2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/jiffies.h>
23 #include <linux/bootmem.h>
24 #include <linux/memblock.h>
25 #include <linux/compiler.h>
26 #include <linux/kernel.h>
27 #include <linux/kmemcheck.h>
28 #include <linux/kasan.h>
29 #include <linux/module.h>
30 #include <linux/suspend.h>
31 #include <linux/pagevec.h>
32 #include <linux/blkdev.h>
33 #include <linux/slab.h>
34 #include <linux/ratelimit.h>
35 #include <linux/oom.h>
36 #include <linux/notifier.h>
37 #include <linux/topology.h>
38 #include <linux/sysctl.h>
39 #include <linux/cpu.h>
40 #include <linux/cpuset.h>
41 #include <linux/memory_hotplug.h>
42 #include <linux/nodemask.h>
43 #include <linux/vmalloc.h>
44 #include <linux/vmstat.h>
45 #include <linux/mempolicy.h>
46 #include <linux/memremap.h>
47 #include <linux/stop_machine.h>
48 #include <linux/sort.h>
49 #include <linux/pfn.h>
50 #include <linux/backing-dev.h>
51 #include <linux/fault-inject.h>
52 #include <linux/page-isolation.h>
53 #include <linux/page_ext.h>
54 #include <linux/debugobjects.h>
55 #include <linux/kmemleak.h>
56 #include <linux/compaction.h>
57 #include <trace/events/kmem.h>
58 #include <linux/prefetch.h>
59 #include <linux/mm_inline.h>
60 #include <linux/migrate.h>
61 #include <linux/page_ext.h>
62 #include <linux/hugetlb.h>
63 #include <linux/sched/rt.h>
64 #include <linux/page_owner.h>
65 #include <linux/kthread.h>
67 #include <asm/sections.h>
68 #include <asm/tlbflush.h>
69 #include <asm/div64.h>
72 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
73 static DEFINE_MUTEX(pcp_batch_high_lock
);
74 #define MIN_PERCPU_PAGELIST_FRACTION (8)
76 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
77 DEFINE_PER_CPU(int, numa_node
);
78 EXPORT_PER_CPU_SYMBOL(numa_node
);
81 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
83 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
84 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
85 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
86 * defined in <linux/topology.h>.
88 DEFINE_PER_CPU(int, _numa_mem_
); /* Kernel "local memory" node */
89 EXPORT_PER_CPU_SYMBOL(_numa_mem_
);
90 int _node_numa_mem_
[MAX_NUMNODES
];
94 * Array of node states.
96 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
97 [N_POSSIBLE
] = NODE_MASK_ALL
,
98 [N_ONLINE
] = { { [0] = 1UL } },
100 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
101 #ifdef CONFIG_HIGHMEM
102 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
104 #ifdef CONFIG_MOVABLE_NODE
105 [N_MEMORY
] = { { [0] = 1UL } },
107 [N_CPU
] = { { [0] = 1UL } },
110 EXPORT_SYMBOL(node_states
);
112 /* Protect totalram_pages and zone->managed_pages */
113 static DEFINE_SPINLOCK(managed_page_count_lock
);
115 unsigned long totalram_pages __read_mostly
;
116 unsigned long totalreserve_pages __read_mostly
;
117 unsigned long totalcma_pages __read_mostly
;
119 int percpu_pagelist_fraction
;
120 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
123 * A cached value of the page's pageblock's migratetype, used when the page is
124 * put on a pcplist. Used to avoid the pageblock migratetype lookup when
125 * freeing from pcplists in most cases, at the cost of possibly becoming stale.
126 * Also the migratetype set in the page does not necessarily match the pcplist
127 * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
128 * other index - this ensures that it will be put on the correct CMA freelist.
130 static inline int get_pcppage_migratetype(struct page
*page
)
135 static inline void set_pcppage_migratetype(struct page
*page
, int migratetype
)
137 page
->index
= migratetype
;
140 #ifdef CONFIG_PM_SLEEP
142 * The following functions are used by the suspend/hibernate code to temporarily
143 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
144 * while devices are suspended. To avoid races with the suspend/hibernate code,
145 * they should always be called with pm_mutex held (gfp_allowed_mask also should
146 * only be modified with pm_mutex held, unless the suspend/hibernate code is
147 * guaranteed not to run in parallel with that modification).
150 static gfp_t saved_gfp_mask
;
152 void pm_restore_gfp_mask(void)
154 WARN_ON(!mutex_is_locked(&pm_mutex
));
155 if (saved_gfp_mask
) {
156 gfp_allowed_mask
= saved_gfp_mask
;
161 void pm_restrict_gfp_mask(void)
163 WARN_ON(!mutex_is_locked(&pm_mutex
));
164 WARN_ON(saved_gfp_mask
);
165 saved_gfp_mask
= gfp_allowed_mask
;
166 gfp_allowed_mask
&= ~(__GFP_IO
| __GFP_FS
);
169 bool pm_suspended_storage(void)
171 if ((gfp_allowed_mask
& (__GFP_IO
| __GFP_FS
)) == (__GFP_IO
| __GFP_FS
))
175 #endif /* CONFIG_PM_SLEEP */
177 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
178 unsigned int pageblock_order __read_mostly
;
181 static void __free_pages_ok(struct page
*page
, unsigned int order
);
184 * results with 256, 32 in the lowmem_reserve sysctl:
185 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
186 * 1G machine -> (16M dma, 784M normal, 224M high)
187 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
188 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
189 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
191 * TBD: should special case ZONE_DMA32 machines here - in those we normally
192 * don't need any ZONE_NORMAL reservation
194 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
195 #ifdef CONFIG_ZONE_DMA
198 #ifdef CONFIG_ZONE_DMA32
201 #ifdef CONFIG_HIGHMEM
207 EXPORT_SYMBOL(totalram_pages
);
209 static char * const zone_names
[MAX_NR_ZONES
] = {
210 #ifdef CONFIG_ZONE_DMA
213 #ifdef CONFIG_ZONE_DMA32
217 #ifdef CONFIG_HIGHMEM
221 #ifdef CONFIG_ZONE_DEVICE
226 char * const migratetype_names
[MIGRATE_TYPES
] = {
234 #ifdef CONFIG_MEMORY_ISOLATION
239 compound_page_dtor
* const compound_page_dtors
[] = {
242 #ifdef CONFIG_HUGETLB_PAGE
245 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
250 int min_free_kbytes
= 1024;
251 int user_min_free_kbytes
= -1;
253 static unsigned long __meminitdata nr_kernel_pages
;
254 static unsigned long __meminitdata nr_all_pages
;
255 static unsigned long __meminitdata dma_reserve
;
257 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
258 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
259 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
260 static unsigned long __initdata required_kernelcore
;
261 static unsigned long __initdata required_movablecore
;
262 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
263 static bool mirrored_kernelcore
;
265 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
267 EXPORT_SYMBOL(movable_zone
);
268 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
271 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
272 int nr_online_nodes __read_mostly
= 1;
273 EXPORT_SYMBOL(nr_node_ids
);
274 EXPORT_SYMBOL(nr_online_nodes
);
277 int page_group_by_mobility_disabled __read_mostly
;
279 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
280 static inline void reset_deferred_meminit(pg_data_t
*pgdat
)
282 pgdat
->first_deferred_pfn
= ULONG_MAX
;
285 /* Returns true if the struct page for the pfn is uninitialised */
286 static inline bool __meminit
early_page_uninitialised(unsigned long pfn
)
288 if (pfn
>= NODE_DATA(early_pfn_to_nid(pfn
))->first_deferred_pfn
)
294 static inline bool early_page_nid_uninitialised(unsigned long pfn
, int nid
)
296 if (pfn
>= NODE_DATA(nid
)->first_deferred_pfn
)
303 * Returns false when the remaining initialisation should be deferred until
304 * later in the boot cycle when it can be parallelised.
306 static inline bool update_defer_init(pg_data_t
*pgdat
,
307 unsigned long pfn
, unsigned long zone_end
,
308 unsigned long *nr_initialised
)
310 /* Always populate low zones for address-contrained allocations */
311 if (zone_end
< pgdat_end_pfn(pgdat
))
314 /* Initialise at least 2G of the highest zone */
316 if (*nr_initialised
> (2UL << (30 - PAGE_SHIFT
)) &&
317 (pfn
& (PAGES_PER_SECTION
- 1)) == 0) {
318 pgdat
->first_deferred_pfn
= pfn
;
325 static inline void reset_deferred_meminit(pg_data_t
*pgdat
)
329 static inline bool early_page_uninitialised(unsigned long pfn
)
334 static inline bool early_page_nid_uninitialised(unsigned long pfn
, int nid
)
339 static inline bool update_defer_init(pg_data_t
*pgdat
,
340 unsigned long pfn
, unsigned long zone_end
,
341 unsigned long *nr_initialised
)
348 void set_pageblock_migratetype(struct page
*page
, int migratetype
)
350 if (unlikely(page_group_by_mobility_disabled
&&
351 migratetype
< MIGRATE_PCPTYPES
))
352 migratetype
= MIGRATE_UNMOVABLE
;
354 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
355 PB_migrate
, PB_migrate_end
);
358 #ifdef CONFIG_DEBUG_VM
359 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
363 unsigned long pfn
= page_to_pfn(page
);
364 unsigned long sp
, start_pfn
;
367 seq
= zone_span_seqbegin(zone
);
368 start_pfn
= zone
->zone_start_pfn
;
369 sp
= zone
->spanned_pages
;
370 if (!zone_spans_pfn(zone
, pfn
))
372 } while (zone_span_seqretry(zone
, seq
));
375 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
376 pfn
, zone_to_nid(zone
), zone
->name
,
377 start_pfn
, start_pfn
+ sp
);
382 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
384 if (!pfn_valid_within(page_to_pfn(page
)))
386 if (zone
!= page_zone(page
))
392 * Temporary debugging check for pages not lying within a given zone.
394 static int bad_range(struct zone
*zone
, struct page
*page
)
396 if (page_outside_zone_boundaries(zone
, page
))
398 if (!page_is_consistent(zone
, page
))
404 static inline int bad_range(struct zone
*zone
, struct page
*page
)
410 static void bad_page(struct page
*page
, const char *reason
,
411 unsigned long bad_flags
)
413 static unsigned long resume
;
414 static unsigned long nr_shown
;
415 static unsigned long nr_unshown
;
417 /* Don't complain about poisoned pages */
418 if (PageHWPoison(page
)) {
419 page_mapcount_reset(page
); /* remove PageBuddy */
424 * Allow a burst of 60 reports, then keep quiet for that minute;
425 * or allow a steady drip of one report per second.
427 if (nr_shown
== 60) {
428 if (time_before(jiffies
, resume
)) {
434 "BUG: Bad page state: %lu messages suppressed\n",
441 resume
= jiffies
+ 60 * HZ
;
443 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
444 current
->comm
, page_to_pfn(page
));
445 dump_page_badflags(page
, reason
, bad_flags
);
446 dump_page_owner(page
);
451 /* Leave bad fields for debug, except PageBuddy could make trouble */
452 page_mapcount_reset(page
); /* remove PageBuddy */
453 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
457 * Higher-order pages are called "compound pages". They are structured thusly:
459 * The first PAGE_SIZE page is called the "head page" and have PG_head set.
461 * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
462 * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
464 * The first tail page's ->compound_dtor holds the offset in array of compound
465 * page destructors. See compound_page_dtors.
467 * The first tail page's ->compound_order holds the order of allocation.
468 * This usage means that zero-order pages may not be compound.
471 void free_compound_page(struct page
*page
)
473 __free_pages_ok(page
, compound_order(page
));
476 void prep_compound_page(struct page
*page
, unsigned int order
)
479 int nr_pages
= 1 << order
;
481 set_compound_page_dtor(page
, COMPOUND_PAGE_DTOR
);
482 set_compound_order(page
, order
);
484 for (i
= 1; i
< nr_pages
; i
++) {
485 struct page
*p
= page
+ i
;
486 set_page_count(p
, 0);
487 p
->mapping
= TAIL_MAPPING
;
488 set_compound_head(p
, page
);
490 atomic_set(compound_mapcount_ptr(page
), -1);
493 #ifdef CONFIG_DEBUG_PAGEALLOC
494 unsigned int _debug_guardpage_minorder
;
495 bool _debug_pagealloc_enabled __read_mostly
496 = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT
);
497 bool _debug_guardpage_enabled __read_mostly
;
499 static int __init
early_debug_pagealloc(char *buf
)
504 if (strcmp(buf
, "on") == 0)
505 _debug_pagealloc_enabled
= true;
507 if (strcmp(buf
, "off") == 0)
508 _debug_pagealloc_enabled
= false;
512 early_param("debug_pagealloc", early_debug_pagealloc
);
514 static bool need_debug_guardpage(void)
516 /* If we don't use debug_pagealloc, we don't need guard page */
517 if (!debug_pagealloc_enabled())
523 static void init_debug_guardpage(void)
525 if (!debug_pagealloc_enabled())
528 _debug_guardpage_enabled
= true;
531 struct page_ext_operations debug_guardpage_ops
= {
532 .need
= need_debug_guardpage
,
533 .init
= init_debug_guardpage
,
536 static int __init
debug_guardpage_minorder_setup(char *buf
)
540 if (kstrtoul(buf
, 10, &res
) < 0 || res
> MAX_ORDER
/ 2) {
541 printk(KERN_ERR
"Bad debug_guardpage_minorder value\n");
544 _debug_guardpage_minorder
= res
;
545 printk(KERN_INFO
"Setting debug_guardpage_minorder to %lu\n", res
);
548 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup
);
550 static inline void set_page_guard(struct zone
*zone
, struct page
*page
,
551 unsigned int order
, int migratetype
)
553 struct page_ext
*page_ext
;
555 if (!debug_guardpage_enabled())
558 page_ext
= lookup_page_ext(page
);
559 __set_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
561 INIT_LIST_HEAD(&page
->lru
);
562 set_page_private(page
, order
);
563 /* Guard pages are not available for any usage */
564 __mod_zone_freepage_state(zone
, -(1 << order
), migratetype
);
567 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
568 unsigned int order
, int migratetype
)
570 struct page_ext
*page_ext
;
572 if (!debug_guardpage_enabled())
575 page_ext
= lookup_page_ext(page
);
576 __clear_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
578 set_page_private(page
, 0);
579 if (!is_migrate_isolate(migratetype
))
580 __mod_zone_freepage_state(zone
, (1 << order
), migratetype
);
583 struct page_ext_operations debug_guardpage_ops
= { NULL
, };
584 static inline void set_page_guard(struct zone
*zone
, struct page
*page
,
585 unsigned int order
, int migratetype
) {}
586 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
587 unsigned int order
, int migratetype
) {}
590 static inline void set_page_order(struct page
*page
, unsigned int order
)
592 set_page_private(page
, order
);
593 __SetPageBuddy(page
);
596 static inline void rmv_page_order(struct page
*page
)
598 __ClearPageBuddy(page
);
599 set_page_private(page
, 0);
603 * This function checks whether a page is free && is the buddy
604 * we can do coalesce a page and its buddy if
605 * (a) the buddy is not in a hole &&
606 * (b) the buddy is in the buddy system &&
607 * (c) a page and its buddy have the same order &&
608 * (d) a page and its buddy are in the same zone.
610 * For recording whether a page is in the buddy system, we set ->_mapcount
611 * PAGE_BUDDY_MAPCOUNT_VALUE.
612 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
613 * serialized by zone->lock.
615 * For recording page's order, we use page_private(page).
617 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
620 if (!pfn_valid_within(page_to_pfn(buddy
)))
623 if (page_is_guard(buddy
) && page_order(buddy
) == order
) {
624 if (page_zone_id(page
) != page_zone_id(buddy
))
627 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
632 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
634 * zone check is done late to avoid uselessly
635 * calculating zone/node ids for pages that could
638 if (page_zone_id(page
) != page_zone_id(buddy
))
641 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
649 * Freeing function for a buddy system allocator.
651 * The concept of a buddy system is to maintain direct-mapped table
652 * (containing bit values) for memory blocks of various "orders".
653 * The bottom level table contains the map for the smallest allocatable
654 * units of memory (here, pages), and each level above it describes
655 * pairs of units from the levels below, hence, "buddies".
656 * At a high level, all that happens here is marking the table entry
657 * at the bottom level available, and propagating the changes upward
658 * as necessary, plus some accounting needed to play nicely with other
659 * parts of the VM system.
660 * At each level, we keep a list of pages, which are heads of continuous
661 * free pages of length of (1 << order) and marked with _mapcount
662 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
664 * So when we are allocating or freeing one, we can derive the state of the
665 * other. That is, if we allocate a small block, and both were
666 * free, the remainder of the region must be split into blocks.
667 * If a block is freed, and its buddy is also free, then this
668 * triggers coalescing into a block of larger size.
673 static inline void __free_one_page(struct page
*page
,
675 struct zone
*zone
, unsigned int order
,
678 unsigned long page_idx
;
679 unsigned long combined_idx
;
680 unsigned long uninitialized_var(buddy_idx
);
682 unsigned int max_order
= MAX_ORDER
;
684 VM_BUG_ON(!zone_is_initialized(zone
));
685 VM_BUG_ON_PAGE(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
, page
);
687 VM_BUG_ON(migratetype
== -1);
688 if (is_migrate_isolate(migratetype
)) {
690 * We restrict max order of merging to prevent merge
691 * between freepages on isolate pageblock and normal
692 * pageblock. Without this, pageblock isolation
693 * could cause incorrect freepage accounting.
695 max_order
= min_t(unsigned int, MAX_ORDER
, pageblock_order
+ 1);
697 __mod_zone_freepage_state(zone
, 1 << order
, migratetype
);
700 page_idx
= pfn
& ((1 << max_order
) - 1);
702 VM_BUG_ON_PAGE(page_idx
& ((1 << order
) - 1), page
);
703 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
705 while (order
< max_order
- 1) {
706 buddy_idx
= __find_buddy_index(page_idx
, order
);
707 buddy
= page
+ (buddy_idx
- page_idx
);
708 if (!page_is_buddy(page
, buddy
, order
))
711 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
712 * merge with it and move up one order.
714 if (page_is_guard(buddy
)) {
715 clear_page_guard(zone
, buddy
, order
, migratetype
);
717 list_del(&buddy
->lru
);
718 zone
->free_area
[order
].nr_free
--;
719 rmv_page_order(buddy
);
721 combined_idx
= buddy_idx
& page_idx
;
722 page
= page
+ (combined_idx
- page_idx
);
723 page_idx
= combined_idx
;
726 set_page_order(page
, order
);
729 * If this is not the largest possible page, check if the buddy
730 * of the next-highest order is free. If it is, it's possible
731 * that pages are being freed that will coalesce soon. In case,
732 * that is happening, add the free page to the tail of the list
733 * so it's less likely to be used soon and more likely to be merged
734 * as a higher order page
736 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
737 struct page
*higher_page
, *higher_buddy
;
738 combined_idx
= buddy_idx
& page_idx
;
739 higher_page
= page
+ (combined_idx
- page_idx
);
740 buddy_idx
= __find_buddy_index(combined_idx
, order
+ 1);
741 higher_buddy
= higher_page
+ (buddy_idx
- combined_idx
);
742 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
743 list_add_tail(&page
->lru
,
744 &zone
->free_area
[order
].free_list
[migratetype
]);
749 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
751 zone
->free_area
[order
].nr_free
++;
754 static inline int free_pages_check(struct page
*page
)
756 const char *bad_reason
= NULL
;
757 unsigned long bad_flags
= 0;
759 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
760 bad_reason
= "nonzero mapcount";
761 if (unlikely(page
->mapping
!= NULL
))
762 bad_reason
= "non-NULL mapping";
763 if (unlikely(atomic_read(&page
->_count
) != 0))
764 bad_reason
= "nonzero _count";
765 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
)) {
766 bad_reason
= "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
767 bad_flags
= PAGE_FLAGS_CHECK_AT_FREE
;
770 if (unlikely(page
->mem_cgroup
))
771 bad_reason
= "page still charged to cgroup";
773 if (unlikely(bad_reason
)) {
774 bad_page(page
, bad_reason
, bad_flags
);
777 page_cpupid_reset_last(page
);
778 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
779 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
784 * Frees a number of pages from the PCP lists
785 * Assumes all pages on list are in same zone, and of same order.
786 * count is the number of pages to free.
788 * If the zone was previously in an "all pages pinned" state then look to
789 * see if this freeing clears that state.
791 * And clear the zone's pages_scanned counter, to hold off the "all pages are
792 * pinned" detection logic.
794 static void free_pcppages_bulk(struct zone
*zone
, int count
,
795 struct per_cpu_pages
*pcp
)
800 unsigned long nr_scanned
;
802 spin_lock(&zone
->lock
);
803 nr_scanned
= zone_page_state(zone
, NR_PAGES_SCANNED
);
805 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, -nr_scanned
);
809 struct list_head
*list
;
812 * Remove pages from lists in a round-robin fashion. A
813 * batch_free count is maintained that is incremented when an
814 * empty list is encountered. This is so more pages are freed
815 * off fuller lists instead of spinning excessively around empty
820 if (++migratetype
== MIGRATE_PCPTYPES
)
822 list
= &pcp
->lists
[migratetype
];
823 } while (list_empty(list
));
825 /* This is the only non-empty list. Free them all. */
826 if (batch_free
== MIGRATE_PCPTYPES
)
827 batch_free
= to_free
;
830 int mt
; /* migratetype of the to-be-freed page */
832 page
= list_last_entry(list
, struct page
, lru
);
833 /* must delete as __free_one_page list manipulates */
834 list_del(&page
->lru
);
836 mt
= get_pcppage_migratetype(page
);
837 /* MIGRATE_ISOLATE page should not go to pcplists */
838 VM_BUG_ON_PAGE(is_migrate_isolate(mt
), page
);
839 /* Pageblock could have been isolated meanwhile */
840 if (unlikely(has_isolate_pageblock(zone
)))
841 mt
= get_pageblock_migratetype(page
);
843 __free_one_page(page
, page_to_pfn(page
), zone
, 0, mt
);
844 trace_mm_page_pcpu_drain(page
, 0, mt
);
845 } while (--to_free
&& --batch_free
&& !list_empty(list
));
847 spin_unlock(&zone
->lock
);
850 static void free_one_page(struct zone
*zone
,
851 struct page
*page
, unsigned long pfn
,
855 unsigned long nr_scanned
;
856 spin_lock(&zone
->lock
);
857 nr_scanned
= zone_page_state(zone
, NR_PAGES_SCANNED
);
859 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, -nr_scanned
);
861 if (unlikely(has_isolate_pageblock(zone
) ||
862 is_migrate_isolate(migratetype
))) {
863 migratetype
= get_pfnblock_migratetype(page
, pfn
);
865 __free_one_page(page
, pfn
, zone
, order
, migratetype
);
866 spin_unlock(&zone
->lock
);
869 static int free_tail_pages_check(struct page
*head_page
, struct page
*page
)
874 * We rely page->lru.next never has bit 0 set, unless the page
875 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
877 BUILD_BUG_ON((unsigned long)LIST_POISON1
& 1);
879 if (!IS_ENABLED(CONFIG_DEBUG_VM
)) {
883 switch (page
- head_page
) {
885 /* the first tail page: ->mapping is compound_mapcount() */
886 if (unlikely(compound_mapcount(page
))) {
887 bad_page(page
, "nonzero compound_mapcount", 0);
893 * the second tail page: ->mapping is
894 * page_deferred_list().next -- ignore value.
