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 pr_alert("BUG: Bad page state in process %s pfn:%05lx\n",
444 current
->comm
, page_to_pfn(page
));
445 __dump_page(page
, reason
);
446 bad_flags
&= page
->flags
;
448 pr_alert("bad because of flags: %#lx(%pGp)\n",
449 bad_flags
, &bad_flags
);
450 dump_page_owner(page
);
455 /* Leave bad fields for debug, except PageBuddy could make trouble */
456 page_mapcount_reset(page
); /* remove PageBuddy */
457 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
461 * Higher-order pages are called "compound pages". They are structured thusly:
463 * The first PAGE_SIZE page is called the "head page" and have PG_head set.
465 * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
466 * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
468 * The first tail page's ->compound_dtor holds the offset in array of compound
469 * page destructors. See compound_page_dtors.
471 * The first tail page's ->compound_order holds the order of allocation.
472 * This usage means that zero-order pages may not be compound.
475 void free_compound_page(struct page
*page
)
477 __free_pages_ok(page
, compound_order(page
));
480 void prep_compound_page(struct page
*page
, unsigned int order
)
483 int nr_pages
= 1 << order
;
485 set_compound_page_dtor(page
, COMPOUND_PAGE_DTOR
);
486 set_compound_order(page
, order
);
488 for (i
= 1; i
< nr_pages
; i
++) {
489 struct page
*p
= page
+ i
;
490 set_page_count(p
, 0);
491 p
->mapping
= TAIL_MAPPING
;
492 set_compound_head(p
, page
);
494 atomic_set(compound_mapcount_ptr(page
), -1);
497 #ifdef CONFIG_DEBUG_PAGEALLOC
498 unsigned int _debug_guardpage_minorder
;
499 bool _debug_pagealloc_enabled __read_mostly
500 = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT
);
501 bool _debug_guardpage_enabled __read_mostly
;
503 static int __init
early_debug_pagealloc(char *buf
)
508 if (strcmp(buf
, "on") == 0)
509 _debug_pagealloc_enabled
= true;
511 if (strcmp(buf
, "off") == 0)
512 _debug_pagealloc_enabled
= false;
516 early_param("debug_pagealloc", early_debug_pagealloc
);
518 static bool need_debug_guardpage(void)
520 /* If we don't use debug_pagealloc, we don't need guard page */
521 if (!debug_pagealloc_enabled())
527 static void init_debug_guardpage(void)
529 if (!debug_pagealloc_enabled())
532 _debug_guardpage_enabled
= true;
535 struct page_ext_operations debug_guardpage_ops
= {
536 .need
= need_debug_guardpage
,
537 .init
= init_debug_guardpage
,
540 static int __init
debug_guardpage_minorder_setup(char *buf
)
544 if (kstrtoul(buf
, 10, &res
) < 0 || res
> MAX_ORDER
/ 2) {
545 printk(KERN_ERR
"Bad debug_guardpage_minorder value\n");
548 _debug_guardpage_minorder
= res
;
549 printk(KERN_INFO
"Setting debug_guardpage_minorder to %lu\n", res
);
552 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup
);
554 static inline void set_page_guard(struct zone
*zone
, struct page
*page
,
555 unsigned int order
, int migratetype
)
557 struct page_ext
*page_ext
;
559 if (!debug_guardpage_enabled())
562 page_ext
= lookup_page_ext(page
);
563 __set_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
565 INIT_LIST_HEAD(&page
->lru
);
566 set_page_private(page
, order
);
567 /* Guard pages are not available for any usage */
568 __mod_zone_freepage_state(zone
, -(1 << order
), migratetype
);
571 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
572 unsigned int order
, int migratetype
)
574 struct page_ext
*page_ext
;
576 if (!debug_guardpage_enabled())
579 page_ext
= lookup_page_ext(page
);
580 __clear_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
582 set_page_private(page
, 0);
583 if (!is_migrate_isolate(migratetype
))
584 __mod_zone_freepage_state(zone
, (1 << order
), migratetype
);
587 struct page_ext_operations debug_guardpage_ops
= { NULL
, };
588 static inline void set_page_guard(struct zone
*zone
, struct page
*page
,
589 unsigned int order
, int migratetype
) {}
590 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
591 unsigned int order
, int migratetype
) {}
594 static inline void set_page_order(struct page
*page
, unsigned int order
)
596 set_page_private(page
, order
);
597 __SetPageBuddy(page
);
600 static inline void rmv_page_order(struct page
*page
)
602 __ClearPageBuddy(page
);
603 set_page_private(page
, 0);
607 * This function checks whether a page is free && is the buddy
608 * we can do coalesce a page and its buddy if
609 * (a) the buddy is not in a hole &&
610 * (b) the buddy is in the buddy system &&
611 * (c) a page and its buddy have the same order &&
612 * (d) a page and its buddy are in the same zone.
614 * For recording whether a page is in the buddy system, we set ->_mapcount
615 * PAGE_BUDDY_MAPCOUNT_VALUE.
616 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
617 * serialized by zone->lock.
619 * For recording page's order, we use page_private(page).
621 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
624 if (!pfn_valid_within(page_to_pfn(buddy
)))
627 if (page_is_guard(buddy
) && page_order(buddy
) == order
) {
628 if (page_zone_id(page
) != page_zone_id(buddy
))
631 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
636 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
638 * zone check is done late to avoid uselessly
639 * calculating zone/node ids for pages that could
642 if (page_zone_id(page
) != page_zone_id(buddy
))
645 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
653 * Freeing function for a buddy system allocator.
655 * The concept of a buddy system is to maintain direct-mapped table
656 * (containing bit values) for memory blocks of various "orders".
657 * The bottom level table contains the map for the smallest allocatable
658 * units of memory (here, pages), and each level above it describes
659 * pairs of units from the levels below, hence, "buddies".
660 * At a high level, all that happens here is marking the table entry
661 * at the bottom level available, and propagating the changes upward
662 * as necessary, plus some accounting needed to play nicely with other
663 * parts of the VM system.
664 * At each level, we keep a list of pages, which are heads of continuous
665 * free pages of length of (1 << order) and marked with _mapcount
666 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
668 * So when we are allocating or freeing one, we can derive the state of the
669 * other. That is, if we allocate a small block, and both were
670 * free, the remainder of the region must be split into blocks.
671 * If a block is freed, and its buddy is also free, then this
672 * triggers coalescing into a block of larger size.
677 static inline void __free_one_page(struct page
*page
,
679 struct zone
*zone
, unsigned int order
,
682 unsigned long page_idx
;
683 unsigned long combined_idx
;
684 unsigned long uninitialized_var(buddy_idx
);
686 unsigned int max_order
= MAX_ORDER
;
688 VM_BUG_ON(!zone_is_initialized(zone
));
689 VM_BUG_ON_PAGE(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
, page
);
691 VM_BUG_ON(migratetype
== -1);
692 if (is_migrate_isolate(migratetype
)) {
694 * We restrict max order of merging to prevent merge
695 * between freepages on isolate pageblock and normal
696 * pageblock. Without this, pageblock isolation
697 * could cause incorrect freepage accounting.
699 max_order
= min_t(unsigned int, MAX_ORDER
, pageblock_order
+ 1);
701 __mod_zone_freepage_state(zone
, 1 << order
, migratetype
);
704 page_idx
= pfn
& ((1 << max_order
) - 1);
706 VM_BUG_ON_PAGE(page_idx
& ((1 << order
) - 1), page
);
707 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
709 while (order
< max_order
- 1) {
710 buddy_idx
= __find_buddy_index(page_idx
, order
);
711 buddy
= page
+ (buddy_idx
- page_idx
);
712 if (!page_is_buddy(page
, buddy
, order
))
715 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
716 * merge with it and move up one order.
718 if (page_is_guard(buddy
)) {
719 clear_page_guard(zone
, buddy
, order
, migratetype
);
721 list_del(&buddy
->lru
);
722 zone
->free_area
[order
].nr_free
--;
723 rmv_page_order(buddy
);
725 combined_idx
= buddy_idx
& page_idx
;
726 page
= page
+ (combined_idx
- page_idx
);
727 page_idx
= combined_idx
;
730 set_page_order(page
, order
);
733 * If this is not the largest possible page, check if the buddy
734 * of the next-highest order is free. If it is, it's possible
735 * that pages are being freed that will coalesce soon. In case,
736 * that is happening, add the free page to the tail of the list
737 * so it's less likely to be used soon and more likely to be merged
738 * as a higher order page
740 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
741 struct page
*higher_page
, *higher_buddy
;
742 combined_idx
= buddy_idx
& page_idx
;
743 higher_page
= page
+ (combined_idx
- page_idx
);
744 buddy_idx
= __find_buddy_index(combined_idx
, order
+ 1);
745 higher_buddy
= higher_page
+ (buddy_idx
- combined_idx
);
746 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
747 list_add_tail(&page
->lru
,
748 &zone
->free_area
[order
].free_list
[migratetype
]);
753 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
755 zone
->free_area
[order
].nr_free
++;
758 static inline int free_pages_check(struct page
*page
)
760 const char *bad_reason
= NULL
;
761 unsigned long bad_flags
= 0;
763 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
764 bad_reason
= "nonzero mapcount";
765 if (unlikely(page
->mapping
!= NULL
))
766 bad_reason
= "non-NULL mapping";
767 if (unlikely(atomic_read(&page
->_count
) != 0))
768 bad_reason
= "nonzero _count";
769 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
)) {
770 bad_reason
= "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
771 bad_flags
= PAGE_FLAGS_CHECK_AT_FREE
;
774 if (unlikely(page
->mem_cgroup
))
775 bad_reason
= "page still charged to cgroup";
777 if (unlikely(bad_reason
)) {
778 bad_page(page
, bad_reason
, bad_flags
);
781 page_cpupid_reset_last(page
);
782 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
783 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
788 * Frees a number of pages from the PCP lists
789 * Assumes all pages on list are in same zone, and of same order.
790 * count is the number of pages to free.
792 * If the zone was previously in an "all pages pinned" state then look to
793 * see if this freeing clears that state.
795 * And clear the zone's pages_scanned counter, to hold off the "all pages are
796 * pinned" detection logic.
