
)) {
2649 pcp
->count
+= rmqueue_bulk(zone
, 0,
2652 if (unlikely(list_empty(list
)))
2657 page
= list_last_entry(list
, struct page
, lru
);
2659 page
= list_first_entry(list
, struct page
, lru
);
2661 list_del(&page
->lru
);
2663 } while (check_new_pcp(page
));
2668 /* Lock and remove page from the per-cpu list */
2669 static struct page
*rmqueue_pcplist(struct zone
*preferred_zone
,
2670 struct zone
*zone
, unsigned int order
,
2671 gfp_t gfp_flags
, int migratetype
)
2673 struct per_cpu_pages
*pcp
;
2674 struct list_head
*list
;
2675 bool cold
= ((gfp_flags
& __GFP_COLD
) != 0);
2679 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2680 list
= &pcp
->lists
[migratetype
];
2681 page
= __rmqueue_pcplist(zone
, migratetype
, cold
, pcp
, list
);
2683 __count_zid_vm_events(PGALLOC
, page_zonenum(page
), 1 << order
);
2684 zone_statistics(preferred_zone
, zone
);
2691 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2694 struct page
*rmqueue(struct zone
*preferred_zone
,
2695 struct zone
*zone
, unsigned int order
,
2696 gfp_t gfp_flags
, unsigned int alloc_flags
,
2699 unsigned long flags
;
2702 if (likely(order
== 0) && !in_interrupt()) {
2703 page
= rmqueue_pcplist(preferred_zone
, zone
, order
,
2704 gfp_flags
, migratetype
);
2709 * We most definitely don't want callers attempting to
2710 * allocate greater than order-1 page units with __GFP_NOFAIL.
2712 WARN_ON_ONCE((gfp_flags
& __GFP_NOFAIL
) && (order
> 1));
2713 spin_lock_irqsave(&zone
->lock
, flags
);
2717 if (alloc_flags
& ALLOC_HARDER
) {
2718 page
= __rmqueue_smallest(zone
, order
, MIGRATE_HIGHATOMIC
);
2720 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
2723 page
= __rmqueue(zone
, order
, migratetype
);
2724 } while (page
&& check_new_pages(page
, order
));
2725 spin_unlock(&zone
->lock
);
2728 __mod_zone_freepage_state(zone
, -(1 << order
),
2729 get_pcppage_migratetype(page
));
2731 __count_zid_vm_events(PGALLOC
, page_zonenum(page
), 1 << order
);
2732 zone_statistics(preferred_zone
, zone
);
2733 local_irq_restore(flags
);
2736 VM_BUG_ON_PAGE(page
&& bad_range(zone
, page
), page
);
2740 local_irq_restore(flags
);
2744 #ifdef CONFIG_FAIL_PAGE_ALLOC
2747 struct fault_attr attr
;
2749 bool ignore_gfp_highmem
;
2750 bool ignore_gfp_reclaim
;
2752 } fail_page_alloc
= {
2753 .attr
= FAULT_ATTR_INITIALIZER
,
2754 .ignore_gfp_reclaim
= true,
2755 .ignore_gfp_highmem
= true,
2759 static int __init
setup_fail_page_alloc(char *str
)
2761 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
2763 __setup("fail_page_alloc=", setup_fail_page_alloc
);
2765 static bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2767 if (order
< fail_page_alloc
.min_order
)
2769 if (gfp_mask
& __GFP_NOFAIL
)
2771 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
2773 if (fail_page_alloc
.ignore_gfp_reclaim
&&
2774 (gfp_mask
& __GFP_DIRECT_RECLAIM
))
2777 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
2780 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2782 static int __init
fail_page_alloc_debugfs(void)
2784 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
2787 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
2788 &fail_page_alloc
.attr
);
2790 return PTR_ERR(dir
);
2792 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
2793 &fail_page_alloc
.ignore_gfp_reclaim
))
2795 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
2796 &fail_page_alloc
.ignore_gfp_highmem
))
2798 if (!debugfs_create_u32("min-order", mode
, dir
,
2799 &fail_page_alloc
.min_order
))
2804 debugfs_remove_recursive(dir
);
2809 late_initcall(fail_page_alloc_debugfs
);
2811 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2813 #else /* CONFIG_FAIL_PAGE_ALLOC */
2815 static inline bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2820 #endif /* CONFIG_FAIL_PAGE_ALLOC */
2823 * Return true if free base pages are above 'mark'. For high-order checks it
2824 * will return true of the order-0 watermark is reached and there is at least
2825 * one free page of a suitable size. Checking now avoids taking the zone lock
2826 * to check in the allocation paths if no pages are free.
2828 bool __zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
2829 int classzone_idx
, unsigned int alloc_flags
,
2834 const bool alloc_harder
= (alloc_flags
& ALLOC_HARDER
);
2836 /* free_pages may go negative - that's OK */
2837 free_pages
-= (1 << order
) - 1;
2839 if (alloc_flags
& ALLOC_HIGH
)
2843 * If the caller does not have rights to ALLOC_HARDER then subtract
2844 * the high-atomic reserves. This will over-estimate the size of the
2845 * atomic reserve but it avoids a search.
2847 if (likely(!alloc_harder
))
2848 free_pages
-= z
->nr_reserved_highatomic
;
2853 /* If allocation can't use CMA areas don't use free CMA pages */
2854 if (!(alloc_flags
& ALLOC_CMA
))
2855 free_pages
-= zone_page_state(z
, NR_FREE_CMA_PAGES
);
2859 * Check watermarks for an order-0 allocation request. If these
2860 * are not met, then a high-order request also cannot go ahead
2861 * even if a suitable page happened to be free.
2863 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
2866 /* If this is an order-0 request then the watermark is fine */
2870 /* For a high-order request, check at least one suitable page is free */
2871 for (o
= order
; o
< MAX_ORDER
; o
++) {
2872 struct free_area
*area
= &z
->free_area
[o
];
2881 for (mt
= 0; mt
< MIGRATE_PCPTYPES
; mt
++) {
2882 if (!list_empty(&area
->free_list
[mt
]))
2887 if ((alloc_flags
& ALLOC_CMA
) &&
2888 !list_empty(&area
->free_list
[MIGRATE_CMA
])) {
2896 bool zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
2897 int classzone_idx
, unsigned int alloc_flags
)
2899 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
2900 zone_page_state(z
, NR_FREE_PAGES
));
2903 static inline bool zone_watermark_fast(struct zone
*z
, unsigned int order
,
2904 unsigned long mark
, int classzone_idx
, unsigned int alloc_flags
)
2906 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
2910 /* If allocation can't use CMA areas don't use free CMA pages */
2911 if (!(alloc_flags
& ALLOC_CMA
))
2912 cma_pages
= zone_page_state(z
, NR_FREE_CMA_PAGES
);
2916 * Fast check for order-0 only. If this fails then the reserves
2917 * need to be calculated. There is a corner case where the check
2918 * passes but only the high-order atomic reserve are free. If
2919 * the caller is !atomic then it'll uselessly search the free
2920 * list. That corner case is then slower but it is harmless.
2922 if (!order
&& (free_pages
- cma_pages
) > mark
+ z
->lowmem_reserve
[classzone_idx
])
2925 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
2929 bool zone_watermark_ok_safe(struct zone
*z
, unsigned int order
,
2930 unsigned long mark
, int classzone_idx
)
2932 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
2934 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
2935 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
2937 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, 0,
2942 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2944 return node_distance(zone_to_nid(local_zone
), zone_to_nid(zone
)) <
2947 #else /* CONFIG_NUMA */
2948 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2952 #endif /* CONFIG_NUMA */
2955 * get_page_from_freelist goes through the zonelist trying to allocate
2958 static struct page
*
2959 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
, int alloc_flags
,
2960 const struct alloc_context
*ac
)
2962 struct zoneref
*z
= ac
->preferred_zoneref
;
2964 struct pglist_data
*last_pgdat_dirty_limit
= NULL
;
2967 * Scan zonelist, looking for a zone with enough free.
2968 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2970 for_next_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
, ac
->high_zoneidx
,
2975 if (cpusets_enabled() &&
2976 (alloc_flags
& ALLOC_CPUSET
) &&
2977 !__cpuset_zone_allowed(zone
, gfp_mask
))
2980 * When allocating a page cache page for writing, we
2981 * want to get it from a node that is within its dirty
2982 * limit, such that no single node holds more than its
2983 * proportional share of globally allowed dirty pages.
2984 * The dirty limits take into account the node's
2985 * lowmem reserves and high watermark so that kswapd
2986 * should be able to balance it without having to
2987 * write pages from its LRU list.
2989 * XXX: For now, allow allocations to potentially
2990 * exceed the per-node dirty limit in the slowpath
2991 * (spread_dirty_pages unset) before going into reclaim,
2992 * which is important when on a NUMA setup the allowed
2993 * nodes are together not big enough to reach the
2994 * global limit. The proper fix for these situations
2995 * will require awareness of nodes in the
2996 * dirty-throttling and the flusher threads.
2998 if (ac
->spread_dirty_pages
) {
2999 if (last_pgdat_dirty_limit
== zone
->zone_pgdat
)
3002 if (!node_dirty_ok(zone
->zone_pgdat
)) {
3003 last_pgdat_dirty_limit
= zone
->zone_pgdat
;
3008 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
3009 if (!zone_watermark_fast(zone
, order
, mark
,
3010 ac_classzone_idx(ac
), alloc_flags
)) {
3013 /* Checked here to keep the fast path fast */
3014 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
3015 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
3018 if (node_reclaim_mode
== 0 ||
3019 !zone_allows_reclaim(ac
->preferred_zoneref
->zone
, zone
))
3022 ret
= node_reclaim(zone
->zone_pgdat
, gfp_mask
, order
);
3024 case NODE_RECLAIM_NOSCAN
:
3027 case NODE_RECLAIM_FULL
:
3028 /* scanned but unreclaimable */
3031 /* did we reclaim enough */
3032 if (zone_watermark_ok(zone
, order
, mark
,
3033 ac_classzone_idx(ac
), alloc_flags
))
3041 page
= rmqueue(ac
->preferred_zoneref
->zone
, zone
, order
,
3042 gfp_mask
, alloc_flags
, ac
->migratetype
);
3044 prep_new_page(page
, order
, gfp_mask
, alloc_flags
);
3047 * If this is a high-order atomic allocation then check
3048 * if the pageblock should be reserved for the future
3050 if (unlikely(order
&& (alloc_flags
& ALLOC_HARDER
)))
3051 reserve_highatomic_pageblock(page
, zone
, order
);
3061 * Large machines with many possible nodes should not always dump per-node
3062 * meminfo in irq context.
3064 static inline bool should_suppress_show_mem(void)
3069 ret
= in_interrupt();
3074 static void warn_alloc_show_mem(gfp_t gfp_mask
, nodemask_t
*nodemask
)
3076 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
3077 static DEFINE_RATELIMIT_STATE(show_mem_rs
, HZ
, 1);
3079 if (should_suppress_show_mem() || !__ratelimit(&show_mem_rs
))
3083 * This documents exceptions given to allocations in certain
3084 * contexts that are allowed to allocate outside current's set
3087 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
3088 if (test_thread_flag(TIF_MEMDIE
) ||
3089 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
3090 filter
&= ~SHOW_MEM_FILTER_NODES
;
3091 if (in_interrupt() || !(gfp_mask
& __GFP_DIRECT_RECLAIM
))
3092 filter
&= ~SHOW_MEM_FILTER_NODES
;
3094 show_mem(filter
, nodemask
);
3097 void warn_alloc(gfp_t gfp_mask
, nodemask_t
*nodemask
, const char *fmt
, ...)
3099 struct va_format vaf
;
3101 static DEFINE_RATELIMIT_STATE(nopage_rs
, DEFAULT_RATELIMIT_INTERVAL
,
3102 DEFAULT_RATELIMIT_BURST
);
3104 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
) ||
3105 debug_guardpage_minorder() > 0)
3108 pr_warn("%s: ", current
->comm
);
3110 va_start(args
, fmt
);
3113 pr_cont("%pV", &vaf
);
3116 pr_cont(", mode:%#x(%pGg), nodemask=", gfp_mask
, &gfp_mask
);
3118 pr_cont("%*pbl\n", nodemask_pr_args(nodemask
));
3120 pr_cont("(null)\n");
3122 cpuset_print_current_mems_allowed();
3125 warn_alloc_show_mem(gfp_mask
, nodemask
);
3128 static inline struct page
*
3129 __alloc_pages_cpuset_fallback(gfp_t gfp_mask
, unsigned int order
,
3130 unsigned int alloc_flags
,
3131 const struct alloc_context
*ac
)
3135 page
= get_page_from_freelist(gfp_mask
, order
,
3136 alloc_flags
|ALLOC_CPUSET
, ac
);
3138 * fallback to ignore cpuset restriction if our nodes
3142 page
= get_page_from_freelist(gfp_mask
, order
,
3148 static inline struct page
*
3149 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
3150 const struct alloc_context
*ac
, unsigned long *did_some_progress
)
3152 struct oom_control oc
= {
3153 .zonelist
= ac
->zonelist
,
3154 .nodemask
= ac
->nodemask
,
3156 .gfp_mask
= gfp_mask
,
3161 *did_some_progress
= 0;
3164 * Acquire the oom lock. If that fails, somebody else is
3165 * making progress for us.
3167 if (!mutex_trylock(&oom_lock
)) {
3168 *did_some_progress
= 1;
3169 schedule_timeout_uninterruptible(1);
3174 * Go through the zonelist yet one more time, keep very high watermark
3175 * here, this is only to catch a parallel oom killing, we must fail if
3176 * we're still under heavy pressure.
3178 page
= get_page_from_freelist(gfp_mask
| __GFP_HARDWALL
, order
,
3179 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
, ac
);
3183 /* Coredumps can quickly deplete all memory reserves */
3184 if (current
->flags
& PF_DUMPCORE
)
3186 /* The OOM killer will not help higher order allocs */
3187 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
3189 /* The OOM killer does not needlessly kill tasks for lowmem */
3190 if (ac
->high_zoneidx
< ZONE_NORMAL
)
3192 if (pm_suspended_storage())
3195 * XXX: GFP_NOFS allocations should rather fail than rely on
3196 * other request to make a forward progress.
3197 * We are in an unfortunate situation where out_of_memory cannot
3198 * do much for this context but let's try it to at least get
3199 * access to memory reserved if the current task is killed (see
3200 * out_of_memory). Once filesystems are ready to handle allocation
3201 * failures more gracefully we should just bail out here.
3204 /* The OOM killer may not free memory on a specific node */
3205 if (gfp_mask
& __GFP_THISNODE
)
3208 /* Exhausted what can be done so it's blamo time */
3209 if (out_of_memory(&oc
) || WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
)) {
3210 *did_some_progress
= 1;
3213 * Help non-failing allocations by giving them access to memory
3216 if (gfp_mask
& __GFP_NOFAIL
)
3217 page
= __alloc_pages_cpuset_fallback(gfp_mask
, order
,
3218 ALLOC_NO_WATERMARKS
, ac
);
3221 mutex_unlock(&oom_lock
);
3226 * Maximum number of compaction retries wit a progress before OOM
3227 * killer is consider as the only way to move forward.
3229 #define MAX_COMPACT_RETRIES 16
3231 #ifdef CONFIG_COMPACTION
3232 /* Try memory compaction for high-order allocations before reclaim */
3233 static struct page
*
3234 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
3235 unsigned int alloc_flags
, const struct alloc_context
*ac
,
3236 enum compact_priority prio
, enum compact_result
*compact_result
)
3243 current
->flags
|= PF_MEMALLOC
;
3244 *compact_result
= try_to_compact_pages(gfp_mask
, order
, alloc_flags
, ac
,
3246 current
->flags
&= ~PF_MEMALLOC
;
3248 if (*compact_result
<= COMPACT_INACTIVE
)
3252 * At least in one zone compaction wasn't deferred or skipped, so let's
3253 * count a compaction stall
3255 count_vm_event(COMPACTSTALL
);
3257 page
= get_page_from_freelist(gfp_mask
, order
, alloc_flags
, ac
);
3260 struct zone
*zone
= page_zone(page
);
3262 zone
->compact_blockskip_flush
= false;
3263 compaction_defer_reset(zone
, order
, true);
3264 count_vm_event(COMPACTSUCCESS
);
3269 * It's bad if compaction run occurs and fails. The most likely reason
3270 * is that pages exist, but not enough to satisfy watermarks.
3272 count_vm_event(COMPACTFAIL
);
3280 should_compact_retry(struct alloc_context
*ac
, int order
, int alloc_flags
,
3281 enum compact_result compact_result
,
3282 enum compact_priority
*compact_priority
,
3283 int *compaction_retries
)
3285 int max_retries
= MAX_COMPACT_RETRIES
;
3288 int retries
= *compaction_retries
;
3289 enum compact_priority priority
= *compact_priority
;
3294 if (compaction_made_progress(compact_result
))
3295 (*compaction_retries
)++;
3298 * compaction considers all the zone as desperately out of memory
3299 * so it doesn't really make much sense to retry except when the
3300 * failure could be caused by insufficient priority
3302 if (compaction_failed(compact_result
))
3303 goto check_priority
;
3306 * make sure the compaction wasn't deferred or didn't bail out early
3307 * due to locks contention before we declare that we should give up.
3308 * But do not retry if the given zonelist is not suitable for
3311 if (compaction_withdrawn(compact_result
)) {
3312 ret
= compaction_zonelist_suitable(ac
, order
, alloc_flags
);
3317 * !costly requests are much more important than __GFP_REPEAT
3318 * costly ones because they are de facto nofail and invoke OOM
3319 * killer to move on while costly can fail and users are ready
3320 * to cope with that. 1/4 retries is rather arbitrary but we
3321 * would need much more detailed feedback from compaction to
3322 * make a better decision.