898 if (page
->mapping
!= TAIL_MAPPING
) {
899 bad_page(page
, "corrupted mapping in tail page", 0);
904 if (unlikely(!PageTail(page
))) {
905 bad_page(page
, "PageTail not set", 0);
908 if (unlikely(compound_head(page
) != head_page
)) {
909 bad_page(page
, "compound_head not consistent", 0);
914 page
->mapping
= NULL
;
915 clear_compound_head(page
);
919 static void __meminit
__init_single_page(struct page
*page
, unsigned long pfn
,
920 unsigned long zone
, int nid
)
922 set_page_links(page
, zone
, nid
, pfn
);
923 init_page_count(page
);
924 page_mapcount_reset(page
);
925 page_cpupid_reset_last(page
);
927 INIT_LIST_HEAD(&page
->lru
);
928 #ifdef WANT_PAGE_VIRTUAL
929 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
930 if (!is_highmem_idx(zone
))
931 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
935 static void __meminit
__init_single_pfn(unsigned long pfn
, unsigned long zone
,
938 return __init_single_page(pfn_to_page(pfn
), pfn
, zone
, nid
);
941 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
942 static void init_reserved_page(unsigned long pfn
)
947 if (!early_page_uninitialised(pfn
))
950 nid
= early_pfn_to_nid(pfn
);
951 pgdat
= NODE_DATA(nid
);
953 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
954 struct zone
*zone
= &pgdat
->node_zones
[zid
];
956 if (pfn
>= zone
->zone_start_pfn
&& pfn
< zone_end_pfn(zone
))
959 __init_single_pfn(pfn
, zid
, nid
);
962 static inline void init_reserved_page(unsigned long pfn
)
965 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
968 * Initialised pages do not have PageReserved set. This function is
969 * called for each range allocated by the bootmem allocator and
970 * marks the pages PageReserved. The remaining valid pages are later
971 * sent to the buddy page allocator.
973 void __meminit
reserve_bootmem_region(unsigned long start
, unsigned long end
)
975 unsigned long start_pfn
= PFN_DOWN(start
);
976 unsigned long end_pfn
= PFN_UP(end
);
978 for (; start_pfn
< end_pfn
; start_pfn
++) {
979 if (pfn_valid(start_pfn
)) {
980 struct page
*page
= pfn_to_page(start_pfn
);
982 init_reserved_page(start_pfn
);
984 /* Avoid false-positive PageTail() */
985 INIT_LIST_HEAD(&page
->lru
);
987 SetPageReserved(page
);
992 static bool free_pages_prepare(struct page
*page
, unsigned int order
)
994 bool compound
= PageCompound(page
);
997 VM_BUG_ON_PAGE(PageTail(page
), page
);
998 VM_BUG_ON_PAGE(compound
&& compound_order(page
) != order
, page
);
1000 trace_mm_page_free(page
, order
);
1001 kmemcheck_free_shadow(page
, order
);
1002 kasan_free_pages(page
, order
);
1005 page
->mapping
= NULL
;
1006 bad
+= free_pages_check(page
);
1007 for (i
= 1; i
< (1 << order
); i
++) {
1009 bad
+= free_tail_pages_check(page
, page
+ i
);
1010 bad
+= free_pages_check(page
+ i
);
1015 reset_page_owner(page
, order
);
1017 if (!PageHighMem(page
)) {
1018 debug_check_no_locks_freed(page_address(page
),
1019 PAGE_SIZE
<< order
);
1020 debug_check_no_obj_freed(page_address(page
),
1021 PAGE_SIZE
<< order
);
1023 arch_free_page(page
, order
);
1024 kernel_map_pages(page
, 1 << order
, 0);
1029 static void __free_pages_ok(struct page
*page
, unsigned int order
)
1031 unsigned long flags
;
1033 unsigned long pfn
= page_to_pfn(page
);
1035 if (!free_pages_prepare(page
, order
))
1038 migratetype
= get_pfnblock_migratetype(page
, pfn
);
1039 local_irq_save(flags
);
1040 __count_vm_events(PGFREE
, 1 << order
);
1041 free_one_page(page_zone(page
), page
, pfn
, order
, migratetype
);
1042 local_irq_restore(flags
);
1045 static void __init
__free_pages_boot_core(struct page
*page
,
1046 unsigned long pfn
, unsigned int order
)
1048 unsigned int nr_pages
= 1 << order
;
1049 struct page
*p
= page
;
1053 for (loop
= 0; loop
< (nr_pages
- 1); loop
++, p
++) {
1055 __ClearPageReserved(p
);
1056 set_page_count(p
, 0);
1058 __ClearPageReserved(p
);
1059 set_page_count(p
, 0);
1061 page_zone(page
)->managed_pages
+= nr_pages
;
1062 set_page_refcounted(page
);
1063 __free_pages(page
, order
);
1066 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1067 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1069 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata
;
1071 int __meminit
early_pfn_to_nid(unsigned long pfn
)
1073 static DEFINE_SPINLOCK(early_pfn_lock
);
1076 spin_lock(&early_pfn_lock
);
1077 nid
= __early_pfn_to_nid(pfn
, &early_pfnnid_cache
);
1080 spin_unlock(&early_pfn_lock
);
1086 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1087 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1088 struct mminit_pfnnid_cache
*state
)
1092 nid
= __early_pfn_to_nid(pfn
, state
);
1093 if (nid
>= 0 && nid
!= node
)
1098 /* Only safe to use early in boot when initialisation is single-threaded */
1099 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1101 return meminit_pfn_in_nid(pfn
, node
, &early_pfnnid_cache
);
1106 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1110 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1111 struct mminit_pfnnid_cache
*state
)
1118 void __init
__free_pages_bootmem(struct page
*page
, unsigned long pfn
,
1121 if (early_page_uninitialised(pfn
))
1123 return __free_pages_boot_core(page
, pfn
, order
);
1126 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1127 static void __init
deferred_free_range(struct page
*page
,
1128 unsigned long pfn
, int nr_pages
)
1135 /* Free a large naturally-aligned chunk if possible */
1136 if (nr_pages
== MAX_ORDER_NR_PAGES
&&
1137 (pfn
& (MAX_ORDER_NR_PAGES
-1)) == 0) {
1138 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
1139 __free_pages_boot_core(page
, pfn
, MAX_ORDER
-1);
1143 for (i
= 0; i
< nr_pages
; i
++, page
++, pfn
++)
1144 __free_pages_boot_core(page
, pfn
, 0);
1147 /* Completion tracking for deferred_init_memmap() threads */
1148 static atomic_t pgdat_init_n_undone __initdata
;
1149 static __initdata
DECLARE_COMPLETION(pgdat_init_all_done_comp
);
1151 static inline void __init
pgdat_init_report_one_done(void)
1153 if (atomic_dec_and_test(&pgdat_init_n_undone
))
1154 complete(&pgdat_init_all_done_comp
);
1157 /* Initialise remaining memory on a node */
1158 static int __init
deferred_init_memmap(void *data
)
1160 pg_data_t
*pgdat
= data
;
1161 int nid
= pgdat
->node_id
;
1162 struct mminit_pfnnid_cache nid_init_state
= { };
1163 unsigned long start
= jiffies
;
1164 unsigned long nr_pages
= 0;
1165 unsigned long walk_start
, walk_end
;
1168 unsigned long first_init_pfn
= pgdat
->first_deferred_pfn
;
1169 const struct cpumask
*cpumask
= cpumask_of_node(pgdat
->node_id
);
1171 if (first_init_pfn
== ULONG_MAX
) {
1172 pgdat_init_report_one_done();
1176 /* Bind memory initialisation thread to a local node if possible */
1177 if (!cpumask_empty(cpumask
))
1178 set_cpus_allowed_ptr(current
, cpumask
);
1180 /* Sanity check boundaries */
1181 BUG_ON(pgdat
->first_deferred_pfn
< pgdat
->node_start_pfn
);
1182 BUG_ON(pgdat
->first_deferred_pfn
> pgdat_end_pfn(pgdat
));
1183 pgdat
->first_deferred_pfn
= ULONG_MAX
;
1185 /* Only the highest zone is deferred so find it */
1186 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
1187 zone
= pgdat
->node_zones
+ zid
;
1188 if (first_init_pfn
< zone_end_pfn(zone
))
1192 for_each_mem_pfn_range(i
, nid
, &walk_start
, &walk_end
, NULL
) {
1193 unsigned long pfn
, end_pfn
;
1194 struct page
*page
= NULL
;
1195 struct page
*free_base_page
= NULL
;
1196 unsigned long free_base_pfn
= 0;
1199 end_pfn
= min(walk_end
, zone_end_pfn(zone
));
1200 pfn
= first_init_pfn
;
1201 if (pfn
< walk_start
)
1203 if (pfn
< zone
->zone_start_pfn
)
1204 pfn
= zone
->zone_start_pfn
;
1206 for (; pfn
< end_pfn
; pfn
++) {
1207 if (!pfn_valid_within(pfn
))
1211 * Ensure pfn_valid is checked every
1212 * MAX_ORDER_NR_PAGES for memory holes
1214 if ((pfn
& (MAX_ORDER_NR_PAGES
- 1)) == 0) {
1215 if (!pfn_valid(pfn
)) {
1221 if (!meminit_pfn_in_nid(pfn
, nid
, &nid_init_state
)) {
1226 /* Minimise pfn page lookups and scheduler checks */
1227 if (page
&& (pfn
& (MAX_ORDER_NR_PAGES
- 1)) != 0) {
1230 nr_pages
+= nr_to_free
;
1231 deferred_free_range(free_base_page
,
1232 free_base_pfn
, nr_to_free
);
1233 free_base_page
= NULL
;
1234 free_base_pfn
= nr_to_free
= 0;
1236 page
= pfn_to_page(pfn
);
1241 VM_BUG_ON(page_zone(page
) != zone
);
1245 __init_single_page(page
, pfn
, zid
, nid
);
1246 if (!free_base_page
) {
1247 free_base_page
= page
;
1248 free_base_pfn
= pfn
;
1253 /* Where possible, batch up pages for a single free */
1256 /* Free the current block of pages to allocator */
1257 nr_pages
+= nr_to_free
;
1258 deferred_free_range(free_base_page
, free_base_pfn
,
1260 free_base_page
= NULL
;
1261 free_base_pfn
= nr_to_free
= 0;
1264 first_init_pfn
= max(end_pfn
, first_init_pfn
);
1267 /* Sanity check that the next zone really is unpopulated */
1268 WARN_ON(++zid
< MAX_NR_ZONES
&& populated_zone(++zone
));
1270 pr_info("node %d initialised, %lu pages in %ums\n", nid
, nr_pages
,
1271 jiffies_to_msecs(jiffies
- start
));
1273 pgdat_init_report_one_done();
1277 void __init
page_alloc_init_late(void)
1281 /* There will be num_node_state(N_MEMORY) threads */
1282 atomic_set(&pgdat_init_n_undone
, num_node_state(N_MEMORY
));
1283 for_each_node_state(nid
, N_MEMORY
) {
1284 kthread_run(deferred_init_memmap
, NODE_DATA(nid
), "pgdatinit%d", nid
);
1287 /* Block until all are initialised */
1288 wait_for_completion(&pgdat_init_all_done_comp
);
1290 /* Reinit limits that are based on free pages after the kernel is up */
1291 files_maxfiles_init();
1293 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1296 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1297 void __init
init_cma_reserved_pageblock(struct page
*page
)
1299 unsigned i
= pageblock_nr_pages
;
1300 struct page
*p
= page
;
1303 __ClearPageReserved(p
);
1304 set_page_count(p
, 0);
1307 set_pageblock_migratetype(page
, MIGRATE_CMA
);
1309 if (pageblock_order
>= MAX_ORDER
) {
1310 i
= pageblock_nr_pages
;
1313 set_page_refcounted(p
);
1314 __free_pages(p
, MAX_ORDER
- 1);
1315 p
+= MAX_ORDER_NR_PAGES
;
1316 } while (i
-= MAX_ORDER_NR_PAGES
);
1318 set_page_refcounted(page
);
1319 __free_pages(page
, pageblock_order
);
1322 adjust_managed_page_count(page
, pageblock_nr_pages
);
1327 * The order of subdivision here is critical for the IO subsystem.
1328 * Please do not alter this order without good reasons and regression
1329 * testing. Specifically, as large blocks of memory are subdivided,
1330 * the order in which smaller blocks are delivered depends on the order
1331 * they're subdivided in this function. This is the primary factor
1332 * influencing the order in which pages are delivered to the IO
1333 * subsystem according to empirical testing, and this is also justified
1334 * by considering the behavior of a buddy system containing a single
1335 * large block of memory acted on by a series of small allocations.
1336 * This behavior is a critical factor in sglist merging's success.
1340 static inline void expand(struct zone
*zone
, struct page
*page
,
1341 int low
, int high
, struct free_area
*area
,
1344 unsigned long size
= 1 << high
;
1346 while (high
> low
) {
1350 VM_BUG_ON_PAGE(bad_range(zone
, &page
[size
]), &page
[size
]);
1352 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC
) &&
1353 debug_guardpage_enabled() &&
1354 high
< debug_guardpage_minorder()) {
1356 * Mark as guard pages (or page), that will allow to
1357 * merge back to allocator when buddy will be freed.
1358 * Corresponding page table entries will not be touched,
1359 * pages will stay not present in virtual address space
1361 set_page_guard(zone
, &page
[size
], high
, migratetype
);
1364 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
1366 set_page_order(&page
[size
], high
);
1371 * This page is about to be returned from the page allocator
1373 static inline int check_new_page(struct page
*page
)
1375 const char *bad_reason
= NULL
;
1376 unsigned long bad_flags
= 0;
1378 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
1379 bad_reason
= "nonzero mapcount";
1380 if (unlikely(page
->mapping
!= NULL
))
1381 bad_reason
= "non-NULL mapping";
1382 if (unlikely(atomic_read(&page
->_count
) != 0))
1383 bad_reason
= "nonzero _count";
1384 if (unlikely(page
->flags
& __PG_HWPOISON
)) {
1385 bad_reason
= "HWPoisoned (hardware-corrupted)";
1386 bad_flags
= __PG_HWPOISON
;
1388 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)) {
1389 bad_reason
= "PAGE_FLAGS_CHECK_AT_PREP flag set";
1390 bad_flags
= PAGE_FLAGS_CHECK_AT_PREP
;
1393 if (unlikely(page
->mem_cgroup
))
1394 bad_reason
= "page still charged to cgroup";
1396 if (unlikely(bad_reason
)) {
1397 bad_page(page
, bad_reason
, bad_flags
);
1403 static int prep_new_page(struct page
*page
, unsigned int order
, gfp_t gfp_flags
,
1408 for (i
= 0; i
< (1 << order
); i
++) {
1409 struct page
*p
= page
+ i
;
1410 if (unlikely(check_new_page(p
)))
1414 set_page_private(page
, 0);
1415 set_page_refcounted(page
);
1417 arch_alloc_page(page
, order
);
1418 kernel_map_pages(page
, 1 << order
, 1);
1419 kasan_alloc_pages(page
, order
);
1421 if (gfp_flags
& __GFP_ZERO
)
1422 for (i
= 0; i
< (1 << order
); i
++)
1423 clear_highpage(page
+ i
);
1425 if (order
&& (gfp_flags
& __GFP_COMP
))
1426 prep_compound_page(page
, order
);
1428 set_page_owner(page
, order
, gfp_flags
);
1431 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1432 * allocate the page. The expectation is that the caller is taking
1433 * steps that will free more memory. The caller should avoid the page
1434 * being used for !PFMEMALLOC purposes.
1436 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
1437 set_page_pfmemalloc(page
);
1439 clear_page_pfmemalloc(page
);
1445 * Go through the free lists for the given migratetype and remove
1446 * the smallest available page from the freelists
1449 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
1452 unsigned int current_order
;
1453 struct free_area
*area
;
1456 /* Find a page of the appropriate size in the preferred list */
1457 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
1458 area
= &(zone
->free_area
[current_order
]);
1459 page
= list_first_entry_or_null(&area
->free_list
[migratetype
],
1463 list_del(&page
->lru
);
1464 rmv_page_order(page
);
1466 expand(zone
, page
, order
, current_order
, area
, migratetype
);
1467 set_pcppage_migratetype(page
, migratetype
);
1476 * This array describes the order lists are fallen back to when
1477 * the free lists for the desirable migrate type are depleted
1479 static int fallbacks
[MIGRATE_TYPES
][4] = {
1480 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1481 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1482 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_TYPES
},
1484 [MIGRATE_CMA
] = { MIGRATE_TYPES
}, /* Never used */
1486 #ifdef CONFIG_MEMORY_ISOLATION
1487 [MIGRATE_ISOLATE
] = { MIGRATE_TYPES
}, /* Never used */
1492 static struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1495 return __rmqueue_smallest(zone
, order
, MIGRATE_CMA
);
1498 static inline struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1499 unsigned int order
) { return NULL
; }
1503 * Move the free pages in a range to the free lists of the requested type.
1504 * Note that start_page and end_pages are not aligned on a pageblock
1505 * boundary. If alignment is required, use move_freepages_block()
1507 int move_freepages(struct zone
*zone
,
1508 struct page
*start_page
, struct page
*end_page
,
1513 int pages_moved
= 0;
1515 #ifndef CONFIG_HOLES_IN_ZONE
1517 * page_zone is not safe to call in this context when
1518 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1519 * anyway as we check zone boundaries in move_freepages_block().
1520 * Remove at a later date when no bug reports exist related to
1521 * grouping pages by mobility
1523 VM_BUG_ON(page_zone(start_page
) != page_zone(end_page
));
1526 for (page
= start_page
; page
<= end_page
;) {
1527 /* Make sure we are not inadvertently changing nodes */
1528 VM_BUG_ON_PAGE(page_to_nid(page
) != zone_to_nid(zone
), page
);
1530 if (!pfn_valid_within(page_to_pfn(page
))) {
1535 if (!PageBuddy(page
)) {
1540 order
= page_order(page
);
1541 list_move(&page
->lru
,
1542 &zone
->free_area
[order
].free_list
[migratetype
]);
1544 pages_moved
+= 1 << order
;
1550 int move_freepages_block(struct zone
*zone
, struct page
*page
,
1553 unsigned long start_pfn
, end_pfn
;
1554 struct page
*start_page
, *end_page
;
1556 start_pfn
= page_to_pfn(page
);
1557 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
1558 start_page
= pfn_to_page(start_pfn
);
1559 end_page
= start_page
+ pageblock_nr_pages
- 1;
1560 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
1562 /* Do not cross zone boundaries */
1563 if (!zone_spans_pfn(zone
, start_pfn
))
1565 if (!zone_spans_pfn(zone
, end_pfn
))
1568 return move_freepages(zone
, start_page
, end_page
, migratetype
);
1571 static void change_pageblock_range(struct page
*pageblock_page
,
1572 int start_order
, int migratetype
)
1574 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
1576 while (nr_pageblocks
--) {
1577 set_pageblock_migratetype(pageblock_page
, migratetype
);
1578 pageblock_page
+= pageblock_nr_pages
;
1583 * When we are falling back to another migratetype during allocation, try to
1584 * steal extra free pages from the same pageblocks to satisfy further
1585 * allocations, instead of polluting multiple pageblocks.
1587 * If we are stealing a relatively large buddy page, it is likely there will
1588 * be more free pages in the pageblock, so try to steal them all. For
1589 * reclaimable and unmovable allocations, we steal regardless of page size,
1590 * as fragmentation caused by those allocations polluting movable pageblocks
1591 * is worse than movable allocations stealing from unmovable and reclaimable
1594 static bool can_steal_fallback(unsigned int order
, int start_mt
)
1597 * Leaving this order check is intended, although there is
1598 * relaxed order check in next check. The reason is that
1599 * we can actually steal whole pageblock if this condition met,
1600 * but, below check doesn't guarantee it and that is just heuristic
1601 * so could be changed anytime.
1603 if (order
>= pageblock_order
)
1606 if (order
>= pageblock_order
/ 2 ||
1607 start_mt
== MIGRATE_RECLAIMABLE
||
1608 start_mt
== MIGRATE_UNMOVABLE
||
1609 page_group_by_mobility_disabled
)
1616 * This function implements actual steal behaviour. If order is large enough,
1617 * we can steal whole pageblock. If not, we first move freepages in this
1618 * pageblock and check whether half of pages are moved or not. If half of
1619 * pages are moved, we can change migratetype of pageblock and permanently
1620 * use it's pages as requested migratetype in the future.
1622 static void steal_suitable_fallback(struct zone
*zone
, struct page
*page
,
1625 unsigned int current_order
= page_order(page
);
1628 /* Take ownership for orders >= pageblock_order */
1629 if (current_order
>= pageblock_order
) {
1630 change_pageblock_range(page
, current_order
, start_type
);
1634 pages
= move_freepages_block(zone
, page
, start_type
);
1636 /* Claim the whole block if over half of it is free */
1637 if (pages
>= (1 << (pageblock_order
-1)) ||
1638 page_group_by_mobility_disabled
)
1639 set_pageblock_migratetype(page
, start_type
);
1643 * Check whether there is a suitable fallback freepage with requested order.
1644 * If only_stealable is true, this function returns fallback_mt only if
1645 * we can steal other freepages all together. This would help to reduce
1646 * fragmentation due to mixed migratetype pages in one pageblock.
1648 int find_suitable_fallback(struct free_area
*area
, unsigned int order
,
1649 int migratetype
, bool only_stealable
, bool *can_steal
)
1654 if (area
->nr_free
== 0)
1659 fallback_mt
= fallbacks
[migratetype
][i
];
1660 if (fallback_mt
== MIGRATE_TYPES
)
1663 if (list_empty(&area
->free_list
[fallback_mt
]))
1666 if (can_steal_fallback(order
, migratetype
))
1669 if (!only_stealable
)
1680 * Reserve a pageblock for exclusive use of high-order atomic allocations if
1681 * there are no empty page blocks that contain a page with a suitable order
1683 static void reserve_highatomic_pageblock(struct page
*page
, struct zone
*zone
,
1684 unsigned int alloc_order
)
1687 unsigned long max_managed
, flags
;
1690 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
1691 * Check is race-prone but harmless.
1693 max_managed
= (zone
->managed_pages
/ 100) + pageblock_nr_pages
;
1694 if (zone
->nr_reserved_highatomic
>= max_managed
)
1697 spin_lock_irqsave(&zone
->lock
, flags
);
1699 /* Recheck the nr_reserved_highatomic limit under the lock */
1700 if (zone
->nr_reserved_highatomic
>= max_managed
)
1704 mt
= get_pageblock_migratetype(page
);
1705 if (mt
!= MIGRATE_HIGHATOMIC
&&
1706 !is_migrate_isolate(mt
) && !is_migrate_cma(mt
)) {
1707 zone
->nr_reserved_highatomic
+= pageblock_nr_pages
;
1708 set_pageblock_migratetype(page
, MIGRATE_HIGHATOMIC
);
1709 move_freepages_block(zone
, page
, MIGRATE_HIGHATOMIC
);
1713 spin_unlock_irqrestore(&zone
->lock
, flags
);
1717 * Used when an allocation is about to fail under memory pressure. This
1718 * potentially hurts the reliability of high-order allocations when under
1719 * intense memory pressure but failed atomic allocations should be easier
1720 * to recover from than an OOM.
1722 static void unreserve_highatomic_pageblock(const struct alloc_context
*ac
)
1724 struct zonelist
*zonelist
= ac
->zonelist
;
1725 unsigned long flags
;
1731 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
, ac
->high_zoneidx
,
1733 /* Preserve at least one pageblock */
1734 if (zone
->nr_reserved_highatomic
<= pageblock_nr_pages
)
1737 spin_lock_irqsave(&zone
->lock
, flags
);
1738 for (order
= 0; order
< MAX_ORDER
; order
++) {
1739 struct free_area
*area
= &(zone
->free_area
[order
]);
1741 page
= list_first_entry_or_null(
1742 &area
->free_list
[MIGRATE_HIGHATOMIC
],
1748 * It should never happen but changes to locking could
1749 * inadvertently allow a per-cpu drain to add pages
1750 * to MIGRATE_HIGHATOMIC while unreserving so be safe
1751 * and watch for underflows.