798 static void free_pcppages_bulk(struct zone
*zone
, int count
,
799 struct per_cpu_pages
*pcp
)
804 unsigned long nr_scanned
;
806 spin_lock(&zone
->lock
);
807 nr_scanned
= zone_page_state(zone
, NR_PAGES_SCANNED
);
809 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, -nr_scanned
);
813 struct list_head
*list
;
816 * Remove pages from lists in a round-robin fashion. A
817 * batch_free count is maintained that is incremented when an
818 * empty list is encountered. This is so more pages are freed
819 * off fuller lists instead of spinning excessively around empty
824 if (++migratetype
== MIGRATE_PCPTYPES
)
826 list
= &pcp
->lists
[migratetype
];
827 } while (list_empty(list
));
829 /* This is the only non-empty list. Free them all. */
830 if (batch_free
== MIGRATE_PCPTYPES
)
831 batch_free
= to_free
;
834 int mt
; /* migratetype of the to-be-freed page */
836 page
= list_last_entry(list
, struct page
, lru
);
837 /* must delete as __free_one_page list manipulates */
838 list_del(&page
->lru
);
840 mt
= get_pcppage_migratetype(page
);
841 /* MIGRATE_ISOLATE page should not go to pcplists */
842 VM_BUG_ON_PAGE(is_migrate_isolate(mt
), page
);
843 /* Pageblock could have been isolated meanwhile */
844 if (unlikely(has_isolate_pageblock(zone
)))
845 mt
= get_pageblock_migratetype(page
);
847 __free_one_page(page
, page_to_pfn(page
), zone
, 0, mt
);
848 trace_mm_page_pcpu_drain(page
, 0, mt
);
849 } while (--to_free
&& --batch_free
&& !list_empty(list
));
851 spin_unlock(&zone
->lock
);
854 static void free_one_page(struct zone
*zone
,
855 struct page
*page
, unsigned long pfn
,
859 unsigned long nr_scanned
;
860 spin_lock(&zone
->lock
);
861 nr_scanned
= zone_page_state(zone
, NR_PAGES_SCANNED
);
863 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, -nr_scanned
);
865 if (unlikely(has_isolate_pageblock(zone
) ||
866 is_migrate_isolate(migratetype
))) {
867 migratetype
= get_pfnblock_migratetype(page
, pfn
);
869 __free_one_page(page
, pfn
, zone
, order
, migratetype
);
870 spin_unlock(&zone
->lock
);
873 static int free_tail_pages_check(struct page
*head_page
, struct page
*page
)
878 * We rely page->lru.next never has bit 0 set, unless the page
879 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
881 BUILD_BUG_ON((unsigned long)LIST_POISON1
& 1);
883 if (!IS_ENABLED(CONFIG_DEBUG_VM
)) {
887 switch (page
- head_page
) {
889 /* the first tail page: ->mapping is compound_mapcount() */
890 if (unlikely(compound_mapcount(page
))) {
891 bad_page(page
, "nonzero compound_mapcount", 0);
897 * the second tail page: ->mapping is
898 * page_deferred_list().next -- ignore value.
902 if (page
->mapping
!= TAIL_MAPPING
) {
903 bad_page(page
, "corrupted mapping in tail page", 0);
908 if (unlikely(!PageTail(page
))) {
909 bad_page(page
, "PageTail not set", 0);
912 if (unlikely(compound_head(page
) != head_page
)) {
913 bad_page(page
, "compound_head not consistent", 0);
918 page
->mapping
= NULL
;
919 clear_compound_head(page
);
923 static void __meminit
__init_single_page(struct page
*page
, unsigned long pfn
,
924 unsigned long zone
, int nid
)
926 set_page_links(page
, zone
, nid
, pfn
);
927 init_page_count(page
);
928 page_mapcount_reset(page
);
929 page_cpupid_reset_last(page
);
931 INIT_LIST_HEAD(&page
->lru
);
932 #ifdef WANT_PAGE_VIRTUAL
933 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
934 if (!is_highmem_idx(zone
))
935 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
939 static void __meminit
__init_single_pfn(unsigned long pfn
, unsigned long zone
,
942 return __init_single_page(pfn_to_page(pfn
), pfn
, zone
, nid
);
945 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
946 static void init_reserved_page(unsigned long pfn
)
951 if (!early_page_uninitialised(pfn
))
954 nid
= early_pfn_to_nid(pfn
);
955 pgdat
= NODE_DATA(nid
);
957 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
958 struct zone
*zone
= &pgdat
->node_zones
[zid
];
960 if (pfn
>= zone
->zone_start_pfn
&& pfn
< zone_end_pfn(zone
))
963 __init_single_pfn(pfn
, zid
, nid
);
966 static inline void init_reserved_page(unsigned long pfn
)
969 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
972 * Initialised pages do not have PageReserved set. This function is
973 * called for each range allocated by the bootmem allocator and
974 * marks the pages PageReserved. The remaining valid pages are later
975 * sent to the buddy page allocator.
977 void __meminit
reserve_bootmem_region(unsigned long start
, unsigned long end
)
979 unsigned long start_pfn
= PFN_DOWN(start
);
980 unsigned long end_pfn
= PFN_UP(end
);
982 for (; start_pfn
< end_pfn
; start_pfn
++) {
983 if (pfn_valid(start_pfn
)) {
984 struct page
*page
= pfn_to_page(start_pfn
);
986 init_reserved_page(start_pfn
);
988 /* Avoid false-positive PageTail() */
989 INIT_LIST_HEAD(&page
->lru
);
991 SetPageReserved(page
);
996 static bool free_pages_prepare(struct page
*page
, unsigned int order
)
998 bool compound
= PageCompound(page
);
1001 VM_BUG_ON_PAGE(PageTail(page
), page
);
1002 VM_BUG_ON_PAGE(compound
&& compound_order(page
) != order
, page
);
1004 trace_mm_page_free(page
, order
);
1005 kmemcheck_free_shadow(page
, order
);
1006 kasan_free_pages(page
, order
);
1009 page
->mapping
= NULL
;
1010 bad
+= free_pages_check(page
);
1011 for (i
= 1; i
< (1 << order
); i
++) {
1013 bad
+= free_tail_pages_check(page
, page
+ i
);
1014 bad
+= free_pages_check(page
+ i
);
1019 reset_page_owner(page
, order
);
1021 if (!PageHighMem(page
)) {
1022 debug_check_no_locks_freed(page_address(page
),
1023 PAGE_SIZE
<< order
);
1024 debug_check_no_obj_freed(page_address(page
),
1025 PAGE_SIZE
<< order
);
1027 arch_free_page(page
, order
);
1028 kernel_poison_pages(page
, 1 << order
, 0);
1029 kernel_map_pages(page
, 1 << order
, 0);
1034 static void __free_pages_ok(struct page
*page
, unsigned int order
)
1036 unsigned long flags
;
1038 unsigned long pfn
= page_to_pfn(page
);
1040 if (!free_pages_prepare(page
, order
))
1043 migratetype
= get_pfnblock_migratetype(page
, pfn
);
1044 local_irq_save(flags
);
1045 __count_vm_events(PGFREE
, 1 << order
);
1046 free_one_page(page_zone(page
), page
, pfn
, order
, migratetype
);
1047 local_irq_restore(flags
);
1050 static void __init
__free_pages_boot_core(struct page
*page
,
1051 unsigned long pfn
, unsigned int order
)
1053 unsigned int nr_pages
= 1 << order
;
1054 struct page
*p
= page
;
1058 for (loop
= 0; loop
< (nr_pages
- 1); loop
++, p
++) {
1060 __ClearPageReserved(p
);
1061 set_page_count(p
, 0);
1063 __ClearPageReserved(p
);
1064 set_page_count(p
, 0);
1066 page_zone(page
)->managed_pages
+= nr_pages
;
1067 set_page_refcounted(page
);
1068 __free_pages(page
, order
);
1071 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1072 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1074 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata
;
1076 int __meminit
early_pfn_to_nid(unsigned long pfn
)
1078 static DEFINE_SPINLOCK(early_pfn_lock
);
1081 spin_lock(&early_pfn_lock
);
1082 nid
= __early_pfn_to_nid(pfn
, &early_pfnnid_cache
);
1085 spin_unlock(&early_pfn_lock
);
1091 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1092 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1093 struct mminit_pfnnid_cache
*state
)
1097 nid
= __early_pfn_to_nid(pfn
, state
);
1098 if (nid
>= 0 && nid
!= node
)
1103 /* Only safe to use early in boot when initialisation is single-threaded */
1104 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1106 return meminit_pfn_in_nid(pfn
, node
, &early_pfnnid_cache
);
1111 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1115 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1116 struct mminit_pfnnid_cache
*state
)
1123 void __init
__free_pages_bootmem(struct page
*page
, unsigned long pfn
,
1126 if (early_page_uninitialised(pfn
))
1128 return __free_pages_boot_core(page
, pfn
, order
);
1132 * Check that the whole (or subset of) a pageblock given by the interval of
1133 * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1134 * with the migration of free compaction scanner. The scanners then need to
1135 * use only pfn_valid_within() check for arches that allow holes within
1138 * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1140 * It's possible on some configurations to have a setup like node0 node1 node0
1141 * i.e. it's possible that all pages within a zones range of pages do not
1142 * belong to a single zone. We assume that a border between node0 and node1
1143 * can occur within a single pageblock, but not a node0 node1 node0
1144 * interleaving within a single pageblock. It is therefore sufficient to check
1145 * the first and last page of a pageblock and avoid checking each individual
1146 * page in a pageblock.