3324 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
3326 if (*compaction_retries
<= max_retries
) {
3332 * Make sure there are attempts at the highest priority if we exhausted
3333 * all retries or failed at the lower priorities.
3336 min_priority
= (order
> PAGE_ALLOC_COSTLY_ORDER
) ?
3337 MIN_COMPACT_COSTLY_PRIORITY
: MIN_COMPACT_PRIORITY
;
3339 if (*compact_priority
> min_priority
) {
3340 (*compact_priority
)--;
3341 *compaction_retries
= 0;
3345 trace_compact_retry(order
, priority
, compact_result
, retries
, max_retries
, ret
);
3349 static inline struct page
*
3350 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
3351 unsigned int alloc_flags
, const struct alloc_context
*ac
,
3352 enum compact_priority prio
, enum compact_result
*compact_result
)
3354 *compact_result
= COMPACT_SKIPPED
;
3359 should_compact_retry(struct alloc_context
*ac
, unsigned int order
, int alloc_flags
,
3360 enum compact_result compact_result
,
3361 enum compact_priority
*compact_priority
,
3362 int *compaction_retries
)
3367 if (!order
|| order
> PAGE_ALLOC_COSTLY_ORDER
)
3371 * There are setups with compaction disabled which would prefer to loop
3372 * inside the allocator rather than hit the oom killer prematurely.
3373 * Let's give them a good hope and keep retrying while the order-0
3374 * watermarks are OK.
3376 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
, ac
->high_zoneidx
,
3378 if (zone_watermark_ok(zone
, 0, min_wmark_pages(zone
),
3379 ac_classzone_idx(ac
), alloc_flags
))
3384 #endif /* CONFIG_COMPACTION */
3386 /* Perform direct synchronous page reclaim */
3388 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
,
3389 const struct alloc_context
*ac
)
3391 struct reclaim_state reclaim_state
;
3396 /* We now go into synchronous reclaim */
3397 cpuset_memory_pressure_bump();
3398 current
->flags
|= PF_MEMALLOC
;
3399 lockdep_set_current_reclaim_state(gfp_mask
);
3400 reclaim_state
.reclaimed_slab
= 0;
3401 current
->reclaim_state
= &reclaim_state
;
3403 progress
= try_to_free_pages(ac
->zonelist
, order
, gfp_mask
,
3406 current
->reclaim_state
= NULL
;
3407 lockdep_clear_current_reclaim_state();
3408 current
->flags
&= ~PF_MEMALLOC
;
3415 /* The really slow allocator path where we enter direct reclaim */
3416 static inline struct page
*
3417 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
3418 unsigned int alloc_flags
, const struct alloc_context
*ac
,
3419 unsigned long *did_some_progress
)
3421 struct page
*page
= NULL
;
3422 bool drained
= false;
3424 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, ac
);
3425 if (unlikely(!(*did_some_progress
)))
3429 page
= get_page_from_freelist(gfp_mask
, order
, alloc_flags
, ac
);
3432 * If an allocation failed after direct reclaim, it could be because
3433 * pages are pinned on the per-cpu lists or in high alloc reserves.
3434 * Shrink them them and try again
3436 if (!page
&& !drained
) {
3437 unreserve_highatomic_pageblock(ac
, false);
3438 drain_all_pages(NULL
);
3446 static void wake_all_kswapds(unsigned int order
, const struct alloc_context
*ac
)
3450 pg_data_t
*last_pgdat
= NULL
;
3452 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
,
3453 ac
->high_zoneidx
, ac
->nodemask
) {
3454 if (last_pgdat
!= zone
->zone_pgdat
)
3455 wakeup_kswapd(zone
, order
, ac
->high_zoneidx
);
3456 last_pgdat
= zone
->zone_pgdat
;
3460 static inline unsigned int
3461 gfp_to_alloc_flags(gfp_t gfp_mask
)
3463 unsigned int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
3465 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
3466 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
3469 * The caller may dip into page reserves a bit more if the caller
3470 * cannot run direct reclaim, or if the caller has realtime scheduling
3471 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
3472 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
3474 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
3476 if (gfp_mask
& __GFP_ATOMIC
) {
3478 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
3479 * if it can't schedule.
3481 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
3482 alloc_flags
|= ALLOC_HARDER
;
3484 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
3485 * comment for __cpuset_node_allowed().
3487 alloc_flags
&= ~ALLOC_CPUSET
;
3488 } else if (unlikely(rt_task(current
)) && !in_interrupt())
3489 alloc_flags
|= ALLOC_HARDER
;
3492 if (gfpflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
3493 alloc_flags
|= ALLOC_CMA
;
3498 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask
)
3500 if (unlikely(gfp_mask
& __GFP_NOMEMALLOC
))
3503 if (gfp_mask
& __GFP_MEMALLOC
)
3505 if (in_serving_softirq() && (current
->flags
& PF_MEMALLOC
))
3507 if (!in_interrupt() &&
3508 ((current
->flags
& PF_MEMALLOC
) ||
3509 unlikely(test_thread_flag(TIF_MEMDIE
))))
3516 * Maximum number of reclaim retries without any progress before OOM killer
3517 * is consider as the only way to move forward.
3519 #define MAX_RECLAIM_RETRIES 16
3522 * Checks whether it makes sense to retry the reclaim to make a forward progress
3523 * for the given allocation request.
3524 * The reclaim feedback represented by did_some_progress (any progress during
3525 * the last reclaim round) and no_progress_loops (number of reclaim rounds without
3526 * any progress in a row) is considered as well as the reclaimable pages on the
3527 * applicable zone list (with a backoff mechanism which is a function of
3528 * no_progress_loops).
3530 * Returns true if a retry is viable or false to enter the oom path.
3533 should_reclaim_retry(gfp_t gfp_mask
, unsigned order
,
3534 struct alloc_context
*ac
, int alloc_flags
,
3535 bool did_some_progress
, int *no_progress_loops
)
3541 * Costly allocations might have made a progress but this doesn't mean
3542 * their order will become available due to high fragmentation so
3543 * always increment the no progress counter for them
3545 if (did_some_progress
&& order
<= PAGE_ALLOC_COSTLY_ORDER
)
3546 *no_progress_loops
= 0;
3548 (*no_progress_loops
)++;
3551 * Make sure we converge to OOM if we cannot make any progress
3552 * several times in the row.
3554 if (*no_progress_loops
> MAX_RECLAIM_RETRIES
) {
3555 /* Before OOM, exhaust highatomic_reserve */
3556 return unreserve_highatomic_pageblock(ac
, true);
3560 * Keep reclaiming pages while there is a chance this will lead
3561 * somewhere. If none of the target zones can satisfy our allocation
3562 * request even if all reclaimable pages are considered then we are
3563 * screwed and have to go OOM.
3565 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
, ac
->high_zoneidx
,
3567 unsigned long available
;
3568 unsigned long reclaimable
;
3569 unsigned long min_wmark
= min_wmark_pages(zone
);
3572 available
= reclaimable
= zone_reclaimable_pages(zone
);
3573 available
-= DIV_ROUND_UP((*no_progress_loops
) * available
,
3574 MAX_RECLAIM_RETRIES
);
3575 available
+= zone_page_state_snapshot(zone
, NR_FREE_PAGES
);
3578 * Would the allocation succeed if we reclaimed the whole
3581 wmark
= __zone_watermark_ok(zone
, order
, min_wmark
,
3582 ac_classzone_idx(ac
), alloc_flags
, available
);
3583 trace_reclaim_retry_zone(z
, order
, reclaimable
,
3584 available
, min_wmark
, *no_progress_loops
, wmark
);
3587 * If we didn't make any progress and have a lot of
3588 * dirty + writeback pages then we should wait for
3589 * an IO to complete to slow down the reclaim and
3590 * prevent from pre mature OOM
3592 if (!did_some_progress
) {
3593 unsigned long write_pending
;
3595 write_pending
= zone_page_state_snapshot(zone
,
3596 NR_ZONE_WRITE_PENDING
);
3598 if (2 * write_pending
> reclaimable
) {
3599 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3605 * Memory allocation/reclaim might be called from a WQ
3606 * context and the current implementation of the WQ
3607 * concurrency control doesn't recognize that
3608 * a particular WQ is congested if the worker thread is
3609 * looping without ever sleeping. Therefore we have to
3610 * do a short sleep here rather than calling
3613 if (current
->flags
& PF_WQ_WORKER
)
3614 schedule_timeout_uninterruptible(1);
3625 static inline struct page
*
3626 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
3627 struct alloc_context
*ac
)
3629 bool can_direct_reclaim
= gfp_mask
& __GFP_DIRECT_RECLAIM
;
3630 struct page
*page
= NULL
;
3631 unsigned int alloc_flags
;
3632 unsigned long did_some_progress
;
3633 enum compact_priority compact_priority
;
3634 enum compact_result compact_result
;
3635 int compaction_retries
;
3636 int no_progress_loops
;
3637 unsigned long alloc_start
= jiffies
;
3638 unsigned int stall_timeout
= 10 * HZ
;
3639 unsigned int cpuset_mems_cookie
;
3642 * In the slowpath, we sanity check order to avoid ever trying to
3643 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3644 * be using allocators in order of preference for an area that is
3647 if (order
>= MAX_ORDER
) {
3648 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
3653 * We also sanity check to catch abuse of atomic reserves being used by
3654 * callers that are not in atomic context.
3656 if (WARN_ON_ONCE((gfp_mask
& (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)) ==
3657 (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)))
3658 gfp_mask
&= ~__GFP_ATOMIC
;
3661 compaction_retries
= 0;
3662 no_progress_loops
= 0;
3663 compact_priority
= DEF_COMPACT_PRIORITY
;
3664 cpuset_mems_cookie
= read_mems_allowed_begin();
3667 * The fast path uses conservative alloc_flags to succeed only until
3668 * kswapd needs to be woken up, and to avoid the cost of setting up
3669 * alloc_flags precisely. So we do that now.
3671 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
3674 * We need to recalculate the starting point for the zonelist iterator
3675 * because we might have used different nodemask in the fast path, or
3676 * there was a cpuset modification and we are retrying - otherwise we
3677 * could end up iterating over non-eligible zones endlessly.
3679 ac
->preferred_zoneref
= first_zones_zonelist(ac
->zonelist
,
3680 ac
->high_zoneidx
, ac
->nodemask
);
3681 if (!ac
->preferred_zoneref
->zone
)
3684 if (gfp_mask
& __GFP_KSWAPD_RECLAIM
)
3685 wake_all_kswapds(order
, ac
);
3688 * The adjusted alloc_flags might result in immediate success, so try
3691 page
= get_page_from_freelist(gfp_mask
, order
, alloc_flags
, ac
);
3696 * For costly allocations, try direct compaction first, as it's likely
3697 * that we have enough base pages and don't need to reclaim. Don't try
3698 * that for allocations that are allowed to ignore watermarks, as the
3699 * ALLOC_NO_WATERMARKS attempt didn't yet happen.
3701 if (can_direct_reclaim
&& order
> PAGE_ALLOC_COSTLY_ORDER
&&
3702 !gfp_pfmemalloc_allowed(gfp_mask
)) {
3703 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
3705 INIT_COMPACT_PRIORITY
,
3711 * Checks for costly allocations with __GFP_NORETRY, which
3712 * includes THP page fault allocations
3714 if (gfp_mask
& __GFP_NORETRY
) {
3716 * If compaction is deferred for high-order allocations,
3717 * it is because sync compaction recently failed. If
3718 * this is the case and the caller requested a THP
3719 * allocation, we do not want to heavily disrupt the
3720 * system, so we fail the allocation instead of entering
3723 if (compact_result
== COMPACT_DEFERRED
)
3727 * Looks like reclaim/compaction is worth trying, but
3728 * sync compaction could be very expensive, so keep
3729 * using async compaction.
3731 compact_priority
= INIT_COMPACT_PRIORITY
;
3736 /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */
3737 if (gfp_mask
& __GFP_KSWAPD_RECLAIM
)
3738 wake_all_kswapds(order
, ac
);
3740 if (gfp_pfmemalloc_allowed(gfp_mask
))
3741 alloc_flags
= ALLOC_NO_WATERMARKS
;
3744 * Reset the zonelist iterators if memory policies can be ignored.
3745 * These allocations are high priority and system rather than user
3748 if (!(alloc_flags
& ALLOC_CPUSET
) || (alloc_flags
& ALLOC_NO_WATERMARKS
)) {
3749 ac
->zonelist
= node_zonelist(numa_node_id(), gfp_mask
);
3750 ac
->preferred_zoneref
= first_zones_zonelist(ac
->zonelist
,
3751 ac
->high_zoneidx
, ac
->nodemask
);
3754 /* Attempt with potentially adjusted zonelist and alloc_flags */
3755 page
= get_page_from_freelist(gfp_mask
, order
, alloc_flags
, ac
);
3759 /* Caller is not willing to reclaim, we can't balance anything */
3760 if (!can_direct_reclaim
)
3763 /* Make sure we know about allocations which stall for too long */
3764 if (time_after(jiffies
, alloc_start
+ stall_timeout
)) {
3765 warn_alloc(gfp_mask
, ac
->nodemask
,
3766 "page allocation stalls for %ums, order:%u",
3767 jiffies_to_msecs(jiffies
-alloc_start
), order
);
3768 stall_timeout
+= 10 * HZ
;
3771 /* Avoid recursion of direct reclaim */
3772 if (current
->flags
& PF_MEMALLOC
)
3775 /* Try direct reclaim and then allocating */
3776 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
, alloc_flags
, ac
,
3777 &did_some_progress
);
3781 /* Try direct compaction and then allocating */
3782 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
, ac
,
3783 compact_priority
, &compact_result
);
3787 /* Do not loop if specifically requested */
3788 if (gfp_mask
& __GFP_NORETRY
)
3792 * Do not retry costly high order allocations unless they are
3795 if (order
> PAGE_ALLOC_COSTLY_ORDER
&& !(gfp_mask
& __GFP_REPEAT
))
3798 if (should_reclaim_retry(gfp_mask
, order
, ac
, alloc_flags
,
3799 did_some_progress
> 0, &no_progress_loops
))
3803 * It doesn't make any sense to retry for the compaction if the order-0
3804 * reclaim is not able to make any progress because the current
3805 * implementation of the compaction depends on the sufficient amount
3806 * of free memory (see __compaction_suitable)
3808 if (did_some_progress
> 0 &&
3809 should_compact_retry(ac
, order
, alloc_flags
,
3810 compact_result
, &compact_priority
,
3811 &compaction_retries
))
3815 * It's possible we raced with cpuset update so the OOM would be
3816 * premature (see below the nopage: label for full explanation).
3818 if (read_mems_allowed_retry(cpuset_mems_cookie
))
3821 /* Reclaim has failed us, start killing things */
3822 page
= __alloc_pages_may_oom(gfp_mask
, order
, ac
, &did_some_progress
);
3826 /* Avoid allocations with no watermarks from looping endlessly */
3827 if (test_thread_flag(TIF_MEMDIE
))
3830 /* Retry as long as the OOM killer is making progress */
3831 if (did_some_progress
) {
3832 no_progress_loops
= 0;
3838 * When updating a task's mems_allowed or mempolicy nodemask, it is
3839 * possible to race with parallel threads in such a way that our
3840 * allocation can fail while the mask is being updated. If we are about
3841 * to fail, check if the cpuset changed during allocation and if so,
3844 if (read_mems_allowed_retry(cpuset_mems_cookie
))
3848 * Make sure that __GFP_NOFAIL request doesn't leak out and make sure
3851 if (gfp_mask
& __GFP_NOFAIL
) {
3853 * All existing users of the __GFP_NOFAIL are blockable, so warn
3854 * of any new users that actually require GFP_NOWAIT
3856 if (WARN_ON_ONCE(!can_direct_reclaim
))
3860 * PF_MEMALLOC request from this context is rather bizarre
3861 * because we cannot reclaim anything and only can loop waiting
3862 * for somebody to do a work for us
3864 WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
);
3867 * non failing costly orders are a hard requirement which we
3868 * are not prepared for much so let's warn about these users
3869 * so that we can identify them and convert them to something
3872 WARN_ON_ONCE(order
> PAGE_ALLOC_COSTLY_ORDER
);
3875 * Help non-failing allocations by giving them access to memory
3876 * reserves but do not use ALLOC_NO_WATERMARKS because this
3877 * could deplete whole memory reserves which would just make
3878 * the situation worse
3880 page
= __alloc_pages_cpuset_fallback(gfp_mask
, order
, ALLOC_HARDER
, ac
);
3888 warn_alloc(gfp_mask
, ac
->nodemask
,
3889 "page allocation failure: order:%u", order
);
3894 static inline bool prepare_alloc_pages(gfp_t gfp_mask
, unsigned int order
,
3895 struct zonelist
*zonelist
, nodemask_t
*nodemask
,
3896 struct alloc_context
*ac
, gfp_t
*alloc_mask
,
3897 unsigned int *alloc_flags
)
3899 ac
->high_zoneidx
= gfp_zone(gfp_mask
);
3900 ac
->zonelist
= zonelist
;
3901 ac
->nodemask
= nodemask
;
3902 ac
->migratetype
= gfpflags_to_migratetype(gfp_mask
);
3904 if (cpusets_enabled()) {
3905 *alloc_mask
|= __GFP_HARDWALL
;
3907 ac
->nodemask
= &cpuset_current_mems_allowed
;
3909 *alloc_flags
|= ALLOC_CPUSET
;
3912 lockdep_trace_alloc(gfp_mask
);
3914 might_sleep_if(gfp_mask
& __GFP_DIRECT_RECLAIM
);
3916 if (should_fail_alloc_page(gfp_mask
, order
))
3919 if (IS_ENABLED(CONFIG_CMA
) && ac
->migratetype
== MIGRATE_MOVABLE
)
3920 *alloc_flags
|= ALLOC_CMA
;
3925 /* Determine whether to spread dirty pages and what the first usable zone */
3926 static inline void finalise_ac(gfp_t gfp_mask
,
3927 unsigned int order
, struct alloc_context
*ac
)
3929 /* Dirty zone balancing only done in the fast path */
3930 ac
->spread_dirty_pages
= (gfp_mask
& __GFP_WRITE
);
3933 * The preferred zone is used for statistics but crucially it is
3934 * also used as the starting point for the zonelist iterator. It
3935 * may get reset for allocations that ignore memory policies.