1753 zone
->nr_reserved_highatomic
-= min(pageblock_nr_pages
,
1754 zone
->nr_reserved_highatomic
);
1757 * Convert to ac->migratetype and avoid the normal
1758 * pageblock stealing heuristics. Minimally, the caller
1759 * is doing the work and needs the pages. More
1760 * importantly, if the block was always converted to
1761 * MIGRATE_UNMOVABLE or another type then the number
1762 * of pageblocks that cannot be completely freed
1765 set_pageblock_migratetype(page
, ac
->migratetype
);
1766 move_freepages_block(zone
, page
, ac
->migratetype
);
1767 spin_unlock_irqrestore(&zone
->lock
, flags
);
1770 spin_unlock_irqrestore(&zone
->lock
, flags
);
1774 /* Remove an element from the buddy allocator from the fallback list */
1775 static inline struct page
*
1776 __rmqueue_fallback(struct zone
*zone
, unsigned int order
, int start_migratetype
)
1778 struct free_area
*area
;
1779 unsigned int current_order
;
1784 /* Find the largest possible block of pages in the other list */
1785 for (current_order
= MAX_ORDER
-1;
1786 current_order
>= order
&& current_order
<= MAX_ORDER
-1;
1788 area
= &(zone
->free_area
[current_order
]);
1789 fallback_mt
= find_suitable_fallback(area
, current_order
,
1790 start_migratetype
, false, &can_steal
);
1791 if (fallback_mt
== -1)
1794 page
= list_first_entry(&area
->free_list
[fallback_mt
],
1797 steal_suitable_fallback(zone
, page
, start_migratetype
);
1799 /* Remove the page from the freelists */
1801 list_del(&page
->lru
);
1802 rmv_page_order(page
);
1804 expand(zone
, page
, order
, current_order
, area
,
1807 * The pcppage_migratetype may differ from pageblock's
1808 * migratetype depending on the decisions in
1809 * find_suitable_fallback(). This is OK as long as it does not
1810 * differ for MIGRATE_CMA pageblocks. Those can be used as
1811 * fallback only via special __rmqueue_cma_fallback() function
1813 set_pcppage_migratetype(page
, start_migratetype
);
1815 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
1816 start_migratetype
, fallback_mt
);
1825 * Do the hard work of removing an element from the buddy allocator.
1826 * Call me with the zone->lock already held.
1828 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
1833 page
= __rmqueue_smallest(zone
, order
, migratetype
);
1834 if (unlikely(!page
)) {
1835 if (migratetype
== MIGRATE_MOVABLE
)
1836 page
= __rmqueue_cma_fallback(zone
, order
);
1839 page
= __rmqueue_fallback(zone
, order
, migratetype
);
1842 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
1847 * Obtain a specified number of elements from the buddy allocator, all under
1848 * a single hold of the lock, for efficiency. Add them to the supplied list.
1849 * Returns the number of new pages which were placed at *list.
1851 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
1852 unsigned long count
, struct list_head
*list
,
1853 int migratetype
, bool cold
)
1857 spin_lock(&zone
->lock
);
1858 for (i
= 0; i
< count
; ++i
) {
1859 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
1860 if (unlikely(page
== NULL
))
1864 * Split buddy pages returned by expand() are received here
1865 * in physical page order. The page is added to the callers and
1866 * list and the list head then moves forward. From the callers
1867 * perspective, the linked list is ordered by page number in
1868 * some conditions. This is useful for IO devices that can
1869 * merge IO requests if the physical pages are ordered
1873 list_add(&page
->lru
, list
);
1875 list_add_tail(&page
->lru
, list
);
1877 if (is_migrate_cma(get_pcppage_migratetype(page
)))
1878 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
,
1881 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
1882 spin_unlock(&zone
->lock
);
1888 * Called from the vmstat counter updater to drain pagesets of this
1889 * currently executing processor on remote nodes after they have
1892 * Note that this function must be called with the thread pinned to
1893 * a single processor.
1895 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
1897 unsigned long flags
;
1898 int to_drain
, batch
;
1900 local_irq_save(flags
);
1901 batch
= READ_ONCE(pcp
->batch
);
1902 to_drain
= min(pcp
->count
, batch
);
1904 free_pcppages_bulk(zone
, to_drain
, pcp
);
1905 pcp
->count
-= to_drain
;
1907 local_irq_restore(flags
);
1912 * Drain pcplists of the indicated processor and zone.
1914 * The processor must either be the current processor and the
1915 * thread pinned to the current processor or a processor that
1918 static void drain_pages_zone(unsigned int cpu
, struct zone
*zone
)
1920 unsigned long flags
;
1921 struct per_cpu_pageset
*pset
;
1922 struct per_cpu_pages
*pcp
;
1924 local_irq_save(flags
);
1925 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
1929 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
1932 local_irq_restore(flags
);
1936 * Drain pcplists of all zones on the indicated processor.
1938 * The processor must either be the current processor and the
1939 * thread pinned to the current processor or a processor that
1942 static void drain_pages(unsigned int cpu
)
1946 for_each_populated_zone(zone
) {
1947 drain_pages_zone(cpu
, zone
);
1952 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1954 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
1955 * the single zone's pages.
1957 void drain_local_pages(struct zone
*zone
)
1959 int cpu
= smp_processor_id();
1962 drain_pages_zone(cpu
, zone
);
1968 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1970 * When zone parameter is non-NULL, spill just the single zone's pages.
1972 * Note that this code is protected against sending an IPI to an offline
1973 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1974 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1975 * nothing keeps CPUs from showing up after we populated the cpumask and
1976 * before the call to on_each_cpu_mask().
1978 void drain_all_pages(struct zone
*zone
)
1983 * Allocate in the BSS so we wont require allocation in
1984 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1986 static cpumask_t cpus_with_pcps
;
1989 * We don't care about racing with CPU hotplug event
1990 * as offline notification will cause the notified
1991 * cpu to drain that CPU pcps and on_each_cpu_mask
1992 * disables preemption as part of its processing
1994 for_each_online_cpu(cpu
) {
1995 struct per_cpu_pageset
*pcp
;
1997 bool has_pcps
= false;
2000 pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
2004 for_each_populated_zone(z
) {
2005 pcp
= per_cpu_ptr(z
->pageset
, cpu
);
2006 if (pcp
->pcp
.count
) {
2014 cpumask_set_cpu(cpu
, &cpus_with_pcps
);
2016 cpumask_clear_cpu(cpu
, &cpus_with_pcps
);
2018 on_each_cpu_mask(&cpus_with_pcps
, (smp_call_func_t
) drain_local_pages
,
2022 #ifdef CONFIG_HIBERNATION
2024 void mark_free_pages(struct zone
*zone
)
2026 unsigned long pfn
, max_zone_pfn
;
2027 unsigned long flags
;
2028 unsigned int order
, t
;
2031 if (zone_is_empty(zone
))
2034 spin_lock_irqsave(&zone
->lock
, flags
);
2036 max_zone_pfn
= zone_end_pfn(zone
);
2037 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
2038 if (pfn_valid(pfn
)) {
2039 page
= pfn_to_page(pfn
);
2040 if (!swsusp_page_is_forbidden(page
))
2041 swsusp_unset_page_free(page
);
2044 for_each_migratetype_order(order
, t
) {
2045 list_for_each_entry(page
,
2046 &zone
->free_area
[order
].free_list
[t
], lru
) {
2049 pfn
= page_to_pfn(page
);
2050 for (i
= 0; i
< (1UL << order
); i
++)
2051 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
2054 spin_unlock_irqrestore(&zone
->lock
, flags
);
2056 #endif /* CONFIG_PM */
2059 * Free a 0-order page
2060 * cold == true ? free a cold page : free a hot page
2062 void free_hot_cold_page(struct page
*page
, bool cold
)
2064 struct zone
*zone
= page_zone(page
);
2065 struct per_cpu_pages
*pcp
;
2066 unsigned long flags
;
2067 unsigned long pfn
= page_to_pfn(page
);
2070 if (!free_pages_prepare(page
, 0))
2073 migratetype
= get_pfnblock_migratetype(page
, pfn
);
2074 set_pcppage_migratetype(page
, migratetype
);
2075 local_irq_save(flags
);
2076 __count_vm_event(PGFREE
);
2079 * We only track unmovable, reclaimable and movable on pcp lists.
2080 * Free ISOLATE pages back to the allocator because they are being
2081 * offlined but treat RESERVE as movable pages so we can get those
2082 * areas back if necessary. Otherwise, we may have to free
2083 * excessively into the page allocator
2085 if (migratetype
>= MIGRATE_PCPTYPES
) {
2086 if (unlikely(is_migrate_isolate(migratetype
))) {
2087 free_one_page(zone
, page
, pfn
, 0, migratetype
);
2090 migratetype
= MIGRATE_MOVABLE
;
2093 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2095 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
2097 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
2099 if (pcp
->count
>= pcp
->high
) {
2100 unsigned long batch
= READ_ONCE(pcp
->batch
);
2101 free_pcppages_bulk(zone
, batch
, pcp
);
2102 pcp
->count
-= batch
;
2106 local_irq_restore(flags
);
2110 * Free a list of 0-order pages
2112 void free_hot_cold_page_list(struct list_head
*list
, bool cold
)
2114 struct page
*page
, *next
;
2116 list_for_each_entry_safe(page
, next
, list
, lru
) {
2117 trace_mm_page_free_batched(page
, cold
);
2118 free_hot_cold_page(page
, cold
);
2123 * split_page takes a non-compound higher-order page, and splits it into
2124 * n (1<<order) sub-pages: page[0..n]
2125 * Each sub-page must be freed individually.
2127 * Note: this is probably too low level an operation for use in drivers.
2128 * Please consult with lkml before using this in your driver.
2130 void split_page(struct page
*page
, unsigned int order
)
2135 VM_BUG_ON_PAGE(PageCompound(page
), page
);
2136 VM_BUG_ON_PAGE(!page_count(page
), page
);
2138 #ifdef CONFIG_KMEMCHECK
2140 * Split shadow pages too, because free(page[0]) would
2141 * otherwise free the whole shadow.
2143 if (kmemcheck_page_is_tracked(page
))
2144 split_page(virt_to_page(page
[0].shadow
), order
);
2147 gfp_mask
= get_page_owner_gfp(page
);
2148 set_page_owner(page
, 0, gfp_mask
);
2149 for (i
= 1; i
< (1 << order
); i
++) {
2150 set_page_refcounted(page
+ i
);
2151 set_page_owner(page
+ i
, 0, gfp_mask
);
2154 EXPORT_SYMBOL_GPL(split_page
);
2156 int __isolate_free_page(struct page
*page
, unsigned int order
)
2158 unsigned long watermark
;
2162 BUG_ON(!PageBuddy(page
));
2164 zone
= page_zone(page
);
2165 mt
= get_pageblock_migratetype(page
);
2167 if (!is_migrate_isolate(mt
)) {
2168 /* Obey watermarks as if the page was being allocated */
2169 watermark
= low_wmark_pages(zone
) + (1 << order
);
2170 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
2173 __mod_zone_freepage_state(zone
, -(1UL << order
), mt
);
2176 /* Remove page from free list */
2177 list_del(&page
->lru
);
2178 zone
->free_area
[order
].nr_free
--;
2179 rmv_page_order(page
);
2181 set_page_owner(page
, order
, __GFP_MOVABLE
);
2183 /* Set the pageblock if the isolated page is at least a pageblock */
2184 if (order
>= pageblock_order
- 1) {
2185 struct page
*endpage
= page
+ (1 << order
) - 1;
2186 for (; page
< endpage
; page
+= pageblock_nr_pages
) {
2187 int mt
= get_pageblock_migratetype(page
);
2188 if (!is_migrate_isolate(mt
) && !is_migrate_cma(mt
))
2189 set_pageblock_migratetype(page
,
2195 return 1UL << order
;
2199 * Similar to split_page except the page is already free. As this is only
2200 * being used for migration, the migratetype of the block also changes.
2201 * As this is called with interrupts disabled, the caller is responsible
2202 * for calling arch_alloc_page() and kernel_map_page() after interrupts
2205 * Note: this is probably too low level an operation for use in drivers.
2206 * Please consult with lkml before using this in your driver.
2208 int split_free_page(struct page
*page
)
2213 order
= page_order(page
);
2215 nr_pages
= __isolate_free_page(page
, order
);
2219 /* Split into individual pages */
2220 set_page_refcounted(page
);
2221 split_page(page
, order
);
2226 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2229 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
2230 struct zone
*zone
, unsigned int order
,
2231 gfp_t gfp_flags
, int alloc_flags
, int migratetype
)
2233 unsigned long flags
;
2235 bool cold
= ((gfp_flags
& __GFP_COLD
) != 0);
2237 if (likely(order
== 0)) {
2238 struct per_cpu_pages
*pcp
;
2239 struct list_head
*list
;
2241 local_irq_save(flags
);
2242 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2243 list
= &pcp
->lists
[migratetype
];
2244 if (list_empty(list
)) {
2245 pcp
->count
+= rmqueue_bulk(zone
, 0,
2248 if (unlikely(list_empty(list
)))
2253 page
= list_last_entry(list
, struct page
, lru
);
2255 page
= list_first_entry(list
, struct page
, lru
);
2257 list_del(&page
->lru
);
2260 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
2262 * __GFP_NOFAIL is not to be used in new code.
2264 * All __GFP_NOFAIL callers should be fixed so that they
2265 * properly detect and handle allocation failures.
2267 * We most definitely don't want callers attempting to
2268 * allocate greater than order-1 page units with
2271 WARN_ON_ONCE(order
> 1);
2273 spin_lock_irqsave(&zone
->lock
, flags
);
2276 if (alloc_flags
& ALLOC_HARDER
) {
2277 page
= __rmqueue_smallest(zone
, order
, MIGRATE_HIGHATOMIC
);
2279 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
2282 page
= __rmqueue(zone
, order
, migratetype
);
2283 spin_unlock(&zone
->lock
);
2286 __mod_zone_freepage_state(zone
, -(1 << order
),
2287 get_pcppage_migratetype(page
));
2290 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
, -(1 << order
));
2291 if (atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]) <= 0 &&
2292 !test_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
))
2293 set_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
);
2295 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
2296 zone_statistics(preferred_zone
, zone
, gfp_flags
);
2297 local_irq_restore(flags
);
2299 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
2303 local_irq_restore(flags
);
2307 #ifdef CONFIG_FAIL_PAGE_ALLOC
2310 struct fault_attr attr
;
2312 bool ignore_gfp_highmem
;
2313 bool ignore_gfp_reclaim
;
2315 } fail_page_alloc
= {
2316 .attr
= FAULT_ATTR_INITIALIZER
,
2317 .ignore_gfp_reclaim
= true,
2318 .ignore_gfp_highmem
= true,
2322 static int __init
setup_fail_page_alloc(char *str
)
2324 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
2326 __setup("fail_page_alloc=", setup_fail_page_alloc
);
2328 static bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2330 if (order
< fail_page_alloc
.min_order
)
2332 if (gfp_mask
& __GFP_NOFAIL
)
2334 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
2336 if (fail_page_alloc
.ignore_gfp_reclaim
&&
2337 (gfp_mask
& __GFP_DIRECT_RECLAIM
))
2340 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
2343 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2345 static int __init
fail_page_alloc_debugfs(void)
2347 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
2350 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
2351 &fail_page_alloc
.attr
);
2353 return PTR_ERR(dir
);
2355 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
2356 &fail_page_alloc
.ignore_gfp_reclaim
))
2358 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
2359 &fail_page_alloc
.ignore_gfp_highmem
))
2361 if (!debugfs_create_u32("min-order", mode
, dir
,
2362 &fail_page_alloc
.min_order
))
2367 debugfs_remove_recursive(dir
);
2372 late_initcall(fail_page_alloc_debugfs
);
2374 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2376 #else /* CONFIG_FAIL_PAGE_ALLOC */
2378 static inline bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2383 #endif /* CONFIG_FAIL_PAGE_ALLOC */
2386 * Return true if free base pages are above 'mark'. For high-order checks it
2387 * will return true of the order-0 watermark is reached and there is at least
2388 * one free page of a suitable size. Checking now avoids taking the zone lock
2389 * to check in the allocation paths if no pages are free.
2391 static bool __zone_watermark_ok(struct zone
*z
, unsigned int order
,
2392 unsigned long mark
, int classzone_idx
, int alloc_flags
,
2397 const int alloc_harder
= (alloc_flags
& ALLOC_HARDER
);
2399 /* free_pages may go negative - that's OK */
2400 free_pages
-= (1 << order
) - 1;
2402 if (alloc_flags
& ALLOC_HIGH
)
2406 * If the caller does not have rights to ALLOC_HARDER then subtract
2407 * the high-atomic reserves. This will over-estimate the size of the
2408 * atomic reserve but it avoids a search.
2410 if (likely(!alloc_harder
))
2411 free_pages
-= z
->nr_reserved_highatomic
;
2416 /* If allocation can't use CMA areas don't use free CMA pages */
2417 if (!(alloc_flags
& ALLOC_CMA
))
2418 free_pages
-= zone_page_state(z
, NR_FREE_CMA_PAGES
);
2422 * Check watermarks for an order-0 allocation request. If these
2423 * are not met, then a high-order request also cannot go ahead
2424 * even if a suitable page happened to be free.
2426 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
2429 /* If this is an order-0 request then the watermark is fine */
2433 /* For a high-order request, check at least one suitable page is free */
2434 for (o
= order
; o
< MAX_ORDER
; o
++) {
2435 struct free_area
*area
= &z
->free_area
[o
];
2444 for (mt
= 0; mt
< MIGRATE_PCPTYPES
; mt
++) {
2445 if (!list_empty(&area
->free_list
[mt
]))
2450 if ((alloc_flags
& ALLOC_CMA
) &&
2451 !list_empty(&area
->free_list
[MIGRATE_CMA
])) {
2459 bool zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
2460 int classzone_idx
, int alloc_flags
)
2462 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
2463 zone_page_state(z
, NR_FREE_PAGES
));
2466 bool zone_watermark_ok_safe(struct zone
*z
, unsigned int order
,
2467 unsigned long mark
, int classzone_idx
)
2469 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
2471 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
2472 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
2474 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, 0,
2479 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
2481 return local_zone
->node
== zone
->node
;
2484 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2486 return node_distance(zone_to_nid(local_zone
), zone_to_nid(zone
)) <
2489 #else /* CONFIG_NUMA */
2490 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
2495 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2499 #endif /* CONFIG_NUMA */
2501 static void reset_alloc_batches(struct zone
*preferred_zone
)
2503 struct zone
*zone
= preferred_zone
->zone_pgdat
->node_zones
;
2506 mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
2507 high_wmark_pages(zone
) - low_wmark_pages(zone
) -
2508 atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]));
2509 clear_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
);
2510 } while (zone
++ != preferred_zone
);
2514 * get_page_from_freelist goes through the zonelist trying to allocate
2517 static struct page
*
2518 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
, int alloc_flags
,
2519 const struct alloc_context
*ac
)
2521 struct zonelist
*zonelist
= ac
->zonelist
;
2523 struct page
*page
= NULL
;
2525 int nr_fair_skipped
= 0;
2526 bool zonelist_rescan
;
2529 zonelist_rescan
= false;
2532 * Scan zonelist, looking for a zone with enough free.
2533 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2535 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
, ac
->high_zoneidx
,
2539 if (cpusets_enabled() &&
2540 (alloc_flags
& ALLOC_CPUSET
) &&
2541 !cpuset_zone_allowed(zone
, gfp_mask
))
2544 * Distribute pages in proportion to the individual
2545 * zone size to ensure fair page aging. The zone a
2546 * page was allocated in should have no effect on the
2547 * time the page has in memory before being reclaimed.
2549 if (alloc_flags
& ALLOC_FAIR
) {
2550 if (!zone_local(ac
->preferred_zone
, zone
))
2552 if (test_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
)) {
2558 * When allocating a page cache page for writing, we
2559 * want to get it from a zone that is within its dirty
2560 * limit, such that no single zone holds more than its
2561 * proportional share of globally allowed dirty pages.
2562 * The dirty limits take into account the zone's
2563 * lowmem reserves and high watermark so that kswapd
2564 * should be able to balance it without having to
2565 * write pages from its LRU list.
2567 * This may look like it could increase pressure on
2568 * lower zones by failing allocations in higher zones
2569 * before they are full. But the pages that do spill
2570 * over are limited as the lower zones are protected
2571 * by this very same mechanism. It should not become
2572 * a practical burden to them.
2574 * XXX: For now, allow allocations to potentially
2575 * exceed the per-zone dirty limit in the slowpath
2576 * (spread_dirty_pages unset) before going into reclaim,
2577 * which is important when on a NUMA setup the allowed
2578 * zones are together not big enough to reach the
2579 * global limit. The proper fix for these situations
2580 * will require awareness of zones in the
2581 * dirty-throttling and the flusher threads.
2583 if (ac
->spread_dirty_pages
&& !zone_dirty_ok(zone
))
2586 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
2587 if (!zone_watermark_ok(zone
, order
, mark
,
2588 ac
->classzone_idx
, alloc_flags
)) {
2591 /* Checked here to keep the fast path fast */
2592 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
2593 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
2596 if (zone_reclaim_mode
== 0 ||
2597 !zone_allows_reclaim(ac
->preferred_zone
, zone
))
2600 ret
= zone_reclaim(zone
, gfp_mask
, order
);
2602 case ZONE_RECLAIM_NOSCAN
:
2605 case ZONE_RECLAIM_FULL
:
2606 /* scanned but unreclaimable */
2609 /* did we reclaim enough */
2610 if (zone_watermark_ok(zone
, order
, mark
,
2611 ac
->classzone_idx
, alloc_flags
))
2619 page
= buffered_rmqueue(ac
->preferred_zone
, zone
, order
,
2620 gfp_mask
, alloc_flags
, ac
->migratetype
);
2622 if (prep_new_page(page
, order
, gfp_mask
, alloc_flags
))
2626 * If this is a high-order atomic allocation then check
2627 * if the pageblock should be reserved for the future
2629 if (unlikely(order
&& (alloc_flags
& ALLOC_HARDER
)))
2630 reserve_highatomic_pageblock(page
, zone
, order
);
2637 * The first pass makes sure allocations are spread fairly within the
2638 * local node. However, the local node might have free pages left
2639 * after the fairness batches are exhausted, and remote zones haven't
2640 * even been considered yet. Try once more without fairness, and
2641 * include remote zones now, before entering the slowpath and waking
2642 * kswapd: prefer spilling to a remote zone over swapping locally.
2644 if (alloc_flags
& ALLOC_FAIR
) {
2645 alloc_flags
&= ~ALLOC_FAIR
;
2646 if (nr_fair_skipped
) {
2647 zonelist_rescan
= true;
2648 reset_alloc_batches(ac
->preferred_zone
);
2650 if (nr_online_nodes
> 1)
2651 zonelist_rescan
= true;
2654 if (zonelist_rescan
)
2661 * Large machines with many possible nodes should not always dump per-node
2662 * meminfo in irq context.
2664 static inline bool should_suppress_show_mem(void)
2669 ret
= in_interrupt();
2674 static DEFINE_RATELIMIT_STATE(nopage_rs
,
2675 DEFAULT_RATELIMIT_INTERVAL
,
2676 DEFAULT_RATELIMIT_BURST
);
2678 void warn_alloc_failed(gfp_t gfp_mask
, unsigned int order
, const char *fmt
, ...)
2680 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
2682 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
) ||
2683 debug_guardpage_minorder() > 0)
2687 * This documents exceptions given to allocations in certain
2688 * contexts that are allowed to allocate outside current's set
2691 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2692 if (test_thread_flag(TIF_MEMDIE
) ||
2693 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
2694 filter
&= ~SHOW_MEM_FILTER_NODES
;
2695 if (in_interrupt() || !(gfp_mask
& __GFP_DIRECT_RECLAIM
))
2696 filter
&= ~SHOW_MEM_FILTER_NODES
;
2699 struct va_format vaf
;
2702 va_start(args
, fmt
);
2707 pr_warn("%pV", &vaf
);
2712 pr_warn("%s: page allocation failure: order:%u, mode:%#x(%pGg)\n",
2713 current
->comm
, order
, gfp_mask
, &gfp_mask
);
2715 if (!should_suppress_show_mem())
2719 static inline struct page
*
2720 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
2721 const struct alloc_context
*ac
, unsigned long *did_some_progress
)
2723 struct oom_control oc
= {
2724 .zonelist
= ac
->zonelist
,
2725 .nodemask
= ac
->nodemask
,
2726 .gfp_mask
= gfp_mask
,
2731 *did_some_progress
= 0;
2734 * Acquire the oom lock. If that fails, somebody else is
2735 * making progress for us.