1148 struct page
*__pageblock_pfn_to_page(unsigned long start_pfn
,
1149 unsigned long end_pfn
, struct zone
*zone
)
1151 struct page
*start_page
;
1152 struct page
*end_page
;
1154 /* end_pfn is one past the range we are checking */
1157 if (!pfn_valid(start_pfn
) || !pfn_valid(end_pfn
))
1160 start_page
= pfn_to_page(start_pfn
);
1162 if (page_zone(start_page
) != zone
)
1165 end_page
= pfn_to_page(end_pfn
);
1167 /* This gives a shorter code than deriving page_zone(end_page) */
1168 if (page_zone_id(start_page
) != page_zone_id(end_page
))
1174 void set_zone_contiguous(struct zone
*zone
)
1176 unsigned long block_start_pfn
= zone
->zone_start_pfn
;
1177 unsigned long block_end_pfn
;
1179 block_end_pfn
= ALIGN(block_start_pfn
+ 1, pageblock_nr_pages
);
1180 for (; block_start_pfn
< zone_end_pfn(zone
);
1181 block_start_pfn
= block_end_pfn
,
1182 block_end_pfn
+= pageblock_nr_pages
) {
1184 block_end_pfn
= min(block_end_pfn
, zone_end_pfn(zone
));
1186 if (!__pageblock_pfn_to_page(block_start_pfn
,
1187 block_end_pfn
, zone
))
1191 /* We confirm that there is no hole */
1192 zone
->contiguous
= true;
1195 void clear_zone_contiguous(struct zone
*zone
)
1197 zone
->contiguous
= false;
1200 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1201 static void __init
deferred_free_range(struct page
*page
,
1202 unsigned long pfn
, int nr_pages
)
1209 /* Free a large naturally-aligned chunk if possible */
1210 if (nr_pages
== MAX_ORDER_NR_PAGES
&&
1211 (pfn
& (MAX_ORDER_NR_PAGES
-1)) == 0) {
1212 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
1213 __free_pages_boot_core(page
, pfn
, MAX_ORDER
-1);
1217 for (i
= 0; i
< nr_pages
; i
++, page
++, pfn
++)
1218 __free_pages_boot_core(page
, pfn
, 0);
1221 /* Completion tracking for deferred_init_memmap() threads */
1222 static atomic_t pgdat_init_n_undone __initdata
;
1223 static __initdata
DECLARE_COMPLETION(pgdat_init_all_done_comp
);
1225 static inline void __init
pgdat_init_report_one_done(void)
1227 if (atomic_dec_and_test(&pgdat_init_n_undone
))
1228 complete(&pgdat_init_all_done_comp
);
1231 /* Initialise remaining memory on a node */
1232 static int __init
deferred_init_memmap(void *data
)
1234 pg_data_t
*pgdat
= data
;
1235 int nid
= pgdat
->node_id
;
1236 struct mminit_pfnnid_cache nid_init_state
= { };
1237 unsigned long start
= jiffies
;
1238 unsigned long nr_pages
= 0;
1239 unsigned long walk_start
, walk_end
;
1242 unsigned long first_init_pfn
= pgdat
->first_deferred_pfn
;
1243 const struct cpumask
*cpumask
= cpumask_of_node(pgdat
->node_id
);
1245 if (first_init_pfn
== ULONG_MAX
) {
1246 pgdat_init_report_one_done();
1250 /* Bind memory initialisation thread to a local node if possible */
1251 if (!cpumask_empty(cpumask
))
1252 set_cpus_allowed_ptr(current
, cpumask
);
1254 /* Sanity check boundaries */
1255 BUG_ON(pgdat
->first_deferred_pfn
< pgdat
->node_start_pfn
);
1256 BUG_ON(pgdat
->first_deferred_pfn
> pgdat_end_pfn(pgdat
));
1257 pgdat
->first_deferred_pfn
= ULONG_MAX
;
1259 /* Only the highest zone is deferred so find it */
1260 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
1261 zone
= pgdat
->node_zones
+ zid
;
1262 if (first_init_pfn
< zone_end_pfn(zone
))
1266 for_each_mem_pfn_range(i
, nid
, &walk_start
, &walk_end
, NULL
) {
1267 unsigned long pfn
, end_pfn
;
1268 struct page
*page
= NULL
;
1269 struct page
*free_base_page
= NULL
;
1270 unsigned long free_base_pfn
= 0;
1273 end_pfn
= min(walk_end
, zone_end_pfn(zone
));
1274 pfn
= first_init_pfn
;
1275 if (pfn
< walk_start
)
1277 if (pfn
< zone
->zone_start_pfn
)
1278 pfn
= zone
->zone_start_pfn
;
1280 for (; pfn
< end_pfn
; pfn
++) {
1281 if (!pfn_valid_within(pfn
))
1285 * Ensure pfn_valid is checked every
1286 * MAX_ORDER_NR_PAGES for memory holes
1288 if ((pfn
& (MAX_ORDER_NR_PAGES
- 1)) == 0) {
1289 if (!pfn_valid(pfn
)) {
1295 if (!meminit_pfn_in_nid(pfn
, nid
, &nid_init_state
)) {
1300 /* Minimise pfn page lookups and scheduler checks */
1301 if (page
&& (pfn
& (MAX_ORDER_NR_PAGES
- 1)) != 0) {
1304 nr_pages
+= nr_to_free
;
1305 deferred_free_range(free_base_page
,
1306 free_base_pfn
, nr_to_free
);
1307 free_base_page
= NULL
;
1308 free_base_pfn
= nr_to_free
= 0;
1310 page
= pfn_to_page(pfn
);
1315 VM_BUG_ON(page_zone(page
) != zone
);
1319 __init_single_page(page
, pfn
, zid
, nid
);
1320 if (!free_base_page
) {
1321 free_base_page
= page
;
1322 free_base_pfn
= pfn
;
1327 /* Where possible, batch up pages for a single free */
1330 /* Free the current block of pages to allocator */
1331 nr_pages
+= nr_to_free
;
1332 deferred_free_range(free_base_page
, free_base_pfn
,
1334 free_base_page
= NULL
;
1335 free_base_pfn
= nr_to_free
= 0;
1338 first_init_pfn
= max(end_pfn
, first_init_pfn
);
1341 /* Sanity check that the next zone really is unpopulated */
1342 WARN_ON(++zid
< MAX_NR_ZONES
&& populated_zone(++zone
));
1344 pr_info("node %d initialised, %lu pages in %ums\n", nid
, nr_pages
,
1345 jiffies_to_msecs(jiffies
- start
));
1347 pgdat_init_report_one_done();
1350 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1352 void __init
page_alloc_init_late(void)
1356 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1359 /* There will be num_node_state(N_MEMORY) threads */
1360 atomic_set(&pgdat_init_n_undone
, num_node_state(N_MEMORY
));
1361 for_each_node_state(nid
, N_MEMORY
) {
1362 kthread_run(deferred_init_memmap
, NODE_DATA(nid
), "pgdatinit%d", nid
);
1365 /* Block until all are initialised */
1366 wait_for_completion(&pgdat_init_all_done_comp
);
1368 /* Reinit limits that are based on free pages after the kernel is up */
1369 files_maxfiles_init();
1372 for_each_populated_zone(zone
)
1373 set_zone_contiguous(zone
);
1377 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1378 void __init
init_cma_reserved_pageblock(struct page
*page
)
1380 unsigned i
= pageblock_nr_pages
;
1381 struct page
*p
= page
;
1384 __ClearPageReserved(p
);
1385 set_page_count(p
, 0);
1388 set_pageblock_migratetype(page
, MIGRATE_CMA
);
1390 if (pageblock_order
>= MAX_ORDER
) {
1391 i
= pageblock_nr_pages
;
1394 set_page_refcounted(p
);
1395 __free_pages(p
, MAX_ORDER
- 1);
1396 p
+= MAX_ORDER_NR_PAGES
;
1397 } while (i
-= MAX_ORDER_NR_PAGES
);
1399 set_page_refcounted(page
);
1400 __free_pages(page
, pageblock_order
);
1403 adjust_managed_page_count(page
, pageblock_nr_pages
);
1408 * The order of subdivision here is critical for the IO subsystem.
1409 * Please do not alter this order without good reasons and regression
1410 * testing. Specifically, as large blocks of memory are subdivided,
1411 * the order in which smaller blocks are delivered depends on the order
1412 * they're subdivided in this function. This is the primary factor
1413 * influencing the order in which pages are delivered to the IO
1414 * subsystem according to empirical testing, and this is also justified
1415 * by considering the behavior of a buddy system containing a single
1416 * large block of memory acted on by a series of small allocations.
1417 * This behavior is a critical factor in sglist merging's success.
1421 static inline void expand(struct zone
*zone
, struct page
*page
,
1422 int low
, int high
, struct free_area
*area
,
1425 unsigned long size
= 1 << high
;
1427 while (high
> low
) {
1431 VM_BUG_ON_PAGE(bad_range(zone
, &page
[size
]), &page
[size
]);
1433 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC
) &&
1434 debug_guardpage_enabled() &&
1435 high
< debug_guardpage_minorder()) {
1437 * Mark as guard pages (or page), that will allow to
1438 * merge back to allocator when buddy will be freed.
1439 * Corresponding page table entries will not be touched,
1440 * pages will stay not present in virtual address space
1442 set_page_guard(zone
, &page
[size
], high
, migratetype
);
1445 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
1447 set_page_order(&page
[size
], high
);
1452 * This page is about to be returned from the page allocator
1454 static inline int check_new_page(struct page
*page
)
1456 const char *bad_reason
= NULL
;
1457 unsigned long bad_flags
= 0;
1459 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
1460 bad_reason
= "nonzero mapcount";
1461 if (unlikely(page
->mapping
!= NULL
))
1462 bad_reason
= "non-NULL mapping";
1463 if (unlikely(atomic_read(&page
->_count
) != 0))
1464 bad_reason
= "nonzero _count";
1465 if (unlikely(page
->flags
& __PG_HWPOISON
)) {
1466 bad_reason
= "HWPoisoned (hardware-corrupted)";
1467 bad_flags
= __PG_HWPOISON
;
1469 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)) {
1470 bad_reason
= "PAGE_FLAGS_CHECK_AT_PREP flag set";
1471 bad_flags
= PAGE_FLAGS_CHECK_AT_PREP
;
1474 if (unlikely(page
->mem_cgroup
))
1475 bad_reason
= "page still charged to cgroup";
1477 if (unlikely(bad_reason
)) {
1478 bad_page(page
, bad_reason
, bad_flags
);
1484 static inline bool free_pages_prezeroed(bool poisoned
)
1486 return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO
) &&
1487 page_poisoning_enabled() && poisoned
;
1490 static int prep_new_page(struct page
*page
, unsigned int order
, gfp_t gfp_flags
,
1494 bool poisoned
= true;
1496 for (i
= 0; i
< (1 << order
); i
++) {
1497 struct page
*p
= page
+ i
;
1498 if (unlikely(check_new_page(p
)))
1501 poisoned
&= page_is_poisoned(p
);
1504 set_page_private(page
, 0);
1505 set_page_refcounted(page
);
1507 arch_alloc_page(page
, order
);
1508 kernel_map_pages(page
, 1 << order
, 1);
1509 kernel_poison_pages(page
, 1 << order
, 1);
1510 kasan_alloc_pages(page
, order
);
1512 if (!free_pages_prezeroed(poisoned
) && (gfp_flags
& __GFP_ZERO
))
1513 for (i
= 0; i
< (1 << order
); i
++)
1514 clear_highpage(page
+ i
);
1516 if (order
&& (gfp_flags
& __GFP_COMP
))
1517 prep_compound_page(page
, order
);
1519 set_page_owner(page
, order
, gfp_flags
);
1522 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1523 * allocate the page. The expectation is that the caller is taking
1524 * steps that will free more memory. The caller should avoid the page
1525 * being used for !PFMEMALLOC purposes.