3937 ac
->preferred_zoneref
= first_zones_zonelist(ac
->zonelist
,
3938 ac
->high_zoneidx
, ac
->nodemask
);
3942 * This is the 'heart' of the zoned buddy allocator.
3945 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
3946 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
3949 unsigned int alloc_flags
= ALLOC_WMARK_LOW
;
3950 gfp_t alloc_mask
= gfp_mask
; /* The gfp_t that was actually used for allocation */
3951 struct alloc_context ac
= { };
3953 gfp_mask
&= gfp_allowed_mask
;
3954 if (!prepare_alloc_pages(gfp_mask
, order
, zonelist
, nodemask
, &ac
, &alloc_mask
, &alloc_flags
))
3957 finalise_ac(gfp_mask
, order
, &ac
);
3959 /* First allocation attempt */
3960 page
= get_page_from_freelist(alloc_mask
, order
, alloc_flags
, &ac
);
3965 * Runtime PM, block IO and its error handling path can deadlock
3966 * because I/O on the device might not complete.
3968 alloc_mask
= memalloc_noio_flags(gfp_mask
);
3969 ac
.spread_dirty_pages
= false;
3972 * Restore the original nodemask if it was potentially replaced with
3973 * &cpuset_current_mems_allowed to optimize the fast-path attempt.
3975 if (unlikely(ac
.nodemask
!= nodemask
))
3976 ac
.nodemask
= nodemask
;
3978 page
= __alloc_pages_slowpath(alloc_mask
, order
, &ac
);
3981 if (memcg_kmem_enabled() && (gfp_mask
& __GFP_ACCOUNT
) && page
&&
3982 unlikely(memcg_kmem_charge(page
, gfp_mask
, order
) != 0)) {
3983 __free_pages(page
, order
);
3987 if (kmemcheck_enabled
&& page
)
3988 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
3990 trace_mm_page_alloc(page
, order
, alloc_mask
, ac
.migratetype
);
3994 EXPORT_SYMBOL(__alloc_pages_nodemask
);
3997 * Common helper functions.
3999 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
4004 * __get_free_pages() returns a 32-bit address, which cannot represent
4007 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
4009 page
= alloc_pages(gfp_mask
, order
);
4012 return (unsigned long) page_address(page
);
4014 EXPORT_SYMBOL(__get_free_pages
);
4016 unsigned long get_zeroed_page(gfp_t gfp_mask
)
4018 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
4020 EXPORT_SYMBOL(get_zeroed_page
);
4022 void __free_pages(struct page
*page
, unsigned int order
)
4024 if (put_page_testzero(page
)) {
4026 free_hot_cold_page(page
, false);
4028 __free_pages_ok(page
, order
);
4032 EXPORT_SYMBOL(__free_pages
);
4034 void free_pages(unsigned long addr
, unsigned int order
)
4037 VM_BUG_ON(!virt_addr_valid((void *)addr
));
4038 __free_pages(virt_to_page((void *)addr
), order
);
4042 EXPORT_SYMBOL(free_pages
);
4046 * An arbitrary-length arbitrary-offset area of memory which resides
4047 * within a 0 or higher order page. Multiple fragments within that page
4048 * are individually refcounted, in the page's reference counter.
4050 * The page_frag functions below provide a simple allocation framework for
4051 * page fragments. This is used by the network stack and network device
4052 * drivers to provide a backing region of memory for use as either an
4053 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
4055 static struct page
*__page_frag_cache_refill(struct page_frag_cache
*nc
,
4058 struct page
*page
= NULL
;
4059 gfp_t gfp
= gfp_mask
;
4061 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4062 gfp_mask
|= __GFP_COMP
| __GFP_NOWARN
| __GFP_NORETRY
|
4064 page
= alloc_pages_node(NUMA_NO_NODE
, gfp_mask
,
4065 PAGE_FRAG_CACHE_MAX_ORDER
);
4066 nc
->size
= page
? PAGE_FRAG_CACHE_MAX_SIZE
: PAGE_SIZE
;
4068 if (unlikely(!page
))
4069 page
= alloc_pages_node(NUMA_NO_NODE
, gfp
, 0);
4071 nc
->va
= page
? page_address(page
) : NULL
;
4076 void __page_frag_cache_drain(struct page
*page
, unsigned int count
)
4078 VM_BUG_ON_PAGE(page_ref_count(page
) == 0, page
);
4080 if (page_ref_sub_and_test(page
, count
)) {
4081 unsigned int order
= compound_order(page
);
4084 free_hot_cold_page(page
, false);
4086 __free_pages_ok(page
, order
);
4089 EXPORT_SYMBOL(__page_frag_cache_drain
);
4091 void *page_frag_alloc(struct page_frag_cache
*nc
,
4092 unsigned int fragsz
, gfp_t gfp_mask
)
4094 unsigned int size
= PAGE_SIZE
;
4098 if (unlikely(!nc
->va
)) {
4100 page
= __page_frag_cache_refill(nc
, gfp_mask
);
4104 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4105 /* if size can vary use size else just use PAGE_SIZE */
4108 /* Even if we own the page, we do not use atomic_set().
4109 * This would break get_page_unless_zero() users.
4111 page_ref_add(page
, size
- 1);
4113 /* reset page count bias and offset to start of new frag */
4114 nc
->pfmemalloc
= page_is_pfmemalloc(page
);
4115 nc
->pagecnt_bias
= size
;
4119 offset
= nc
->offset
- fragsz
;
4120 if (unlikely(offset
< 0)) {
4121 page
= virt_to_page(nc
->va
);
4123 if (!page_ref_sub_and_test(page
, nc
->pagecnt_bias
))
4126 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4127 /* if size can vary use size else just use PAGE_SIZE */
4130 /* OK, page count is 0, we can safely set it */
4131 set_page_count(page
, size
);
4133 /* reset page count bias and offset to start of new frag */
4134 nc
->pagecnt_bias
= size
;
4135 offset
= size
- fragsz
;
4139 nc
->offset
= offset
;
4141 return nc
->va
+ offset
;
4143 EXPORT_SYMBOL(page_frag_alloc
);
4146 * Frees a page fragment allocated out of either a compound or order 0 page.
4148 void page_frag_free(void *addr
)
4150 struct page
*page
= virt_to_head_page(addr
);
4152 if (unlikely(put_page_testzero(page
)))
4153 __free_pages_ok(page
, compound_order(page
));
4155 EXPORT_SYMBOL(page_frag_free
);
4157 static void *make_alloc_exact(unsigned long addr
, unsigned int order
,
4161 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
4162 unsigned long used
= addr
+ PAGE_ALIGN(size
);
4164 split_page(virt_to_page((void *)addr
), order
);
4165 while (used
< alloc_end
) {
4170 return (void *)addr
;
4174 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
4175 * @size: the number of bytes to allocate
4176 * @gfp_mask: GFP flags for the allocation
4178 * This function is similar to alloc_pages(), except that it allocates the
4179 * minimum number of pages to satisfy the request. alloc_pages() can only
4180 * allocate memory in power-of-two pages.
4182 * This function is also limited by MAX_ORDER.
4184 * Memory allocated by this function must be released by free_pages_exact().
4186 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
4188 unsigned int order
= get_order(size
);
4191 addr
= __get_free_pages(gfp_mask
, order
);
4192 return make_alloc_exact(addr
, order
, size
);
4194 EXPORT_SYMBOL(alloc_pages_exact
);
4197 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
4199 * @nid: the preferred node ID where memory should be allocated
4200 * @size: the number of bytes to allocate
4201 * @gfp_mask: GFP flags for the allocation
4203 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
4206 void * __meminit
alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
4208 unsigned int order
= get_order(size
);
4209 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
4212 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
4216 * free_pages_exact - release memory allocated via alloc_pages_exact()
4217 * @virt: the value returned by alloc_pages_exact.
4218 * @size: size of allocation, same value as passed to alloc_pages_exact().
4220 * Release the memory allocated by a previous call to alloc_pages_exact.
4222 void free_pages_exact(void *virt
, size_t size
)
4224 unsigned long addr
= (unsigned long)virt
;
4225 unsigned long end
= addr
+ PAGE_ALIGN(size
);
4227 while (addr
< end
) {
4232 EXPORT_SYMBOL(free_pages_exact
);
4235 * nr_free_zone_pages - count number of pages beyond high watermark
4236 * @offset: The zone index of the highest zone
4238 * nr_free_zone_pages() counts the number of counts pages which are beyond the
4239 * high watermark within all zones at or below a given zone index. For each
4240 * zone, the number of pages is calculated as:
4241 * managed_pages - high_pages
4243 static unsigned long nr_free_zone_pages(int offset
)
4248 /* Just pick one node, since fallback list is circular */
4249 unsigned long sum
= 0;
4251 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
4253 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
4254 unsigned long size
= zone
->managed_pages
;
4255 unsigned long high
= high_wmark_pages(zone
);
4264 * nr_free_buffer_pages - count number of pages beyond high watermark
4266 * nr_free_buffer_pages() counts the number of pages which are beyond the high
4267 * watermark within ZONE_DMA and ZONE_NORMAL.
4269 unsigned long nr_free_buffer_pages(void)
4271 return nr_free_zone_pages(gfp_zone(GFP_USER
));
4273 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
4276 * nr_free_pagecache_pages - count number of pages beyond high watermark
4278 * nr_free_pagecache_pages() counts the number of pages which are beyond the
4279 * high watermark within all zones.
4281 unsigned long nr_free_pagecache_pages(void)
4283 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
4286 static inline void show_node(struct zone
*zone
)
4288 if (IS_ENABLED(CONFIG_NUMA
))
4289 printk("Node %d ", zone_to_nid(zone
));
4292 long si_mem_available(void)
4295 unsigned long pagecache
;
4296 unsigned long wmark_low
= 0;
4297 unsigned long pages
[NR_LRU_LISTS
];
4301 for (lru
= LRU_BASE
; lru
< NR_LRU_LISTS
; lru
++)
4302 pages
[lru
] = global_node_page_state(NR_LRU_BASE
+ lru
);
4305 wmark_low
+= zone
->watermark
[WMARK_LOW
];
4308 * Estimate the amount of memory available for userspace allocations,
4309 * without causing swapping.
4311 available
= global_page_state(NR_FREE_PAGES
) - totalreserve_pages
;
4314 * Not all the page cache can be freed, otherwise the system will
4315 * start swapping. Assume at least half of the page cache, or the
4316 * low watermark worth of cache, needs to stay.
4318 pagecache
= pages
[LRU_ACTIVE_FILE
] + pages
[LRU_INACTIVE_FILE
];
4319 pagecache
-= min(pagecache
/ 2, wmark_low
);
4320 available
+= pagecache
;
4323 * Part of the reclaimable slab consists of items that are in use,
4324 * and cannot be freed. Cap this estimate at the low watermark.
4326 available
+= global_page_state(NR_SLAB_RECLAIMABLE
) -
4327 min(global_page_state(NR_SLAB_RECLAIMABLE
) / 2, wmark_low
);
4333 EXPORT_SYMBOL_GPL(si_mem_available
);
4335 void si_meminfo(struct sysinfo
*val
)
4337 val
->totalram
= totalram_pages
;
4338 val
->sharedram
= global_node_page_state(NR_SHMEM
);
4339 val
->freeram
= global_page_state(NR_FREE_PAGES
);
4340 val
->bufferram
= nr_blockdev_pages();
4341 val
->totalhigh
= totalhigh_pages
;
4342 val
->freehigh
= nr_free_highpages();
4343 val
->mem_unit
= PAGE_SIZE
;
4346 EXPORT_SYMBOL(si_meminfo
);
4349 void si_meminfo_node(struct sysinfo
*val
, int nid
)
4351 int zone_type
; /* needs to be signed */
4352 unsigned long managed_pages
= 0;
4353 unsigned long managed_highpages
= 0;
4354 unsigned long free_highpages
= 0;
4355 pg_data_t
*pgdat
= NODE_DATA(nid
);
4357 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++)
4358 managed_pages
+= pgdat
->node_zones
[zone_type
].managed_pages
;
4359 val
->totalram
= managed_pages
;
4360 val
->sharedram
= node_page_state(pgdat
, NR_SHMEM
);
4361 val
->freeram
= sum_zone_node_page_state(nid
, NR_FREE_PAGES
);
4362 #ifdef CONFIG_HIGHMEM
4363 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
4364 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4366 if (is_highmem(zone
)) {
4367 managed_highpages
+= zone
->managed_pages
;
4368 free_highpages
+= zone_page_state(zone
, NR_FREE_PAGES
);
4371 val
->totalhigh
= managed_highpages
;
4372 val
->freehigh
= free_highpages
;
4374 val
->totalhigh
= managed_highpages
;
4375 val
->freehigh
= free_highpages
;
4377 val
->mem_unit
= PAGE_SIZE
;
4382 * Determine whether the node should be displayed or not, depending on whether
4383 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
4385 static bool show_mem_node_skip(unsigned int flags
, int nid
, nodemask_t
*nodemask
)
4387 if (!(flags
& SHOW_MEM_FILTER_NODES
))
4391 * no node mask - aka implicit memory numa policy. Do not bother with
4392 * the synchronization - read_mems_allowed_begin - because we do not
4393 * have to be precise here.
4396 nodemask
= &cpuset_current_mems_allowed
;
4398 return !node_isset(nid
, *nodemask
);
4401 #define K(x) ((x) << (PAGE_SHIFT-10))
4403 static void show_migration_types(unsigned char type
)
4405 static const char types
[MIGRATE_TYPES
] = {
4406 [MIGRATE_UNMOVABLE
] = 'U',
4407 [MIGRATE_MOVABLE
] = 'M',
4408 [MIGRATE_RECLAIMABLE
] = 'E',
4409 [MIGRATE_HIGHATOMIC
] = 'H',
4411 [MIGRATE_CMA
] = 'C',
4413 #ifdef CONFIG_MEMORY_ISOLATION
4414 [MIGRATE_ISOLATE
] = 'I',
4417 char tmp
[MIGRATE_TYPES
+ 1];
4421 for (i
= 0; i
< MIGRATE_TYPES
; i
++) {
4422 if (type
& (1 << i
))
4427 printk(KERN_CONT
"(%s) ", tmp
);
4431 * Show free area list (used inside shift_scroll-lock stuff)
4432 * We also calculate the percentage fragmentation. We do this by counting the
4433 * memory on each free list with the exception of the first item on the list.