2737 if (!mutex_trylock(&oom_lock
)) {
2738 *did_some_progress
= 1;
2739 schedule_timeout_uninterruptible(1);
2744 * Go through the zonelist yet one more time, keep very high watermark
2745 * here, this is only to catch a parallel oom killing, we must fail if
2746 * we're still under heavy pressure.
2748 page
= get_page_from_freelist(gfp_mask
| __GFP_HARDWALL
, order
,
2749 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
, ac
);
2753 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2754 /* Coredumps can quickly deplete all memory reserves */
2755 if (current
->flags
& PF_DUMPCORE
)
2757 /* The OOM killer will not help higher order allocs */
2758 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2760 /* The OOM killer does not needlessly kill tasks for lowmem */
2761 if (ac
->high_zoneidx
< ZONE_NORMAL
)
2763 /* The OOM killer does not compensate for IO-less reclaim */
2764 if (!(gfp_mask
& __GFP_FS
)) {
2766 * XXX: Page reclaim didn't yield anything,
2767 * and the OOM killer can't be invoked, but
2768 * keep looping as per tradition.
2770 *did_some_progress
= 1;
2773 if (pm_suspended_storage())
2775 /* The OOM killer may not free memory on a specific node */
2776 if (gfp_mask
& __GFP_THISNODE
)
2779 /* Exhausted what can be done so it's blamo time */
2780 if (out_of_memory(&oc
) || WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
)) {
2781 *did_some_progress
= 1;
2783 if (gfp_mask
& __GFP_NOFAIL
) {
2784 page
= get_page_from_freelist(gfp_mask
, order
,
2785 ALLOC_NO_WATERMARKS
|ALLOC_CPUSET
, ac
);
2787 * fallback to ignore cpuset restriction if our nodes
2791 page
= get_page_from_freelist(gfp_mask
, order
,
2792 ALLOC_NO_WATERMARKS
, ac
);
2796 mutex_unlock(&oom_lock
);
2800 #ifdef CONFIG_COMPACTION
2801 /* Try memory compaction for high-order allocations before reclaim */
2802 static struct page
*
2803 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2804 int alloc_flags
, const struct alloc_context
*ac
,
2805 enum migrate_mode mode
, int *contended_compaction
,
2806 bool *deferred_compaction
)
2808 unsigned long compact_result
;
2814 current
->flags
|= PF_MEMALLOC
;
2815 compact_result
= try_to_compact_pages(gfp_mask
, order
, alloc_flags
, ac
,
2816 mode
, contended_compaction
);
2817 current
->flags
&= ~PF_MEMALLOC
;
2819 switch (compact_result
) {
2820 case COMPACT_DEFERRED
:
2821 *deferred_compaction
= true;
2823 case COMPACT_SKIPPED
:
2830 * At least in one zone compaction wasn't deferred or skipped, so let's
2831 * count a compaction stall
2833 count_vm_event(COMPACTSTALL
);
2835 page
= get_page_from_freelist(gfp_mask
, order
,
2836 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
2839 struct zone
*zone
= page_zone(page
);
2841 zone
->compact_blockskip_flush
= false;
2842 compaction_defer_reset(zone
, order
, true);
2843 count_vm_event(COMPACTSUCCESS
);
2848 * It's bad if compaction run occurs and fails. The most likely reason
2849 * is that pages exist, but not enough to satisfy watermarks.
2851 count_vm_event(COMPACTFAIL
);
2858 static inline struct page
*
2859 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2860 int alloc_flags
, const struct alloc_context
*ac
,
2861 enum migrate_mode mode
, int *contended_compaction
,
2862 bool *deferred_compaction
)
2866 #endif /* CONFIG_COMPACTION */
2868 /* Perform direct synchronous page reclaim */
2870 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
,
2871 const struct alloc_context
*ac
)
2873 struct reclaim_state reclaim_state
;
2878 /* We now go into synchronous reclaim */
2879 cpuset_memory_pressure_bump();
2880 current
->flags
|= PF_MEMALLOC
;
2881 lockdep_set_current_reclaim_state(gfp_mask
);
2882 reclaim_state
.reclaimed_slab
= 0;
2883 current
->reclaim_state
= &reclaim_state
;
2885 progress
= try_to_free_pages(ac
->zonelist
, order
, gfp_mask
,
2888 current
->reclaim_state
= NULL
;
2889 lockdep_clear_current_reclaim_state();
2890 current
->flags
&= ~PF_MEMALLOC
;
2897 /* The really slow allocator path where we enter direct reclaim */
2898 static inline struct page
*
2899 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
2900 int alloc_flags
, const struct alloc_context
*ac
,
2901 unsigned long *did_some_progress
)
2903 struct page
*page
= NULL
;
2904 bool drained
= false;
2906 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, ac
);
2907 if (unlikely(!(*did_some_progress
)))
2911 page
= get_page_from_freelist(gfp_mask
, order
,
2912 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
2915 * If an allocation failed after direct reclaim, it could be because
2916 * pages are pinned on the per-cpu lists or in high alloc reserves.
2917 * Shrink them them and try again
2919 if (!page
&& !drained
) {
2920 unreserve_highatomic_pageblock(ac
);
2921 drain_all_pages(NULL
);
2929 static void wake_all_kswapds(unsigned int order
, const struct alloc_context
*ac
)
2934 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
,
2935 ac
->high_zoneidx
, ac
->nodemask
)
2936 wakeup_kswapd(zone
, order
, zone_idx(ac
->preferred_zone
));
2940 gfp_to_alloc_flags(gfp_t gfp_mask
)
2942 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
2944 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2945 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
2948 * The caller may dip into page reserves a bit more if the caller
2949 * cannot run direct reclaim, or if the caller has realtime scheduling
2950 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2951 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
2953 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
2955 if (gfp_mask
& __GFP_ATOMIC
) {
2957 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2958 * if it can't schedule.
2960 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2961 alloc_flags
|= ALLOC_HARDER
;
2963 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
2964 * comment for __cpuset_node_allowed().
2966 alloc_flags
&= ~ALLOC_CPUSET
;
2967 } else if (unlikely(rt_task(current
)) && !in_interrupt())
2968 alloc_flags
|= ALLOC_HARDER
;
2970 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
2971 if (gfp_mask
& __GFP_MEMALLOC
)
2972 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2973 else if (in_serving_softirq() && (current
->flags
& PF_MEMALLOC
))
2974 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2975 else if (!in_interrupt() &&
2976 ((current
->flags
& PF_MEMALLOC
) ||
2977 unlikely(test_thread_flag(TIF_MEMDIE
))))
2978 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2981 if (gfpflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
2982 alloc_flags
|= ALLOC_CMA
;
2987 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask
)
2989 return !!(gfp_to_alloc_flags(gfp_mask
) & ALLOC_NO_WATERMARKS
);
2992 static inline bool is_thp_gfp_mask(gfp_t gfp_mask
)
2994 return (gfp_mask
& (GFP_TRANSHUGE
| __GFP_KSWAPD_RECLAIM
)) == GFP_TRANSHUGE
;
2997 static inline struct page
*
2998 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
2999 struct alloc_context
*ac
)
3001 bool can_direct_reclaim
= gfp_mask
& __GFP_DIRECT_RECLAIM
;
3002 struct page
*page
= NULL
;
3004 unsigned long pages_reclaimed
= 0;
3005 unsigned long did_some_progress
;
3006 enum migrate_mode migration_mode
= MIGRATE_ASYNC
;
3007 bool deferred_compaction
= false;
3008 int contended_compaction
= COMPACT_CONTENDED_NONE
;
3011 * In the slowpath, we sanity check order to avoid ever trying to
3012 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3013 * be using allocators in order of preference for an area that is
3016 if (order
>= MAX_ORDER
) {
3017 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
3022 * We also sanity check to catch abuse of atomic reserves being used by
3023 * callers that are not in atomic context.
3025 if (WARN_ON_ONCE((gfp_mask
& (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)) ==
3026 (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)))
3027 gfp_mask
&= ~__GFP_ATOMIC
;
3030 * If this allocation cannot block and it is for a specific node, then
3031 * fail early. There's no need to wakeup kswapd or retry for a
3032 * speculative node-specific allocation.
3034 if (IS_ENABLED(CONFIG_NUMA
) && (gfp_mask
& __GFP_THISNODE
) && !can_direct_reclaim
)
3038 if (gfp_mask
& __GFP_KSWAPD_RECLAIM
)
3039 wake_all_kswapds(order
, ac
);
3042 * OK, we're below the kswapd watermark and have kicked background
3043 * reclaim. Now things get more complex, so set up alloc_flags according
3044 * to how we want to proceed.
3046 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
3049 * Find the true preferred zone if the allocation is unconstrained by
3052 if (!(alloc_flags
& ALLOC_CPUSET
) && !ac
->nodemask
) {
3053 struct zoneref
*preferred_zoneref
;
3054 preferred_zoneref
= first_zones_zonelist(ac
->zonelist
,
3055 ac
->high_zoneidx
, NULL
, &ac
->preferred_zone
);
3056 ac
->classzone_idx
= zonelist_zone_idx(preferred_zoneref
);
3059 /* This is the last chance, in general, before the goto nopage. */
3060 page
= get_page_from_freelist(gfp_mask
, order
,
3061 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
3065 /* Allocate without watermarks if the context allows */
3066 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
3068 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3069 * the allocation is high priority and these type of
3070 * allocations are system rather than user orientated
3072 ac
->zonelist
= node_zonelist(numa_node_id(), gfp_mask
);
3073 page
= get_page_from_freelist(gfp_mask
, order
,
3074 ALLOC_NO_WATERMARKS
, ac
);
3079 /* Caller is not willing to reclaim, we can't balance anything */
3080 if (!can_direct_reclaim
) {
3082 * All existing users of the __GFP_NOFAIL are blockable, so warn
3083 * of any new users that actually allow this type of allocation
3086 WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
);
3090 /* Avoid recursion of direct reclaim */
3091 if (current
->flags
& PF_MEMALLOC
) {
3093 * __GFP_NOFAIL request from this context is rather bizarre
3094 * because we cannot reclaim anything and only can loop waiting
3095 * for somebody to do a work for us.
3097 if (WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
)) {
3104 /* Avoid allocations with no watermarks from looping endlessly */
3105 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
3109 * Try direct compaction. The first pass is asynchronous. Subsequent
3110 * attempts after direct reclaim are synchronous
3112 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
, ac
,
3114 &contended_compaction
,
3115 &deferred_compaction
);
3119 /* Checks for THP-specific high-order allocations */
3120 if (is_thp_gfp_mask(gfp_mask
)) {
3122 * If compaction is deferred for high-order allocations, it is
3123 * because sync compaction recently failed. If this is the case
3124 * and the caller requested a THP allocation, we do not want
3125 * to heavily disrupt the system, so we fail the allocation
3126 * instead of entering direct reclaim.
3128 if (deferred_compaction
)
3132 * In all zones where compaction was attempted (and not
3133 * deferred or skipped), lock contention has been detected.
3134 * For THP allocation we do not want to disrupt the others
3135 * so we fallback to base pages instead.
3137 if (contended_compaction
== COMPACT_CONTENDED_LOCK
)
3141 * If compaction was aborted due to need_resched(), we do not
3142 * want to further increase allocation latency, unless it is
3143 * khugepaged trying to collapse.
3145 if (contended_compaction
== COMPACT_CONTENDED_SCHED
3146 && !(current
->flags
& PF_KTHREAD
))
3151 * It can become very expensive to allocate transparent hugepages at
3152 * fault, so use asynchronous memory compaction for THP unless it is
3153 * khugepaged trying to collapse.
3155 if (!is_thp_gfp_mask(gfp_mask
) || (current
->flags
& PF_KTHREAD
))
3156 migration_mode
= MIGRATE_SYNC_LIGHT
;
3158 /* Try direct reclaim and then allocating */
3159 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
, alloc_flags
, ac
,
3160 &did_some_progress
);
3164 /* Do not loop if specifically requested */
3165 if (gfp_mask
& __GFP_NORETRY
)
3168 /* Keep reclaiming pages as long as there is reasonable progress */
3169 pages_reclaimed
+= did_some_progress
;
3170 if ((did_some_progress
&& order
<= PAGE_ALLOC_COSTLY_ORDER
) ||
3171 ((gfp_mask
& __GFP_REPEAT
) && pages_reclaimed
< (1 << order
))) {
3172 /* Wait for some write requests to complete then retry */
3173 wait_iff_congested(ac
->preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
3177 /* Reclaim has failed us, start killing things */
3178 page
= __alloc_pages_may_oom(gfp_mask
, order
, ac
, &did_some_progress
);
3182 /* Retry as long as the OOM killer is making progress */
3183 if (did_some_progress
)
3188 * High-order allocations do not necessarily loop after
3189 * direct reclaim and reclaim/compaction depends on compaction
3190 * being called after reclaim so call directly if necessary
3192 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
,
3194 &contended_compaction
,
3195 &deferred_compaction
);
3199 warn_alloc_failed(gfp_mask
, order
, NULL
);
3205 * This is the 'heart' of the zoned buddy allocator.
3208 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
3209 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
3211 struct zoneref
*preferred_zoneref
;
3212 struct page
*page
= NULL
;
3213 unsigned int cpuset_mems_cookie
;
3214 int alloc_flags
= ALLOC_WMARK_LOW
|ALLOC_CPUSET
|ALLOC_FAIR
;
3215 gfp_t alloc_mask
; /* The gfp_t that was actually used for allocation */
3216 struct alloc_context ac
= {
3217 .high_zoneidx
= gfp_zone(gfp_mask
),
3218 .nodemask
= nodemask
,
3219 .migratetype
= gfpflags_to_migratetype(gfp_mask
),
3222 gfp_mask
&= gfp_allowed_mask
;
3224 lockdep_trace_alloc(gfp_mask
);
3226 might_sleep_if(gfp_mask
& __GFP_DIRECT_RECLAIM
);
3228 if (should_fail_alloc_page(gfp_mask
, order
))
3232 * Check the zones suitable for the gfp_mask contain at least one
3233 * valid zone. It's possible to have an empty zonelist as a result
3234 * of __GFP_THISNODE and a memoryless node
3236 if (unlikely(!zonelist
->_zonerefs
->zone
))
3239 if (IS_ENABLED(CONFIG_CMA
) && ac
.migratetype
== MIGRATE_MOVABLE
)
3240 alloc_flags
|= ALLOC_CMA
;
3243 cpuset_mems_cookie
= read_mems_allowed_begin();
3245 /* We set it here, as __alloc_pages_slowpath might have changed it */
3246 ac
.zonelist
= zonelist
;
3248 /* Dirty zone balancing only done in the fast path */
3249 ac
.spread_dirty_pages
= (gfp_mask
& __GFP_WRITE
);
3251 /* The preferred zone is used for statistics later */
3252 preferred_zoneref
= first_zones_zonelist(ac
.zonelist
, ac
.high_zoneidx
,
3253 ac
.nodemask
? : &cpuset_current_mems_allowed
,
3254 &ac
.preferred_zone
);
3255 if (!ac
.preferred_zone
)
3257 ac
.classzone_idx
= zonelist_zone_idx(preferred_zoneref
);
3259 /* First allocation attempt */
3260 alloc_mask
= gfp_mask
|__GFP_HARDWALL
;
3261 page
= get_page_from_freelist(alloc_mask
, order
, alloc_flags
, &ac
);
3262 if (unlikely(!page
)) {
3264 * Runtime PM, block IO and its error handling path
3265 * can deadlock because I/O on the device might not
3268 alloc_mask
= memalloc_noio_flags(gfp_mask
);
3269 ac
.spread_dirty_pages
= false;
3271 page
= __alloc_pages_slowpath(alloc_mask
, order
, &ac
);
3274 if (kmemcheck_enabled
&& page
)
3275 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
3277 trace_mm_page_alloc(page
, order
, alloc_mask
, ac
.migratetype
);
3281 * When updating a task's mems_allowed, it is possible to race with
3282 * parallel threads in such a way that an allocation can fail while
3283 * the mask is being updated. If a page allocation is about to fail,
3284 * check if the cpuset changed during allocation and if so, retry.
3286 if (unlikely(!page
&& read_mems_allowed_retry(cpuset_mems_cookie
)))
3291 EXPORT_SYMBOL(__alloc_pages_nodemask
);
3294 * Common helper functions.
3296 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
3301 * __get_free_pages() returns a 32-bit address, which cannot represent
3304 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
3306 page
= alloc_pages(gfp_mask
, order
);
3309 return (unsigned long) page_address(page
);
3311 EXPORT_SYMBOL(__get_free_pages
);
3313 unsigned long get_zeroed_page(gfp_t gfp_mask
)
3315 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
3317 EXPORT_SYMBOL(get_zeroed_page
);
3319 void __free_pages(struct page
*page
, unsigned int order
)
3321 if (put_page_testzero(page
)) {
3323 free_hot_cold_page(page
, false);
3325 __free_pages_ok(page
, order
);
3329 EXPORT_SYMBOL(__free_pages
);
3331 void free_pages(unsigned long addr
, unsigned int order
)
3334 VM_BUG_ON(!virt_addr_valid((void *)addr
));
3335 __free_pages(virt_to_page((void *)addr
), order
);
3339 EXPORT_SYMBOL(free_pages
);
3343 * An arbitrary-length arbitrary-offset area of memory which resides
3344 * within a 0 or higher order page. Multiple fragments within that page
3345 * are individually refcounted, in the page's reference counter.
3347 * The page_frag functions below provide a simple allocation framework for
3348 * page fragments. This is used by the network stack and network device
3349 * drivers to provide a backing region of memory for use as either an
3350 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3352 static struct page
*__page_frag_refill(struct page_frag_cache
*nc
,
3355 struct page
*page
= NULL
;
3356 gfp_t gfp
= gfp_mask
;
3358 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3359 gfp_mask
|= __GFP_COMP
| __GFP_NOWARN
| __GFP_NORETRY
|
3361 page
= alloc_pages_node(NUMA_NO_NODE
, gfp_mask
,
3362 PAGE_FRAG_CACHE_MAX_ORDER
);
3363 nc
->size
= page
? PAGE_FRAG_CACHE_MAX_SIZE
: PAGE_SIZE
;
3365 if (unlikely(!page
))
3366 page
= alloc_pages_node(NUMA_NO_NODE
, gfp
, 0);
3368 nc
->va
= page
? page_address(page
) : NULL
;
3373 void *__alloc_page_frag(struct page_frag_cache
*nc
,
3374 unsigned int fragsz
, gfp_t gfp_mask
)
3376 unsigned int size
= PAGE_SIZE
;
3380 if (unlikely(!nc
->va
)) {
3382 page
= __page_frag_refill(nc
, gfp_mask
);
3386 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3387 /* if size can vary use size else just use PAGE_SIZE */
3390 /* Even if we own the page, we do not use atomic_set().
3391 * This would break get_page_unless_zero() users.
3393 atomic_add(size
- 1, &page
->_count
);
3395 /* reset page count bias and offset to start of new frag */
3396 nc
->pfmemalloc
= page_is_pfmemalloc(page
);
3397 nc
->pagecnt_bias
= size
;
3401 offset
= nc
->offset
- fragsz
;
3402 if (unlikely(offset
< 0)) {
3403 page
= virt_to_page(nc
->va
);
3405 if (!atomic_sub_and_test(nc
->pagecnt_bias
, &page
->_count
))
3408 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3409 /* if size can vary use size else just use PAGE_SIZE */
3412 /* OK, page count is 0, we can safely set it */
3413 atomic_set(&page
->_count
, size
);
3415 /* reset page count bias and offset to start of new frag */
3416 nc
->pagecnt_bias
= size
;
3417 offset
= size
- fragsz
;
3421 nc
->offset
= offset
;
3423 return nc
->va
+ offset
;
3425 EXPORT_SYMBOL(__alloc_page_frag
);
3428 * Frees a page fragment allocated out of either a compound or order 0 page.
3430 void __free_page_frag(void *addr
)
3432 struct page
*page
= virt_to_head_page(addr
);
3434 if (unlikely(put_page_testzero(page
)))
3435 __free_pages_ok(page
, compound_order(page
));
3437 EXPORT_SYMBOL(__free_page_frag
);
3440 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
3441 * of the current memory cgroup if __GFP_ACCOUNT is set, other than that it is
3442 * equivalent to alloc_pages.
3444 * It should be used when the caller would like to use kmalloc, but since the
3445 * allocation is large, it has to fall back to the page allocator.
3447 struct page
*alloc_kmem_pages(gfp_t gfp_mask
, unsigned int order
)
3451 page
= alloc_pages(gfp_mask
, order
);
3452 if (page
&& memcg_kmem_charge(page
, gfp_mask
, order
) != 0) {
3453 __free_pages(page
, order
);
3459 struct page
*alloc_kmem_pages_node(int nid
, gfp_t gfp_mask
, unsigned int order
)
3463 page
= alloc_pages_node(nid
, gfp_mask
, order
);
3464 if (page
&& memcg_kmem_charge(page
, gfp_mask
, order
) != 0) {
3465 __free_pages(page
, order
);
3472 * __free_kmem_pages and free_kmem_pages will free pages allocated with
3475 void __free_kmem_pages(struct page
*page
, unsigned int order
)
3477 memcg_kmem_uncharge(page
, order
);
3478 __free_pages(page
, order
);
3481 void free_kmem_pages(unsigned long addr
, unsigned int order
)
3484 VM_BUG_ON(!virt_addr_valid((void *)addr
));
3485 __free_kmem_pages(virt_to_page((void *)addr
), order
);
3489 static void *make_alloc_exact(unsigned long addr
, unsigned int order
,
3493 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
3494 unsigned long used
= addr
+ PAGE_ALIGN(size
);
3496 split_page(virt_to_page((void *)addr
), order
);
3497 while (used
< alloc_end
) {
3502 return (void *)addr
;
3506 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3507 * @size: the number of bytes to allocate
3508 * @gfp_mask: GFP flags for the allocation
3510 * This function is similar to alloc_pages(), except that it allocates the
3511 * minimum number of pages to satisfy the request. alloc_pages() can only
3512 * allocate memory in power-of-two pages.
3514 * This function is also limited by MAX_ORDER.
3516 * Memory allocated by this function must be released by free_pages_exact().
3518 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
3520 unsigned int order
= get_order(size
);
3523 addr
= __get_free_pages(gfp_mask
, order
);
3524 return make_alloc_exact(addr
, order
, size
);
3526 EXPORT_SYMBOL(alloc_pages_exact
);
3529 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3531 * @nid: the preferred node ID where memory should be allocated
3532 * @size: the number of bytes to allocate
3533 * @gfp_mask: GFP flags for the allocation
3535 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3538 void * __meminit
alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
3540 unsigned int order
= get_order(size
);
3541 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
3544 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
3548 * free_pages_exact - release memory allocated via alloc_pages_exact()
3549 * @virt: the value returned by alloc_pages_exact.
3550 * @size: size of allocation, same value as passed to alloc_pages_exact().
3552 * Release the memory allocated by a previous call to alloc_pages_exact.