1527 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
1528 set_page_pfmemalloc(page
);
1530 clear_page_pfmemalloc(page
);
1536 * Go through the free lists for the given migratetype and remove
1537 * the smallest available page from the freelists
1540 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
1543 unsigned int current_order
;
1544 struct free_area
*area
;
1547 /* Find a page of the appropriate size in the preferred list */
1548 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
1549 area
= &(zone
->free_area
[current_order
]);
1550 page
= list_first_entry_or_null(&area
->free_list
[migratetype
],
1554 list_del(&page
->lru
);
1555 rmv_page_order(page
);
1557 expand(zone
, page
, order
, current_order
, area
, migratetype
);
1558 set_pcppage_migratetype(page
, migratetype
);
1567 * This array describes the order lists are fallen back to when
1568 * the free lists for the desirable migrate type are depleted
1570 static int fallbacks
[MIGRATE_TYPES
][4] = {
1571 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1572 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1573 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_TYPES
},
1575 [MIGRATE_CMA
] = { MIGRATE_TYPES
}, /* Never used */
1577 #ifdef CONFIG_MEMORY_ISOLATION
1578 [MIGRATE_ISOLATE
] = { MIGRATE_TYPES
}, /* Never used */
1583 static struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1586 return __rmqueue_smallest(zone
, order
, MIGRATE_CMA
);
1589 static inline struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1590 unsigned int order
) { return NULL
; }
1594 * Move the free pages in a range to the free lists of the requested type.
1595 * Note that start_page and end_pages are not aligned on a pageblock
1596 * boundary. If alignment is required, use move_freepages_block()
1598 int move_freepages(struct zone
*zone
,
1599 struct page
*start_page
, struct page
*end_page
,
1604 int pages_moved
= 0;
1606 #ifndef CONFIG_HOLES_IN_ZONE
1608 * page_zone is not safe to call in this context when
1609 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1610 * anyway as we check zone boundaries in move_freepages_block().
1611 * Remove at a later date when no bug reports exist related to
1612 * grouping pages by mobility
1614 VM_BUG_ON(page_zone(start_page
) != page_zone(end_page
));
1617 for (page
= start_page
; page
<= end_page
;) {
1618 /* Make sure we are not inadvertently changing nodes */
1619 VM_BUG_ON_PAGE(page_to_nid(page
) != zone_to_nid(zone
), page
);
1621 if (!pfn_valid_within(page_to_pfn(page
))) {
1626 if (!PageBuddy(page
)) {
1631 order
= page_order(page
);
1632 list_move(&page
->lru
,
1633 &zone
->free_area
[order
].free_list
[migratetype
]);
1635 pages_moved
+= 1 << order
;
1641 int move_freepages_block(struct zone
*zone
, struct page
*page
,
1644 unsigned long start_pfn
, end_pfn
;
1645 struct page
*start_page
, *end_page
;
1647 start_pfn
= page_to_pfn(page
);
1648 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
1649 start_page
= pfn_to_page(start_pfn
);
1650 end_page
= start_page
+ pageblock_nr_pages
- 1;
1651 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
1653 /* Do not cross zone boundaries */
1654 if (!zone_spans_pfn(zone
, start_pfn
))
1656 if (!zone_spans_pfn(zone
, end_pfn
))
1659 return move_freepages(zone
, start_page
, end_page
, migratetype
);
1662 static void change_pageblock_range(struct page
*pageblock_page
,
1663 int start_order
, int migratetype
)
1665 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
1667 while (nr_pageblocks
--) {
1668 set_pageblock_migratetype(pageblock_page
, migratetype
);
1669 pageblock_page
+= pageblock_nr_pages
;
1674 * When we are falling back to another migratetype during allocation, try to
1675 * steal extra free pages from the same pageblocks to satisfy further
1676 * allocations, instead of polluting multiple pageblocks.
1678 * If we are stealing a relatively large buddy page, it is likely there will
1679 * be more free pages in the pageblock, so try to steal them all. For
1680 * reclaimable and unmovable allocations, we steal regardless of page size,
1681 * as fragmentation caused by those allocations polluting movable pageblocks
1682 * is worse than movable allocations stealing from unmovable and reclaimable
1685 static bool can_steal_fallback(unsigned int order
, int start_mt
)
1688 * Leaving this order check is intended, although there is
1689 * relaxed order check in next check. The reason is that
1690 * we can actually steal whole pageblock if this condition met,
1691 * but, below check doesn't guarantee it and that is just heuristic
1692 * so could be changed anytime.
1694 if (order
>= pageblock_order
)
1697 if (order
>= pageblock_order
/ 2 ||
1698 start_mt
== MIGRATE_RECLAIMABLE
||
1699 start_mt
== MIGRATE_UNMOVABLE
||
1700 page_group_by_mobility_disabled
)
1707 * This function implements actual steal behaviour. If order is large enough,
1708 * we can steal whole pageblock. If not, we first move freepages in this
1709 * pageblock and check whether half of pages are moved or not. If half of
1710 * pages are moved, we can change migratetype of pageblock and permanently
1711 * use it's pages as requested migratetype in the future.
1713 static void steal_suitable_fallback(struct zone
*zone
, struct page
*page
,
1716 unsigned int current_order
= page_order(page
);
1719 /* Take ownership for orders >= pageblock_order */
1720 if (current_order
>= pageblock_order
) {
1721 change_pageblock_range(page
, current_order
, start_type
);
1725 pages
= move_freepages_block(zone
, page
, start_type
);
1727 /* Claim the whole block if over half of it is free */
1728 if (pages
>= (1 << (pageblock_order
-1)) ||
1729 page_group_by_mobility_disabled
)
1730 set_pageblock_migratetype(page
, start_type
);
1734 * Check whether there is a suitable fallback freepage with requested order.
1735 * If only_stealable is true, this function returns fallback_mt only if
1736 * we can steal other freepages all together. This would help to reduce
1737 * fragmentation due to mixed migratetype pages in one pageblock.
1739 int find_suitable_fallback(struct free_area
*area
, unsigned int order
,
1740 int migratetype
, bool only_stealable
, bool *can_steal
)
1745 if (area
->nr_free
== 0)
1750 fallback_mt
= fallbacks
[migratetype
][i
];
1751 if (fallback_mt
== MIGRATE_TYPES
)
1754 if (list_empty(&area
->free_list
[fallback_mt
]))
1757 if (can_steal_fallback(order
, migratetype
))
1760 if (!only_stealable
)
1771 * Reserve a pageblock for exclusive use of high-order atomic allocations if
1772 * there are no empty page blocks that contain a page with a suitable order
1774 static void reserve_highatomic_pageblock(struct page
*page
, struct zone
*zone
,
1775 unsigned int alloc_order
)
1778 unsigned long max_managed
, flags
;
1781 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
1782 * Check is race-prone but harmless.
1784 max_managed
= (zone
->managed_pages
/ 100) + pageblock_nr_pages
;
1785 if (zone
->nr_reserved_highatomic
>= max_managed
)
1788 spin_lock_irqsave(&zone
->lock
, flags
);
1790 /* Recheck the nr_reserved_highatomic limit under the lock */
1791 if (zone
->nr_reserved_highatomic
>= max_managed
)
1795 mt
= get_pageblock_migratetype(page
);
1796 if (mt
!= MIGRATE_HIGHATOMIC
&&
1797 !is_migrate_isolate(mt
) && !is_migrate_cma(mt
)) {
1798 zone
->nr_reserved_highatomic
+= pageblock_nr_pages
;
1799 set_pageblock_migratetype(page
, MIGRATE_HIGHATOMIC
);
1800 move_freepages_block(zone
, page
, MIGRATE_HIGHATOMIC
);
1804 spin_unlock_irqrestore(&zone
->lock
, flags
);
1808 * Used when an allocation is about to fail under memory pressure. This
1809 * potentially hurts the reliability of high-order allocations when under
1810 * intense memory pressure but failed atomic allocations should be easier
1811 * to recover from than an OOM.
1813 static void unreserve_highatomic_pageblock(const struct alloc_context
*ac
)
1815 struct zonelist
*zonelist
= ac
->zonelist
;
1816 unsigned long flags
;
1822 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
, ac
->high_zoneidx
,
1824 /* Preserve at least one pageblock */
1825 if (zone
->nr_reserved_highatomic
<= pageblock_nr_pages
)
1828 spin_lock_irqsave(&zone
->lock
, flags
);
1829 for (order
= 0; order
< MAX_ORDER
; order
++) {
1830 struct free_area
*area
= &(zone
->free_area
[order
]);
1832 page
= list_first_entry_or_null(
1833 &area
->free_list
[MIGRATE_HIGHATOMIC
],
1839 * It should never happen but changes to locking could
1840 * inadvertently allow a per-cpu drain to add pages
1841 * to MIGRATE_HIGHATOMIC while unreserving so be safe
1842 * and watch for underflows.
1844 zone
->nr_reserved_highatomic
-= min(pageblock_nr_pages
,
1845 zone
->nr_reserved_highatomic
);
1848 * Convert to ac->migratetype and avoid the normal
1849 * pageblock stealing heuristics. Minimally, the caller
1850 * is doing the work and needs the pages. More
1851 * importantly, if the block was always converted to
1852 * MIGRATE_UNMOVABLE or another type then the number
1853 * of pageblocks that cannot be completely freed
1856 set_pageblock_migratetype(page
, ac
->migratetype
);
1857 move_freepages_block(zone
, page
, ac
->migratetype
);
1858 spin_unlock_irqrestore(&zone
->lock
, flags
);
1861 spin_unlock_irqrestore(&zone
->lock
, flags
);
1865 /* Remove an element from the buddy allocator from the fallback list */
1866 static inline struct page
*
1867 __rmqueue_fallback(struct zone
*zone
, unsigned int order
, int start_migratetype
)
1869 struct free_area
*area
;
1870 unsigned int current_order
;
1875 /* Find the largest possible block of pages in the other list */
1876 for (current_order
= MAX_ORDER
-1;
1877 current_order
>= order
&& current_order
<= MAX_ORDER
-1;
1879 area
= &(zone
->free_area
[current_order
]);
1880 fallback_mt
= find_suitable_fallback(area
, current_order
,
1881 start_migratetype
, false, &can_steal
);
1882 if (fallback_mt
== -1)
1885 page
= list_first_entry(&area
->free_list
[fallback_mt
],
1888 steal_suitable_fallback(zone
, page
, start_migratetype
);
1890 /* Remove the page from the freelists */
1892 list_del(&page
->lru
);
1893 rmv_page_order(page
);
1895 expand(zone
, page
, order
, current_order
, area
,
1898 * The pcppage_migratetype may differ from pageblock's
1899 * migratetype depending on the decisions in
1900 * find_suitable_fallback(). This is OK as long as it does not
1901 * differ for MIGRATE_CMA pageblocks. Those can be used as
1902 * fallback only via special __rmqueue_cma_fallback() function
1904 set_pcppage_migratetype(page
, start_migratetype
);
1906 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
1907 start_migratetype
, fallback_mt
);
1916 * Do the hard work of removing an element from the buddy allocator.