4436 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
4439 void show_free_areas(unsigned int filter
, nodemask_t
*nodemask
)
4441 unsigned long free_pcp
= 0;
4446 for_each_populated_zone(zone
) {
4447 if (show_mem_node_skip(filter
, zone_to_nid(zone
), nodemask
))
4450 for_each_online_cpu(cpu
)
4451 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
4454 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
4455 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
4456 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
4457 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
4458 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
4459 " free:%lu free_pcp:%lu free_cma:%lu\n",
4460 global_node_page_state(NR_ACTIVE_ANON
),
4461 global_node_page_state(NR_INACTIVE_ANON
),
4462 global_node_page_state(NR_ISOLATED_ANON
),
4463 global_node_page_state(NR_ACTIVE_FILE
),
4464 global_node_page_state(NR_INACTIVE_FILE
),
4465 global_node_page_state(NR_ISOLATED_FILE
),
4466 global_node_page_state(NR_UNEVICTABLE
),
4467 global_node_page_state(NR_FILE_DIRTY
),
4468 global_node_page_state(NR_WRITEBACK
),
4469 global_node_page_state(NR_UNSTABLE_NFS
),
4470 global_page_state(NR_SLAB_RECLAIMABLE
),
4471 global_page_state(NR_SLAB_UNRECLAIMABLE
),
4472 global_node_page_state(NR_FILE_MAPPED
),
4473 global_node_page_state(NR_SHMEM
),
4474 global_page_state(NR_PAGETABLE
),
4475 global_page_state(NR_BOUNCE
),
4476 global_page_state(NR_FREE_PAGES
),
4478 global_page_state(NR_FREE_CMA_PAGES
));
4480 for_each_online_pgdat(pgdat
) {
4481 if (show_mem_node_skip(filter
, pgdat
->node_id
, nodemask
))
4485 " active_anon:%lukB"
4486 " inactive_anon:%lukB"
4487 " active_file:%lukB"
4488 " inactive_file:%lukB"
4489 " unevictable:%lukB"
4490 " isolated(anon):%lukB"
4491 " isolated(file):%lukB"
4496 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4498 " shmem_pmdmapped: %lukB"
4501 " writeback_tmp:%lukB"
4503 " pages_scanned:%lu"
4504 " all_unreclaimable? %s"
4507 K(node_page_state(pgdat
, NR_ACTIVE_ANON
)),
4508 K(node_page_state(pgdat
, NR_INACTIVE_ANON
)),
4509 K(node_page_state(pgdat
, NR_ACTIVE_FILE
)),
4510 K(node_page_state(pgdat
, NR_INACTIVE_FILE
)),
4511 K(node_page_state(pgdat
, NR_UNEVICTABLE
)),
4512 K(node_page_state(pgdat
, NR_ISOLATED_ANON
)),
4513 K(node_page_state(pgdat
, NR_ISOLATED_FILE
)),
4514 K(node_page_state(pgdat
, NR_FILE_MAPPED
)),
4515 K(node_page_state(pgdat
, NR_FILE_DIRTY
)),
4516 K(node_page_state(pgdat
, NR_WRITEBACK
)),
4517 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4518 K(node_page_state(pgdat
, NR_SHMEM_THPS
) * HPAGE_PMD_NR
),
4519 K(node_page_state(pgdat
, NR_SHMEM_PMDMAPPED
)
4521 K(node_page_state(pgdat
, NR_ANON_THPS
) * HPAGE_PMD_NR
),
4523 K(node_page_state(pgdat
, NR_SHMEM
)),
4524 K(node_page_state(pgdat
, NR_WRITEBACK_TEMP
)),
4525 K(node_page_state(pgdat
, NR_UNSTABLE_NFS
)),
4526 node_page_state(pgdat
, NR_PAGES_SCANNED
),
4527 !pgdat_reclaimable(pgdat
) ? "yes" : "no");
4530 for_each_populated_zone(zone
) {
4533 if (show_mem_node_skip(filter
, zone_to_nid(zone
), nodemask
))
4537 for_each_online_cpu(cpu
)
4538 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
4547 " active_anon:%lukB"
4548 " inactive_anon:%lukB"
4549 " active_file:%lukB"
4550 " inactive_file:%lukB"
4551 " unevictable:%lukB"
4552 " writepending:%lukB"
4556 " slab_reclaimable:%lukB"
4557 " slab_unreclaimable:%lukB"
4558 " kernel_stack:%lukB"
4566 K(zone_page_state(zone
, NR_FREE_PAGES
)),
4567 K(min_wmark_pages(zone
)),
4568 K(low_wmark_pages(zone
)),
4569 K(high_wmark_pages(zone
)),
4570 K(zone_page_state(zone
, NR_ZONE_ACTIVE_ANON
)),
4571 K(zone_page_state(zone
, NR_ZONE_INACTIVE_ANON
)),
4572 K(zone_page_state(zone
, NR_ZONE_ACTIVE_FILE
)),
4573 K(zone_page_state(zone
, NR_ZONE_INACTIVE_FILE
)),
4574 K(zone_page_state(zone
, NR_ZONE_UNEVICTABLE
)),
4575 K(zone_page_state(zone
, NR_ZONE_WRITE_PENDING
)),
4576 K(zone
->present_pages
),
4577 K(zone
->managed_pages
),
4578 K(zone_page_state(zone
, NR_MLOCK
)),
4579 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
4580 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
4581 zone_page_state(zone
, NR_KERNEL_STACK_KB
),
4582 K(zone_page_state(zone
, NR_PAGETABLE
)),
4583 K(zone_page_state(zone
, NR_BOUNCE
)),
4585 K(this_cpu_read(zone
->pageset
->pcp
.count
)),
4586 K(zone_page_state(zone
, NR_FREE_CMA_PAGES
)));
4587 printk("lowmem_reserve[]:");
4588 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4589 printk(KERN_CONT
" %ld", zone
->lowmem_reserve
[i
]);
4590 printk(KERN_CONT
"\n");
4593 for_each_populated_zone(zone
) {
4595 unsigned long nr
[MAX_ORDER
], flags
, total
= 0;
4596 unsigned char types
[MAX_ORDER
];
4598 if (show_mem_node_skip(filter
, zone_to_nid(zone
), nodemask
))
4601 printk(KERN_CONT
"%s: ", zone
->name
);
4603 spin_lock_irqsave(&zone
->lock
, flags
);
4604 for (order
= 0; order
< MAX_ORDER
; order
++) {
4605 struct free_area
*area
= &zone
->free_area
[order
];
4608 nr
[order
] = area
->nr_free
;
4609 total
+= nr
[order
] << order
;
4612 for (type
= 0; type
< MIGRATE_TYPES
; type
++) {
4613 if (!list_empty(&area
->free_list
[type
]))
4614 types
[order
] |= 1 << type
;
4617 spin_unlock_irqrestore(&zone
->lock
, flags
);
4618 for (order
= 0; order
< MAX_ORDER
; order
++) {
4619 printk(KERN_CONT
"%lu*%lukB ",
4620 nr
[order
], K(1UL) << order
);
4622 show_migration_types(types
[order
]);
4624 printk(KERN_CONT
"= %lukB\n", K(total
));
4627 hugetlb_show_meminfo();
4629 printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES
));
4631 show_swap_cache_info();
4634 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
4636 zoneref
->zone
= zone
;
4637 zoneref
->zone_idx
= zone_idx(zone
);
4641 * Builds allocation fallback zone lists.
4643 * Add all populated zones of a node to the zonelist.
4645 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
4649 enum zone_type zone_type
= MAX_NR_ZONES
;
4653 zone
= pgdat
->node_zones
+ zone_type
;
4654 if (managed_zone(zone
)) {
4655 zoneref_set_zone(zone
,
4656 &zonelist
->_zonerefs
[nr_zones
++]);
4657 check_highest_zone(zone_type
);
4659 } while (zone_type
);
4667 * 0 = automatic detection of better ordering.
4668 * 1 = order by ([node] distance, -zonetype)
4669 * 2 = order by (-zonetype, [node] distance)
4671 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
4672 * the same zonelist. So only NUMA can configure this param.
4674 #define ZONELIST_ORDER_DEFAULT 0
4675 #define ZONELIST_ORDER_NODE 1
4676 #define ZONELIST_ORDER_ZONE 2
4678 /* zonelist order in the kernel.
4679 * set_zonelist_order() will set this to NODE or ZONE.
4681 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4682 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
4686 /* The value user specified ....changed by config */
4687 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4688 /* string for sysctl */
4689 #define NUMA_ZONELIST_ORDER_LEN 16
4690 char numa_zonelist_order
[16] = "default";
4693 * interface for configure zonelist ordering.
4694 * command line option "numa_zonelist_order"
4695 * = "[dD]efault - default, automatic configuration.
4696 * = "[nN]ode - order by node locality, then by zone within node
4697 * = "[zZ]one - order by zone, then by locality within zone
4700 static int __parse_numa_zonelist_order(char *s
)
4702 if (*s
== 'd' || *s
== 'D') {
4703 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4704 } else if (*s
== 'n' || *s
== 'N') {
4705 user_zonelist_order
= ZONELIST_ORDER_NODE
;
4706 } else if (*s
== 'z' || *s
== 'Z') {
4707 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
4709 pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s
);
4715 static __init
int setup_numa_zonelist_order(char *s
)
4722 ret
= __parse_numa_zonelist_order(s
);
4724 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
4728 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
4731 * sysctl handler for numa_zonelist_order
4733 int numa_zonelist_order_handler(struct ctl_table
*table
, int write
,
4734 void __user
*buffer
, size_t *length
,
4737 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
4739 static DEFINE_MUTEX(zl_order_mutex
);
4741 mutex_lock(&zl_order_mutex
);
4743 if (strlen((char *)table
->data
) >= NUMA_ZONELIST_ORDER_LEN
) {
4747 strcpy(saved_string
, (char *)table
->data
);
4749 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
4753 int oldval
= user_zonelist_order
;
4755 ret
= __parse_numa_zonelist_order((char *)table
->data
);
4758 * bogus value. restore saved string
4760 strncpy((char *)table
->data
, saved_string
,
4761 NUMA_ZONELIST_ORDER_LEN
);
4762 user_zonelist_order
= oldval
;
4763 } else if (oldval
!= user_zonelist_order
) {
4764 mutex_lock(&zonelists_mutex
);
4765 build_all_zonelists(NULL
, NULL
);
4766 mutex_unlock(&zonelists_mutex
);
4770 mutex_unlock(&zl_order_mutex
);
4775 #define MAX_NODE_LOAD (nr_online_nodes)
4776 static int node_load
[MAX_NUMNODES
];
4779 * find_next_best_node - find the next node that should appear in a given node's fallback list
4780 * @node: node whose fallback list we're appending
4781 * @used_node_mask: nodemask_t of already used nodes
4783 * We use a number of factors to determine which is the next node that should
4784 * appear on a given node's fallback list. The node should not have appeared
4785 * already in @node's fallback list, and it should be the next closest node
4786 * according to the distance array (which contains arbitrary distance values
4787 * from each node to each node in the system), and should also prefer nodes
4788 * with no CPUs, since presumably they'll have very little allocation pressure
4789 * on them otherwise.
4790 * It returns -1 if no node is found.
4792 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
4795 int min_val
= INT_MAX
;
4796 int best_node
= NUMA_NO_NODE
;
4797 const struct cpumask
*tmp
= cpumask_of_node(0);
4799 /* Use the local node if we haven't already */
4800 if (!node_isset(node
, *used_node_mask
)) {
4801 node_set(node
, *used_node_mask
);
4805 for_each_node_state(n
, N_MEMORY
) {
4807 /* Don't want a node to appear more than once */
4808 if (node_isset(n
, *used_node_mask
))
4811 /* Use the distance array to find the distance */
4812 val
= node_distance(node
, n
);
4814 /* Penalize nodes under us ("prefer the next node") */
4817 /* Give preference to headless and unused nodes */
4818 tmp
= cpumask_of_node(n
);
4819 if (!cpumask_empty(tmp
))
4820 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
4822 /* Slight preference for less loaded node */
4823 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
4824 val
+= node_load
[n
];
4826 if (val
< min_val
) {
4833 node_set(best_node
, *used_node_mask
);
4840 * Build zonelists ordered by node and zones within node.
4841 * This results in maximum locality--normal zone overflows into local
4842 * DMA zone, if any--but risks exhausting DMA zone.
4844 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
4847 struct zonelist
*zonelist
;
4849 zonelist
= &pgdat
->node_zonelists
[ZONELIST_FALLBACK
];
4850 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
4852 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4853 zonelist
->_zonerefs
[j
].zone
= NULL
;
4854 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4858 * Build gfp_thisnode zonelists
4860 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
4863 struct zonelist
*zonelist
;
4865 zonelist
= &pgdat
->node_zonelists
[ZONELIST_NOFALLBACK
];
4866 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4867 zonelist
->_zonerefs
[j
].zone
= NULL
;
4868 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4872 * Build zonelists ordered by zone and nodes within zones.
4873 * This results in conserving DMA zone[s] until all Normal memory is
4874 * exhausted, but results in overflowing to remote node while memory
4875 * may still exist in local DMA zone.
4877 static int node_order
[MAX_NUMNODES
];
4879 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
4882 int zone_type
; /* needs to be signed */
4884 struct zonelist
*zonelist
;
4886 zonelist
= &pgdat
->node_zonelists
[ZONELIST_FALLBACK
];
4888 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
4889 for (j
= 0; j
< nr_nodes
; j
++) {
4890 node
= node_order
[j
];
4891 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
4892 if (managed_zone(z
)) {
4894 &zonelist
->_zonerefs
[pos
++]);
4895 check_highest_zone(zone_type
);
4899 zonelist
->_zonerefs
[pos
].zone
= NULL
;
4900 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
4903 #if defined(CONFIG_64BIT)
4905 * Devices that require DMA32/DMA are relatively rare and do not justify a
4906 * penalty to every machine in case the specialised case applies. Default
4907 * to Node-ordering on 64-bit NUMA machines
4909 static int default_zonelist_order(void)
4911 return ZONELIST_ORDER_NODE
;
4915 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4916 * by the kernel. If processes running on node 0 deplete the low memory zone
4917 * then reclaim will occur more frequency increasing stalls and potentially
4918 * be easier to OOM if a large percentage of the zone is under writeback or
4919 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4920 * Hence, default to zone ordering on 32-bit.
4922 static int default_zonelist_order(void)
4924 return ZONELIST_ORDER_ZONE
;
4926 #endif /* CONFIG_64BIT */
4928 static void set_zonelist_order(void)
4930 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
4931 current_zonelist_order
= default_zonelist_order();
4933 current_zonelist_order
= user_zonelist_order
;
4936 static void build_zonelists(pg_data_t
*pgdat
)
4939 nodemask_t used_mask
;
4940 int local_node
, prev_node
;
4941 struct zonelist
*zonelist
;
4942 unsigned int order
= current_zonelist_order
;
4944 /* initialize zonelists */
4945 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
4946 zonelist
= pgdat
->node_zonelists
+ i
;
4947 zonelist
->_zonerefs
[0].zone
= NULL
;
4948 zonelist
->_zonerefs
[0].zone_idx
= 0;
4951 /* NUMA-aware ordering of nodes */
4952 local_node
= pgdat
->node_id
;
4953 load
= nr_online_nodes
;
4954 prev_node
= local_node
;
4955 nodes_clear(used_mask
);
4957 memset(node_order
, 0, sizeof(node_order
));
4960 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
4962 * We don't want to pressure a particular node.
4963 * So adding penalty to the first node in same
4964 * distance group to make it round-robin.
4966 if (node_distance(local_node
, node
) !=
4967 node_distance(local_node
, prev_node
))
4968 node_load
[node
] = load
;
4972 if (order
== ZONELIST_ORDER_NODE
)
4973 build_zonelists_in_node_order(pgdat
, node
);
4975 node_order
[i
++] = node
; /* remember order */
4978 if (order
== ZONELIST_ORDER_ZONE
) {
4979 /* calculate node order -- i.e., DMA last! */
4980 build_zonelists_in_zone_order(pgdat
, i
);
4983 build_thisnode_zonelists(pgdat
);
4986 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4988 * Return node id of node used for "local" allocations.
4989 * I.e., first node id of first zone in arg node's generic zonelist.
4990 * Used for initializing percpu 'numa_mem', which is used primarily
4991 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4993 int local_memory_node(int node
)
4997 z
= first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
4998 gfp_zone(GFP_KERNEL
),
5000 return z
->zone
->node
;
5004 static void setup_min_unmapped_ratio(void);
5005 static void setup_min_slab_ratio(void);
5006 #else /* CONFIG_NUMA */
5008 static void set_zonelist_order(void)
5010 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
5013 static void build_zonelists(pg_data_t
*pgdat
)
5015 int node
, local_node
;
5017 struct zonelist
*zonelist
;
5019 local_node
= pgdat
->node_id
;
5021 zonelist
= &pgdat
->node_zonelists
[ZONELIST_FALLBACK
];
5022 j
= build_zonelists_node(pgdat
, zonelist
, 0);
5025 * Now we build the zonelist so that it contains the zones
5026 * of all the other nodes.
5027 * We don't want to pressure a particular node, so when
5028 * building the zones for node N, we make sure that the
5029 * zones coming right after the local ones are those from
5030 * node N+1 (modulo N)
5032 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
5033 if (!node_online(node
))
5035 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
5037 for (node
= 0; node
< local_node
; node
++) {
5038 if (!node_online(node
))
5040 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
5043 zonelist
->_zonerefs
[j
].zone
= NULL
;
5044 zonelist
->_zonerefs
[j
].zone_idx
= 0;
5047 #endif /* CONFIG_NUMA */
5050 * Boot pageset table. One per cpu which is going to be used for all
5051 * zones and all nodes. The parameters will be set in such a way
5052 * that an item put on a list will immediately be handed over to
5053 * the buddy list. This is safe since pageset manipulation is done
5054 * with interrupts disabled.
5056 * The boot_pagesets must be kept even after bootup is complete for
5057 * unused processors and/or zones. They do play a role for bootstrapping
5058 * hotplugged processors.
5060 * zoneinfo_show() and maybe other functions do
5061 * not check if the processor is online before following the pageset pointer.
5062 * Other parts of the kernel may not check if the zone is available.
5064 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
5065 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
5066 static void setup_zone_pageset(struct zone
*zone
);
5069 * Global mutex to protect against size modification of zonelists
5070 * as well as to serialize pageset setup for the new populated zone.
5072 DEFINE_MUTEX(zonelists_mutex
);
5074 /* return values int ....just for stop_machine() */
5075 static int __build_all_zonelists(void *data
)
5079 pg_data_t
*self
= data
;
5082 memset(node_load
, 0, sizeof(node_load
));
5085 if (self
&& !node_online(self
->node_id
)) {
5086 build_zonelists(self
);
5089 for_each_online_node(nid
) {
5090 pg_data_t
*pgdat
= NODE_DATA(nid
);
5092 build_zonelists(pgdat
);
5096 * Initialize the boot_pagesets that are going to be used
5097 * for bootstrapping processors. The real pagesets for
5098 * each zone will be allocated later when the per cpu
5099 * allocator is available.
5101 * boot_pagesets are used also for bootstrapping offline
5102 * cpus if the system is already booted because the pagesets
5103 * are needed to initialize allocators on a specific cpu too.
5104 * F.e. the percpu allocator needs the page allocator which
5105 * needs the percpu allocator in order to allocate its pagesets
5106 * (a chicken-egg dilemma).
5108 for_each_possible_cpu(cpu
) {
5109 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
5111 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
5113 * We now know the "local memory node" for each node--
5114 * i.e., the node of the first zone in the generic zonelist.
5115 * Set up numa_mem percpu variable for on-line cpus. During
5116 * boot, only the boot cpu should be on-line; we'll init the
5117 * secondary cpus' numa_mem as they come on-line. During
5118 * node/memory hotplug, we'll fixup all on-line cpus.