3554 void free_pages_exact(void *virt
, size_t size
)
3556 unsigned long addr
= (unsigned long)virt
;
3557 unsigned long end
= addr
+ PAGE_ALIGN(size
);
3559 while (addr
< end
) {
3564 EXPORT_SYMBOL(free_pages_exact
);
3567 * nr_free_zone_pages - count number of pages beyond high watermark
3568 * @offset: The zone index of the highest zone
3570 * nr_free_zone_pages() counts the number of counts pages which are beyond the
3571 * high watermark within all zones at or below a given zone index. For each
3572 * zone, the number of pages is calculated as:
3573 * managed_pages - high_pages
3575 static unsigned long nr_free_zone_pages(int offset
)
3580 /* Just pick one node, since fallback list is circular */
3581 unsigned long sum
= 0;
3583 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
3585 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
3586 unsigned long size
= zone
->managed_pages
;
3587 unsigned long high
= high_wmark_pages(zone
);
3596 * nr_free_buffer_pages - count number of pages beyond high watermark
3598 * nr_free_buffer_pages() counts the number of pages which are beyond the high
3599 * watermark within ZONE_DMA and ZONE_NORMAL.
3601 unsigned long nr_free_buffer_pages(void)
3603 return nr_free_zone_pages(gfp_zone(GFP_USER
));
3605 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
3608 * nr_free_pagecache_pages - count number of pages beyond high watermark
3610 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3611 * high watermark within all zones.
3613 unsigned long nr_free_pagecache_pages(void)
3615 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
3618 static inline void show_node(struct zone
*zone
)
3620 if (IS_ENABLED(CONFIG_NUMA
))
3621 printk("Node %d ", zone_to_nid(zone
));
3624 void si_meminfo(struct sysinfo
*val
)
3626 val
->totalram
= totalram_pages
;
3627 val
->sharedram
= global_page_state(NR_SHMEM
);
3628 val
->freeram
= global_page_state(NR_FREE_PAGES
);
3629 val
->bufferram
= nr_blockdev_pages();
3630 val
->totalhigh
= totalhigh_pages
;
3631 val
->freehigh
= nr_free_highpages();
3632 val
->mem_unit
= PAGE_SIZE
;
3635 EXPORT_SYMBOL(si_meminfo
);
3638 void si_meminfo_node(struct sysinfo
*val
, int nid
)
3640 int zone_type
; /* needs to be signed */
3641 unsigned long managed_pages
= 0;
3642 pg_data_t
*pgdat
= NODE_DATA(nid
);
3644 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++)
3645 managed_pages
+= pgdat
->node_zones
[zone_type
].managed_pages
;
3646 val
->totalram
= managed_pages
;
3647 val
->sharedram
= node_page_state(nid
, NR_SHMEM
);
3648 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
3649 #ifdef CONFIG_HIGHMEM
3650 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].managed_pages
;
3651 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
3657 val
->mem_unit
= PAGE_SIZE
;
3662 * Determine whether the node should be displayed or not, depending on whether
3663 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
3665 bool skip_free_areas_node(unsigned int flags
, int nid
)
3668 unsigned int cpuset_mems_cookie
;
3670 if (!(flags
& SHOW_MEM_FILTER_NODES
))
3674 cpuset_mems_cookie
= read_mems_allowed_begin();
3675 ret
= !node_isset(nid
, cpuset_current_mems_allowed
);
3676 } while (read_mems_allowed_retry(cpuset_mems_cookie
));
3681 #define K(x) ((x) << (PAGE_SHIFT-10))
3683 static void show_migration_types(unsigned char type
)
3685 static const char types
[MIGRATE_TYPES
] = {
3686 [MIGRATE_UNMOVABLE
] = 'U',
3687 [MIGRATE_MOVABLE
] = 'M',
3688 [MIGRATE_RECLAIMABLE
] = 'E',
3689 [MIGRATE_HIGHATOMIC
] = 'H',
3691 [MIGRATE_CMA
] = 'C',
3693 #ifdef CONFIG_MEMORY_ISOLATION
3694 [MIGRATE_ISOLATE
] = 'I',
3697 char tmp
[MIGRATE_TYPES
+ 1];
3701 for (i
= 0; i
< MIGRATE_TYPES
; i
++) {
3702 if (type
& (1 << i
))
3707 printk("(%s) ", tmp
);
3711 * Show free area list (used inside shift_scroll-lock stuff)
3712 * We also calculate the percentage fragmentation. We do this by counting the
3713 * memory on each free list with the exception of the first item on the list.
3716 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
3719 void show_free_areas(unsigned int filter
)
3721 unsigned long free_pcp
= 0;
3725 for_each_populated_zone(zone
) {
3726 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3729 for_each_online_cpu(cpu
)
3730 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
3733 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3734 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3735 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
3736 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3737 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3738 " free:%lu free_pcp:%lu free_cma:%lu\n",
3739 global_page_state(NR_ACTIVE_ANON
),
3740 global_page_state(NR_INACTIVE_ANON
),
3741 global_page_state(NR_ISOLATED_ANON
),
3742 global_page_state(NR_ACTIVE_FILE
),
3743 global_page_state(NR_INACTIVE_FILE
),
3744 global_page_state(NR_ISOLATED_FILE
),
3745 global_page_state(NR_UNEVICTABLE
),
3746 global_page_state(NR_FILE_DIRTY
),
3747 global_page_state(NR_WRITEBACK
),
3748 global_page_state(NR_UNSTABLE_NFS
),
3749 global_page_state(NR_SLAB_RECLAIMABLE
),
3750 global_page_state(NR_SLAB_UNRECLAIMABLE
),
3751 global_page_state(NR_FILE_MAPPED
),
3752 global_page_state(NR_SHMEM
),
3753 global_page_state(NR_PAGETABLE
),
3754 global_page_state(NR_BOUNCE
),
3755 global_page_state(NR_FREE_PAGES
),
3757 global_page_state(NR_FREE_CMA_PAGES
));
3759 for_each_populated_zone(zone
) {
3762 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3766 for_each_online_cpu(cpu
)
3767 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
3775 " active_anon:%lukB"
3776 " inactive_anon:%lukB"
3777 " active_file:%lukB"
3778 " inactive_file:%lukB"
3779 " unevictable:%lukB"
3780 " isolated(anon):%lukB"
3781 " isolated(file):%lukB"
3789 " slab_reclaimable:%lukB"
3790 " slab_unreclaimable:%lukB"
3791 " kernel_stack:%lukB"
3798 " writeback_tmp:%lukB"
3799 " pages_scanned:%lu"
3800 " all_unreclaimable? %s"
3803 K(zone_page_state(zone
, NR_FREE_PAGES
)),
3804 K(min_wmark_pages(zone
)),
3805 K(low_wmark_pages(zone
)),
3806 K(high_wmark_pages(zone
)),
3807 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
3808 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
3809 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
3810 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
3811 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
3812 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
3813 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
3814 K(zone
->present_pages
),
3815 K(zone
->managed_pages
),
3816 K(zone_page_state(zone
, NR_MLOCK
)),
3817 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
3818 K(zone_page_state(zone
, NR_WRITEBACK
)),
3819 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
3820 K(zone_page_state(zone
, NR_SHMEM
)),
3821 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
3822 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
3823 zone_page_state(zone
, NR_KERNEL_STACK
) *
3825 K(zone_page_state(zone
, NR_PAGETABLE
)),
3826 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
3827 K(zone_page_state(zone
, NR_BOUNCE
)),
3829 K(this_cpu_read(zone
->pageset
->pcp
.count
)),
3830 K(zone_page_state(zone
, NR_FREE_CMA_PAGES
)),
3831 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
3832 K(zone_page_state(zone
, NR_PAGES_SCANNED
)),
3833 (!zone_reclaimable(zone
) ? "yes" : "no")
3835 printk("lowmem_reserve[]:");
3836 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3837 printk(" %ld", zone
->lowmem_reserve
[i
]);
3841 for_each_populated_zone(zone
) {
3843 unsigned long nr
[MAX_ORDER
], flags
, total
= 0;
3844 unsigned char types
[MAX_ORDER
];
3846 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3849 printk("%s: ", zone
->name
);
3851 spin_lock_irqsave(&zone
->lock
, flags
);
3852 for (order
= 0; order
< MAX_ORDER
; order
++) {
3853 struct free_area
*area
= &zone
->free_area
[order
];
3856 nr
[order
] = area
->nr_free
;
3857 total
+= nr
[order
] << order
;
3860 for (type
= 0; type
< MIGRATE_TYPES
; type
++) {
3861 if (!list_empty(&area
->free_list
[type
]))
3862 types
[order
] |= 1 << type
;
3865 spin_unlock_irqrestore(&zone
->lock
, flags
);
3866 for (order
= 0; order
< MAX_ORDER
; order
++) {
3867 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
3869 show_migration_types(types
[order
]);
3871 printk("= %lukB\n", K(total
));
3874 hugetlb_show_meminfo();
3876 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
3878 show_swap_cache_info();
3881 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
3883 zoneref
->zone
= zone
;
3884 zoneref
->zone_idx
= zone_idx(zone
);
3888 * Builds allocation fallback zone lists.
3890 * Add all populated zones of a node to the zonelist.
3892 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
3896 enum zone_type zone_type
= MAX_NR_ZONES
;
3900 zone
= pgdat
->node_zones
+ zone_type
;
3901 if (populated_zone(zone
)) {
3902 zoneref_set_zone(zone
,
3903 &zonelist
->_zonerefs
[nr_zones
++]);
3904 check_highest_zone(zone_type
);
3906 } while (zone_type
);
3914 * 0 = automatic detection of better ordering.
3915 * 1 = order by ([node] distance, -zonetype)
3916 * 2 = order by (-zonetype, [node] distance)
3918 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3919 * the same zonelist. So only NUMA can configure this param.
3921 #define ZONELIST_ORDER_DEFAULT 0
3922 #define ZONELIST_ORDER_NODE 1
3923 #define ZONELIST_ORDER_ZONE 2
3925 /* zonelist order in the kernel.
3926 * set_zonelist_order() will set this to NODE or ZONE.
3928 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3929 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
3933 /* The value user specified ....changed by config */
3934 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3935 /* string for sysctl */
3936 #define NUMA_ZONELIST_ORDER_LEN 16
3937 char numa_zonelist_order
[16] = "default";
3940 * interface for configure zonelist ordering.
3941 * command line option "numa_zonelist_order"
3942 * = "[dD]efault - default, automatic configuration.
3943 * = "[nN]ode - order by node locality, then by zone within node
3944 * = "[zZ]one - order by zone, then by locality within zone
3947 static int __parse_numa_zonelist_order(char *s
)
3949 if (*s
== 'd' || *s
== 'D') {
3950 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3951 } else if (*s
== 'n' || *s
== 'N') {
3952 user_zonelist_order
= ZONELIST_ORDER_NODE
;
3953 } else if (*s
== 'z' || *s
== 'Z') {
3954 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
3957 "Ignoring invalid numa_zonelist_order value: "
3964 static __init
int setup_numa_zonelist_order(char *s
)
3971 ret
= __parse_numa_zonelist_order(s
);
3973 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
3977 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
3980 * sysctl handler for numa_zonelist_order
3982 int numa_zonelist_order_handler(struct ctl_table
*table
, int write
,
3983 void __user
*buffer
, size_t *length
,
3986 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
3988 static DEFINE_MUTEX(zl_order_mutex
);
3990 mutex_lock(&zl_order_mutex
);
3992 if (strlen((char *)table
->data
) >= NUMA_ZONELIST_ORDER_LEN
) {
3996 strcpy(saved_string
, (char *)table
->data
);
3998 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
4002 int oldval
= user_zonelist_order
;
4004 ret
= __parse_numa_zonelist_order((char *)table
->data
);
4007 * bogus value. restore saved string
4009 strncpy((char *)table
->data
, saved_string
,
4010 NUMA_ZONELIST_ORDER_LEN
);
4011 user_zonelist_order
= oldval
;
4012 } else if (oldval
!= user_zonelist_order
) {
4013 mutex_lock(&zonelists_mutex
);
4014 build_all_zonelists(NULL
, NULL
);
4015 mutex_unlock(&zonelists_mutex
);
4019 mutex_unlock(&zl_order_mutex
);
4024 #define MAX_NODE_LOAD (nr_online_nodes)
4025 static int node_load
[MAX_NUMNODES
];
4028 * find_next_best_node - find the next node that should appear in a given node's fallback list
4029 * @node: node whose fallback list we're appending
4030 * @used_node_mask: nodemask_t of already used nodes
4032 * We use a number of factors to determine which is the next node that should
4033 * appear on a given node's fallback list. The node should not have appeared
4034 * already in @node's fallback list, and it should be the next closest node
4035 * according to the distance array (which contains arbitrary distance values
4036 * from each node to each node in the system), and should also prefer nodes
4037 * with no CPUs, since presumably they'll have very little allocation pressure
4038 * on them otherwise.
4039 * It returns -1 if no node is found.
4041 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
4044 int min_val
= INT_MAX
;
4045 int best_node
= NUMA_NO_NODE
;
4046 const struct cpumask
*tmp
= cpumask_of_node(0);
4048 /* Use the local node if we haven't already */
4049 if (!node_isset(node
, *used_node_mask
)) {
4050 node_set(node
, *used_node_mask
);
4054 for_each_node_state(n
, N_MEMORY
) {
4056 /* Don't want a node to appear more than once */
4057 if (node_isset(n
, *used_node_mask
))
4060 /* Use the distance array to find the distance */
4061 val
= node_distance(node
, n
);
4063 /* Penalize nodes under us ("prefer the next node") */
4066 /* Give preference to headless and unused nodes */
4067 tmp
= cpumask_of_node(n
);
4068 if (!cpumask_empty(tmp
))
4069 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
4071 /* Slight preference for less loaded node */
4072 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
4073 val
+= node_load
[n
];
4075 if (val
< min_val
) {
4082 node_set(best_node
, *used_node_mask
);
4089 * Build zonelists ordered by node and zones within node.
4090 * This results in maximum locality--normal zone overflows into local
4091 * DMA zone, if any--but risks exhausting DMA zone.
4093 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
4096 struct zonelist
*zonelist
;
4098 zonelist
= &pgdat
->node_zonelists
[0];
4099 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
4101 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4102 zonelist
->_zonerefs
[j
].zone
= NULL
;
4103 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4107 * Build gfp_thisnode zonelists
4109 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
4112 struct zonelist
*zonelist
;
4114 zonelist
= &pgdat
->node_zonelists
[1];
4115 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4116 zonelist
->_zonerefs
[j
].zone
= NULL
;
4117 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4121 * Build zonelists ordered by zone and nodes within zones.
4122 * This results in conserving DMA zone[s] until all Normal memory is
4123 * exhausted, but results in overflowing to remote node while memory
4124 * may still exist in local DMA zone.
4126 static int node_order
[MAX_NUMNODES
];
4128 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
4131 int zone_type
; /* needs to be signed */
4133 struct zonelist
*zonelist
;
4135 zonelist
= &pgdat
->node_zonelists
[0];
4137 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
4138 for (j
= 0; j
< nr_nodes
; j
++) {
4139 node
= node_order
[j
];
4140 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
4141 if (populated_zone(z
)) {
4143 &zonelist
->_zonerefs
[pos
++]);
4144 check_highest_zone(zone_type
);
4148 zonelist
->_zonerefs
[pos
].zone
= NULL
;
4149 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
4152 #if defined(CONFIG_64BIT)
4154 * Devices that require DMA32/DMA are relatively rare and do not justify a
4155 * penalty to every machine in case the specialised case applies. Default
4156 * to Node-ordering on 64-bit NUMA machines
4158 static int default_zonelist_order(void)
4160 return ZONELIST_ORDER_NODE
;
4164 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4165 * by the kernel. If processes running on node 0 deplete the low memory zone
4166 * then reclaim will occur more frequency increasing stalls and potentially
4167 * be easier to OOM if a large percentage of the zone is under writeback or
4168 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4169 * Hence, default to zone ordering on 32-bit.
4171 static int default_zonelist_order(void)
4173 return ZONELIST_ORDER_ZONE
;
4175 #endif /* CONFIG_64BIT */
4177 static void set_zonelist_order(void)
4179 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
4180 current_zonelist_order
= default_zonelist_order();
4182 current_zonelist_order
= user_zonelist_order
;
4185 static void build_zonelists(pg_data_t
*pgdat
)
4188 nodemask_t used_mask
;
4189 int local_node
, prev_node
;
4190 struct zonelist
*zonelist
;
4191 unsigned int order
= current_zonelist_order
;
4193 /* initialize zonelists */
4194 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
4195 zonelist
= pgdat
->node_zonelists
+ i
;
4196 zonelist
->_zonerefs
[0].zone
= NULL
;
4197 zonelist
->_zonerefs
[0].zone_idx
= 0;
4200 /* NUMA-aware ordering of nodes */
4201 local_node
= pgdat
->node_id
;
4202 load
= nr_online_nodes
;
4203 prev_node
= local_node
;
4204 nodes_clear(used_mask
);
4206 memset(node_order
, 0, sizeof(node_order
));
4209 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
4211 * We don't want to pressure a particular node.
4212 * So adding penalty to the first node in same
4213 * distance group to make it round-robin.
4215 if (node_distance(local_node
, node
) !=
4216 node_distance(local_node
, prev_node
))
4217 node_load
[node
] = load
;
4221 if (order
== ZONELIST_ORDER_NODE
)
4222 build_zonelists_in_node_order(pgdat
, node
);
4224 node_order
[i
++] = node
; /* remember order */
4227 if (order
== ZONELIST_ORDER_ZONE
) {
4228 /* calculate node order -- i.e., DMA last! */
4229 build_zonelists_in_zone_order(pgdat
, i
);
4232 build_thisnode_zonelists(pgdat
);
4235 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4237 * Return node id of node used for "local" allocations.
4238 * I.e., first node id of first zone in arg node's generic zonelist.
4239 * Used for initializing percpu 'numa_mem', which is used primarily
4240 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4242 int local_memory_node(int node
)
4246 (void)first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
4247 gfp_zone(GFP_KERNEL
),
4254 #else /* CONFIG_NUMA */
4256 static void set_zonelist_order(void)
4258 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
4261 static void build_zonelists(pg_data_t
*pgdat
)
4263 int node
, local_node
;
4265 struct zonelist
*zonelist
;
4267 local_node
= pgdat
->node_id
;
4269 zonelist
= &pgdat
->node_zonelists
[0];
4270 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4273 * Now we build the zonelist so that it contains the zones
4274 * of all the other nodes.
4275 * We don't want to pressure a particular node, so when
4276 * building the zones for node N, we make sure that the
4277 * zones coming right after the local ones are those from
4278 * node N+1 (modulo N)
4280 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
4281 if (!node_online(node
))
4283 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4285 for (node
= 0; node
< local_node
; node
++) {
4286 if (!node_online(node
))
4288 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4291 zonelist
->_zonerefs
[j
].zone
= NULL
;
4292 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4295 #endif /* CONFIG_NUMA */
4298 * Boot pageset table. One per cpu which is going to be used for all
4299 * zones and all nodes. The parameters will be set in such a way
4300 * that an item put on a list will immediately be handed over to
4301 * the buddy list. This is safe since pageset manipulation is done
4302 * with interrupts disabled.
4304 * The boot_pagesets must be kept even after bootup is complete for
4305 * unused processors and/or zones. They do play a role for bootstrapping
4306 * hotplugged processors.
4308 * zoneinfo_show() and maybe other functions do
4309 * not check if the processor is online before following the pageset pointer.
4310 * Other parts of the kernel may not check if the zone is available.
4312 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
4313 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
4314 static void setup_zone_pageset(struct zone
*zone
);
4317 * Global mutex to protect against size modification of zonelists
4318 * as well as to serialize pageset setup for the new populated zone.
4320 DEFINE_MUTEX(zonelists_mutex
);
4322 /* return values int ....just for stop_machine() */
4323 static int __build_all_zonelists(void *data
)
4327 pg_data_t
*self
= data
;
4330 memset(node_load
, 0, sizeof(node_load
));
4333 if (self
&& !node_online(self
->node_id
)) {
4334 build_zonelists(self
);
4337 for_each_online_node(nid
) {
4338 pg_data_t
*pgdat
= NODE_DATA(nid
);
4340 build_zonelists(pgdat
);
4344 * Initialize the boot_pagesets that are going to be used
4345 * for bootstrapping processors. The real pagesets for
4346 * each zone will be allocated later when the per cpu
4347 * allocator is available.
4349 * boot_pagesets are used also for bootstrapping offline
4350 * cpus if the system is already booted because the pagesets
4351 * are needed to initialize allocators on a specific cpu too.
4352 * F.e. the percpu allocator needs the page allocator which
4353 * needs the percpu allocator in order to allocate its pagesets
4354 * (a chicken-egg dilemma).
4356 for_each_possible_cpu(cpu
) {
4357 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
4359 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4361 * We now know the "local memory node" for each node--
4362 * i.e., the node of the first zone in the generic zonelist.
4363 * Set up numa_mem percpu variable for on-line cpus. During
4364 * boot, only the boot cpu should be on-line; we'll init the
4365 * secondary cpus' numa_mem as they come on-line. During
4366 * node/memory hotplug, we'll fixup all on-line cpus.
4368 if (cpu_online(cpu
))
4369 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
4376 static noinline
void __init
4377 build_all_zonelists_init(void)
4379 __build_all_zonelists(NULL
);
4380 mminit_verify_zonelist();
4381 cpuset_init_current_mems_allowed();
4385 * Called with zonelists_mutex held always
4386 * unless system_state == SYSTEM_BOOTING.
4388 * __ref due to (1) call of __meminit annotated setup_zone_pageset
4389 * [we're only called with non-NULL zone through __meminit paths] and
4390 * (2) call of __init annotated helper build_all_zonelists_init
4391 * [protected by SYSTEM_BOOTING].
4393 void __ref
build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
)
4395 set_zonelist_order();
4397 if (system_state
== SYSTEM_BOOTING
) {
4398 build_all_zonelists_init();
4400 #ifdef CONFIG_MEMORY_HOTPLUG
4402 setup_zone_pageset(zone
);
4404 /* we have to stop all cpus to guarantee there is no user
4406 stop_machine(__build_all_zonelists
, pgdat
, NULL
);
4407 /* cpuset refresh routine should be here */
4409 vm_total_pages
= nr_free_pagecache_pages();
4411 * Disable grouping by mobility if the number of pages in the
4412 * system is too low to allow the mechanism to work. It would be
4413 * more accurate, but expensive to check per-zone. This check is
4414 * made on memory-hotadd so a system can start with mobility
4415 * disabled and enable it later
4417 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
4418 page_group_by_mobility_disabled
= 1;
4420 page_group_by_mobility_disabled
= 0;
4422 pr_info("Built %i zonelists in %s order, mobility grouping %s. "
4423 "Total pages: %ld\n",
4425 zonelist_order_name
[current_zonelist_order
],
4426 page_group_by_mobility_disabled
? "off" : "on",
4429 pr_info("Policy zone: %s\n", zone_names
[policy_zone
]);
4434 * Helper functions to size the waitqueue hash table.
4435 * Essentially these want to choose hash table sizes sufficiently
4436 * large so that collisions trying to wait on pages are rare.
4437 * But in fact, the number of active page waitqueues on typical
4438 * systems is ridiculously low, less than 200. So this is even
4439 * conservative, even though it seems large.
4441 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
4442 * waitqueues, i.e. the size of the waitq table given the number of pages.
4444 #define PAGES_PER_WAITQUEUE 256
4446 #ifndef CONFIG_MEMORY_HOTPLUG
4447 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
4449 unsigned long size
= 1;
4451 pages
/= PAGES_PER_WAITQUEUE
;
4453 while (size
< pages
)
4457 * Once we have dozens or even hundreds of threads sleeping
4458 * on IO we've got bigger problems than wait queue collision.
4459 * Limit the size of the wait table to a reasonable size.
4461 size
= min(size
, 4096UL);
4463 return max(size
, 4UL);
4467 * A zone's size might be changed by hot-add, so it is not possible to determine
4468 * a suitable size for its wait_table. So we use the maximum size now.
4470 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
4472 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
4473 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4474 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
4476 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4477 * or more by the traditional way. (See above). It equals:
4479 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
4480 * ia64(16K page size) : = ( 8G + 4M)byte.