1917 * Call me with the zone->lock already held.
1919 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
1924 page
= __rmqueue_smallest(zone
, order
, migratetype
);
1925 if (unlikely(!page
)) {
1926 if (migratetype
== MIGRATE_MOVABLE
)
1927 page
= __rmqueue_cma_fallback(zone
, order
);
1930 page
= __rmqueue_fallback(zone
, order
, migratetype
);
1933 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
1938 * Obtain a specified number of elements from the buddy allocator, all under
1939 * a single hold of the lock, for efficiency. Add them to the supplied list.
1940 * Returns the number of new pages which were placed at *list.
1942 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
1943 unsigned long count
, struct list_head
*list
,
1944 int migratetype
, bool cold
)
1948 spin_lock(&zone
->lock
);
1949 for (i
= 0; i
< count
; ++i
) {
1950 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
1951 if (unlikely(page
== NULL
))
1955 * Split buddy pages returned by expand() are received here
1956 * in physical page order. The page is added to the callers and
1957 * list and the list head then moves forward. From the callers
1958 * perspective, the linked list is ordered by page number in
1959 * some conditions. This is useful for IO devices that can
1960 * merge IO requests if the physical pages are ordered
1964 list_add(&page
->lru
, list
);
1966 list_add_tail(&page
->lru
, list
);
1968 if (is_migrate_cma(get_pcppage_migratetype(page
)))
1969 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
,
1972 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
1973 spin_unlock(&zone
->lock
);
1979 * Called from the vmstat counter updater to drain pagesets of this
1980 * currently executing processor on remote nodes after they have
1983 * Note that this function must be called with the thread pinned to
1984 * a single processor.
1986 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
1988 unsigned long flags
;
1989 int to_drain
, batch
;
1991 local_irq_save(flags
);
1992 batch
= READ_ONCE(pcp
->batch
);
1993 to_drain
= min(pcp
->count
, batch
);
1995 free_pcppages_bulk(zone
, to_drain
, pcp
);
1996 pcp
->count
-= to_drain
;
1998 local_irq_restore(flags
);
2003 * Drain pcplists of the indicated processor and zone.
2005 * The processor must either be the current processor and the
2006 * thread pinned to the current processor or a processor that
2009 static void drain_pages_zone(unsigned int cpu
, struct zone
*zone
)
2011 unsigned long flags
;
2012 struct per_cpu_pageset
*pset
;
2013 struct per_cpu_pages
*pcp
;
2015 local_irq_save(flags
);
2016 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
2020 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
2023 local_irq_restore(flags
);
2027 * Drain pcplists of all zones on the indicated processor.
2029 * The processor must either be the current processor and the
2030 * thread pinned to the current processor or a processor that
2033 static void drain_pages(unsigned int cpu
)
2037 for_each_populated_zone(zone
) {
2038 drain_pages_zone(cpu
, zone
);
2043 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
2045 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2046 * the single zone's pages.
2048 void drain_local_pages(struct zone
*zone
)
2050 int cpu
= smp_processor_id();
2053 drain_pages_zone(cpu
, zone
);
2059 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2061 * When zone parameter is non-NULL, spill just the single zone's pages.
2063 * Note that this code is protected against sending an IPI to an offline
2064 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
2065 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
2066 * nothing keeps CPUs from showing up after we populated the cpumask and
2067 * before the call to on_each_cpu_mask().
2069 void drain_all_pages(struct zone
*zone
)
2074 * Allocate in the BSS so we wont require allocation in
2075 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2077 static cpumask_t cpus_with_pcps
;
2080 * We don't care about racing with CPU hotplug event
2081 * as offline notification will cause the notified
2082 * cpu to drain that CPU pcps and on_each_cpu_mask
2083 * disables preemption as part of its processing
2085 for_each_online_cpu(cpu
) {
2086 struct per_cpu_pageset
*pcp
;
2088 bool has_pcps
= false;
2091 pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
2095 for_each_populated_zone(z
) {
2096 pcp
= per_cpu_ptr(z
->pageset
, cpu
);
2097 if (pcp
->pcp
.count
) {
2105 cpumask_set_cpu(cpu
, &cpus_with_pcps
);
2107 cpumask_clear_cpu(cpu
, &cpus_with_pcps
);
2109 on_each_cpu_mask(&cpus_with_pcps
, (smp_call_func_t
) drain_local_pages
,
2113 #ifdef CONFIG_HIBERNATION
2115 void mark_free_pages(struct zone
*zone
)
2117 unsigned long pfn
, max_zone_pfn
;
2118 unsigned long flags
;
2119 unsigned int order
, t
;
2122 if (zone_is_empty(zone
))
2125 spin_lock_irqsave(&zone
->lock
, flags
);
2127 max_zone_pfn
= zone_end_pfn(zone
);
2128 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
2129 if (pfn_valid(pfn
)) {
2130 page
= pfn_to_page(pfn
);
2131 if (!swsusp_page_is_forbidden(page
))
2132 swsusp_unset_page_free(page
);
2135 for_each_migratetype_order(order
, t
) {
2136 list_for_each_entry(page
,
2137 &zone
->free_area
[order
].free_list
[t
], lru
) {
2140 pfn
= page_to_pfn(page
);
2141 for (i
= 0; i
< (1UL << order
); i
++)
2142 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
2145 spin_unlock_irqrestore(&zone
->lock
, flags
);
2147 #endif /* CONFIG_PM */
2150 * Free a 0-order page
2151 * cold == true ? free a cold page : free a hot page
2153 void free_hot_cold_page(struct page
*page
, bool cold
)
2155 struct zone
*zone
= page_zone(page
);
2156 struct per_cpu_pages
*pcp
;
2157 unsigned long flags
;
2158 unsigned long pfn
= page_to_pfn(page
);
2161 if (!free_pages_prepare(page
, 0))
2164 migratetype
= get_pfnblock_migratetype(page
, pfn
);
2165 set_pcppage_migratetype(page
, migratetype
);
2166 local_irq_save(flags
);
2167 __count_vm_event(PGFREE
);
2170 * We only track unmovable, reclaimable and movable on pcp lists.
2171 * Free ISOLATE pages back to the allocator because they are being
2172 * offlined but treat RESERVE as movable pages so we can get those
2173 * areas back if necessary. Otherwise, we may have to free
2174 * excessively into the page allocator
2176 if (migratetype
>= MIGRATE_PCPTYPES
) {
2177 if (unlikely(is_migrate_isolate(migratetype
))) {
2178 free_one_page(zone
, page
, pfn
, 0, migratetype
);
2181 migratetype
= MIGRATE_MOVABLE
;
2184 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2186 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
2188 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
2190 if (pcp
->count
>= pcp
->high
) {
2191 unsigned long batch
= READ_ONCE(pcp
->batch
);
2192 free_pcppages_bulk(zone
, batch
, pcp
);
2193 pcp
->count
-= batch
;
2197 local_irq_restore(flags
);
2201 * Free a list of 0-order pages
2203 void free_hot_cold_page_list(struct list_head
*list
, bool cold
)
2205 struct page
*page
, *next
;
2207 list_for_each_entry_safe(page
, next
, list
, lru
) {
2208 trace_mm_page_free_batched(page
, cold
);
2209 free_hot_cold_page(page
, cold
);
2214 * split_page takes a non-compound higher-order page, and splits it into
2215 * n (1<<order) sub-pages: page[0..n]
2216 * Each sub-page must be freed individually.
2218 * Note: this is probably too low level an operation for use in drivers.
2219 * Please consult with lkml before using this in your driver.
2221 void split_page(struct page
*page
, unsigned int order
)
2226 VM_BUG_ON_PAGE(PageCompound(page
), page
);
2227 VM_BUG_ON_PAGE(!page_count(page
), page
);
2229 #ifdef CONFIG_KMEMCHECK
2231 * Split shadow pages too, because free(page[0]) would
2232 * otherwise free the whole shadow.
2234 if (kmemcheck_page_is_tracked(page
))
2235 split_page(virt_to_page(page
[0].shadow
), order
);
2238 gfp_mask
= get_page_owner_gfp(page
);
2239 set_page_owner(page
, 0, gfp_mask
);
2240 for (i
= 1; i
< (1 << order
); i
++) {
2241 set_page_refcounted(page
+ i
);
2242 set_page_owner(page
+ i
, 0, gfp_mask
);
2245 EXPORT_SYMBOL_GPL(split_page
);
2247 int __isolate_free_page(struct page
*page
, unsigned int order
)
2249 unsigned long watermark
;
2253 BUG_ON(!PageBuddy(page
));
2255 zone
= page_zone(page
);
2256 mt
= get_pageblock_migratetype(page
);
2258 if (!is_migrate_isolate(mt
)) {
2259 /* Obey watermarks as if the page was being allocated */
2260 watermark
= low_wmark_pages(zone
) + (1 << order
);
2261 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
2264 __mod_zone_freepage_state(zone
, -(1UL << order
), mt
);
2267 /* Remove page from free list */
2268 list_del(&page
->lru
);
2269 zone
->free_area
[order
].nr_free
--;
2270 rmv_page_order(page
);
2272 set_page_owner(page
, order
, __GFP_MOVABLE
);
2274 /* Set the pageblock if the isolated page is at least a pageblock */
2275 if (order
>= pageblock_order
- 1) {
2276 struct page
*endpage
= page
+ (1 << order
) - 1;
2277 for (; page
< endpage
; page
+= pageblock_nr_pages
) {
2278 int mt
= get_pageblock_migratetype(page
);
2279 if (!is_migrate_isolate(mt
) && !is_migrate_cma(mt
))
2280 set_pageblock_migratetype(page
,
2286 return 1UL << order
;
2290 * Similar to split_page except the page is already free. As this is only
2291 * being used for migration, the migratetype of the block also changes.
2292 * As this is called with interrupts disabled, the caller is responsible
2293 * for calling arch_alloc_page() and kernel_map_page() after interrupts
2296 * Note: this is probably too low level an operation for use in drivers.
2297 * Please consult with lkml before using this in your driver.