5120 if (cpu_online(cpu
))
5121 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
5128 static noinline
void __init
5129 build_all_zonelists_init(void)
5131 __build_all_zonelists(NULL
);
5132 mminit_verify_zonelist();
5133 cpuset_init_current_mems_allowed();
5137 * Called with zonelists_mutex held always
5138 * unless system_state == SYSTEM_BOOTING.
5140 * __ref due to (1) call of __meminit annotated setup_zone_pageset
5141 * [we're only called with non-NULL zone through __meminit paths] and
5142 * (2) call of __init annotated helper build_all_zonelists_init
5143 * [protected by SYSTEM_BOOTING].
5145 void __ref
build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
)
5147 set_zonelist_order();
5149 if (system_state
== SYSTEM_BOOTING
) {
5150 build_all_zonelists_init();
5152 #ifdef CONFIG_MEMORY_HOTPLUG
5154 setup_zone_pageset(zone
);
5156 /* we have to stop all cpus to guarantee there is no user
5158 stop_machine(__build_all_zonelists
, pgdat
, NULL
);
5159 /* cpuset refresh routine should be here */
5161 vm_total_pages
= nr_free_pagecache_pages();
5163 * Disable grouping by mobility if the number of pages in the
5164 * system is too low to allow the mechanism to work. It would be
5165 * more accurate, but expensive to check per-zone. This check is
5166 * made on memory-hotadd so a system can start with mobility
5167 * disabled and enable it later
5169 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
5170 page_group_by_mobility_disabled
= 1;
5172 page_group_by_mobility_disabled
= 0;
5174 pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n",
5176 zonelist_order_name
[current_zonelist_order
],
5177 page_group_by_mobility_disabled
? "off" : "on",
5180 pr_info("Policy zone: %s\n", zone_names
[policy_zone
]);
5185 * Initially all pages are reserved - free ones are freed
5186 * up by free_all_bootmem() once the early boot process is
5187 * done. Non-atomic initialization, single-pass.
5189 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
5190 unsigned long start_pfn
, enum memmap_context context
)
5192 struct vmem_altmap
*altmap
= to_vmem_altmap(__pfn_to_phys(start_pfn
));
5193 unsigned long end_pfn
= start_pfn
+ size
;
5194 pg_data_t
*pgdat
= NODE_DATA(nid
);
5196 unsigned long nr_initialised
= 0;
5197 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5198 struct memblock_region
*r
= NULL
, *tmp
;
5201 if (highest_memmap_pfn
< end_pfn
- 1)
5202 highest_memmap_pfn
= end_pfn
- 1;
5205 * Honor reservation requested by the driver for this ZONE_DEVICE
5208 if (altmap
&& start_pfn
== altmap
->base_pfn
)
5209 start_pfn
+= altmap
->reserve
;
5211 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
5213 * There can be holes in boot-time mem_map[]s handed to this
5214 * function. They do not exist on hotplugged memory.
5216 if (context
!= MEMMAP_EARLY
)
5219 if (!early_pfn_valid(pfn
)) {
5220 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5222 * Skip to the pfn preceding the next valid one (or
5223 * end_pfn), such that we hit a valid pfn (or end_pfn)
5224 * on our next iteration of the loop.
5226 pfn
= memblock_next_valid_pfn(pfn
, end_pfn
) - 1;
5230 if (!early_pfn_in_nid(pfn
, nid
))
5232 if (!update_defer_init(pgdat
, pfn
, end_pfn
, &nr_initialised
))
5235 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5237 * Check given memblock attribute by firmware which can affect
5238 * kernel memory layout. If zone==ZONE_MOVABLE but memory is
5239 * mirrored, it's an overlapped memmap init. skip it.
5241 if (mirrored_kernelcore
&& zone
== ZONE_MOVABLE
) {
5242 if (!r
|| pfn
>= memblock_region_memory_end_pfn(r
)) {
5243 for_each_memblock(memory
, tmp
)
5244 if (pfn
< memblock_region_memory_end_pfn(tmp
))
5248 if (pfn
>= memblock_region_memory_base_pfn(r
) &&
5249 memblock_is_mirror(r
)) {
5250 /* already initialized as NORMAL */
5251 pfn
= memblock_region_memory_end_pfn(r
);
5259 * Mark the block movable so that blocks are reserved for
5260 * movable at startup. This will force kernel allocations
5261 * to reserve their blocks rather than leaking throughout
5262 * the address space during boot when many long-lived
5263 * kernel allocations are made.
5265 * bitmap is created for zone's valid pfn range. but memmap
5266 * can be created for invalid pages (for alignment)
5267 * check here not to call set_pageblock_migratetype() against
5270 if (!(pfn
& (pageblock_nr_pages
- 1))) {
5271 struct page
*page
= pfn_to_page(pfn
);
5273 __init_single_page(page
, pfn
, zone
, nid
);
5274 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
5276 __init_single_pfn(pfn
, zone
, nid
);
5281 static void __meminit
zone_init_free_lists(struct zone
*zone
)
5283 unsigned int order
, t
;
5284 for_each_migratetype_order(order
, t
) {
5285 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
5286 zone
->free_area
[order
].nr_free
= 0;
5290 #ifndef __HAVE_ARCH_MEMMAP_INIT
5291 #define memmap_init(size, nid, zone, start_pfn) \
5292 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
5295 static int zone_batchsize(struct zone
*zone
)
5301 * The per-cpu-pages pools are set to around 1000th of the
5302 * size of the zone. But no more than 1/2 of a meg.
5304 * OK, so we don't know how big the cache is. So guess.
5306 batch
= zone
->managed_pages
/ 1024;
5307 if (batch
* PAGE_SIZE
> 512 * 1024)
5308 batch
= (512 * 1024) / PAGE_SIZE
;
5309 batch
/= 4; /* We effectively *= 4 below */
5314 * Clamp the batch to a 2^n - 1 value. Having a power
5315 * of 2 value was found to be more likely to have
5316 * suboptimal cache aliasing properties in some cases.
5318 * For example if 2 tasks are alternately allocating
5319 * batches of pages, one task can end up with a lot
5320 * of pages of one half of the possible page colors
5321 * and the other with pages of the other colors.
5323 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
5328 /* The deferral and batching of frees should be suppressed under NOMMU
5331 * The problem is that NOMMU needs to be able to allocate large chunks
5332 * of contiguous memory as there's no hardware page translation to
5333 * assemble apparent contiguous memory from discontiguous pages.
5335 * Queueing large contiguous runs of pages for batching, however,
5336 * causes the pages to actually be freed in smaller chunks. As there
5337 * can be a significant delay between the individual batches being
5338 * recycled, this leads to the once large chunks of space being
5339 * fragmented and becoming unavailable for high-order allocations.
5346 * pcp->high and pcp->batch values are related and dependent on one another:
5347 * ->batch must never be higher then ->high.
5348 * The following function updates them in a safe manner without read side
5351 * Any new users of pcp->batch and pcp->high should ensure they can cope with
5352 * those fields changing asynchronously (acording the the above rule).
5354 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
5355 * outside of boot time (or some other assurance that no concurrent updaters
5358 static void pageset_update(struct per_cpu_pages
*pcp
, unsigned long high
,
5359 unsigned long batch
)
5361 /* start with a fail safe value for batch */
5365 /* Update high, then batch, in order */
5372 /* a companion to pageset_set_high() */
5373 static void pageset_set_batch(struct per_cpu_pageset
*p
, unsigned long batch
)
5375 pageset_update(&p
->pcp
, 6 * batch
, max(1UL, 1 * batch
));
5378 static void pageset_init(struct per_cpu_pageset
*p
)
5380 struct per_cpu_pages
*pcp
;
5383 memset(p
, 0, sizeof(*p
));
5387 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
5388 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
5391 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
5394 pageset_set_batch(p
, batch
);
5398 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
5399 * to the value high for the pageset p.
5401 static void pageset_set_high(struct per_cpu_pageset
*p
,
5404 unsigned long batch
= max(1UL, high
/ 4);
5405 if ((high
/ 4) > (PAGE_SHIFT
* 8))
5406 batch
= PAGE_SHIFT
* 8;
5408 pageset_update(&p
->pcp
, high
, batch
);
5411 static void pageset_set_high_and_batch(struct zone
*zone
,
5412 struct per_cpu_pageset
*pcp
)
5414 if (percpu_pagelist_fraction
)
5415 pageset_set_high(pcp
,
5416 (zone
->managed_pages
/
5417 percpu_pagelist_fraction
));
5419 pageset_set_batch(pcp
, zone_batchsize(zone
));
5422 static void __meminit
zone_pageset_init(struct zone
*zone
, int cpu
)
5424 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
5427 pageset_set_high_and_batch(zone
, pcp
);
5430 static void __meminit
setup_zone_pageset(struct zone
*zone
)
5433 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
5434 for_each_possible_cpu(cpu
)
5435 zone_pageset_init(zone
, cpu
);
5439 * Allocate per cpu pagesets and initialize them.
5440 * Before this call only boot pagesets were available.
5442 void __init
setup_per_cpu_pageset(void)
5444 struct pglist_data
*pgdat
;
5447 for_each_populated_zone(zone
)
5448 setup_zone_pageset(zone
);
5450 for_each_online_pgdat(pgdat
)
5451 pgdat
->per_cpu_nodestats
=
5452 alloc_percpu(struct per_cpu_nodestat
);
5455 static __meminit
void zone_pcp_init(struct zone
*zone
)
5458 * per cpu subsystem is not up at this point. The following code
5459 * relies on the ability of the linker to provide the
5460 * offset of a (static) per cpu variable into the per cpu area.
5462 zone
->pageset
= &boot_pageset
;
5464 if (populated_zone(zone
))
5465 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
5466 zone
->name
, zone
->present_pages
,
5467 zone_batchsize(zone
));
5470 int __meminit
init_currently_empty_zone(struct zone
*zone
,
5471 unsigned long zone_start_pfn
,
5474 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
5476 pgdat
->nr_zones
= zone_idx(zone
) + 1;
5478 zone
->zone_start_pfn
= zone_start_pfn
;
5480 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
5481 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
5483 (unsigned long)zone_idx(zone
),
5484 zone_start_pfn
, (zone_start_pfn
+ size
));
5486 zone_init_free_lists(zone
);
5487 zone
->initialized
= 1;
5492 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5493 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
5496 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
5498 int __meminit
__early_pfn_to_nid(unsigned long pfn
,
5499 struct mminit_pfnnid_cache
*state
)
5501 unsigned long start_pfn
, end_pfn
;
5504 if (state
->last_start
<= pfn
&& pfn
< state
->last_end
)
5505 return state
->last_nid
;
5507 nid
= memblock_search_pfn_nid(pfn
, &start_pfn
, &end_pfn
);
5509 state
->last_start
= start_pfn
;
5510 state
->last_end
= end_pfn
;
5511 state
->last_nid
= nid
;
5516 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
5519 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
5520 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
5521 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
5523 * If an architecture guarantees that all ranges registered contain no holes
5524 * and may be freed, this this function may be used instead of calling
5525 * memblock_free_early_nid() manually.
5527 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
5529 unsigned long start_pfn
, end_pfn
;
5532 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
5533 start_pfn
= min(start_pfn
, max_low_pfn
);
5534 end_pfn
= min(end_pfn
, max_low_pfn
);
5536 if (start_pfn
< end_pfn
)
5537 memblock_free_early_nid(PFN_PHYS(start_pfn
),
5538 (end_pfn
- start_pfn
) << PAGE_SHIFT
,
5544 * sparse_memory_present_with_active_regions - Call memory_present for each active range
5545 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
5547 * If an architecture guarantees that all ranges registered contain no holes and may
5548 * be freed, this function may be used instead of calling memory_present() manually.
5550 void __init
sparse_memory_present_with_active_regions(int nid
)
5552 unsigned long start_pfn
, end_pfn
;
5555 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
5556 memory_present(this_nid
, start_pfn
, end_pfn
);
5560 * get_pfn_range_for_nid - Return the start and end page frames for a node
5561 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
5562 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
5563 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
5565 * It returns the start and end page frame of a node based on information
5566 * provided by memblock_set_node(). If called for a node
5567 * with no available memory, a warning is printed and the start and end
5570 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
5571 unsigned long *start_pfn
, unsigned long *end_pfn
)
5573 unsigned long this_start_pfn
, this_end_pfn
;
5579 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
5580 *start_pfn
= min(*start_pfn
, this_start_pfn
);
5581 *end_pfn
= max(*end_pfn
, this_end_pfn
);
5584 if (*start_pfn
== -1UL)
5589 * This finds a zone that can be used for ZONE_MOVABLE pages. The
5590 * assumption is made that zones within a node are ordered in monotonic
5591 * increasing memory addresses so that the "highest" populated zone is used
5593 static void __init
find_usable_zone_for_movable(void)
5596 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
5597 if (zone_index
== ZONE_MOVABLE
)
5600 if (arch_zone_highest_possible_pfn
[zone_index
] >
5601 arch_zone_lowest_possible_pfn
[zone_index
])
5605 VM_BUG_ON(zone_index
== -1);
5606 movable_zone
= zone_index
;
5610 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
5611 * because it is sized independent of architecture. Unlike the other zones,
5612 * the starting point for ZONE_MOVABLE is not fixed. It may be different
5613 * in each node depending on the size of each node and how evenly kernelcore
5614 * is distributed. This helper function adjusts the zone ranges
5615 * provided by the architecture for a given node by using the end of the
5616 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5617 * zones within a node are in order of monotonic increases memory addresses
5619 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
5620 unsigned long zone_type
,
5621 unsigned long node_start_pfn
,
5622 unsigned long node_end_pfn
,
5623 unsigned long *zone_start_pfn
,
5624 unsigned long *zone_end_pfn
)
5626 /* Only adjust if ZONE_MOVABLE is on this node */
5627 if (zone_movable_pfn
[nid
]) {
5628 /* Size ZONE_MOVABLE */
5629 if (zone_type
== ZONE_MOVABLE
) {
5630 *zone_start_pfn
= zone_movable_pfn
[nid
];
5631 *zone_end_pfn
= min(node_end_pfn
,
5632 arch_zone_highest_possible_pfn
[movable_zone
]);
5634 /* Adjust for ZONE_MOVABLE starting within this range */
5635 } else if (!mirrored_kernelcore
&&
5636 *zone_start_pfn
< zone_movable_pfn
[nid
] &&
5637 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
5638 *zone_end_pfn
= zone_movable_pfn
[nid
];
5640 /* Check if this whole range is within ZONE_MOVABLE */
5641 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
5642 *zone_start_pfn
= *zone_end_pfn
;
5647 * Return the number of pages a zone spans in a node, including holes
5648 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5650 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5651 unsigned long zone_type
,
5652 unsigned long node_start_pfn
,
5653 unsigned long node_end_pfn
,
5654 unsigned long *zone_start_pfn
,
5655 unsigned long *zone_end_pfn
,
5656 unsigned long *ignored
)
5658 /* When hotadd a new node from cpu_up(), the node should be empty */
5659 if (!node_start_pfn
&& !node_end_pfn
)
5662 /* Get the start and end of the zone */
5663 *zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
5664 *zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
5665 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5666 node_start_pfn
, node_end_pfn
,
5667 zone_start_pfn
, zone_end_pfn
);
5669 /* Check that this node has pages within the zone's required range */
5670 if (*zone_end_pfn
< node_start_pfn
|| *zone_start_pfn
> node_end_pfn
)
5673 /* Move the zone boundaries inside the node if necessary */
5674 *zone_end_pfn
= min(*zone_end_pfn
, node_end_pfn
);
5675 *zone_start_pfn
= max(*zone_start_pfn
, node_start_pfn
);
5677 /* Return the spanned pages */
5678 return *zone_end_pfn
- *zone_start_pfn
;
5682 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
5683 * then all holes in the requested range will be accounted for.
5685 unsigned long __meminit
__absent_pages_in_range(int nid
,
5686 unsigned long range_start_pfn
,
5687 unsigned long range_end_pfn
)
5689 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
5690 unsigned long start_pfn
, end_pfn
;
5693 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5694 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
5695 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
5696 nr_absent
-= end_pfn
- start_pfn
;
5702 * absent_pages_in_range - Return number of page frames in holes within a range
5703 * @start_pfn: The start PFN to start searching for holes
5704 * @end_pfn: The end PFN to stop searching for holes
5706 * It returns the number of pages frames in memory holes within a range.
5708 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
5709 unsigned long end_pfn
)
5711 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
5714 /* Return the number of page frames in holes in a zone on a node */
5715 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5716 unsigned long zone_type
,
5717 unsigned long node_start_pfn
,
5718 unsigned long node_end_pfn
,
5719 unsigned long *ignored
)
5721 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
5722 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
5723 unsigned long zone_start_pfn
, zone_end_pfn
;
5724 unsigned long nr_absent
;
5726 /* When hotadd a new node from cpu_up(), the node should be empty */
5727 if (!node_start_pfn
&& !node_end_pfn
)
5730 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
5731 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
5733 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5734 node_start_pfn
, node_end_pfn
,
5735 &zone_start_pfn
, &zone_end_pfn
);
5736 nr_absent
= __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
5739 * ZONE_MOVABLE handling.