4481 * powerpc (64K page size) : = (32G +16M)byte.
4483 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
4490 * This is an integer logarithm so that shifts can be used later
4491 * to extract the more random high bits from the multiplicative
4492 * hash function before the remainder is taken.
4494 static inline unsigned long wait_table_bits(unsigned long size
)
4500 * Initially all pages are reserved - free ones are freed
4501 * up by free_all_bootmem() once the early boot process is
4502 * done. Non-atomic initialization, single-pass.
4504 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
4505 unsigned long start_pfn
, enum memmap_context context
)
4507 struct vmem_altmap
*altmap
= to_vmem_altmap(__pfn_to_phys(start_pfn
));
4508 unsigned long end_pfn
= start_pfn
+ size
;
4509 pg_data_t
*pgdat
= NODE_DATA(nid
);
4511 unsigned long nr_initialised
= 0;
4512 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4513 struct memblock_region
*r
= NULL
, *tmp
;
4516 if (highest_memmap_pfn
< end_pfn
- 1)
4517 highest_memmap_pfn
= end_pfn
- 1;
4520 * Honor reservation requested by the driver for this ZONE_DEVICE
4523 if (altmap
&& start_pfn
== altmap
->base_pfn
)
4524 start_pfn
+= altmap
->reserve
;
4526 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
4528 * There can be holes in boot-time mem_map[]s handed to this
4529 * function. They do not exist on hotplugged memory.
4531 if (context
!= MEMMAP_EARLY
)
4534 if (!early_pfn_valid(pfn
))
4536 if (!early_pfn_in_nid(pfn
, nid
))
4538 if (!update_defer_init(pgdat
, pfn
, end_pfn
, &nr_initialised
))
4541 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4543 * If not mirrored_kernelcore and ZONE_MOVABLE exists, range
4544 * from zone_movable_pfn[nid] to end of each node should be
4545 * ZONE_MOVABLE not ZONE_NORMAL. skip it.
4547 if (!mirrored_kernelcore
&& zone_movable_pfn
[nid
])
4548 if (zone
== ZONE_NORMAL
&& pfn
>= zone_movable_pfn
[nid
])
4552 * Check given memblock attribute by firmware which can affect
4553 * kernel memory layout. If zone==ZONE_MOVABLE but memory is
4554 * mirrored, it's an overlapped memmap init. skip it.
4556 if (mirrored_kernelcore
&& zone
== ZONE_MOVABLE
) {
4557 if (!r
|| pfn
>= memblock_region_memory_end_pfn(r
)) {
4558 for_each_memblock(memory
, tmp
)
4559 if (pfn
< memblock_region_memory_end_pfn(tmp
))
4563 if (pfn
>= memblock_region_memory_base_pfn(r
) &&
4564 memblock_is_mirror(r
)) {
4565 /* already initialized as NORMAL */
4566 pfn
= memblock_region_memory_end_pfn(r
);
4574 * Mark the block movable so that blocks are reserved for
4575 * movable at startup. This will force kernel allocations
4576 * to reserve their blocks rather than leaking throughout
4577 * the address space during boot when many long-lived
4578 * kernel allocations are made.
4580 * bitmap is created for zone's valid pfn range. but memmap
4581 * can be created for invalid pages (for alignment)
4582 * check here not to call set_pageblock_migratetype() against
4585 if (!(pfn
& (pageblock_nr_pages
- 1))) {
4586 struct page
*page
= pfn_to_page(pfn
);
4588 __init_single_page(page
, pfn
, zone
, nid
);
4589 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4591 __init_single_pfn(pfn
, zone
, nid
);
4596 static void __meminit
zone_init_free_lists(struct zone
*zone
)
4598 unsigned int order
, t
;
4599 for_each_migratetype_order(order
, t
) {
4600 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
4601 zone
->free_area
[order
].nr_free
= 0;
4605 #ifndef __HAVE_ARCH_MEMMAP_INIT
4606 #define memmap_init(size, nid, zone, start_pfn) \
4607 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4610 static int zone_batchsize(struct zone
*zone
)
4616 * The per-cpu-pages pools are set to around 1000th of the
4617 * size of the zone. But no more than 1/2 of a meg.
4619 * OK, so we don't know how big the cache is. So guess.
4621 batch
= zone
->managed_pages
/ 1024;
4622 if (batch
* PAGE_SIZE
> 512 * 1024)
4623 batch
= (512 * 1024) / PAGE_SIZE
;
4624 batch
/= 4; /* We effectively *= 4 below */
4629 * Clamp the batch to a 2^n - 1 value. Having a power
4630 * of 2 value was found to be more likely to have
4631 * suboptimal cache aliasing properties in some cases.
4633 * For example if 2 tasks are alternately allocating
4634 * batches of pages, one task can end up with a lot
4635 * of pages of one half of the possible page colors
4636 * and the other with pages of the other colors.
4638 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
4643 /* The deferral and batching of frees should be suppressed under NOMMU
4646 * The problem is that NOMMU needs to be able to allocate large chunks
4647 * of contiguous memory as there's no hardware page translation to
4648 * assemble apparent contiguous memory from discontiguous pages.
4650 * Queueing large contiguous runs of pages for batching, however,
4651 * causes the pages to actually be freed in smaller chunks. As there
4652 * can be a significant delay between the individual batches being
4653 * recycled, this leads to the once large chunks of space being
4654 * fragmented and becoming unavailable for high-order allocations.
4661 * pcp->high and pcp->batch values are related and dependent on one another:
4662 * ->batch must never be higher then ->high.
4663 * The following function updates them in a safe manner without read side
4666 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4667 * those fields changing asynchronously (acording the the above rule).
4669 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4670 * outside of boot time (or some other assurance that no concurrent updaters
4673 static void pageset_update(struct per_cpu_pages
*pcp
, unsigned long high
,
4674 unsigned long batch
)
4676 /* start with a fail safe value for batch */
4680 /* Update high, then batch, in order */
4687 /* a companion to pageset_set_high() */
4688 static void pageset_set_batch(struct per_cpu_pageset
*p
, unsigned long batch
)
4690 pageset_update(&p
->pcp
, 6 * batch
, max(1UL, 1 * batch
));
4693 static void pageset_init(struct per_cpu_pageset
*p
)
4695 struct per_cpu_pages
*pcp
;
4698 memset(p
, 0, sizeof(*p
));
4702 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
4703 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
4706 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
4709 pageset_set_batch(p
, batch
);
4713 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
4714 * to the value high for the pageset p.
4716 static void pageset_set_high(struct per_cpu_pageset
*p
,
4719 unsigned long batch
= max(1UL, high
/ 4);
4720 if ((high
/ 4) > (PAGE_SHIFT
* 8))
4721 batch
= PAGE_SHIFT
* 8;
4723 pageset_update(&p
->pcp
, high
, batch
);
4726 static void pageset_set_high_and_batch(struct zone
*zone
,
4727 struct per_cpu_pageset
*pcp
)
4729 if (percpu_pagelist_fraction
)
4730 pageset_set_high(pcp
,
4731 (zone
->managed_pages
/
4732 percpu_pagelist_fraction
));
4734 pageset_set_batch(pcp
, zone_batchsize(zone
));
4737 static void __meminit
zone_pageset_init(struct zone
*zone
, int cpu
)
4739 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
4742 pageset_set_high_and_batch(zone
, pcp
);
4745 static void __meminit
setup_zone_pageset(struct zone
*zone
)
4748 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
4749 for_each_possible_cpu(cpu
)
4750 zone_pageset_init(zone
, cpu
);
4754 * Allocate per cpu pagesets and initialize them.
4755 * Before this call only boot pagesets were available.
4757 void __init
setup_per_cpu_pageset(void)
4761 for_each_populated_zone(zone
)
4762 setup_zone_pageset(zone
);
4765 static noinline __init_refok
4766 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
4772 * The per-page waitqueue mechanism uses hashed waitqueues
4775 zone
->wait_table_hash_nr_entries
=
4776 wait_table_hash_nr_entries(zone_size_pages
);
4777 zone
->wait_table_bits
=
4778 wait_table_bits(zone
->wait_table_hash_nr_entries
);
4779 alloc_size
= zone
->wait_table_hash_nr_entries
4780 * sizeof(wait_queue_head_t
);
4782 if (!slab_is_available()) {
4783 zone
->wait_table
= (wait_queue_head_t
*)
4784 memblock_virt_alloc_node_nopanic(
4785 alloc_size
, zone
->zone_pgdat
->node_id
);
4788 * This case means that a zone whose size was 0 gets new memory
4789 * via memory hot-add.
4790 * But it may be the case that a new node was hot-added. In
4791 * this case vmalloc() will not be able to use this new node's
4792 * memory - this wait_table must be initialized to use this new
4793 * node itself as well.
4794 * To use this new node's memory, further consideration will be
4797 zone
->wait_table
= vmalloc(alloc_size
);
4799 if (!zone
->wait_table
)
4802 for (i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
4803 init_waitqueue_head(zone
->wait_table
+ i
);
4808 static __meminit
void zone_pcp_init(struct zone
*zone
)
4811 * per cpu subsystem is not up at this point. The following code
4812 * relies on the ability of the linker to provide the
4813 * offset of a (static) per cpu variable into the per cpu area.
4815 zone
->pageset
= &boot_pageset
;
4817 if (populated_zone(zone
))
4818 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
4819 zone
->name
, zone
->present_pages
,
4820 zone_batchsize(zone
));
4823 int __meminit
init_currently_empty_zone(struct zone
*zone
,
4824 unsigned long zone_start_pfn
,
4827 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
4829 ret
= zone_wait_table_init(zone
, size
);
4832 pgdat
->nr_zones
= zone_idx(zone
) + 1;
4834 zone
->zone_start_pfn
= zone_start_pfn
;
4836 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
4837 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4839 (unsigned long)zone_idx(zone
),
4840 zone_start_pfn
, (zone_start_pfn
+ size
));
4842 zone_init_free_lists(zone
);
4847 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4848 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4851 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4853 int __meminit
__early_pfn_to_nid(unsigned long pfn
,
4854 struct mminit_pfnnid_cache
*state
)
4856 unsigned long start_pfn
, end_pfn
;
4859 if (state
->last_start
<= pfn
&& pfn
< state
->last_end
)
4860 return state
->last_nid
;
4862 nid
= memblock_search_pfn_nid(pfn
, &start_pfn
, &end_pfn
);
4864 state
->last_start
= start_pfn
;
4865 state
->last_end
= end_pfn
;
4866 state
->last_nid
= nid
;
4871 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4874 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
4875 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4876 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
4878 * If an architecture guarantees that all ranges registered contain no holes
4879 * and may be freed, this this function may be used instead of calling
4880 * memblock_free_early_nid() manually.
4882 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
4884 unsigned long start_pfn
, end_pfn
;
4887 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
4888 start_pfn
= min(start_pfn
, max_low_pfn
);
4889 end_pfn
= min(end_pfn
, max_low_pfn
);
4891 if (start_pfn
< end_pfn
)
4892 memblock_free_early_nid(PFN_PHYS(start_pfn
),
4893 (end_pfn
- start_pfn
) << PAGE_SHIFT
,
4899 * sparse_memory_present_with_active_regions - Call memory_present for each active range
4900 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4902 * If an architecture guarantees that all ranges registered contain no holes and may
4903 * be freed, this function may be used instead of calling memory_present() manually.
4905 void __init
sparse_memory_present_with_active_regions(int nid
)
4907 unsigned long start_pfn
, end_pfn
;
4910 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
4911 memory_present(this_nid
, start_pfn
, end_pfn
);
4915 * get_pfn_range_for_nid - Return the start and end page frames for a node
4916 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4917 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4918 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4920 * It returns the start and end page frame of a node based on information
4921 * provided by memblock_set_node(). If called for a node
4922 * with no available memory, a warning is printed and the start and end
4925 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
4926 unsigned long *start_pfn
, unsigned long *end_pfn
)
4928 unsigned long this_start_pfn
, this_end_pfn
;
4934 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
4935 *start_pfn
= min(*start_pfn
, this_start_pfn
);
4936 *end_pfn
= max(*end_pfn
, this_end_pfn
);
4939 if (*start_pfn
== -1UL)
4944 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4945 * assumption is made that zones within a node are ordered in monotonic
4946 * increasing memory addresses so that the "highest" populated zone is used
4948 static void __init
find_usable_zone_for_movable(void)
4951 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
4952 if (zone_index
== ZONE_MOVABLE
)
4955 if (arch_zone_highest_possible_pfn
[zone_index
] >
4956 arch_zone_lowest_possible_pfn
[zone_index
])
4960 VM_BUG_ON(zone_index
== -1);
4961 movable_zone
= zone_index
;
4965 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4966 * because it is sized independent of architecture. Unlike the other zones,
4967 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4968 * in each node depending on the size of each node and how evenly kernelcore
4969 * is distributed. This helper function adjusts the zone ranges
4970 * provided by the architecture for a given node by using the end of the
4971 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4972 * zones within a node are in order of monotonic increases memory addresses
4974 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
4975 unsigned long zone_type
,
4976 unsigned long node_start_pfn
,
4977 unsigned long node_end_pfn
,
4978 unsigned long *zone_start_pfn
,
4979 unsigned long *zone_end_pfn
)
4981 /* Only adjust if ZONE_MOVABLE is on this node */
4982 if (zone_movable_pfn
[nid
]) {
4983 /* Size ZONE_MOVABLE */
4984 if (zone_type
== ZONE_MOVABLE
) {
4985 *zone_start_pfn
= zone_movable_pfn
[nid
];
4986 *zone_end_pfn
= min(node_end_pfn
,
4987 arch_zone_highest_possible_pfn
[movable_zone
]);
4989 /* Check if this whole range is within ZONE_MOVABLE */
4990 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
4991 *zone_start_pfn
= *zone_end_pfn
;
4996 * Return the number of pages a zone spans in a node, including holes
4997 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4999 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5000 unsigned long zone_type
,
5001 unsigned long node_start_pfn
,
5002 unsigned long node_end_pfn
,
5003 unsigned long *zone_start_pfn
,
5004 unsigned long *zone_end_pfn
,
5005 unsigned long *ignored
)
5007 /* When hotadd a new node from cpu_up(), the node should be empty */
5008 if (!node_start_pfn
&& !node_end_pfn
)
5011 /* Get the start and end of the zone */
5012 *zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
5013 *zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
5014 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5015 node_start_pfn
, node_end_pfn
,
5016 zone_start_pfn
, zone_end_pfn
);
5018 /* Check that this node has pages within the zone's required range */
5019 if (*zone_end_pfn
< node_start_pfn
|| *zone_start_pfn
> node_end_pfn
)
5022 /* Move the zone boundaries inside the node if necessary */
5023 *zone_end_pfn
= min(*zone_end_pfn
, node_end_pfn
);
5024 *zone_start_pfn
= max(*zone_start_pfn
, node_start_pfn
);
5026 /* Return the spanned pages */
5027 return *zone_end_pfn
- *zone_start_pfn
;
5031 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
5032 * then all holes in the requested range will be accounted for.
5034 unsigned long __meminit
__absent_pages_in_range(int nid
,
5035 unsigned long range_start_pfn
,
5036 unsigned long range_end_pfn
)
5038 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
5039 unsigned long start_pfn
, end_pfn
;
5042 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5043 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
5044 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
5045 nr_absent
-= end_pfn
- start_pfn
;
5051 * absent_pages_in_range - Return number of page frames in holes within a range
5052 * @start_pfn: The start PFN to start searching for holes
5053 * @end_pfn: The end PFN to stop searching for holes
5055 * It returns the number of pages frames in memory holes within a range.
5057 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
5058 unsigned long end_pfn
)
5060 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
5063 /* Return the number of page frames in holes in a zone on a node */
5064 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5065 unsigned long zone_type
,
5066 unsigned long node_start_pfn
,
5067 unsigned long node_end_pfn
,
5068 unsigned long *ignored
)
5070 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
5071 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
5072 unsigned long zone_start_pfn
, zone_end_pfn
;
5073 unsigned long nr_absent
;
5075 /* When hotadd a new node from cpu_up(), the node should be empty */
5076 if (!node_start_pfn
&& !node_end_pfn
)
5079 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
5080 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
5082 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5083 node_start_pfn
, node_end_pfn
,
5084 &zone_start_pfn
, &zone_end_pfn
);
5085 nr_absent
= __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
5088 * ZONE_MOVABLE handling.
5089 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
5092 if (zone_movable_pfn
[nid
]) {
5093 if (mirrored_kernelcore
) {
5094 unsigned long start_pfn
, end_pfn
;
5095 struct memblock_region
*r
;
5097 for_each_memblock(memory
, r
) {
5098 start_pfn
= clamp(memblock_region_memory_base_pfn(r
),
5099 zone_start_pfn
, zone_end_pfn
);
5100 end_pfn
= clamp(memblock_region_memory_end_pfn(r
),
5101 zone_start_pfn
, zone_end_pfn
);
5103 if (zone_type
== ZONE_MOVABLE
&&
5104 memblock_is_mirror(r
))
5105 nr_absent
+= end_pfn
- start_pfn
;
5107 if (zone_type
== ZONE_NORMAL
&&
5108 !memblock_is_mirror(r
))
5109 nr_absent
+= end_pfn
- start_pfn
;
5112 if (zone_type
== ZONE_NORMAL
)
5113 nr_absent
+= node_end_pfn
- zone_movable_pfn
[nid
];
5120 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5121 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5122 unsigned long zone_type
,
5123 unsigned long node_start_pfn
,
5124 unsigned long node_end_pfn
,
5125 unsigned long *zone_start_pfn
,
5126 unsigned long *zone_end_pfn
,
5127 unsigned long *zones_size
)
5131 *zone_start_pfn
= node_start_pfn
;
5132 for (zone
= 0; zone
< zone_type
; zone
++)
5133 *zone_start_pfn
+= zones_size
[zone
];
5135 *zone_end_pfn
= *zone_start_pfn
+ zones_size
[zone_type
];
5137 return zones_size
[zone_type
];
5140 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5141 unsigned long zone_type
,
5142 unsigned long node_start_pfn
,
5143 unsigned long node_end_pfn
,
5144 unsigned long *zholes_size
)
5149 return zholes_size
[zone_type
];
5152 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5154 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
5155 unsigned long node_start_pfn
,
5156 unsigned long node_end_pfn
,
5157 unsigned long *zones_size
,
5158 unsigned long *zholes_size
)
5160 unsigned long realtotalpages
= 0, totalpages
= 0;
5163 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5164 struct zone
*zone
= pgdat
->node_zones
+ i
;
5165 unsigned long zone_start_pfn
, zone_end_pfn
;
5166 unsigned long size
, real_size
;
5168 size
= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
5174 real_size
= size
- zone_absent_pages_in_node(pgdat
->node_id
, i
,
5175 node_start_pfn
, node_end_pfn
,
5178 zone
->zone_start_pfn
= zone_start_pfn
;
5180 zone
->zone_start_pfn
= 0;
5181 zone
->spanned_pages
= size
;
5182 zone
->present_pages
= real_size
;
5185 realtotalpages
+= real_size
;
5188 pgdat
->node_spanned_pages
= totalpages
;
5189 pgdat
->node_present_pages
= realtotalpages
;
5190 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
5194 #ifndef CONFIG_SPARSEMEM
5196 * Calculate the size of the zone->blockflags rounded to an unsigned long
5197 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5198 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
5199 * round what is now in bits to nearest long in bits, then return it in
5202 static unsigned long __init
usemap_size(unsigned long zone_start_pfn
, unsigned long zonesize
)
5204 unsigned long usemapsize
;
5206 zonesize
+= zone_start_pfn
& (pageblock_nr_pages
-1);
5207 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
5208 usemapsize
= usemapsize
>> pageblock_order
;
5209 usemapsize
*= NR_PAGEBLOCK_BITS
;
5210 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
5212 return usemapsize
/ 8;
5215 static void __init
setup_usemap(struct pglist_data
*pgdat
,
5217 unsigned long zone_start_pfn
,
5218 unsigned long zonesize
)
5220 unsigned long usemapsize
= usemap_size(zone_start_pfn
, zonesize
);
5221 zone
->pageblock_flags
= NULL
;
5223 zone
->pageblock_flags
=
5224 memblock_virt_alloc_node_nopanic(usemapsize
,
5228 static inline void setup_usemap(struct pglist_data
*pgdat
, struct zone
*zone
,
5229 unsigned long zone_start_pfn
, unsigned long zonesize
) {}
5230 #endif /* CONFIG_SPARSEMEM */
5232 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
5234 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
5235 void __paginginit
set_pageblock_order(void)
5239 /* Check that pageblock_nr_pages has not already been setup */
5240 if (pageblock_order
)
5243 if (HPAGE_SHIFT
> PAGE_SHIFT
)
5244 order
= HUGETLB_PAGE_ORDER
;
5246 order
= MAX_ORDER
- 1;
5249 * Assume the largest contiguous order of interest is a huge page.
5250 * This value may be variable depending on boot parameters on IA64 and
5253 pageblock_order
= order
;
5255 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5258 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
5259 * is unused as pageblock_order is set at compile-time. See
5260 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5263 void __paginginit
set_pageblock_order(void)
5267 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5269 static unsigned long __paginginit
calc_memmap_size(unsigned long spanned_pages
,
5270 unsigned long present_pages
)
5272 unsigned long pages
= spanned_pages
;
5275 * Provide a more accurate estimation if there are holes within
5276 * the zone and SPARSEMEM is in use. If there are holes within the
5277 * zone, each populated memory region may cost us one or two extra
5278 * memmap pages due to alignment because memmap pages for each
5279 * populated regions may not naturally algined on page boundary.
5280 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5282 if (spanned_pages
> present_pages
+ (present_pages
>> 4) &&
5283 IS_ENABLED(CONFIG_SPARSEMEM
))
5284 pages
= present_pages
;
5286 return PAGE_ALIGN(pages
* sizeof(struct page
)) >> PAGE_SHIFT
;
5290 * Set up the zone data structures:
5291 * - mark all pages reserved
5292 * - mark all memory queues empty
5293 * - clear the memory bitmaps
5295 * NOTE: pgdat should get zeroed by caller.
5297 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
)
5300 int nid
= pgdat
->node_id
;
5303 pgdat_resize_init(pgdat
);
5304 #ifdef CONFIG_NUMA_BALANCING
5305 spin_lock_init(&pgdat
->numabalancing_migrate_lock
);
5306 pgdat
->numabalancing_migrate_nr_pages
= 0;
5307 pgdat
->numabalancing_migrate_next_window
= jiffies
;
5309 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5310 spin_lock_init(&pgdat
->split_queue_lock
);
5311 INIT_LIST_HEAD(&pgdat
->split_queue
);
5312 pgdat
->split_queue_len
= 0;
5314 init_waitqueue_head(&pgdat
->kswapd_wait
);
5315 init_waitqueue_head(&pgdat
->pfmemalloc_wait
);
5316 pgdat_page_ext_init(pgdat
);
5318 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5319 struct zone
*zone
= pgdat
->node_zones
+ j
;
5320 unsigned long size
, realsize
, freesize
, memmap_pages
;
5321 unsigned long zone_start_pfn
= zone
->zone_start_pfn
;
5323 size
= zone
->spanned_pages
;
5324 realsize
= freesize
= zone
->present_pages
;
5327 * Adjust freesize so that it accounts for how much memory
5328 * is used by this zone for memmap. This affects the watermark
5329 * and per-cpu initialisations
5331 memmap_pages
= calc_memmap_size(size
, realsize
);
5332 if (!is_highmem_idx(j
)) {
5333 if (freesize
>= memmap_pages
) {
5334 freesize
-= memmap_pages
;
5337 " %s zone: %lu pages used for memmap\n",
5338 zone_names
[j
], memmap_pages
);
5341 " %s zone: %lu pages exceeds freesize %lu\n",
5342 zone_names
[j
], memmap_pages
, freesize
);
5345 /* Account for reserved pages */
5346 if (j
== 0 && freesize
> dma_reserve
) {
5347 freesize
-= dma_reserve
;
5348 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
5349 zone_names
[0], dma_reserve
);
5352 if (!is_highmem_idx(j
))
5353 nr_kernel_pages
+= freesize
;
5354 /* Charge for highmem memmap if there are enough kernel pages */
5355 else if (nr_kernel_pages
> memmap_pages
* 2)
5356 nr_kernel_pages
-= memmap_pages
;
5357 nr_all_pages
+= freesize
;
5360 * Set an approximate value for lowmem here, it will be adjusted
5361 * when the bootmem allocator frees pages into the buddy system.