2299 int split_free_page(struct page
*page
)
2304 order
= page_order(page
);
2306 nr_pages
= __isolate_free_page(page
, order
);
2310 /* Split into individual pages */
2311 set_page_refcounted(page
);
2312 split_page(page
, order
);
2317 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2320 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
2321 struct zone
*zone
, unsigned int order
,
2322 gfp_t gfp_flags
, int alloc_flags
, int migratetype
)
2324 unsigned long flags
;
2326 bool cold
= ((gfp_flags
& __GFP_COLD
) != 0);
2328 if (likely(order
== 0)) {
2329 struct per_cpu_pages
*pcp
;
2330 struct list_head
*list
;
2332 local_irq_save(flags
);
2333 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2334 list
= &pcp
->lists
[migratetype
];
2335 if (list_empty(list
)) {
2336 pcp
->count
+= rmqueue_bulk(zone
, 0,
2339 if (unlikely(list_empty(list
)))
2344 page
= list_last_entry(list
, struct page
, lru
);
2346 page
= list_first_entry(list
, struct page
, lru
);
2348 list_del(&page
->lru
);
2351 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
2353 * __GFP_NOFAIL is not to be used in new code.
2355 * All __GFP_NOFAIL callers should be fixed so that they
2356 * properly detect and handle allocation failures.
2358 * We most definitely don't want callers attempting to
2359 * allocate greater than order-1 page units with
2362 WARN_ON_ONCE(order
> 1);
2364 spin_lock_irqsave(&zone
->lock
, flags
);
2367 if (alloc_flags
& ALLOC_HARDER
) {
2368 page
= __rmqueue_smallest(zone
, order
, MIGRATE_HIGHATOMIC
);
2370 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
2373 page
= __rmqueue(zone
, order
, migratetype
);
2374 spin_unlock(&zone
->lock
);
2377 __mod_zone_freepage_state(zone
, -(1 << order
),
2378 get_pcppage_migratetype(page
));
2381 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
, -(1 << order
));
2382 if (atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]) <= 0 &&
2383 !test_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
))
2384 set_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
);
2386 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
2387 zone_statistics(preferred_zone
, zone
, gfp_flags
);
2388 local_irq_restore(flags
);
2390 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
2394 local_irq_restore(flags
);
2398 #ifdef CONFIG_FAIL_PAGE_ALLOC
2401 struct fault_attr attr
;
2403 bool ignore_gfp_highmem
;
2404 bool ignore_gfp_reclaim
;
2406 } fail_page_alloc
= {
2407 .attr
= FAULT_ATTR_INITIALIZER
,
2408 .ignore_gfp_reclaim
= true,
2409 .ignore_gfp_highmem
= true,
2413 static int __init
setup_fail_page_alloc(char *str
)
2415 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
2417 __setup("fail_page_alloc=", setup_fail_page_alloc
);
2419 static bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2421 if (order
< fail_page_alloc
.min_order
)
2423 if (gfp_mask
& __GFP_NOFAIL
)
2425 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
2427 if (fail_page_alloc
.ignore_gfp_reclaim
&&
2428 (gfp_mask
& __GFP_DIRECT_RECLAIM
))
2431 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
2434 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2436 static int __init
fail_page_alloc_debugfs(void)
2438 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
2441 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
2442 &fail_page_alloc
.attr
);
2444 return PTR_ERR(dir
);
2446 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
2447 &fail_page_alloc
.ignore_gfp_reclaim
))
2449 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
2450 &fail_page_alloc
.ignore_gfp_highmem
))
2452 if (!debugfs_create_u32("min-order", mode
, dir
,
2453 &fail_page_alloc
.min_order
))
2458 debugfs_remove_recursive(dir
);
2463 late_initcall(fail_page_alloc_debugfs
);
2465 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2467 #else /* CONFIG_FAIL_PAGE_ALLOC */
2469 static inline bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2474 #endif /* CONFIG_FAIL_PAGE_ALLOC */
2477 * Return true if free base pages are above 'mark'. For high-order checks it
2478 * will return true of the order-0 watermark is reached and there is at least
2479 * one free page of a suitable size. Checking now avoids taking the zone lock
2480 * to check in the allocation paths if no pages are free.
2482 static bool __zone_watermark_ok(struct zone
*z
, unsigned int order
,
2483 unsigned long mark
, int classzone_idx
, int alloc_flags
,
2488 const int alloc_harder
= (alloc_flags
& ALLOC_HARDER
);
2490 /* free_pages may go negative - that's OK */
2491 free_pages
-= (1 << order
) - 1;
2493 if (alloc_flags
& ALLOC_HIGH
)
2497 * If the caller does not have rights to ALLOC_HARDER then subtract
2498 * the high-atomic reserves. This will over-estimate the size of the
2499 * atomic reserve but it avoids a search.
2501 if (likely(!alloc_harder
))
2502 free_pages
-= z
->nr_reserved_highatomic
;
2507 /* If allocation can't use CMA areas don't use free CMA pages */
2508 if (!(alloc_flags
& ALLOC_CMA
))
2509 free_pages
-= zone_page_state(z
, NR_FREE_CMA_PAGES
);
2513 * Check watermarks for an order-0 allocation request. If these
2514 * are not met, then a high-order request also cannot go ahead
2515 * even if a suitable page happened to be free.
2517 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
2520 /* If this is an order-0 request then the watermark is fine */
2524 /* For a high-order request, check at least one suitable page is free */
2525 for (o
= order
; o
< MAX_ORDER
; o
++) {
2526 struct free_area
*area
= &z
->free_area
[o
];
2535 for (mt
= 0; mt
< MIGRATE_PCPTYPES
; mt
++) {
2536 if (!list_empty(&area
->free_list
[mt
]))
2541 if ((alloc_flags
& ALLOC_CMA
) &&
2542 !list_empty(&area
->free_list
[MIGRATE_CMA
])) {
2550 bool zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
2551 int classzone_idx
, int alloc_flags
)
2553 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
2554 zone_page_state(z
, NR_FREE_PAGES
));
2557 bool zone_watermark_ok_safe(struct zone
*z
, unsigned int order
,
2558 unsigned long mark
, int classzone_idx
)
2560 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
2562 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
2563 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
2565 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, 0,
2570 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
2572 return local_zone
->node
== zone
->node
;
2575 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2577 return node_distance(zone_to_nid(local_zone
), zone_to_nid(zone
)) <
2580 #else /* CONFIG_NUMA */
2581 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
2586 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2590 #endif /* CONFIG_NUMA */
2592 static void reset_alloc_batches(struct zone
*preferred_zone
)
2594 struct zone
*zone
= preferred_zone
->zone_pgdat
->node_zones
;
2597 mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
2598 high_wmark_pages(zone
) - low_wmark_pages(zone
) -
2599 atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]));
2600 clear_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
);
2601 } while (zone
++ != preferred_zone
);
2605 * get_page_from_freelist goes through the zonelist trying to allocate
2608 static struct page
*
2609 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
, int alloc_flags
,
2610 const struct alloc_context
*ac
)
2612 struct zonelist
*zonelist
= ac
->zonelist
;
2614 struct page
*page
= NULL
;
2616 int nr_fair_skipped
= 0;
2617 bool zonelist_rescan
;
2620 zonelist_rescan
= false;
2623 * Scan zonelist, looking for a zone with enough free.
2624 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2626 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
, ac
->high_zoneidx
,
2630 if (cpusets_enabled() &&
2631 (alloc_flags
& ALLOC_CPUSET
) &&
2632 !cpuset_zone_allowed(zone
, gfp_mask
))
2635 * Distribute pages in proportion to the individual
2636 * zone size to ensure fair page aging. The zone a
2637 * page was allocated in should have no effect on the
2638 * time the page has in memory before being reclaimed.
2640 if (alloc_flags
& ALLOC_FAIR
) {
2641 if (!zone_local(ac
->preferred_zone
, zone
))
2643 if (test_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
)) {
2649 * When allocating a page cache page for writing, we
2650 * want to get it from a zone that is within its dirty
2651 * limit, such that no single zone holds more than its
2652 * proportional share of globally allowed dirty pages.
2653 * The dirty limits take into account the zone's
2654 * lowmem reserves and high watermark so that kswapd
2655 * should be able to balance it without having to
2656 * write pages from its LRU list.
2658 * This may look like it could increase pressure on
2659 * lower zones by failing allocations in higher zones
2660 * before they are full. But the pages that do spill
2661 * over are limited as the lower zones are protected
2662 * by this very same mechanism. It should not become
2663 * a practical burden to them.
2665 * XXX: For now, allow allocations to potentially
2666 * exceed the per-zone dirty limit in the slowpath
2667 * (spread_dirty_pages unset) before going into reclaim,
2668 * which is important when on a NUMA setup the allowed
2669 * zones are together not big enough to reach the
2670 * global limit. The proper fix for these situations
2671 * will require awareness of zones in the
2672 * dirty-throttling and the flusher threads.
2674 if (ac
->spread_dirty_pages
&& !zone_dirty_ok(zone
))
2677 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
2678 if (!zone_watermark_ok(zone
, order
, mark
,
2679 ac
->classzone_idx
, alloc_flags
)) {
2682 /* Checked here to keep the fast path fast */
2683 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
2684 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
2687 if (zone_reclaim_mode
== 0 ||
2688 !zone_allows_reclaim(ac
->preferred_zone
, zone
))
2691 ret
= zone_reclaim(zone
, gfp_mask
, order
);
2693 case ZONE_RECLAIM_NOSCAN
:
2696 case ZONE_RECLAIM_FULL
:
2697 /* scanned but unreclaimable */
2700 /* did we reclaim enough */
2701 if (zone_watermark_ok(zone
, order
, mark
,
2702 ac
->classzone_idx
, alloc_flags
))
2710 page
= buffered_rmqueue(ac
->preferred_zone
, zone
, order
,
2711 gfp_mask
, alloc_flags
, ac
->migratetype
);
2713 if (prep_new_page(page
, order
, gfp_mask
, alloc_flags
))
2717 * If this is a high-order atomic allocation then check
2718 * if the pageblock should be reserved for the future
2720 if (unlikely(order
&& (alloc_flags
& ALLOC_HARDER
)))
2721 reserve_highatomic_pageblock(page
, zone
, order
);
2728 * The first pass makes sure allocations are spread fairly within the
2729 * local node. However, the local node might have free pages left
2730 * after the fairness batches are exhausted, and remote zones haven't
2731 * even been considered yet. Try once more without fairness, and
2732 * include remote zones now, before entering the slowpath and waking
2733 * kswapd: prefer spilling to a remote zone over swapping locally.
2735 if (alloc_flags
& ALLOC_FAIR
) {
2736 alloc_flags
&= ~ALLOC_FAIR
;
2737 if (nr_fair_skipped
) {
2738 zonelist_rescan
= true;
2739 reset_alloc_batches(ac
->preferred_zone
);
2741 if (nr_online_nodes
> 1)
2742 zonelist_rescan
= true;
2745 if (zonelist_rescan
)
2752 * Large machines with many possible nodes should not always dump per-node
2753 * meminfo in irq context.