5740 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
5743 if (mirrored_kernelcore
&& zone_movable_pfn
[nid
]) {
5744 unsigned long start_pfn
, end_pfn
;
5745 struct memblock_region
*r
;
5747 for_each_memblock(memory
, r
) {
5748 start_pfn
= clamp(memblock_region_memory_base_pfn(r
),
5749 zone_start_pfn
, zone_end_pfn
);
5750 end_pfn
= clamp(memblock_region_memory_end_pfn(r
),
5751 zone_start_pfn
, zone_end_pfn
);
5753 if (zone_type
== ZONE_MOVABLE
&&
5754 memblock_is_mirror(r
))
5755 nr_absent
+= end_pfn
- start_pfn
;
5757 if (zone_type
== ZONE_NORMAL
&&
5758 !memblock_is_mirror(r
))
5759 nr_absent
+= end_pfn
- start_pfn
;
5766 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5767 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5768 unsigned long zone_type
,
5769 unsigned long node_start_pfn
,
5770 unsigned long node_end_pfn
,
5771 unsigned long *zone_start_pfn
,
5772 unsigned long *zone_end_pfn
,
5773 unsigned long *zones_size
)
5777 *zone_start_pfn
= node_start_pfn
;
5778 for (zone
= 0; zone
< zone_type
; zone
++)
5779 *zone_start_pfn
+= zones_size
[zone
];
5781 *zone_end_pfn
= *zone_start_pfn
+ zones_size
[zone_type
];
5783 return zones_size
[zone_type
];
5786 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5787 unsigned long zone_type
,
5788 unsigned long node_start_pfn
,
5789 unsigned long node_end_pfn
,
5790 unsigned long *zholes_size
)
5795 return zholes_size
[zone_type
];
5798 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5800 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
5801 unsigned long node_start_pfn
,
5802 unsigned long node_end_pfn
,
5803 unsigned long *zones_size
,
5804 unsigned long *zholes_size
)
5806 unsigned long realtotalpages
= 0, totalpages
= 0;
5809 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5810 struct zone
*zone
= pgdat
->node_zones
+ i
;
5811 unsigned long zone_start_pfn
, zone_end_pfn
;
5812 unsigned long size
, real_size
;
5814 size
= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
5820 real_size
= size
- zone_absent_pages_in_node(pgdat
->node_id
, i
,
5821 node_start_pfn
, node_end_pfn
,
5824 zone
->zone_start_pfn
= zone_start_pfn
;
5826 zone
->zone_start_pfn
= 0;
5827 zone
->spanned_pages
= size
;
5828 zone
->present_pages
= real_size
;
5831 realtotalpages
+= real_size
;
5834 pgdat
->node_spanned_pages
= totalpages
;
5835 pgdat
->node_present_pages
= realtotalpages
;
5836 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
5840 #ifndef CONFIG_SPARSEMEM
5842 * Calculate the size of the zone->blockflags rounded to an unsigned long
5843 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5844 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
5845 * round what is now in bits to nearest long in bits, then return it in
5848 static unsigned long __init
usemap_size(unsigned long zone_start_pfn
, unsigned long zonesize
)
5850 unsigned long usemapsize
;
5852 zonesize
+= zone_start_pfn
& (pageblock_nr_pages
-1);
5853 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
5854 usemapsize
= usemapsize
>> pageblock_order
;
5855 usemapsize
*= NR_PAGEBLOCK_BITS
;
5856 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
5858 return usemapsize
/ 8;
5861 static void __init
setup_usemap(struct pglist_data
*pgdat
,
5863 unsigned long zone_start_pfn
,
5864 unsigned long zonesize
)
5866 unsigned long usemapsize
= usemap_size(zone_start_pfn
, zonesize
);
5867 zone
->pageblock_flags
= NULL
;
5869 zone
->pageblock_flags
=
5870 memblock_virt_alloc_node_nopanic(usemapsize
,
5874 static inline void setup_usemap(struct pglist_data
*pgdat
, struct zone
*zone
,
5875 unsigned long zone_start_pfn
, unsigned long zonesize
) {}
5876 #endif /* CONFIG_SPARSEMEM */
5878 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
5880 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
5881 void __paginginit
set_pageblock_order(void)
5885 /* Check that pageblock_nr_pages has not already been setup */
5886 if (pageblock_order
)
5889 if (HPAGE_SHIFT
> PAGE_SHIFT
)
5890 order
= HUGETLB_PAGE_ORDER
;
5892 order
= MAX_ORDER
- 1;
5895 * Assume the largest contiguous order of interest is a huge page.
5896 * This value may be variable depending on boot parameters on IA64 and
5899 pageblock_order
= order
;
5901 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5904 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
5905 * is unused as pageblock_order is set at compile-time. See
5906 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5909 void __paginginit
set_pageblock_order(void)
5913 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5915 static unsigned long __paginginit
calc_memmap_size(unsigned long spanned_pages
,
5916 unsigned long present_pages
)
5918 unsigned long pages
= spanned_pages
;
5921 * Provide a more accurate estimation if there are holes within
5922 * the zone and SPARSEMEM is in use. If there are holes within the
5923 * zone, each populated memory region may cost us one or two extra
5924 * memmap pages due to alignment because memmap pages for each
5925 * populated regions may not naturally algined on page boundary.
5926 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5928 if (spanned_pages
> present_pages
+ (present_pages
>> 4) &&
5929 IS_ENABLED(CONFIG_SPARSEMEM
))
5930 pages
= present_pages
;
5932 return PAGE_ALIGN(pages
* sizeof(struct page
)) >> PAGE_SHIFT
;
5936 * Set up the zone data structures:
5937 * - mark all pages reserved
5938 * - mark all memory queues empty
5939 * - clear the memory bitmaps
5941 * NOTE: pgdat should get zeroed by caller.
5943 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
)
5946 int nid
= pgdat
->node_id
;
5949 pgdat_resize_init(pgdat
);
5950 #ifdef CONFIG_NUMA_BALANCING
5951 spin_lock_init(&pgdat
->numabalancing_migrate_lock
);
5952 pgdat
->numabalancing_migrate_nr_pages
= 0;
5953 pgdat
->numabalancing_migrate_next_window
= jiffies
;
5955 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5956 spin_lock_init(&pgdat
->split_queue_lock
);
5957 INIT_LIST_HEAD(&pgdat
->split_queue
);
5958 pgdat
->split_queue_len
= 0;
5960 init_waitqueue_head(&pgdat
->kswapd_wait
);
5961 init_waitqueue_head(&pgdat
->pfmemalloc_wait
);
5962 #ifdef CONFIG_COMPACTION
5963 init_waitqueue_head(&pgdat
->kcompactd_wait
);
5965 pgdat_page_ext_init(pgdat
);
5966 spin_lock_init(&pgdat
->lru_lock
);
5967 lruvec_init(node_lruvec(pgdat
));
5969 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5970 struct zone
*zone
= pgdat
->node_zones
+ j
;
5971 unsigned long size
, realsize
, freesize
, memmap_pages
;
5972 unsigned long zone_start_pfn
= zone
->zone_start_pfn
;
5974 size
= zone
->spanned_pages
;
5975 realsize
= freesize
= zone
->present_pages
;
5978 * Adjust freesize so that it accounts for how much memory
5979 * is used by this zone for memmap. This affects the watermark
5980 * and per-cpu initialisations
5982 memmap_pages
= calc_memmap_size(size
, realsize
);
5983 if (!is_highmem_idx(j
)) {
5984 if (freesize
>= memmap_pages
) {
5985 freesize
-= memmap_pages
;
5988 " %s zone: %lu pages used for memmap\n",
5989 zone_names
[j
], memmap_pages
);
5991 pr_warn(" %s zone: %lu pages exceeds freesize %lu\n",
5992 zone_names
[j
], memmap_pages
, freesize
);
5995 /* Account for reserved pages */
5996 if (j
== 0 && freesize
> dma_reserve
) {
5997 freesize
-= dma_reserve
;
5998 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
5999 zone_names
[0], dma_reserve
);
6002 if (!is_highmem_idx(j
))
6003 nr_kernel_pages
+= freesize
;
6004 /* Charge for highmem memmap if there are enough kernel pages */
6005 else if (nr_kernel_pages
> memmap_pages
* 2)
6006 nr_kernel_pages
-= memmap_pages
;
6007 nr_all_pages
+= freesize
;
6010 * Set an approximate value for lowmem here, it will be adjusted
6011 * when the bootmem allocator frees pages into the buddy system.
6012 * And all highmem pages will be managed by the buddy system.
6014 zone
->managed_pages
= is_highmem_idx(j
) ? realsize
: freesize
;
6018 zone
->name
= zone_names
[j
];
6019 zone
->zone_pgdat
= pgdat
;
6020 spin_lock_init(&zone
->lock
);
6021 zone_seqlock_init(zone
);
6022 zone_pcp_init(zone
);
6027 set_pageblock_order();
6028 setup_usemap(pgdat
, zone
, zone_start_pfn
, size
);
6029 ret
= init_currently_empty_zone(zone
, zone_start_pfn
, size
);
6031 memmap_init(size
, nid
, j
, zone_start_pfn
);
6035 static void __ref
alloc_node_mem_map(struct pglist_data
*pgdat
)
6037 unsigned long __maybe_unused start
= 0;
6038 unsigned long __maybe_unused offset
= 0;
6040 /* Skip empty nodes */
6041 if (!pgdat
->node_spanned_pages
)
6044 #ifdef CONFIG_FLAT_NODE_MEM_MAP
6045 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
6046 offset
= pgdat
->node_start_pfn
- start
;
6047 /* ia64 gets its own node_mem_map, before this, without bootmem */
6048 if (!pgdat
->node_mem_map
) {
6049 unsigned long size
, end
;
6053 * The zone's endpoints aren't required to be MAX_ORDER
6054 * aligned but the node_mem_map endpoints must be in order
6055 * for the buddy allocator to function correctly.
6057 end
= pgdat_end_pfn(pgdat
);
6058 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
6059 size
= (end
- start
) * sizeof(struct page
);
6060 map
= alloc_remap(pgdat
->node_id
, size
);
6062 map
= memblock_virt_alloc_node_nopanic(size
,
6064 pgdat
->node_mem_map
= map
+ offset
;
6066 #ifndef CONFIG_NEED_MULTIPLE_NODES
6068 * With no DISCONTIG, the global mem_map is just set as node 0's
6070 if (pgdat
== NODE_DATA(0)) {
6071 mem_map
= NODE_DATA(0)->node_mem_map
;
6072 #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
6073 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
6075 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6078 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
6081 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
6082 unsigned long node_start_pfn
, unsigned long *zholes_size
)
6084 pg_data_t
*pgdat
= NODE_DATA(nid
);
6085 unsigned long start_pfn
= 0;
6086 unsigned long end_pfn
= 0;
6088 /* pg_data_t should be reset to zero when it's allocated */
6089 WARN_ON(pgdat
->nr_zones
|| pgdat
->kswapd_classzone_idx
);
6091 reset_deferred_meminit(pgdat
);
6092 pgdat
->node_id
= nid
;
6093 pgdat
->node_start_pfn
= node_start_pfn
;
6094 pgdat
->per_cpu_nodestats
= NULL
;
6095 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6096 get_pfn_range_for_nid(nid
, &start_pfn
, &end_pfn
);
6097 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid
,
6098 (u64
)start_pfn
<< PAGE_SHIFT
,
6099 end_pfn
? ((u64
)end_pfn
<< PAGE_SHIFT
) - 1 : 0);
6101 start_pfn
= node_start_pfn
;
6103 calculate_node_totalpages(pgdat
, start_pfn
, end_pfn
,
6104 zones_size
, zholes_size
);
6106 alloc_node_mem_map(pgdat
);
6107 #ifdef CONFIG_FLAT_NODE_MEM_MAP
6108 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
6109 nid
, (unsigned long)pgdat
,
6110 (unsigned long)pgdat
->node_mem_map
);
6113 free_area_init_core(pgdat
);
6116 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6118 #if MAX_NUMNODES > 1
6120 * Figure out the number of possible node ids.
6122 void __init
setup_nr_node_ids(void)
6124 unsigned int highest
;
6126 highest
= find_last_bit(node_possible_map
.bits
, MAX_NUMNODES
);
6127 nr_node_ids
= highest
+ 1;
6132 * node_map_pfn_alignment - determine the maximum internode alignment
6134 * This function should be called after node map is populated and sorted.
6135 * It calculates the maximum power of two alignment which can distinguish
6138 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
6139 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
6140 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
6141 * shifted, 1GiB is enough and this function will indicate so.
6143 * This is used to test whether pfn -> nid mapping of the chosen memory
6144 * model has fine enough granularity to avoid incorrect mapping for the
6145 * populated node map.
6147 * Returns the determined alignment in pfn's. 0 if there is no alignment
6148 * requirement (single node).
6150 unsigned long __init
node_map_pfn_alignment(void)
6152 unsigned long accl_mask
= 0, last_end
= 0;
6153 unsigned long start
, end
, mask
;
6157 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
6158 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
6165 * Start with a mask granular enough to pin-point to the
6166 * start pfn and tick off bits one-by-one until it becomes
6167 * too coarse to separate the current node from the last.
6169 mask
= ~((1 << __ffs(start
)) - 1);
6170 while (mask
&& last_end
<= (start
& (mask
<< 1)))
6173 /* accumulate all internode masks */
6177 /* convert mask to number of pages */
6178 return ~accl_mask
+ 1;
6181 /* Find the lowest pfn for a node */
6182 static unsigned long __init
find_min_pfn_for_node(int nid
)
6184 unsigned long min_pfn
= ULONG_MAX
;
6185 unsigned long start_pfn
;
6188 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
6189 min_pfn
= min(min_pfn
, start_pfn
);
6191 if (min_pfn
== ULONG_MAX
) {
6192 pr_warn("Could not find start_pfn for node %d\n", nid
);
6200 * find_min_pfn_with_active_regions - Find the minimum PFN registered
6202 * It returns the minimum PFN based on information provided via
6203 * memblock_set_node().
6205 unsigned long __init
find_min_pfn_with_active_regions(void)
6207 return find_min_pfn_for_node(MAX_NUMNODES
);
6211 * early_calculate_totalpages()
6212 * Sum pages in active regions for movable zone.
6213 * Populate N_MEMORY for calculating usable_nodes.
6215 static unsigned long __init
early_calculate_totalpages(void)
6217 unsigned long totalpages
= 0;
6218 unsigned long start_pfn
, end_pfn
;
6221 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
6222 unsigned long pages
= end_pfn
- start_pfn
;
6224 totalpages
+= pages
;
6226 node_set_state(nid
, N_MEMORY
);
6232 * Find the PFN the Movable zone begins in each node. Kernel memory
6233 * is spread evenly between nodes as long as the nodes have enough
6234 * memory. When they don't, some nodes will have more kernelcore than
6237 static void __init
find_zone_movable_pfns_for_nodes(void)
6240 unsigned long usable_startpfn
;
6241 unsigned long kernelcore_node
, kernelcore_remaining
;
6242 /* save the state before borrow the nodemask */
6243 nodemask_t saved_node_state
= node_states
[N_MEMORY
];
6244 unsigned long totalpages
= early_calculate_totalpages();
6245 int usable_nodes
= nodes_weight(node_states
[N_MEMORY
]);
6246 struct memblock_region
*r
;
6248 /* Need to find movable_zone earlier when movable_node is specified. */
6249 find_usable_zone_for_movable();
6252 * If movable_node is specified, ignore kernelcore and movablecore
6255 if (movable_node_is_enabled()) {
6256 for_each_memblock(memory
, r
) {
6257 if (!memblock_is_hotpluggable(r
))
6262 usable_startpfn
= PFN_DOWN(r
->base
);
6263 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
6264 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
6272 * If kernelcore=mirror is specified, ignore movablecore option
6274 if (mirrored_kernelcore
) {
6275 bool mem_below_4gb_not_mirrored
= false;
6277 for_each_memblock(memory
, r
) {
6278 if (memblock_is_mirror(r
))
6283 usable_startpfn
= memblock_region_memory_base_pfn(r
);
6285 if (usable_startpfn
< 0x100000) {
6286 mem_below_4gb_not_mirrored
= true;
6290 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
6291 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
6295 if (mem_below_4gb_not_mirrored
)
6296 pr_warn("This configuration results in unmirrored kernel memory.");
6302 * If movablecore=nn[KMG] was specified, calculate what size of
6303 * kernelcore that corresponds so that memory usable for
6304 * any allocation type is evenly spread. If both kernelcore
6305 * and movablecore are specified, then the value of kernelcore
6306 * will be used for required_kernelcore if it's greater than
6307 * what movablecore would have allowed.
6309 if (required_movablecore
) {
6310 unsigned long corepages
;
6313 * Round-up so that ZONE_MOVABLE is at least as large as what
6314 * was requested by the user
6316 required_movablecore
=
6317 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
6318 required_movablecore
= min(totalpages
, required_movablecore
);
6319 corepages
= totalpages
- required_movablecore
;
6321 required_kernelcore
= max(required_kernelcore
, corepages
);
6325 * If kernelcore was not specified or kernelcore size is larger
6326 * than totalpages, there is no ZONE_MOVABLE.