5362 * And all highmem pages will be managed by the buddy system.
5364 zone
->managed_pages
= is_highmem_idx(j
) ? realsize
: freesize
;
5367 zone
->min_unmapped_pages
= (freesize
*sysctl_min_unmapped_ratio
)
5369 zone
->min_slab_pages
= (freesize
* sysctl_min_slab_ratio
) / 100;
5371 zone
->name
= zone_names
[j
];
5372 spin_lock_init(&zone
->lock
);
5373 spin_lock_init(&zone
->lru_lock
);
5374 zone_seqlock_init(zone
);
5375 zone
->zone_pgdat
= pgdat
;
5376 zone_pcp_init(zone
);
5378 /* For bootup, initialized properly in watermark setup */
5379 mod_zone_page_state(zone
, NR_ALLOC_BATCH
, zone
->managed_pages
);
5381 lruvec_init(&zone
->lruvec
);
5385 set_pageblock_order();
5386 setup_usemap(pgdat
, zone
, zone_start_pfn
, size
);
5387 ret
= init_currently_empty_zone(zone
, zone_start_pfn
, size
);
5389 memmap_init(size
, nid
, j
, zone_start_pfn
);
5393 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
5395 unsigned long __maybe_unused start
= 0;
5396 unsigned long __maybe_unused offset
= 0;
5398 /* Skip empty nodes */
5399 if (!pgdat
->node_spanned_pages
)
5402 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5403 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
5404 offset
= pgdat
->node_start_pfn
- start
;
5405 /* ia64 gets its own node_mem_map, before this, without bootmem */
5406 if (!pgdat
->node_mem_map
) {
5407 unsigned long size
, end
;
5411 * The zone's endpoints aren't required to be MAX_ORDER
5412 * aligned but the node_mem_map endpoints must be in order
5413 * for the buddy allocator to function correctly.
5415 end
= pgdat_end_pfn(pgdat
);
5416 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
5417 size
= (end
- start
) * sizeof(struct page
);
5418 map
= alloc_remap(pgdat
->node_id
, size
);
5420 map
= memblock_virt_alloc_node_nopanic(size
,
5422 pgdat
->node_mem_map
= map
+ offset
;
5424 #ifndef CONFIG_NEED_MULTIPLE_NODES
5426 * With no DISCONTIG, the global mem_map is just set as node 0's
5428 if (pgdat
== NODE_DATA(0)) {
5429 mem_map
= NODE_DATA(0)->node_mem_map
;
5430 #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
5431 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
5433 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5436 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
5439 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
5440 unsigned long node_start_pfn
, unsigned long *zholes_size
)
5442 pg_data_t
*pgdat
= NODE_DATA(nid
);
5443 unsigned long start_pfn
= 0;
5444 unsigned long end_pfn
= 0;
5446 /* pg_data_t should be reset to zero when it's allocated */
5447 WARN_ON(pgdat
->nr_zones
|| pgdat
->classzone_idx
);
5449 reset_deferred_meminit(pgdat
);
5450 pgdat
->node_id
= nid
;
5451 pgdat
->node_start_pfn
= node_start_pfn
;
5452 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5453 get_pfn_range_for_nid(nid
, &start_pfn
, &end_pfn
);
5454 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid
,
5455 (u64
)start_pfn
<< PAGE_SHIFT
,
5456 end_pfn
? ((u64
)end_pfn
<< PAGE_SHIFT
) - 1 : 0);
5458 start_pfn
= node_start_pfn
;
5460 calculate_node_totalpages(pgdat
, start_pfn
, end_pfn
,
5461 zones_size
, zholes_size
);
5463 alloc_node_mem_map(pgdat
);
5464 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5465 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5466 nid
, (unsigned long)pgdat
,
5467 (unsigned long)pgdat
->node_mem_map
);
5470 free_area_init_core(pgdat
);
5473 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5475 #if MAX_NUMNODES > 1
5477 * Figure out the number of possible node ids.
5479 void __init
setup_nr_node_ids(void)
5481 unsigned int highest
;
5483 highest
= find_last_bit(node_possible_map
.bits
, MAX_NUMNODES
);
5484 nr_node_ids
= highest
+ 1;
5489 * node_map_pfn_alignment - determine the maximum internode alignment
5491 * This function should be called after node map is populated and sorted.
5492 * It calculates the maximum power of two alignment which can distinguish
5495 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5496 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
5497 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
5498 * shifted, 1GiB is enough and this function will indicate so.
5500 * This is used to test whether pfn -> nid mapping of the chosen memory
5501 * model has fine enough granularity to avoid incorrect mapping for the
5502 * populated node map.
5504 * Returns the determined alignment in pfn's. 0 if there is no alignment
5505 * requirement (single node).
5507 unsigned long __init
node_map_pfn_alignment(void)
5509 unsigned long accl_mask
= 0, last_end
= 0;
5510 unsigned long start
, end
, mask
;
5514 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
5515 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
5522 * Start with a mask granular enough to pin-point to the
5523 * start pfn and tick off bits one-by-one until it becomes
5524 * too coarse to separate the current node from the last.
5526 mask
= ~((1 << __ffs(start
)) - 1);
5527 while (mask
&& last_end
<= (start
& (mask
<< 1)))
5530 /* accumulate all internode masks */
5534 /* convert mask to number of pages */
5535 return ~accl_mask
+ 1;
5538 /* Find the lowest pfn for a node */
5539 static unsigned long __init
find_min_pfn_for_node(int nid
)
5541 unsigned long min_pfn
= ULONG_MAX
;
5542 unsigned long start_pfn
;
5545 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
5546 min_pfn
= min(min_pfn
, start_pfn
);
5548 if (min_pfn
== ULONG_MAX
) {
5550 "Could not find start_pfn for node %d\n", nid
);
5558 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5560 * It returns the minimum PFN based on information provided via
5561 * memblock_set_node().
5563 unsigned long __init
find_min_pfn_with_active_regions(void)
5565 return find_min_pfn_for_node(MAX_NUMNODES
);
5569 * early_calculate_totalpages()
5570 * Sum pages in active regions for movable zone.
5571 * Populate N_MEMORY for calculating usable_nodes.
5573 static unsigned long __init
early_calculate_totalpages(void)
5575 unsigned long totalpages
= 0;
5576 unsigned long start_pfn
, end_pfn
;
5579 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
5580 unsigned long pages
= end_pfn
- start_pfn
;
5582 totalpages
+= pages
;
5584 node_set_state(nid
, N_MEMORY
);
5590 * Find the PFN the Movable zone begins in each node. Kernel memory
5591 * is spread evenly between nodes as long as the nodes have enough
5592 * memory. When they don't, some nodes will have more kernelcore than
5595 static void __init
find_zone_movable_pfns_for_nodes(void)
5598 unsigned long usable_startpfn
;
5599 unsigned long kernelcore_node
, kernelcore_remaining
;
5600 /* save the state before borrow the nodemask */
5601 nodemask_t saved_node_state
= node_states
[N_MEMORY
];
5602 unsigned long totalpages
= early_calculate_totalpages();
5603 int usable_nodes
= nodes_weight(node_states
[N_MEMORY
]);
5604 struct memblock_region
*r
;
5606 /* Need to find movable_zone earlier when movable_node is specified. */
5607 find_usable_zone_for_movable();
5610 * If movable_node is specified, ignore kernelcore and movablecore
5613 if (movable_node_is_enabled()) {
5614 for_each_memblock(memory
, r
) {
5615 if (!memblock_is_hotpluggable(r
))
5620 usable_startpfn
= PFN_DOWN(r
->base
);
5621 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
5622 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
5630 * If kernelcore=mirror is specified, ignore movablecore option
5632 if (mirrored_kernelcore
) {
5633 bool mem_below_4gb_not_mirrored
= false;
5635 for_each_memblock(memory
, r
) {
5636 if (memblock_is_mirror(r
))
5641 usable_startpfn
= memblock_region_memory_base_pfn(r
);
5643 if (usable_startpfn
< 0x100000) {
5644 mem_below_4gb_not_mirrored
= true;
5648 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
5649 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
5653 if (mem_below_4gb_not_mirrored
)
5654 pr_warn("This configuration results in unmirrored kernel memory.");
5660 * If movablecore=nn[KMG] was specified, calculate what size of
5661 * kernelcore that corresponds so that memory usable for
5662 * any allocation type is evenly spread. If both kernelcore
5663 * and movablecore are specified, then the value of kernelcore
5664 * will be used for required_kernelcore if it's greater than
5665 * what movablecore would have allowed.
5667 if (required_movablecore
) {
5668 unsigned long corepages
;
5671 * Round-up so that ZONE_MOVABLE is at least as large as what
5672 * was requested by the user
5674 required_movablecore
=
5675 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
5676 required_movablecore
= min(totalpages
, required_movablecore
);
5677 corepages
= totalpages
- required_movablecore
;
5679 required_kernelcore
= max(required_kernelcore
, corepages
);
5683 * If kernelcore was not specified or kernelcore size is larger
5684 * than totalpages, there is no ZONE_MOVABLE.
5686 if (!required_kernelcore
|| required_kernelcore
>= totalpages
)
5689 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5690 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
5693 /* Spread kernelcore memory as evenly as possible throughout nodes */
5694 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5695 for_each_node_state(nid
, N_MEMORY
) {
5696 unsigned long start_pfn
, end_pfn
;
5699 * Recalculate kernelcore_node if the division per node
5700 * now exceeds what is necessary to satisfy the requested
5701 * amount of memory for the kernel
5703 if (required_kernelcore
< kernelcore_node
)
5704 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5707 * As the map is walked, we track how much memory is usable
5708 * by the kernel using kernelcore_remaining. When it is
5709 * 0, the rest of the node is usable by ZONE_MOVABLE
5711 kernelcore_remaining
= kernelcore_node
;
5713 /* Go through each range of PFNs within this node */
5714 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5715 unsigned long size_pages
;
5717 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
5718 if (start_pfn
>= end_pfn
)
5721 /* Account for what is only usable for kernelcore */
5722 if (start_pfn
< usable_startpfn
) {
5723 unsigned long kernel_pages
;
5724 kernel_pages
= min(end_pfn
, usable_startpfn
)
5727 kernelcore_remaining
-= min(kernel_pages
,
5728 kernelcore_remaining
);
5729 required_kernelcore
-= min(kernel_pages
,
5730 required_kernelcore
);
5732 /* Continue if range is now fully accounted */
5733 if (end_pfn
<= usable_startpfn
) {
5736 * Push zone_movable_pfn to the end so
5737 * that if we have to rebalance
5738 * kernelcore across nodes, we will
5739 * not double account here
5741 zone_movable_pfn
[nid
] = end_pfn
;
5744 start_pfn
= usable_startpfn
;
5748 * The usable PFN range for ZONE_MOVABLE is from
5749 * start_pfn->end_pfn. Calculate size_pages as the
5750 * number of pages used as kernelcore
5752 size_pages
= end_pfn
- start_pfn
;
5753 if (size_pages
> kernelcore_remaining
)
5754 size_pages
= kernelcore_remaining
;
5755 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
5758 * Some kernelcore has been met, update counts and
5759 * break if the kernelcore for this node has been
5762 required_kernelcore
-= min(required_kernelcore
,
5764 kernelcore_remaining
-= size_pages
;
5765 if (!kernelcore_remaining
)
5771 * If there is still required_kernelcore, we do another pass with one
5772 * less node in the count. This will push zone_movable_pfn[nid] further
5773 * along on the nodes that still have memory until kernelcore is
5777 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
5781 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5782 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
5783 zone_movable_pfn
[nid
] =
5784 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
5787 /* restore the node_state */
5788 node_states
[N_MEMORY
] = saved_node_state
;
5791 /* Any regular or high memory on that node ? */
5792 static void check_for_memory(pg_data_t
*pgdat
, int nid
)
5794 enum zone_type zone_type
;
5796 if (N_MEMORY
== N_NORMAL_MEMORY
)
5799 for (zone_type
= 0; zone_type
<= ZONE_MOVABLE
- 1; zone_type
++) {
5800 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
5801 if (populated_zone(zone
)) {
5802 node_set_state(nid
, N_HIGH_MEMORY
);
5803 if (N_NORMAL_MEMORY
!= N_HIGH_MEMORY
&&
5804 zone_type
<= ZONE_NORMAL
)
5805 node_set_state(nid
, N_NORMAL_MEMORY
);
5812 * free_area_init_nodes - Initialise all pg_data_t and zone data
5813 * @max_zone_pfn: an array of max PFNs for each zone
5815 * This will call free_area_init_node() for each active node in the system.
5816 * Using the page ranges provided by memblock_set_node(), the size of each
5817 * zone in each node and their holes is calculated. If the maximum PFN
5818 * between two adjacent zones match, it is assumed that the zone is empty.
5819 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5820 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5821 * starts where the previous one ended. For example, ZONE_DMA32 starts
5822 * at arch_max_dma_pfn.
5824 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
5826 unsigned long start_pfn
, end_pfn
;
5829 /* Record where the zone boundaries are */
5830 memset(arch_zone_lowest_possible_pfn
, 0,
5831 sizeof(arch_zone_lowest_possible_pfn
));
5832 memset(arch_zone_highest_possible_pfn
, 0,
5833 sizeof(arch_zone_highest_possible_pfn
));
5834 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
5835 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
5836 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
5837 if (i
== ZONE_MOVABLE
)
5839 arch_zone_lowest_possible_pfn
[i
] =
5840 arch_zone_highest_possible_pfn
[i
-1];
5841 arch_zone_highest_possible_pfn
[i
] =
5842 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
5844 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
5845 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
5847 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5848 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
5849 find_zone_movable_pfns_for_nodes();
5851 /* Print out the zone ranges */
5852 pr_info("Zone ranges:\n");
5853 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5854 if (i
== ZONE_MOVABLE
)
5856 pr_info(" %-8s ", zone_names
[i
]);
5857 if (arch_zone_lowest_possible_pfn
[i
] ==
5858 arch_zone_highest_possible_pfn
[i
])
5861 pr_cont("[mem %#018Lx-%#018Lx]\n",
5862 (u64
)arch_zone_lowest_possible_pfn
[i
]
5864 ((u64
)arch_zone_highest_possible_pfn
[i
]
5865 << PAGE_SHIFT
) - 1);
5868 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5869 pr_info("Movable zone start for each node\n");
5870 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
5871 if (zone_movable_pfn
[i
])
5872 pr_info(" Node %d: %#018Lx\n", i
,
5873 (u64
)zone_movable_pfn
[i
] << PAGE_SHIFT
);
5876 /* Print out the early node map */
5877 pr_info("Early memory node ranges\n");
5878 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
5879 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid
,
5880 (u64
)start_pfn
<< PAGE_SHIFT
,
5881 ((u64
)end_pfn
<< PAGE_SHIFT
) - 1);
5883 /* Initialise every node */
5884 mminit_verify_pageflags_layout();
5885 setup_nr_node_ids();
5886 for_each_online_node(nid
) {
5887 pg_data_t
*pgdat
= NODE_DATA(nid
);
5888 free_area_init_node(nid
, NULL
,
5889 find_min_pfn_for_node(nid
), NULL
);
5891 /* Any memory on that node */
5892 if (pgdat
->node_present_pages
)
5893 node_set_state(nid
, N_MEMORY
);
5894 check_for_memory(pgdat
, nid
);
5898 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
5900 unsigned long long coremem
;
5904 coremem
= memparse(p
, &p
);
5905 *core
= coremem
>> PAGE_SHIFT
;
5907 /* Paranoid check that UL is enough for the coremem value */
5908 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
5914 * kernelcore=size sets the amount of memory for use for allocations that
5915 * cannot be reclaimed or migrated.
5917 static int __init
cmdline_parse_kernelcore(char *p
)
5919 /* parse kernelcore=mirror */
5920 if (parse_option_str(p
, "mirror")) {
5921 mirrored_kernelcore
= true;
5925 return cmdline_parse_core(p
, &required_kernelcore
);
5929 * movablecore=size sets the amount of memory for use for allocations that
5930 * can be reclaimed or migrated.
5932 static int __init
cmdline_parse_movablecore(char *p
)
5934 return cmdline_parse_core(p
, &required_movablecore
);
5937 early_param("kernelcore", cmdline_parse_kernelcore
);
5938 early_param("movablecore", cmdline_parse_movablecore
);
5940 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5942 void adjust_managed_page_count(struct page
*page
, long count
)
5944 spin_lock(&managed_page_count_lock
);
5945 page_zone(page
)->managed_pages
+= count
;
5946 totalram_pages
+= count
;
5947 #ifdef CONFIG_HIGHMEM
5948 if (PageHighMem(page
))
5949 totalhigh_pages
+= count
;
5951 spin_unlock(&managed_page_count_lock
);
5953 EXPORT_SYMBOL(adjust_managed_page_count
);
5955 unsigned long free_reserved_area(void *start
, void *end
, int poison
, char *s
)
5958 unsigned long pages
= 0;
5960 start
= (void *)PAGE_ALIGN((unsigned long)start
);
5961 end
= (void *)((unsigned long)end
& PAGE_MASK
);
5962 for (pos
= start
; pos
< end
; pos
+= PAGE_SIZE
, pages
++) {
5963 if ((unsigned int)poison
<= 0xFF)
5964 memset(pos
, poison
, PAGE_SIZE
);
5965 free_reserved_page(virt_to_page(pos
));
5969 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
5970 s
, pages
<< (PAGE_SHIFT
- 10), start
, end
);
5974 EXPORT_SYMBOL(free_reserved_area
);
5976 #ifdef CONFIG_HIGHMEM
5977 void free_highmem_page(struct page
*page
)
5979 __free_reserved_page(page
);
5981 page_zone(page
)->managed_pages
++;
5987 void __init
mem_init_print_info(const char *str
)
5989 unsigned long physpages
, codesize
, datasize
, rosize
, bss_size
;
5990 unsigned long init_code_size
, init_data_size
;
5992 physpages
= get_num_physpages();
5993 codesize
= _etext
- _stext
;
5994 datasize
= _edata
- _sdata
;
5995 rosize
= __end_rodata
- __start_rodata
;
5996 bss_size
= __bss_stop
- __bss_start
;
5997 init_data_size
= __init_end
- __init_begin
;
5998 init_code_size
= _einittext
- _sinittext
;
6001 * Detect special cases and adjust section sizes accordingly:
6002 * 1) .init.* may be embedded into .data sections
6003 * 2) .init.text.* may be out of [__init_begin, __init_end],
6004 * please refer to arch/tile/kernel/vmlinux.lds.S.
6005 * 3) .rodata.* may be embedded into .text or .data sections.
6007 #define adj_init_size(start, end, size, pos, adj) \
6009 if (start <= pos && pos < end && size > adj) \
6013 adj_init_size(__init_begin
, __init_end
, init_data_size
,
6014 _sinittext
, init_code_size
);
6015 adj_init_size(_stext
, _etext
, codesize
, _sinittext
, init_code_size
);
6016 adj_init_size(_sdata
, _edata
, datasize
, __init_begin
, init_data_size
);
6017 adj_init_size(_stext
, _etext
, codesize
, __start_rodata
, rosize
);
6018 adj_init_size(_sdata
, _edata
, datasize
, __start_rodata
, rosize
);
6020 #undef adj_init_size
6022 pr_info("Memory: %luK/%luK available "
6023 "(%luK kernel code, %luK rwdata, %luK rodata, "
6024 "%luK init, %luK bss, %luK reserved, %luK cma-reserved"
6025 #ifdef CONFIG_HIGHMEM
6029 nr_free_pages() << (PAGE_SHIFT
-10), physpages
<< (PAGE_SHIFT
-10),
6030 codesize
>> 10, datasize
>> 10, rosize
>> 10,
6031 (init_data_size
+ init_code_size
) >> 10, bss_size
>> 10,
6032 (physpages
- totalram_pages
- totalcma_pages
) << (PAGE_SHIFT
-10),
6033 totalcma_pages
<< (PAGE_SHIFT
-10),
6034 #ifdef CONFIG_HIGHMEM
6035 totalhigh_pages
<< (PAGE_SHIFT
-10),
6037 str
? ", " : "", str
? str
: "");
6041 * set_dma_reserve - set the specified number of pages reserved in the first zone
6042 * @new_dma_reserve: The number of pages to mark reserved
6044 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
6045 * In the DMA zone, a significant percentage may be consumed by kernel image
6046 * and other unfreeable allocations which can skew the watermarks badly. This
6047 * function may optionally be used to account for unfreeable pages in the
6048 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6049 * smaller per-cpu batchsize.
6051 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
6053 dma_reserve
= new_dma_reserve
;
6056 void __init
free_area_init(unsigned long *zones_size
)
6058 free_area_init_node(0, zones_size
,
6059 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
6062 static int page_alloc_cpu_notify(struct notifier_block
*self
,
6063 unsigned long action
, void *hcpu
)
6065 int cpu
= (unsigned long)hcpu
;
6067 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
6068 lru_add_drain_cpu(cpu
);
6072 * Spill the event counters of the dead processor
6073 * into the current processors event counters.
6074 * This artificially elevates the count of the current
6077 vm_events_fold_cpu(cpu
);
6080 * Zero the differential counters of the dead processor
6081 * so that the vm statistics are consistent.
6083 * This is only okay since the processor is dead and cannot
6084 * race with what we are doing.
6086 cpu_vm_stats_fold(cpu
);
6091 void __init
page_alloc_init(void)
6093 hotcpu_notifier(page_alloc_cpu_notify
, 0);
6097 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
6098 * or min_free_kbytes changes.
6100 static void calculate_totalreserve_pages(void)
6102 struct pglist_data
*pgdat
;
6103 unsigned long reserve_pages
= 0;
6104 enum zone_type i
, j
;
6106 for_each_online_pgdat(pgdat
) {
6107 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
6108 struct zone
*zone
= pgdat
->node_zones
+ i
;
6111 /* Find valid and maximum lowmem_reserve in the zone */
6112 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
6113 if (zone
->lowmem_reserve
[j
] > max
)
6114 max
= zone
->lowmem_reserve
[j
];
6117 /* we treat the high watermark as reserved pages. */
6118 max
+= high_wmark_pages(zone
);
6120 if (max
> zone
->managed_pages
)
6121 max
= zone
->managed_pages
;
6123 zone
->totalreserve_pages
= max
;
6125 reserve_pages
+= max
;
6128 totalreserve_pages
= reserve_pages
;
6132 * setup_per_zone_lowmem_reserve - called whenever
6133 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
6134 * has a correct pages reserved value, so an adequate number of
6135 * pages are left in the zone after a successful __alloc_pages().