2755 static inline bool should_suppress_show_mem(void)
2760 ret
= in_interrupt();
2765 static DEFINE_RATELIMIT_STATE(nopage_rs
,
2766 DEFAULT_RATELIMIT_INTERVAL
,
2767 DEFAULT_RATELIMIT_BURST
);
2769 void warn_alloc_failed(gfp_t gfp_mask
, unsigned int order
, const char *fmt
, ...)
2771 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
2773 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
) ||
2774 debug_guardpage_minorder() > 0)
2778 * This documents exceptions given to allocations in certain
2779 * contexts that are allowed to allocate outside current's set
2782 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2783 if (test_thread_flag(TIF_MEMDIE
) ||
2784 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
2785 filter
&= ~SHOW_MEM_FILTER_NODES
;
2786 if (in_interrupt() || !(gfp_mask
& __GFP_DIRECT_RECLAIM
))
2787 filter
&= ~SHOW_MEM_FILTER_NODES
;
2790 struct va_format vaf
;
2793 va_start(args
, fmt
);
2798 pr_warn("%pV", &vaf
);
2803 pr_warn("%s: page allocation failure: order:%u, mode:%#x(%pGg)\n",
2804 current
->comm
, order
, gfp_mask
, &gfp_mask
);
2806 if (!should_suppress_show_mem())
2810 static inline struct page
*
2811 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
2812 const struct alloc_context
*ac
, unsigned long *did_some_progress
)
2814 struct oom_control oc
= {
2815 .zonelist
= ac
->zonelist
,
2816 .nodemask
= ac
->nodemask
,
2817 .gfp_mask
= gfp_mask
,
2822 *did_some_progress
= 0;
2825 * Acquire the oom lock. If that fails, somebody else is
2826 * making progress for us.
2828 if (!mutex_trylock(&oom_lock
)) {
2829 *did_some_progress
= 1;
2830 schedule_timeout_uninterruptible(1);
2835 * Go through the zonelist yet one more time, keep very high watermark
2836 * here, this is only to catch a parallel oom killing, we must fail if
2837 * we're still under heavy pressure.
2839 page
= get_page_from_freelist(gfp_mask
| __GFP_HARDWALL
, order
,
2840 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
, ac
);
2844 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2845 /* Coredumps can quickly deplete all memory reserves */
2846 if (current
->flags
& PF_DUMPCORE
)
2848 /* The OOM killer will not help higher order allocs */
2849 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2851 /* The OOM killer does not needlessly kill tasks for lowmem */
2852 if (ac
->high_zoneidx
< ZONE_NORMAL
)
2854 /* The OOM killer does not compensate for IO-less reclaim */
2855 if (!(gfp_mask
& __GFP_FS
)) {
2857 * XXX: Page reclaim didn't yield anything,
2858 * and the OOM killer can't be invoked, but
2859 * keep looping as per tradition.
2861 *did_some_progress
= 1;
2864 if (pm_suspended_storage())
2866 /* The OOM killer may not free memory on a specific node */
2867 if (gfp_mask
& __GFP_THISNODE
)
2870 /* Exhausted what can be done so it's blamo time */
2871 if (out_of_memory(&oc
) || WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
)) {
2872 *did_some_progress
= 1;
2874 if (gfp_mask
& __GFP_NOFAIL
) {
2875 page
= get_page_from_freelist(gfp_mask
, order
,
2876 ALLOC_NO_WATERMARKS
|ALLOC_CPUSET
, ac
);
2878 * fallback to ignore cpuset restriction if our nodes
2882 page
= get_page_from_freelist(gfp_mask
, order
,
2883 ALLOC_NO_WATERMARKS
, ac
);
2887 mutex_unlock(&oom_lock
);
2891 #ifdef CONFIG_COMPACTION
2892 /* Try memory compaction for high-order allocations before reclaim */
2893 static struct page
*
2894 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2895 int alloc_flags
, const struct alloc_context
*ac
,
2896 enum migrate_mode mode
, int *contended_compaction
,
2897 bool *deferred_compaction
)
2899 unsigned long compact_result
;
2905 current
->flags
|= PF_MEMALLOC
;
2906 compact_result
= try_to_compact_pages(gfp_mask
, order
, alloc_flags
, ac
,
2907 mode
, contended_compaction
);
2908 current
->flags
&= ~PF_MEMALLOC
;
2910 switch (compact_result
) {
2911 case COMPACT_DEFERRED
:
2912 *deferred_compaction
= true;
2914 case COMPACT_SKIPPED
:
2921 * At least in one zone compaction wasn't deferred or skipped, so let's
2922 * count a compaction stall
2924 count_vm_event(COMPACTSTALL
);
2926 page
= get_page_from_freelist(gfp_mask
, order
,
2927 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
2930 struct zone
*zone
= page_zone(page
);
2932 zone
->compact_blockskip_flush
= false;
2933 compaction_defer_reset(zone
, order
, true);
2934 count_vm_event(COMPACTSUCCESS
);
2939 * It's bad if compaction run occurs and fails. The most likely reason
2940 * is that pages exist, but not enough to satisfy watermarks.
2942 count_vm_event(COMPACTFAIL
);
2949 static inline struct page
*
2950 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2951 int alloc_flags
, const struct alloc_context
*ac
,
2952 enum migrate_mode mode
, int *contended_compaction
,
2953 bool *deferred_compaction
)
2957 #endif /* CONFIG_COMPACTION */
2959 /* Perform direct synchronous page reclaim */
2961 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
,
2962 const struct alloc_context
*ac
)
2964 struct reclaim_state reclaim_state
;
2969 /* We now go into synchronous reclaim */
2970 cpuset_memory_pressure_bump();
2971 current
->flags
|= PF_MEMALLOC
;
2972 lockdep_set_current_reclaim_state(gfp_mask
);
2973 reclaim_state
.reclaimed_slab
= 0;
2974 current
->reclaim_state
= &reclaim_state
;
2976 progress
= try_to_free_pages(ac
->zonelist
, order
, gfp_mask
,
2979 current
->reclaim_state
= NULL
;
2980 lockdep_clear_current_reclaim_state();
2981 current
->flags
&= ~PF_MEMALLOC
;
2988 /* The really slow allocator path where we enter direct reclaim */
2989 static inline struct page
*
2990 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
2991 int alloc_flags
, const struct alloc_context
*ac
,
2992 unsigned long *did_some_progress
)
2994 struct page
*page
= NULL
;
2995 bool drained
= false;
2997 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, ac
);
2998 if (unlikely(!(*did_some_progress
)))
3002 page
= get_page_from_freelist(gfp_mask
, order
,
3003 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
3006 * If an allocation failed after direct reclaim, it could be because
3007 * pages are pinned on the per-cpu lists or in high alloc reserves.
3008 * Shrink them them and try again
3010 if (!page
&& !drained
) {
3011 unreserve_highatomic_pageblock(ac
);
3012 drain_all_pages(NULL
);
3020 static void wake_all_kswapds(unsigned int order
, const struct alloc_context
*ac
)
3025 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
,
3026 ac
->high_zoneidx
, ac
->nodemask
)
3027 wakeup_kswapd(zone
, order
, zone_idx(ac
->preferred_zone
));
3031 gfp_to_alloc_flags(gfp_t gfp_mask
)
3033 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
3035 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
3036 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
3039 * The caller may dip into page reserves a bit more if the caller
3040 * cannot run direct reclaim, or if the caller has realtime scheduling
3041 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
3042 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
3044 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
3046 if (gfp_mask
& __GFP_ATOMIC
) {
3048 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
3049 * if it can't schedule.
3051 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
3052 alloc_flags
|= ALLOC_HARDER
;
3054 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
3055 * comment for __cpuset_node_allowed().
3057 alloc_flags
&= ~ALLOC_CPUSET
;
3058 } else if (unlikely(rt_task(current
)) && !in_interrupt())
3059 alloc_flags
|= ALLOC_HARDER
;
3061 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
3062 if (gfp_mask
& __GFP_MEMALLOC
)
3063 alloc_flags
|= ALLOC_NO_WATERMARKS
;
3064 else if (in_serving_softirq() && (current
->flags
& PF_MEMALLOC
))
3065 alloc_flags
|= ALLOC_NO_WATERMARKS
;
3066 else if (!in_interrupt() &&
3067 ((current
->flags
& PF_MEMALLOC
) ||
3068 unlikely(test_thread_flag(TIF_MEMDIE
))))
3069 alloc_flags
|= ALLOC_NO_WATERMARKS
;
3072 if (gfpflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
3073 alloc_flags
|= ALLOC_CMA
;
3078 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask
)
3080 return !!(gfp_to_alloc_flags(gfp_mask
) & ALLOC_NO_WATERMARKS
);
3083 static inline bool is_thp_gfp_mask(gfp_t gfp_mask
)
3085 return (gfp_mask
& (GFP_TRANSHUGE
| __GFP_KSWAPD_RECLAIM
)) == GFP_TRANSHUGE
;
3088 static inline struct page
*
3089 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
3090 struct alloc_context
*ac
)
3092 bool can_direct_reclaim
= gfp_mask
& __GFP_DIRECT_RECLAIM
;
3093 struct page
*page
= NULL
;
3095 unsigned long pages_reclaimed
= 0;
3096 unsigned long did_some_progress
;
3097 enum migrate_mode migration_mode
= MIGRATE_ASYNC
;
3098 bool deferred_compaction
= false;
3099 int contended_compaction
= COMPACT_CONTENDED_NONE
;
3102 * In the slowpath, we sanity check order to avoid ever trying to
3103 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3104 * be using allocators in order of preference for an area that is
3107 if (order
>= MAX_ORDER
) {
3108 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
3113 * We also sanity check to catch abuse of atomic reserves being used by
3114 * callers that are not in atomic context.
3116 if (WARN_ON_ONCE((gfp_mask
& (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)) ==
3117 (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)))
3118 gfp_mask
&= ~__GFP_ATOMIC
;
3121 * If this allocation cannot block and it is for a specific node, then
3122 * fail early. There's no need to wakeup kswapd or retry for a
3123 * speculative node-specific allocation.
3125 if (IS_ENABLED(CONFIG_NUMA
) && (gfp_mask
& __GFP_THISNODE
) && !can_direct_reclaim
)
3129 if (gfp_mask
& __GFP_KSWAPD_RECLAIM
)
3130 wake_all_kswapds(order
, ac
);
3133 * OK, we're below the kswapd watermark and have kicked background
3134 * reclaim. Now things get more complex, so set up alloc_flags according
3135 * to how we want to proceed.