6328 if (!required_kernelcore
|| required_kernelcore
>= totalpages
)
6331 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
6332 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
6335 /* Spread kernelcore memory as evenly as possible throughout nodes */
6336 kernelcore_node
= required_kernelcore
/ usable_nodes
;
6337 for_each_node_state(nid
, N_MEMORY
) {
6338 unsigned long start_pfn
, end_pfn
;
6341 * Recalculate kernelcore_node if the division per node
6342 * now exceeds what is necessary to satisfy the requested
6343 * amount of memory for the kernel
6345 if (required_kernelcore
< kernelcore_node
)
6346 kernelcore_node
= required_kernelcore
/ usable_nodes
;
6349 * As the map is walked, we track how much memory is usable
6350 * by the kernel using kernelcore_remaining. When it is
6351 * 0, the rest of the node is usable by ZONE_MOVABLE
6353 kernelcore_remaining
= kernelcore_node
;
6355 /* Go through each range of PFNs within this node */
6356 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
6357 unsigned long size_pages
;
6359 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
6360 if (start_pfn
>= end_pfn
)
6363 /* Account for what is only usable for kernelcore */
6364 if (start_pfn
< usable_startpfn
) {
6365 unsigned long kernel_pages
;
6366 kernel_pages
= min(end_pfn
, usable_startpfn
)
6369 kernelcore_remaining
-= min(kernel_pages
,
6370 kernelcore_remaining
);
6371 required_kernelcore
-= min(kernel_pages
,
6372 required_kernelcore
);
6374 /* Continue if range is now fully accounted */
6375 if (end_pfn
<= usable_startpfn
) {
6378 * Push zone_movable_pfn to the end so
6379 * that if we have to rebalance
6380 * kernelcore across nodes, we will
6381 * not double account here
6383 zone_movable_pfn
[nid
] = end_pfn
;
6386 start_pfn
= usable_startpfn
;
6390 * The usable PFN range for ZONE_MOVABLE is from
6391 * start_pfn->end_pfn. Calculate size_pages as the
6392 * number of pages used as kernelcore
6394 size_pages
= end_pfn
- start_pfn
;
6395 if (size_pages
> kernelcore_remaining
)
6396 size_pages
= kernelcore_remaining
;
6397 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
6400 * Some kernelcore has been met, update counts and
6401 * break if the kernelcore for this node has been
6404 required_kernelcore
-= min(required_kernelcore
,
6406 kernelcore_remaining
-= size_pages
;
6407 if (!kernelcore_remaining
)
6413 * If there is still required_kernelcore, we do another pass with one
6414 * less node in the count. This will push zone_movable_pfn[nid] further
6415 * along on the nodes that still have memory until kernelcore is
6419 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
6423 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
6424 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
6425 zone_movable_pfn
[nid
] =
6426 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
6429 /* restore the node_state */
6430 node_states
[N_MEMORY
] = saved_node_state
;
6433 /* Any regular or high memory on that node ? */
6434 static void check_for_memory(pg_data_t
*pgdat
, int nid
)
6436 enum zone_type zone_type
;
6438 if (N_MEMORY
== N_NORMAL_MEMORY
)
6441 for (zone_type
= 0; zone_type
<= ZONE_MOVABLE
- 1; zone_type
++) {
6442 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
6443 if (populated_zone(zone
)) {
6444 node_set_state(nid
, N_HIGH_MEMORY
);
6445 if (N_NORMAL_MEMORY
!= N_HIGH_MEMORY
&&
6446 zone_type
<= ZONE_NORMAL
)
6447 node_set_state(nid
, N_NORMAL_MEMORY
);
6454 * free_area_init_nodes - Initialise all pg_data_t and zone data
6455 * @max_zone_pfn: an array of max PFNs for each zone
6457 * This will call free_area_init_node() for each active node in the system.
6458 * Using the page ranges provided by memblock_set_node(), the size of each
6459 * zone in each node and their holes is calculated. If the maximum PFN
6460 * between two adjacent zones match, it is assumed that the zone is empty.
6461 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
6462 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
6463 * starts where the previous one ended. For example, ZONE_DMA32 starts
6464 * at arch_max_dma_pfn.
6466 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
6468 unsigned long start_pfn
, end_pfn
;
6471 /* Record where the zone boundaries are */
6472 memset(arch_zone_lowest_possible_pfn
, 0,
6473 sizeof(arch_zone_lowest_possible_pfn
));
6474 memset(arch_zone_highest_possible_pfn
, 0,
6475 sizeof(arch_zone_highest_possible_pfn
));
6477 start_pfn
= find_min_pfn_with_active_regions();
6479 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
6480 if (i
== ZONE_MOVABLE
)
6483 end_pfn
= max(max_zone_pfn
[i
], start_pfn
);
6484 arch_zone_lowest_possible_pfn
[i
] = start_pfn
;
6485 arch_zone_highest_possible_pfn
[i
] = end_pfn
;
6487 start_pfn
= end_pfn
;
6489 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
6490 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
6492 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
6493 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
6494 find_zone_movable_pfns_for_nodes();
6496 /* Print out the zone ranges */
6497 pr_info("Zone ranges:\n");
6498 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
6499 if (i
== ZONE_MOVABLE
)
6501 pr_info(" %-8s ", zone_names
[i
]);
6502 if (arch_zone_lowest_possible_pfn
[i
] ==
6503 arch_zone_highest_possible_pfn
[i
])
6506 pr_cont("[mem %#018Lx-%#018Lx]\n",
6507 (u64
)arch_zone_lowest_possible_pfn
[i
]
6509 ((u64
)arch_zone_highest_possible_pfn
[i
]
6510 << PAGE_SHIFT
) - 1);
6513 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
6514 pr_info("Movable zone start for each node\n");
6515 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
6516 if (zone_movable_pfn
[i
])
6517 pr_info(" Node %d: %#018Lx\n", i
,
6518 (u64
)zone_movable_pfn
[i
] << PAGE_SHIFT
);
6521 /* Print out the early node map */
6522 pr_info("Early memory node ranges\n");
6523 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
6524 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid
,
6525 (u64
)start_pfn
<< PAGE_SHIFT
,
6526 ((u64
)end_pfn
<< PAGE_SHIFT
) - 1);
6528 /* Initialise every node */
6529 mminit_verify_pageflags_layout();
6530 setup_nr_node_ids();
6531 for_each_online_node(nid
) {
6532 pg_data_t
*pgdat
= NODE_DATA(nid
);
6533 free_area_init_node(nid
, NULL
,
6534 find_min_pfn_for_node(nid
), NULL
);
6536 /* Any memory on that node */
6537 if (pgdat
->node_present_pages
)
6538 node_set_state(nid
, N_MEMORY
);
6539 check_for_memory(pgdat
, nid
);
6543 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
6545 unsigned long long coremem
;
6549 coremem
= memparse(p
, &p
);
6550 *core
= coremem
>> PAGE_SHIFT
;
6552 /* Paranoid check that UL is enough for the coremem value */
6553 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
6559 * kernelcore=size sets the amount of memory for use for allocations that
6560 * cannot be reclaimed or migrated.
6562 static int __init
cmdline_parse_kernelcore(char *p
)
6564 /* parse kernelcore=mirror */
6565 if (parse_option_str(p
, "mirror")) {
6566 mirrored_kernelcore
= true;
6570 return cmdline_parse_core(p
, &required_kernelcore
);
6574 * movablecore=size sets the amount of memory for use for allocations that
6575 * can be reclaimed or migrated.
6577 static int __init
cmdline_parse_movablecore(char *p
)
6579 return cmdline_parse_core(p
, &required_movablecore
);
6582 early_param("kernelcore", cmdline_parse_kernelcore
);
6583 early_param("movablecore", cmdline_parse_movablecore
);
6585 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6587 void adjust_managed_page_count(struct page
*page
, long count
)
6589 spin_lock(&managed_page_count_lock
);
6590 page_zone(page
)->managed_pages
+= count
;
6591 totalram_pages
+= count
;
6592 #ifdef CONFIG_HIGHMEM
6593 if (PageHighMem(page
))
6594 totalhigh_pages
+= count
;
6596 spin_unlock(&managed_page_count_lock
);
6598 EXPORT_SYMBOL(adjust_managed_page_count
);
6600 unsigned long free_reserved_area(void *start
, void *end
, int poison
, char *s
)
6603 unsigned long pages
= 0;
6605 start
= (void *)PAGE_ALIGN((unsigned long)start
);
6606 end
= (void *)((unsigned long)end
& PAGE_MASK
);
6607 for (pos
= start
; pos
< end
; pos
+= PAGE_SIZE
, pages
++) {
6608 if ((unsigned int)poison
<= 0xFF)
6609 memset(pos
, poison
, PAGE_SIZE
);
6610 free_reserved_page(virt_to_page(pos
));
6614 pr_info("Freeing %s memory: %ldK\n",
6615 s
, pages
<< (PAGE_SHIFT
- 10));
6619 EXPORT_SYMBOL(free_reserved_area
);
6621 #ifdef CONFIG_HIGHMEM
6622 void free_highmem_page(struct page
*page
)
6624 __free_reserved_page(page
);
6626 page_zone(page
)->managed_pages
++;
6632 void __init
mem_init_print_info(const char *str
)
6634 unsigned long physpages
, codesize
, datasize
, rosize
, bss_size
;
6635 unsigned long init_code_size
, init_data_size
;
6637 physpages
= get_num_physpages();
6638 codesize
= _etext
- _stext
;
6639 datasize
= _edata
- _sdata
;
6640 rosize
= __end_rodata
- __start_rodata
;
6641 bss_size
= __bss_stop
- __bss_start
;
6642 init_data_size
= __init_end
- __init_begin
;
6643 init_code_size
= _einittext
- _sinittext
;
6646 * Detect special cases and adjust section sizes accordingly:
6647 * 1) .init.* may be embedded into .data sections
6648 * 2) .init.text.* may be out of [__init_begin, __init_end],
6649 * please refer to arch/tile/kernel/vmlinux.lds.S.
6650 * 3) .rodata.* may be embedded into .text or .data sections.
6652 #define adj_init_size(start, end, size, pos, adj) \
6654 if (start <= pos && pos < end && size > adj) \
6658 adj_init_size(__init_begin
, __init_end
, init_data_size
,
6659 _sinittext
, init_code_size
);
6660 adj_init_size(_stext
, _etext
, codesize
, _sinittext
, init_code_size
);
6661 adj_init_size(_sdata
, _edata
, datasize
, __init_begin
, init_data_size
);
6662 adj_init_size(_stext
, _etext
, codesize
, __start_rodata
, rosize
);
6663 adj_init_size(_sdata
, _edata
, datasize
, __start_rodata
, rosize
);
6665 #undef adj_init_size
6667 pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
6668 #ifdef CONFIG_HIGHMEM
6672 nr_free_pages() << (PAGE_SHIFT
- 10),
6673 physpages
<< (PAGE_SHIFT
- 10),
6674 codesize
>> 10, datasize
>> 10, rosize
>> 10,
6675 (init_data_size
+ init_code_size
) >> 10, bss_size
>> 10,
6676 (physpages
- totalram_pages
- totalcma_pages
) << (PAGE_SHIFT
- 10),
6677 totalcma_pages
<< (PAGE_SHIFT
- 10),
6678 #ifdef CONFIG_HIGHMEM
6679 totalhigh_pages
<< (PAGE_SHIFT
- 10),
6681 str
? ", " : "", str
? str
: "");
6685 * set_dma_reserve - set the specified number of pages reserved in the first zone
6686 * @new_dma_reserve: The number of pages to mark reserved
6688 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
6689 * In the DMA zone, a significant percentage may be consumed by kernel image
6690 * and other unfreeable allocations which can skew the watermarks badly. This
6691 * function may optionally be used to account for unfreeable pages in the
6692 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6693 * smaller per-cpu batchsize.
6695 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
6697 dma_reserve
= new_dma_reserve
;
6700 void __init
free_area_init(unsigned long *zones_size
)
6702 free_area_init_node(0, zones_size
,
6703 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
6706 static int page_alloc_cpu_dead(unsigned int cpu
)
6709 lru_add_drain_cpu(cpu
);
6713 * Spill the event counters of the dead processor
6714 * into the current processors event counters.
6715 * This artificially elevates the count of the current
6718 vm_events_fold_cpu(cpu
);
6721 * Zero the differential counters of the dead processor
6722 * so that the vm statistics are consistent.
6724 * This is only okay since the processor is dead and cannot
6725 * race with what we are doing.
6727 cpu_vm_stats_fold(cpu
);
6731 void __init
page_alloc_init(void)
6735 ret
= cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD
,
6736 "mm/page_alloc:dead", NULL
,
6737 page_alloc_cpu_dead
);
6742 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
6743 * or min_free_kbytes changes.
6745 static void calculate_totalreserve_pages(void)
6747 struct pglist_data
*pgdat
;
6748 unsigned long reserve_pages
= 0;
6749 enum zone_type i
, j
;
6751 for_each_online_pgdat(pgdat
) {
6753 pgdat
->totalreserve_pages
= 0;
6755 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
6756 struct zone
*zone
= pgdat
->node_zones
+ i
;
6759 /* Find valid and maximum lowmem_reserve in the zone */
6760 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
6761 if (zone
->lowmem_reserve
[j
] > max
)
6762 max
= zone
->lowmem_reserve
[j
];
6765 /* we treat the high watermark as reserved pages. */
6766 max
+= high_wmark_pages(zone
);
6768 if (max
> zone
->managed_pages
)
6769 max
= zone
->managed_pages
;
6771 pgdat
->totalreserve_pages
+= max
;
6773 reserve_pages
+= max
;
6776 totalreserve_pages
= reserve_pages
;
6780 * setup_per_zone_lowmem_reserve - called whenever
6781 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
6782 * has a correct pages reserved value, so an adequate number of
6783 * pages are left in the zone after a successful __alloc_pages().
6785 static void setup_per_zone_lowmem_reserve(void)
6787 struct pglist_data
*pgdat
;
6788 enum zone_type j
, idx
;
6790 for_each_online_pgdat(pgdat
) {
6791 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
6792 struct zone
*zone
= pgdat
->node_zones
+ j
;
6793 unsigned long managed_pages
= zone
->managed_pages
;
6795 zone
->lowmem_reserve
[j
] = 0;
6799 struct zone
*lower_zone
;
6803 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
6804 sysctl_lowmem_reserve_ratio
[idx
] = 1;
6806 lower_zone
= pgdat
->node_zones
+ idx
;
6807 lower_zone
->lowmem_reserve
[j
] = managed_pages
/
6808 sysctl_lowmem_reserve_ratio
[idx
];
6809 managed_pages
+= lower_zone
->managed_pages
;
6814 /* update totalreserve_pages */
6815 calculate_totalreserve_pages();
6818 static void __setup_per_zone_wmarks(void)
6820 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
6821 unsigned long lowmem_pages
= 0;
6823 unsigned long flags
;
6825 /* Calculate total number of !ZONE_HIGHMEM pages */
6826 for_each_zone(zone
) {
6827 if (!is_highmem(zone
))
6828 lowmem_pages
+= zone
->managed_pages
;
6831 for_each_zone(zone
) {
6834 spin_lock_irqsave(&zone
->lock
, flags
);
6835 tmp
= (u64
)pages_min
* zone
->managed_pages
;
6836 do_div(tmp
, lowmem_pages
);
6837 if (is_highmem(zone
)) {
6839 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6840 * need highmem pages, so cap pages_min to a small
6843 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
6844 * deltas control asynch page reclaim, and so should
6845 * not be capped for highmem.
6847 unsigned long min_pages
;
6849 min_pages
= zone
->managed_pages
/ 1024;
6850 min_pages
= clamp(min_pages
, SWAP_CLUSTER_MAX
, 128UL);
6851 zone
->watermark
[WMARK_MIN
] = min_pages
;
6854 * If it's a lowmem zone, reserve a number of pages
6855 * proportionate to the zone's size.
6857 zone
->watermark
[WMARK_MIN
] = tmp
;
6861 * Set the kswapd watermarks distance according to the
6862 * scale factor in proportion to available memory, but
6863 * ensure a minimum size on small systems.
6865 tmp
= max_t(u64
, tmp
>> 2,
6866 mult_frac(zone
->managed_pages
,
6867 watermark_scale_factor
, 10000));
6869 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + tmp
;
6870 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + tmp
* 2;
6872 spin_unlock_irqrestore(&zone
->lock
, flags
);
6875 /* update totalreserve_pages */
6876 calculate_totalreserve_pages();
6880 * setup_per_zone_wmarks - called when min_free_kbytes changes
6881 * or when memory is hot-{added|removed}
6883 * Ensures that the watermark[min,low,high] values for each zone are set
6884 * correctly with respect to min_free_kbytes.
6886 void setup_per_zone_wmarks(void)
6888 mutex_lock(&zonelists_mutex
);
6889 __setup_per_zone_wmarks();
6890 mutex_unlock(&zonelists_mutex
);
6894 * Initialise min_free_kbytes.
6896 * For small machines we want it small (128k min). For large machines
6897 * we want it large (64MB max). But it is not linear, because network
6898 * bandwidth does not increase linearly with machine size. We use
6900 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
6901 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6917 int __meminit
init_per_zone_wmark_min(void)
6919 unsigned long lowmem_kbytes
;
6920 int new_min_free_kbytes
;
6922 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
6923 new_min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
6925 if (new_min_free_kbytes
> user_min_free_kbytes
) {
6926 min_free_kbytes
= new_min_free_kbytes
;
6927 if (min_free_kbytes
< 128)
6928 min_free_kbytes
= 128;
6929 if (min_free_kbytes
> 65536)
6930 min_free_kbytes
= 65536;
6932 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6933 new_min_free_kbytes
, user_min_free_kbytes
);
6935 setup_per_zone_wmarks();
6936 refresh_zone_stat_thresholds();
6937 setup_per_zone_lowmem_reserve();
6940 setup_min_unmapped_ratio();
6941 setup_min_slab_ratio();
6946 core_initcall(init_per_zone_wmark_min
)
6949 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
6950 * that we can call two helper functions whenever min_free_kbytes
6953 int min_free_kbytes_sysctl_handler(struct ctl_table
*table
, int write
,
6954 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6958 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6963 user_min_free_kbytes
= min_free_kbytes
;
6964 setup_per_zone_wmarks();
6969 int watermark_scale_factor_sysctl_handler(struct ctl_table
*table
, int write
,
6970 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6974 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6979 setup_per_zone_wmarks();
6985 static void setup_min_unmapped_ratio(void)
6990 for_each_online_pgdat(pgdat
)
6991 pgdat
->min_unmapped_pages
= 0;
6994 zone
->zone_pgdat
->min_unmapped_pages
+= (zone
->managed_pages
*
6995 sysctl_min_unmapped_ratio
) / 100;
6999 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
7000 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
7004 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
7008 setup_min_unmapped_ratio();
7013 static void setup_min_slab_ratio(void)
7018 for_each_online_pgdat(pgdat
)
7019 pgdat
->min_slab_pages
= 0;
7022 zone
->zone_pgdat
->min_slab_pages
+= (zone
->managed_pages
*
7023 sysctl_min_slab_ratio
) / 100;
7026 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
7027 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
7031 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
7035 setup_min_slab_ratio();
7042 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
7043 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
7044 * whenever sysctl_lowmem_reserve_ratio changes.