6137 static void setup_per_zone_lowmem_reserve(void)
6139 struct pglist_data
*pgdat
;
6140 enum zone_type j
, idx
;
6142 for_each_online_pgdat(pgdat
) {
6143 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
6144 struct zone
*zone
= pgdat
->node_zones
+ j
;
6145 unsigned long managed_pages
= zone
->managed_pages
;
6147 zone
->lowmem_reserve
[j
] = 0;
6151 struct zone
*lower_zone
;
6155 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
6156 sysctl_lowmem_reserve_ratio
[idx
] = 1;
6158 lower_zone
= pgdat
->node_zones
+ idx
;
6159 lower_zone
->lowmem_reserve
[j
] = managed_pages
/
6160 sysctl_lowmem_reserve_ratio
[idx
];
6161 managed_pages
+= lower_zone
->managed_pages
;
6166 /* update totalreserve_pages */
6167 calculate_totalreserve_pages();
6170 static void __setup_per_zone_wmarks(void)
6172 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
6173 unsigned long lowmem_pages
= 0;
6175 unsigned long flags
;
6177 /* Calculate total number of !ZONE_HIGHMEM pages */
6178 for_each_zone(zone
) {
6179 if (!is_highmem(zone
))
6180 lowmem_pages
+= zone
->managed_pages
;
6183 for_each_zone(zone
) {
6186 spin_lock_irqsave(&zone
->lock
, flags
);
6187 tmp
= (u64
)pages_min
* zone
->managed_pages
;
6188 do_div(tmp
, lowmem_pages
);
6189 if (is_highmem(zone
)) {
6191 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6192 * need highmem pages, so cap pages_min to a small
6195 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
6196 * deltas control asynch page reclaim, and so should
6197 * not be capped for highmem.
6199 unsigned long min_pages
;
6201 min_pages
= zone
->managed_pages
/ 1024;
6202 min_pages
= clamp(min_pages
, SWAP_CLUSTER_MAX
, 128UL);
6203 zone
->watermark
[WMARK_MIN
] = min_pages
;
6206 * If it's a lowmem zone, reserve a number of pages
6207 * proportionate to the zone's size.
6209 zone
->watermark
[WMARK_MIN
] = tmp
;
6212 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + (tmp
>> 2);
6213 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + (tmp
>> 1);
6215 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
6216 high_wmark_pages(zone
) - low_wmark_pages(zone
) -
6217 atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]));
6219 spin_unlock_irqrestore(&zone
->lock
, flags
);
6222 /* update totalreserve_pages */
6223 calculate_totalreserve_pages();
6227 * setup_per_zone_wmarks - called when min_free_kbytes changes
6228 * or when memory is hot-{added|removed}
6230 * Ensures that the watermark[min,low,high] values for each zone are set
6231 * correctly with respect to min_free_kbytes.
6233 void setup_per_zone_wmarks(void)
6235 mutex_lock(&zonelists_mutex
);
6236 __setup_per_zone_wmarks();
6237 mutex_unlock(&zonelists_mutex
);
6241 * The inactive anon list should be small enough that the VM never has to
6242 * do too much work, but large enough that each inactive page has a chance
6243 * to be referenced again before it is swapped out.
6245 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
6246 * INACTIVE_ANON pages on this zone's LRU, maintained by the
6247 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
6248 * the anonymous pages are kept on the inactive list.
6251 * memory ratio inactive anon
6252 * -------------------------------------
6261 static void __meminit
calculate_zone_inactive_ratio(struct zone
*zone
)
6263 unsigned int gb
, ratio
;
6265 /* Zone size in gigabytes */
6266 gb
= zone
->managed_pages
>> (30 - PAGE_SHIFT
);
6268 ratio
= int_sqrt(10 * gb
);
6272 zone
->inactive_ratio
= ratio
;
6275 static void __meminit
setup_per_zone_inactive_ratio(void)
6280 calculate_zone_inactive_ratio(zone
);
6284 * Initialise min_free_kbytes.
6286 * For small machines we want it small (128k min). For large machines
6287 * we want it large (64MB max). But it is not linear, because network
6288 * bandwidth does not increase linearly with machine size. We use
6290 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
6291 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6307 int __meminit
init_per_zone_wmark_min(void)
6309 unsigned long lowmem_kbytes
;
6310 int new_min_free_kbytes
;
6312 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
6313 new_min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
6315 if (new_min_free_kbytes
> user_min_free_kbytes
) {
6316 min_free_kbytes
= new_min_free_kbytes
;
6317 if (min_free_kbytes
< 128)
6318 min_free_kbytes
= 128;
6319 if (min_free_kbytes
> 65536)
6320 min_free_kbytes
= 65536;
6322 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6323 new_min_free_kbytes
, user_min_free_kbytes
);
6325 setup_per_zone_wmarks();
6326 refresh_zone_stat_thresholds();
6327 setup_per_zone_lowmem_reserve();
6328 setup_per_zone_inactive_ratio();
6331 module_init(init_per_zone_wmark_min
)
6334 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
6335 * that we can call two helper functions whenever min_free_kbytes
6338 int min_free_kbytes_sysctl_handler(struct ctl_table
*table
, int write
,
6339 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6343 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6348 user_min_free_kbytes
= min_free_kbytes
;
6349 setup_per_zone_wmarks();
6355 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6356 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6361 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6366 zone
->min_unmapped_pages
= (zone
->managed_pages
*
6367 sysctl_min_unmapped_ratio
) / 100;
6371 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6372 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6377 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6382 zone
->min_slab_pages
= (zone
->managed_pages
*
6383 sysctl_min_slab_ratio
) / 100;
6389 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6390 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6391 * whenever sysctl_lowmem_reserve_ratio changes.
6393 * The reserve ratio obviously has absolutely no relation with the
6394 * minimum watermarks. The lowmem reserve ratio can only make sense
6395 * if in function of the boot time zone sizes.
6397 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6398 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6400 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6401 setup_per_zone_lowmem_reserve();
6406 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
6407 * cpu. It is the fraction of total pages in each zone that a hot per cpu
6408 * pagelist can have before it gets flushed back to buddy allocator.
6410 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table
*table
, int write
,
6411 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6414 int old_percpu_pagelist_fraction
;
6417 mutex_lock(&pcp_batch_high_lock
);
6418 old_percpu_pagelist_fraction
= percpu_pagelist_fraction
;
6420 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6421 if (!write
|| ret
< 0)
6424 /* Sanity checking to avoid pcp imbalance */
6425 if (percpu_pagelist_fraction
&&
6426 percpu_pagelist_fraction
< MIN_PERCPU_PAGELIST_FRACTION
) {
6427 percpu_pagelist_fraction
= old_percpu_pagelist_fraction
;
6433 if (percpu_pagelist_fraction
== old_percpu_pagelist_fraction
)
6436 for_each_populated_zone(zone
) {
6439 for_each_possible_cpu(cpu
)
6440 pageset_set_high_and_batch(zone
,
6441 per_cpu_ptr(zone
->pageset
, cpu
));
6444 mutex_unlock(&pcp_batch_high_lock
);
6449 int hashdist
= HASHDIST_DEFAULT
;
6451 static int __init
set_hashdist(char *str
)
6455 hashdist
= simple_strtoul(str
, &str
, 0);
6458 __setup("hashdist=", set_hashdist
);
6462 * allocate a large system hash table from bootmem
6463 * - it is assumed that the hash table must contain an exact power-of-2
6464 * quantity of entries
6465 * - limit is the number of hash buckets, not the total allocation size
6467 void *__init
alloc_large_system_hash(const char *tablename
,
6468 unsigned long bucketsize
,
6469 unsigned long numentries
,
6472 unsigned int *_hash_shift
,
6473 unsigned int *_hash_mask
,
6474 unsigned long low_limit
,
6475 unsigned long high_limit
)
6477 unsigned long long max
= high_limit
;
6478 unsigned long log2qty
, size
;
6481 /* allow the kernel cmdline to have a say */
6483 /* round applicable memory size up to nearest megabyte */
6484 numentries
= nr_kernel_pages
;
6486 /* It isn't necessary when PAGE_SIZE >= 1MB */
6487 if (PAGE_SHIFT
< 20)
6488 numentries
= round_up(numentries
, (1<<20)/PAGE_SIZE
);
6490 /* limit to 1 bucket per 2^scale bytes of low memory */
6491 if (scale
> PAGE_SHIFT
)
6492 numentries
>>= (scale
- PAGE_SHIFT
);
6494 numentries
<<= (PAGE_SHIFT
- scale
);
6496 /* Make sure we've got at least a 0-order allocation.. */
6497 if (unlikely(flags
& HASH_SMALL
)) {
6498 /* Makes no sense without HASH_EARLY */
6499 WARN_ON(!(flags
& HASH_EARLY
));
6500 if (!(numentries
>> *_hash_shift
)) {
6501 numentries
= 1UL << *_hash_shift
;
6502 BUG_ON(!numentries
);
6504 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
6505 numentries
= PAGE_SIZE
/ bucketsize
;
6507 numentries
= roundup_pow_of_two(numentries
);
6509 /* limit allocation size to 1/16 total memory by default */
6511 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
6512 do_div(max
, bucketsize
);
6514 max
= min(max
, 0x80000000ULL
);
6516 if (numentries
< low_limit
)
6517 numentries
= low_limit
;
6518 if (numentries
> max
)
6521 log2qty
= ilog2(numentries
);
6524 size
= bucketsize
<< log2qty
;
6525 if (flags
& HASH_EARLY
)
6526 table
= memblock_virt_alloc_nopanic(size
, 0);
6528 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
6531 * If bucketsize is not a power-of-two, we may free
6532 * some pages at the end of hash table which
6533 * alloc_pages_exact() automatically does
6535 if (get_order(size
) < MAX_ORDER
) {
6536 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
6537 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
6540 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
6543 panic("Failed to allocate %s hash table\n", tablename
);
6545 printk(KERN_INFO
"%s hash table entries: %ld (order: %d, %lu bytes)\n",
6548 ilog2(size
) - PAGE_SHIFT
,
6552 *_hash_shift
= log2qty
;
6554 *_hash_mask
= (1 << log2qty
) - 1;
6559 /* Return a pointer to the bitmap storing bits affecting a block of pages */
6560 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
6563 #ifdef CONFIG_SPARSEMEM
6564 return __pfn_to_section(pfn
)->pageblock_flags
;
6566 return zone
->pageblock_flags
;
6567 #endif /* CONFIG_SPARSEMEM */
6570 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
6572 #ifdef CONFIG_SPARSEMEM
6573 pfn
&= (PAGES_PER_SECTION
-1);
6574 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
6576 pfn
= pfn
- round_down(zone
->zone_start_pfn
, pageblock_nr_pages
);
6577 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
6578 #endif /* CONFIG_SPARSEMEM */
6582 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
6583 * @page: The page within the block of interest
6584 * @pfn: The target page frame number
6585 * @end_bitidx: The last bit of interest to retrieve
6586 * @mask: mask of bits that the caller is interested in
6588 * Return: pageblock_bits flags
6590 unsigned long get_pfnblock_flags_mask(struct page
*page
, unsigned long pfn
,
6591 unsigned long end_bitidx
,
6595 unsigned long *bitmap
;
6596 unsigned long bitidx
, word_bitidx
;
6599 zone
= page_zone(page
);
6600 bitmap
= get_pageblock_bitmap(zone
, pfn
);
6601 bitidx
= pfn_to_bitidx(zone
, pfn
);
6602 word_bitidx
= bitidx
/ BITS_PER_LONG
;
6603 bitidx
&= (BITS_PER_LONG
-1);
6605 word
= bitmap
[word_bitidx
];
6606 bitidx
+= end_bitidx
;
6607 return (word
>> (BITS_PER_LONG
- bitidx
- 1)) & mask
;
6611 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
6612 * @page: The page within the block of interest
6613 * @flags: The flags to set
6614 * @pfn: The target page frame number
6615 * @end_bitidx: The last bit of interest
6616 * @mask: mask of bits that the caller is interested in
6618 void set_pfnblock_flags_mask(struct page
*page
, unsigned long flags
,
6620 unsigned long end_bitidx
,
6624 unsigned long *bitmap
;
6625 unsigned long bitidx
, word_bitidx
;
6626 unsigned long old_word
, word
;
6628 BUILD_BUG_ON(NR_PAGEBLOCK_BITS
!= 4);
6630 zone
= page_zone(page
);
6631 bitmap
= get_pageblock_bitmap(zone
, pfn
);
6632 bitidx
= pfn_to_bitidx(zone
, pfn
);
6633 word_bitidx
= bitidx
/ BITS_PER_LONG
;
6634 bitidx
&= (BITS_PER_LONG
-1);
6636 VM_BUG_ON_PAGE(!zone_spans_pfn(zone
, pfn
), page
);
6638 bitidx
+= end_bitidx
;
6639 mask
<<= (BITS_PER_LONG
- bitidx
- 1);
6640 flags
<<= (BITS_PER_LONG
- bitidx
- 1);
6642 word
= READ_ONCE(bitmap
[word_bitidx
]);
6644 old_word
= cmpxchg(&bitmap
[word_bitidx
], word
, (word
& ~mask
) | flags
);
6645 if (word
== old_word
)
6652 * This function checks whether pageblock includes unmovable pages or not.
6653 * If @count is not zero, it is okay to include less @count unmovable pages
6655 * PageLRU check without isolation or lru_lock could race so that
6656 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6657 * expect this function should be exact.
6659 bool has_unmovable_pages(struct zone
*zone
, struct page
*page
, int count
,
6660 bool skip_hwpoisoned_pages
)
6662 unsigned long pfn
, iter
, found
;
6666 * For avoiding noise data, lru_add_drain_all() should be called
6667 * If ZONE_MOVABLE, the zone never contains unmovable pages
6669 if (zone_idx(zone
) == ZONE_MOVABLE
)
6671 mt
= get_pageblock_migratetype(page
);
6672 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
6675 pfn
= page_to_pfn(page
);
6676 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
6677 unsigned long check
= pfn
+ iter
;
6679 if (!pfn_valid_within(check
))
6682 page
= pfn_to_page(check
);
6685 * Hugepages are not in LRU lists, but they're movable.
6686 * We need not scan over tail pages bacause we don't
6687 * handle each tail page individually in migration.
6689 if (PageHuge(page
)) {
6690 iter
= round_up(iter
+ 1, 1<<compound_order(page
)) - 1;
6695 * We can't use page_count without pin a page
6696 * because another CPU can free compound page.
6697 * This check already skips compound tails of THP
6698 * because their page->_count is zero at all time.
6700 if (!atomic_read(&page
->_count
)) {
6701 if (PageBuddy(page
))
6702 iter
+= (1 << page_order(page
)) - 1;
6707 * The HWPoisoned page may be not in buddy system, and
6708 * page_count() is not 0.
6710 if (skip_hwpoisoned_pages
&& PageHWPoison(page
))
6716 * If there are RECLAIMABLE pages, we need to check
6717 * it. But now, memory offline itself doesn't call
6718 * shrink_node_slabs() and it still to be fixed.
6721 * If the page is not RAM, page_count()should be 0.
6722 * we don't need more check. This is an _used_ not-movable page.
6724 * The problematic thing here is PG_reserved pages. PG_reserved
6725 * is set to both of a memory hole page and a _used_ kernel
6734 bool is_pageblock_removable_nolock(struct page
*page
)
6740 * We have to be careful here because we are iterating over memory
6741 * sections which are not zone aware so we might end up outside of
6742 * the zone but still within the section.
6743 * We have to take care about the node as well. If the node is offline
6744 * its NODE_DATA will be NULL - see page_zone.
6746 if (!node_online(page_to_nid(page
)))
6749 zone
= page_zone(page
);
6750 pfn
= page_to_pfn(page
);
6751 if (!zone_spans_pfn(zone
, pfn
))
6754 return !has_unmovable_pages(zone
, page
, 0, true);
6757 #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
6759 static unsigned long pfn_max_align_down(unsigned long pfn
)
6761 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
6762 pageblock_nr_pages
) - 1);
6765 static unsigned long pfn_max_align_up(unsigned long pfn
)
6767 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
6768 pageblock_nr_pages
));
6771 /* [start, end) must belong to a single zone. */
6772 static int __alloc_contig_migrate_range(struct compact_control
*cc
,
6773 unsigned long start
, unsigned long end
)
6775 /* This function is based on compact_zone() from compaction.c. */
6776 unsigned long nr_reclaimed
;
6777 unsigned long pfn
= start
;
6778 unsigned int tries
= 0;
6783 while (pfn
< end
|| !list_empty(&cc
->migratepages
)) {
6784 if (fatal_signal_pending(current
)) {
6789 if (list_empty(&cc
->migratepages
)) {
6790 cc
->nr_migratepages
= 0;
6791 pfn
= isolate_migratepages_range(cc
, pfn
, end
);
6797 } else if (++tries
== 5) {
6798 ret
= ret
< 0 ? ret
: -EBUSY
;
6802 nr_reclaimed
= reclaim_clean_pages_from_list(cc
->zone
,
6804 cc
->nr_migratepages
-= nr_reclaimed
;
6806 ret
= migrate_pages(&cc
->migratepages
, alloc_migrate_target
,
6807 NULL
, 0, cc
->mode
, MR_CMA
);
6810 putback_movable_pages(&cc
->migratepages
);
6817 * alloc_contig_range() -- tries to allocate given range of pages
6818 * @start: start PFN to allocate
6819 * @end: one-past-the-last PFN to allocate
6820 * @migratetype: migratetype of the underlaying pageblocks (either
6821 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6822 * in range must have the same migratetype and it must
6823 * be either of the two.
6825 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6826 * aligned, however it's the caller's responsibility to guarantee that
6827 * we are the only thread that changes migrate type of pageblocks the
6830 * The PFN range must belong to a single zone.
6832 * Returns zero on success or negative error code. On success all
6833 * pages which PFN is in [start, end) are allocated for the caller and
6834 * need to be freed with free_contig_range().
6836 int alloc_contig_range(unsigned long start
, unsigned long end
,
6837 unsigned migratetype
)
6839 unsigned long outer_start
, outer_end
;
6843 struct compact_control cc
= {
6844 .nr_migratepages
= 0,
6846 .zone
= page_zone(pfn_to_page(start
)),
6847 .mode
= MIGRATE_SYNC
,
6848 .ignore_skip_hint
= true,
6850 INIT_LIST_HEAD(&cc
.migratepages
);
6853 * What we do here is we mark all pageblocks in range as
6854 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6855 * have different sizes, and due to the way page allocator
6856 * work, we align the range to biggest of the two pages so
6857 * that page allocator won't try to merge buddies from
6858 * different pageblocks and change MIGRATE_ISOLATE to some
6859 * other migration type.
6861 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6862 * migrate the pages from an unaligned range (ie. pages that
6863 * we are interested in). This will put all the pages in
6864 * range back to page allocator as MIGRATE_ISOLATE.
6866 * When this is done, we take the pages in range from page
6867 * allocator removing them from the buddy system. This way
6868 * page allocator will never consider using them.
6870 * This lets us mark the pageblocks back as
6871 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6872 * aligned range but not in the unaligned, original range are
6873 * put back to page allocator so that buddy can use them.
6876 ret
= start_isolate_page_range(pfn_max_align_down(start
),
6877 pfn_max_align_up(end
), migratetype
,
6883 * In case of -EBUSY, we'd like to know which page causes problem.
6884 * So, just fall through. We will check it in test_pages_isolated().
6886 ret
= __alloc_contig_migrate_range(&cc
, start
, end
);
6887 if (ret
&& ret
!= -EBUSY
)
6891 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6892 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6893 * more, all pages in [start, end) are free in page allocator.
6894 * What we are going to do is to allocate all pages from
6895 * [start, end) (that is remove them from page allocator).
6897 * The only problem is that pages at the beginning and at the
6898 * end of interesting range may be not aligned with pages that
6899 * page allocator holds, ie. they can be part of higher order
6900 * pages. Because of this, we reserve the bigger range and
6901 * once this is done free the pages we are not interested in.
6903 * We don't have to hold zone->lock here because the pages are
6904 * isolated thus they won't get removed from buddy.
6907 lru_add_drain_all();
6908 drain_all_pages(cc
.zone
);
6911 outer_start
= start
;
6912 while (!PageBuddy(pfn_to_page(outer_start
))) {
6913 if (++order
>= MAX_ORDER
) {
6914 outer_start
= start
;
6917 outer_start
&= ~0UL << order
;
6920 if (outer_start
!= start
) {
6921 order
= page_order(pfn_to_page(outer_start
));
6924 * outer_start page could be small order buddy page and
6925 * it doesn't include start page. Adjust outer_start
6926 * in this case to report failed page properly
6927 * on tracepoint in test_pages_isolated()
6929 if (outer_start
+ (1UL << order
) <= start
)
6930 outer_start
= start
;
6933 /* Make sure the range is really isolated. */
6934 if (test_pages_isolated(outer_start
, end
, false)) {
6935 pr_info("%s: [%lx, %lx) PFNs busy\n",
6936 __func__
, outer_start
, end
);
6941 /* Grab isolated pages from freelists. */
6942 outer_end
= isolate_freepages_range(&cc
, outer_start
, end
);
6948 /* Free head and tail (if any) */
6949 if (start
!= outer_start
)
6950 free_contig_range(outer_start
, start
- outer_start
);
6951 if (end
!= outer_end
)
6952 free_contig_range(end
, outer_end
- end
);
6955 undo_isolate_page_range(pfn_max_align_down(start
),
6956 pfn_max_align_up(end
), migratetype
);
6960 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
6962 unsigned int count
= 0;
6964 for (; nr_pages
--; pfn
++) {
6965 struct page
*page
= pfn_to_page(pfn
);
6967 count
+= page_count(page
) != 1;
6970 WARN(count
!= 0, "%d pages are still in use!\n", count
);
6974 #ifdef CONFIG_MEMORY_HOTPLUG
6976 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6977 * page high values need to be recalulated.
6979 void __meminit
zone_pcp_update(struct zone
*zone
)
6982 mutex_lock(&pcp_batch_high_lock
);
6983 for_each_possible_cpu(cpu
)
6984 pageset_set_high_and_batch(zone
,
6985 per_cpu_ptr(zone
->pageset
, cpu
));
6986 mutex_unlock(&pcp_batch_high_lock
);
6990 void zone_pcp_reset(struct zone
*zone
)
6992 unsigned long flags
;
6994 struct per_cpu_pageset
*pset
;
6996 /* avoid races with drain_pages() */
6997 local_irq_save(flags
);
6998 if (zone
->pageset
!= &boot_pageset
) {
6999 for_each_online_cpu(cpu
) {
7000 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
7001 drain_zonestat(zone
, pset
);
7003 free_percpu(zone
->pageset
);
7004 zone
->pageset
= &boot_pageset
;
7006 local_irq_restore(flags
);
7009 #ifdef CONFIG_MEMORY_HOTREMOVE
7011 * All pages in the range must be isolated before calling this.
7014 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
7018 unsigned int order
, i
;
7020 unsigned long flags
;
7021 /* find the first valid pfn */
7022 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
7027 zone
= page_zone(pfn_to_page(pfn
));
7028 spin_lock_irqsave(&zone
->lock
, flags
);
7030 while (pfn
< end_pfn
) {
7031 if (!pfn_valid(pfn
)) {
7035 page
= pfn_to_page(pfn
);
7037 * The HWPoisoned page may be not in buddy system, and
7038 * page_count() is not 0.
7040 if (unlikely(!PageBuddy(page
) && PageHWPoison(page
))) {
7042 SetPageReserved(page
);
7046 BUG_ON(page_count(page
));
7047 BUG_ON(!PageBuddy(page
));
7048 order
= page_order(page
);
7049 #ifdef CONFIG_DEBUG_VM
7050 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
7051 pfn
, 1 << order
, end_pfn
);
7053 list_del(&page
->lru
);
7054 rmv_page_order(page
);
7055 zone
->free_area
[order
].nr_free
--;
7056 for (i
= 0; i
< (1 << order
); i
++)
7057 SetPageReserved((page
+i
));
7058 pfn
+= (1 << order
);
7060 spin_unlock_irqrestore(&zone
->lock
, flags
);
7064 #ifdef CONFIG_MEMORY_FAILURE
7065 bool is_free_buddy_page(struct page
*page
)
7067 struct zone
*zone
= page_zone(page
);
7068 unsigned long pfn
= page_to_pfn(page
);
7069 unsigned long flags
;
7072 spin_lock_irqsave(&zone
->lock
, flags
);
7073 for (order
= 0; order
< MAX_ORDER
; order
++) {
7074 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
7076 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
7079 spin_unlock_irqrestore(&zone
->lock
, flags
);
7081 return order
< MAX_ORDER
;