7046 * The reserve ratio obviously has absolutely no relation with the
7047 * minimum watermarks. The lowmem reserve ratio can only make sense
7048 * if in function of the boot time zone sizes.
7050 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
7051 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
7053 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
7054 setup_per_zone_lowmem_reserve();
7059 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
7060 * cpu. It is the fraction of total pages in each zone that a hot per cpu
7061 * pagelist can have before it gets flushed back to buddy allocator.
7063 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table
*table
, int write
,
7064 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
7067 int old_percpu_pagelist_fraction
;
7070 mutex_lock(&pcp_batch_high_lock
);
7071 old_percpu_pagelist_fraction
= percpu_pagelist_fraction
;
7073 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
7074 if (!write
|| ret
< 0)
7077 /* Sanity checking to avoid pcp imbalance */
7078 if (percpu_pagelist_fraction
&&
7079 percpu_pagelist_fraction
< MIN_PERCPU_PAGELIST_FRACTION
) {
7080 percpu_pagelist_fraction
= old_percpu_pagelist_fraction
;
7086 if (percpu_pagelist_fraction
== old_percpu_pagelist_fraction
)
7089 for_each_populated_zone(zone
) {
7092 for_each_possible_cpu(cpu
)
7093 pageset_set_high_and_batch(zone
,
7094 per_cpu_ptr(zone
->pageset
, cpu
));
7097 mutex_unlock(&pcp_batch_high_lock
);
7102 int hashdist
= HASHDIST_DEFAULT
;
7104 static int __init
set_hashdist(char *str
)
7108 hashdist
= simple_strtoul(str
, &str
, 0);
7111 __setup("hashdist=", set_hashdist
);
7114 #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES
7116 * Returns the number of pages that arch has reserved but
7117 * is not known to alloc_large_system_hash().
7119 static unsigned long __init
arch_reserved_kernel_pages(void)
7126 * allocate a large system hash table from bootmem
7127 * - it is assumed that the hash table must contain an exact power-of-2
7128 * quantity of entries
7129 * - limit is the number of hash buckets, not the total allocation size
7131 void *__init
alloc_large_system_hash(const char *tablename
,
7132 unsigned long bucketsize
,
7133 unsigned long numentries
,
7136 unsigned int *_hash_shift
,
7137 unsigned int *_hash_mask
,
7138 unsigned long low_limit
,
7139 unsigned long high_limit
)
7141 unsigned long long max
= high_limit
;
7142 unsigned long log2qty
, size
;
7145 /* allow the kernel cmdline to have a say */
7147 /* round applicable memory size up to nearest megabyte */
7148 numentries
= nr_kernel_pages
;
7149 numentries
-= arch_reserved_kernel_pages();
7151 /* It isn't necessary when PAGE_SIZE >= 1MB */
7152 if (PAGE_SHIFT
< 20)
7153 numentries
= round_up(numentries
, (1<<20)/PAGE_SIZE
);
7155 /* limit to 1 bucket per 2^scale bytes of low memory */
7156 if (scale
> PAGE_SHIFT
)
7157 numentries
>>= (scale
- PAGE_SHIFT
);
7159 numentries
<<= (PAGE_SHIFT
- scale
);
7161 /* Make sure we've got at least a 0-order allocation.. */
7162 if (unlikely(flags
& HASH_SMALL
)) {
7163 /* Makes no sense without HASH_EARLY */
7164 WARN_ON(!(flags
& HASH_EARLY
));
7165 if (!(numentries
>> *_hash_shift
)) {
7166 numentries
= 1UL << *_hash_shift
;
7167 BUG_ON(!numentries
);
7169 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
7170 numentries
= PAGE_SIZE
/ bucketsize
;
7172 numentries
= roundup_pow_of_two(numentries
);
7174 /* limit allocation size to 1/16 total memory by default */
7176 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
7177 do_div(max
, bucketsize
);
7179 max
= min(max
, 0x80000000ULL
);
7181 if (numentries
< low_limit
)
7182 numentries
= low_limit
;
7183 if (numentries
> max
)
7186 log2qty
= ilog2(numentries
);
7189 size
= bucketsize
<< log2qty
;
7190 if (flags
& HASH_EARLY
)
7191 table
= memblock_virt_alloc_nopanic(size
, 0);
7193 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
7196 * If bucketsize is not a power-of-two, we may free
7197 * some pages at the end of hash table which
7198 * alloc_pages_exact() automatically does
7200 if (get_order(size
) < MAX_ORDER
) {
7201 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
7202 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
7205 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
7208 panic("Failed to allocate %s hash table\n", tablename
);
7210 pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n",
7211 tablename
, 1UL << log2qty
, ilog2(size
) - PAGE_SHIFT
, size
);
7214 *_hash_shift
= log2qty
;
7216 *_hash_mask
= (1 << log2qty
) - 1;
7222 * This function checks whether pageblock includes unmovable pages or not.
7223 * If @count is not zero, it is okay to include less @count unmovable pages
7225 * PageLRU check without isolation or lru_lock could race so that
7226 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
7227 * expect this function should be exact.
7229 bool has_unmovable_pages(struct zone
*zone
, struct page
*page
, int count
,
7230 bool skip_hwpoisoned_pages
)
7232 unsigned long pfn
, iter
, found
;
7236 * For avoiding noise data, lru_add_drain_all() should be called
7237 * If ZONE_MOVABLE, the zone never contains unmovable pages
7239 if (zone_idx(zone
) == ZONE_MOVABLE
)
7241 mt
= get_pageblock_migratetype(page
);
7242 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
7245 pfn
= page_to_pfn(page
);
7246 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
7247 unsigned long check
= pfn
+ iter
;
7249 if (!pfn_valid_within(check
))
7252 page
= pfn_to_page(check
);
7255 * Hugepages are not in LRU lists, but they're movable.
7256 * We need not scan over tail pages bacause we don't
7257 * handle each tail page individually in migration.
7259 if (PageHuge(page
)) {
7260 iter
= round_up(iter
+ 1, 1<<compound_order(page
)) - 1;
7265 * We can't use page_count without pin a page
7266 * because another CPU can free compound page.
7267 * This check already skips compound tails of THP
7268 * because their page->_refcount is zero at all time.
7270 if (!page_ref_count(page
)) {
7271 if (PageBuddy(page
))
7272 iter
+= (1 << page_order(page
)) - 1;
7277 * The HWPoisoned page may be not in buddy system, and
7278 * page_count() is not 0.
7280 if (skip_hwpoisoned_pages
&& PageHWPoison(page
))
7286 * If there are RECLAIMABLE pages, we need to check
7287 * it. But now, memory offline itself doesn't call
7288 * shrink_node_slabs() and it still to be fixed.
7291 * If the page is not RAM, page_count()should be 0.
7292 * we don't need more check. This is an _used_ not-movable page.
7294 * The problematic thing here is PG_reserved pages. PG_reserved
7295 * is set to both of a memory hole page and a _used_ kernel
7304 bool is_pageblock_removable_nolock(struct page
*page
)
7310 * We have to be careful here because we are iterating over memory
7311 * sections which are not zone aware so we might end up outside of
7312 * the zone but still within the section.
7313 * We have to take care about the node as well. If the node is offline
7314 * its NODE_DATA will be NULL - see page_zone.
7316 if (!node_online(page_to_nid(page
)))
7319 zone
= page_zone(page
);
7320 pfn
= page_to_pfn(page
);
7321 if (!zone_spans_pfn(zone
, pfn
))
7324 return !has_unmovable_pages(zone
, page
, 0, true);
7327 #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
7329 static unsigned long pfn_max_align_down(unsigned long pfn
)
7331 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
7332 pageblock_nr_pages
) - 1);
7335 static unsigned long pfn_max_align_up(unsigned long pfn
)
7337 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
7338 pageblock_nr_pages
));
7341 /* [start, end) must belong to a single zone. */
7342 static int __alloc_contig_migrate_range(struct compact_control
*cc
,
7343 unsigned long start
, unsigned long end
)
7345 /* This function is based on compact_zone() from compaction.c. */
7346 unsigned long nr_reclaimed
;
7347 unsigned long pfn
= start
;
7348 unsigned int tries
= 0;
7353 while (pfn
< end
|| !list_empty(&cc
->migratepages
)) {
7354 if (fatal_signal_pending(current
)) {
7359 if (list_empty(&cc
->migratepages
)) {
7360 cc
->nr_migratepages
= 0;
7361 pfn
= isolate_migratepages_range(cc
, pfn
, end
);
7367 } else if (++tries
== 5) {
7368 ret
= ret
< 0 ? ret
: -EBUSY
;
7372 nr_reclaimed
= reclaim_clean_pages_from_list(cc
->zone
,
7374 cc
->nr_migratepages
-= nr_reclaimed
;
7376 ret
= migrate_pages(&cc
->migratepages
, alloc_migrate_target
,
7377 NULL
, 0, cc
->mode
, MR_CMA
);
7380 putback_movable_pages(&cc
->migratepages
);
7387 * alloc_contig_range() -- tries to allocate given range of pages
7388 * @start: start PFN to allocate
7389 * @end: one-past-the-last PFN to allocate
7390 * @migratetype: migratetype of the underlaying pageblocks (either
7391 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
7392 * in range must have the same migratetype and it must
7393 * be either of the two.
7395 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
7396 * aligned, however it's the caller's responsibility to guarantee that
7397 * we are the only thread that changes migrate type of pageblocks the
7400 * The PFN range must belong to a single zone.
7402 * Returns zero on success or negative error code. On success all
7403 * pages which PFN is in [start, end) are allocated for the caller and
7404 * need to be freed with free_contig_range().
7406 int alloc_contig_range(unsigned long start
, unsigned long end
,
7407 unsigned migratetype
)
7409 unsigned long outer_start
, outer_end
;
7413 struct compact_control cc
= {
7414 .nr_migratepages
= 0,
7416 .zone
= page_zone(pfn_to_page(start
)),
7417 .mode
= MIGRATE_SYNC
,
7418 .ignore_skip_hint
= true,
7419 .gfp_mask
= GFP_KERNEL
,
7421 INIT_LIST_HEAD(&cc
.migratepages
);
7424 * What we do here is we mark all pageblocks in range as
7425 * MIGRATE_ISOLATE. Because pageblock and max order pages may
7426 * have different sizes, and due to the way page allocator
7427 * work, we align the range to biggest of the two pages so
7428 * that page allocator won't try to merge buddies from
7429 * different pageblocks and change MIGRATE_ISOLATE to some
7430 * other migration type.
7432 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
7433 * migrate the pages from an unaligned range (ie. pages that
7434 * we are interested in). This will put all the pages in
7435 * range back to page allocator as MIGRATE_ISOLATE.
7437 * When this is done, we take the pages in range from page
7438 * allocator removing them from the buddy system. This way
7439 * page allocator will never consider using them.
7441 * This lets us mark the pageblocks back as
7442 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
7443 * aligned range but not in the unaligned, original range are
7444 * put back to page allocator so that buddy can use them.
7447 ret
= start_isolate_page_range(pfn_max_align_down(start
),
7448 pfn_max_align_up(end
), migratetype
,
7454 * In case of -EBUSY, we'd like to know which page causes problem.
7455 * So, just fall through. We will check it in test_pages_isolated().
7457 ret
= __alloc_contig_migrate_range(&cc
, start
, end
);
7458 if (ret
&& ret
!= -EBUSY
)
7462 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
7463 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
7464 * more, all pages in [start, end) are free in page allocator.
7465 * What we are going to do is to allocate all pages from
7466 * [start, end) (that is remove them from page allocator).
7468 * The only problem is that pages at the beginning and at the
7469 * end of interesting range may be not aligned with pages that
7470 * page allocator holds, ie. they can be part of higher order
7471 * pages. Because of this, we reserve the bigger range and
7472 * once this is done free the pages we are not interested in.
7474 * We don't have to hold zone->lock here because the pages are
7475 * isolated thus they won't get removed from buddy.
7478 lru_add_drain_all();
7479 drain_all_pages(cc
.zone
);
7482 outer_start
= start
;
7483 while (!PageBuddy(pfn_to_page(outer_start
))) {
7484 if (++order
>= MAX_ORDER
) {
7485 outer_start
= start
;
7488 outer_start
&= ~0UL << order
;
7491 if (outer_start
!= start
) {
7492 order
= page_order(pfn_to_page(outer_start
));
7495 * outer_start page could be small order buddy page and
7496 * it doesn't include start page. Adjust outer_start
7497 * in this case to report failed page properly
7498 * on tracepoint in test_pages_isolated()
7500 if (outer_start
+ (1UL << order
) <= start
)
7501 outer_start
= start
;
7504 /* Make sure the range is really isolated. */
7505 if (test_pages_isolated(outer_start
, end
, false)) {
7506 pr_info("%s: [%lx, %lx) PFNs busy\n",
7507 __func__
, outer_start
, end
);
7512 /* Grab isolated pages from freelists. */
7513 outer_end
= isolate_freepages_range(&cc
, outer_start
, end
);
7519 /* Free head and tail (if any) */
7520 if (start
!= outer_start
)
7521 free_contig_range(outer_start
, start
- outer_start
);
7522 if (end
!= outer_end
)
7523 free_contig_range(end
, outer_end
- end
);
7526 undo_isolate_page_range(pfn_max_align_down(start
),
7527 pfn_max_align_up(end
), migratetype
);
7531 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
7533 unsigned int count
= 0;
7535 for (; nr_pages
--; pfn
++) {
7536 struct page
*page
= pfn_to_page(pfn
);
7538 count
+= page_count(page
) != 1;
7541 WARN(count
!= 0, "%d pages are still in use!\n", count
);
7545 #ifdef CONFIG_MEMORY_HOTPLUG
7547 * The zone indicated has a new number of managed_pages; batch sizes and percpu
7548 * page high values need to be recalulated.
7550 void __meminit
zone_pcp_update(struct zone
*zone
)
7553 mutex_lock(&pcp_batch_high_lock
);
7554 for_each_possible_cpu(cpu
)
7555 pageset_set_high_and_batch(zone
,
7556 per_cpu_ptr(zone
->pageset
, cpu
));
7557 mutex_unlock(&pcp_batch_high_lock
);
7561 void zone_pcp_reset(struct zone
*zone
)
7563 unsigned long flags
;
7565 struct per_cpu_pageset
*pset
;
7567 /* avoid races with drain_pages() */
7568 local_irq_save(flags
);
7569 if (zone
->pageset
!= &boot_pageset
) {
7570 for_each_online_cpu(cpu
) {
7571 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
7572 drain_zonestat(zone
, pset
);
7574 free_percpu(zone
->pageset
);
7575 zone
->pageset
= &boot_pageset
;
7577 local_irq_restore(flags
);
7580 #ifdef CONFIG_MEMORY_HOTREMOVE
7582 * All pages in the range must be in a single zone and isolated
7583 * before calling this.
7586 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
7590 unsigned int order
, i
;
7592 unsigned long flags
;
7593 /* find the first valid pfn */
7594 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
7599 zone
= page_zone(pfn_to_page(pfn
));
7600 spin_lock_irqsave(&zone
->lock
, flags
);
7602 while (pfn
< end_pfn
) {
7603 if (!pfn_valid(pfn
)) {
7607 page
= pfn_to_page(pfn
);
7609 * The HWPoisoned page may be not in buddy system, and
7610 * page_count() is not 0.
7612 if (unlikely(!PageBuddy(page
) && PageHWPoison(page
))) {
7614 SetPageReserved(page
);
7618 BUG_ON(page_count(page
));
7619 BUG_ON(!PageBuddy(page
));
7620 order
= page_order(page
);
7621 #ifdef CONFIG_DEBUG_VM
7622 pr_info("remove from free list %lx %d %lx\n",
7623 pfn
, 1 << order
, end_pfn
);
7625 list_del(&page
->lru
);
7626 rmv_page_order(page
);
7627 zone
->free_area
[order
].nr_free
--;
7628 for (i
= 0; i
< (1 << order
); i
++)
7629 SetPageReserved((page
+i
));
7630 pfn
+= (1 << order
);
7632 spin_unlock_irqrestore(&zone
->lock
, flags
);
7636 bool is_free_buddy_page(struct page
*page
)
7638 struct zone
*zone
= page_zone(page
);
7639 unsigned long pfn
= page_to_pfn(page
);
7640 unsigned long flags
;
7643 spin_lock_irqsave(&zone
->lock
, flags
);
7644 for (order
= 0; order
< MAX_ORDER
; order
++) {
7645 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
7647 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
7650 spin_unlock_irqrestore(&zone
->lock
, flags
);
7652 return order
< MAX_ORDER
;