3137 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
3140 * Find the true preferred zone if the allocation is unconstrained by
3143 if (!(alloc_flags
& ALLOC_CPUSET
) && !ac
->nodemask
) {
3144 struct zoneref
*preferred_zoneref
;
3145 preferred_zoneref
= first_zones_zonelist(ac
->zonelist
,
3146 ac
->high_zoneidx
, NULL
, &ac
->preferred_zone
);
3147 ac
->classzone_idx
= zonelist_zone_idx(preferred_zoneref
);
3150 /* This is the last chance, in general, before the goto nopage. */
3151 page
= get_page_from_freelist(gfp_mask
, order
,
3152 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
3156 /* Allocate without watermarks if the context allows */
3157 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
3159 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3160 * the allocation is high priority and these type of
3161 * allocations are system rather than user orientated
3163 ac
->zonelist
= node_zonelist(numa_node_id(), gfp_mask
);
3164 page
= get_page_from_freelist(gfp_mask
, order
,
3165 ALLOC_NO_WATERMARKS
, ac
);
3170 /* Caller is not willing to reclaim, we can't balance anything */
3171 if (!can_direct_reclaim
) {
3173 * All existing users of the __GFP_NOFAIL are blockable, so warn
3174 * of any new users that actually allow this type of allocation
3177 WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
);
3181 /* Avoid recursion of direct reclaim */
3182 if (current
->flags
& PF_MEMALLOC
) {
3184 * __GFP_NOFAIL request from this context is rather bizarre
3185 * because we cannot reclaim anything and only can loop waiting
3186 * for somebody to do a work for us.
3188 if (WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
)) {
3195 /* Avoid allocations with no watermarks from looping endlessly */
3196 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
3200 * Try direct compaction. The first pass is asynchronous. Subsequent
3201 * attempts after direct reclaim are synchronous
3203 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
, ac
,
3205 &contended_compaction
,
3206 &deferred_compaction
);
3210 /* Checks for THP-specific high-order allocations */
3211 if (is_thp_gfp_mask(gfp_mask
)) {
3213 * If compaction is deferred for high-order allocations, it is
3214 * because sync compaction recently failed. If this is the case
3215 * and the caller requested a THP allocation, we do not want
3216 * to heavily disrupt the system, so we fail the allocation
3217 * instead of entering direct reclaim.
3219 if (deferred_compaction
)
3223 * In all zones where compaction was attempted (and not
3224 * deferred or skipped), lock contention has been detected.
3225 * For THP allocation we do not want to disrupt the others
3226 * so we fallback to base pages instead.
3228 if (contended_compaction
== COMPACT_CONTENDED_LOCK
)
3232 * If compaction was aborted due to need_resched(), we do not
3233 * want to further increase allocation latency, unless it is
3234 * khugepaged trying to collapse.
3236 if (contended_compaction
== COMPACT_CONTENDED_SCHED
3237 && !(current
->flags
& PF_KTHREAD
))
3242 * It can become very expensive to allocate transparent hugepages at
3243 * fault, so use asynchronous memory compaction for THP unless it is
3244 * khugepaged trying to collapse.
3246 if (!is_thp_gfp_mask(gfp_mask
) || (current
->flags
& PF_KTHREAD
))
3247 migration_mode
= MIGRATE_SYNC_LIGHT
;
3249 /* Try direct reclaim and then allocating */
3250 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
, alloc_flags
, ac
,
3251 &did_some_progress
);
3255 /* Do not loop if specifically requested */
3256 if (gfp_mask
& __GFP_NORETRY
)
3259 /* Keep reclaiming pages as long as there is reasonable progress */
3260 pages_reclaimed
+= did_some_progress
;
3261 if ((did_some_progress
&& order
<= PAGE_ALLOC_COSTLY_ORDER
) ||
3262 ((gfp_mask
& __GFP_REPEAT
) && pages_reclaimed
< (1 << order
))) {
3263 /* Wait for some write requests to complete then retry */
3264 wait_iff_congested(ac
->preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
3268 /* Reclaim has failed us, start killing things */
3269 page
= __alloc_pages_may_oom(gfp_mask
, order
, ac
, &did_some_progress
);
3273 /* Retry as long as the OOM killer is making progress */
3274 if (did_some_progress
)
3279 * High-order allocations do not necessarily loop after
3280 * direct reclaim and reclaim/compaction depends on compaction
3281 * being called after reclaim so call directly if necessary
3283 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
,
3285 &contended_compaction
,
3286 &deferred_compaction
);
3290 warn_alloc_failed(gfp_mask
, order
, NULL
);
3296 * This is the 'heart' of the zoned buddy allocator.
3299 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
3300 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
3302 struct zoneref
*preferred_zoneref
;
3303 struct page
*page
= NULL
;
3304 unsigned int cpuset_mems_cookie
;
3305 int alloc_flags
= ALLOC_WMARK_LOW
|ALLOC_CPUSET
|ALLOC_FAIR
;
3306 gfp_t alloc_mask
; /* The gfp_t that was actually used for allocation */
3307 struct alloc_context ac
= {
3308 .high_zoneidx
= gfp_zone(gfp_mask
),
3309 .nodemask
= nodemask
,
3310 .migratetype
= gfpflags_to_migratetype(gfp_mask
),
3313 gfp_mask
&= gfp_allowed_mask
;
3315 lockdep_trace_alloc(gfp_mask
);
3317 might_sleep_if(gfp_mask
& __GFP_DIRECT_RECLAIM
);
3319 if (should_fail_alloc_page(gfp_mask
, order
))
3323 * Check the zones suitable for the gfp_mask contain at least one
3324 * valid zone. It's possible to have an empty zonelist as a result
3325 * of __GFP_THISNODE and a memoryless node
3327 if (unlikely(!zonelist
->_zonerefs
->zone
))
3330 if (IS_ENABLED(CONFIG_CMA
) && ac
.migratetype
== MIGRATE_MOVABLE
)
3331 alloc_flags
|= ALLOC_CMA
;
3334 cpuset_mems_cookie
= read_mems_allowed_begin();
3336 /* We set it here, as __alloc_pages_slowpath might have changed it */
3337 ac
.zonelist
= zonelist
;
3339 /* Dirty zone balancing only done in the fast path */
3340 ac
.spread_dirty_pages
= (gfp_mask
& __GFP_WRITE
);
3342 /* The preferred zone is used for statistics later */
3343 preferred_zoneref
= first_zones_zonelist(ac
.zonelist
, ac
.high_zoneidx
,
3344 ac
.nodemask
? : &cpuset_current_mems_allowed
,
3345 &ac
.preferred_zone
);
3346 if (!ac
.preferred_zone
)
3348 ac
.classzone_idx
= zonelist_zone_idx(preferred_zoneref
);
3350 /* First allocation attempt */
3351 alloc_mask
= gfp_mask
|__GFP_HARDWALL
;
3352 page
= get_page_from_freelist(alloc_mask
, order
, alloc_flags
, &ac
);
3353 if (unlikely(!page
)) {
3355 * Runtime PM, block IO and its error handling path
3356 * can deadlock because I/O on the device might not
3359 alloc_mask
= memalloc_noio_flags(gfp_mask
);
3360 ac
.spread_dirty_pages
= false;
3362 page
= __alloc_pages_slowpath(alloc_mask
, order
, &ac
);
3365 if (kmemcheck_enabled
&& page
)
3366 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
3368 trace_mm_page_alloc(page
, order
, alloc_mask
, ac
.migratetype
);
3372 * When updating a task's mems_allowed, it is possible to race with
3373 * parallel threads in such a way that an allocation can fail while
3374 * the mask is being updated. If a page allocation is about to fail,
3375 * check if the cpuset changed during allocation and if so, retry.
3377 if (unlikely(!page
&& read_mems_allowed_retry(cpuset_mems_cookie
)))
3382 EXPORT_SYMBOL(__alloc_pages_nodemask
);
3385 * Common helper functions.
3387 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
3392 * __get_free_pages() returns a 32-bit address, which cannot represent
3395 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
3397 page
= alloc_pages(gfp_mask
, order
);
3400 return (unsigned long) page_address(page
);
3402 EXPORT_SYMBOL(__get_free_pages
);
3404 unsigned long get_zeroed_page(gfp_t gfp_mask
)
3406 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
3408 EXPORT_SYMBOL(get_zeroed_page
);
3410 void __free_pages(struct page
*page
, unsigned int order
)
3412 if (put_page_testzero(page
)) {
3414 free_hot_cold_page(page
, false);
3416 __free_pages_ok(page
, order
);
3420 EXPORT_SYMBOL(__free_pages
);
3422 void free_pages(unsigned long addr
, unsigned int order
)
3425 VM_BUG_ON(!virt_addr_valid((void *)addr
));
3426 __free_pages(virt_to_page((void *)addr
), order
);
3430 EXPORT_SYMBOL(free_pages
);
3434 * An arbitrary-length arbitrary-offset area of memory which resides
3435 * within a 0 or higher order page. Multiple fragments within that page
3436 * are individually refcounted, in the page's reference counter.
3438 * The page_frag functions below provide a simple allocation framework for
3439 * page fragments. This is used by the network stack and network device
3440 * drivers to provide a backing region of memory for use as either an
3441 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3443 static struct page
*__page_frag_refill(struct page_frag_cache
*nc
,
3446 struct page
*page
= NULL
;
3447 gfp_t gfp
= gfp_mask
;
3449 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3450 gfp_mask
|= __GFP_COMP
| __GFP_NOWARN
| __GFP_NORETRY
|
3452 page
= alloc_pages_node(NUMA_NO_NODE
, gfp_mask
,
3453 PAGE_FRAG_CACHE_MAX_ORDER
);
3454 nc
->size
= page
? PAGE_FRAG_CACHE_MAX_SIZE
: PAGE_SIZE
;
3456 if (unlikely(!page
))
3457 page
= alloc_pages_node(NUMA_NO_NODE
, gfp
, 0);
3459 nc
->va
= page
? page_address(page
) : NULL
;
3464 void *__alloc_page_frag(struct page_frag_cache
*nc
,
3465 unsigned int fragsz
, gfp_t gfp_mask
)
3467 unsigned int size
= PAGE_SIZE
;
3471 if (unlikely(!nc
->va
)) {
3473 page
= __page_frag_refill(nc
, gfp_mask
);
3477 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3478 /* if size can vary use size else just use PAGE_SIZE */
3481 /* Even if we own the page, we do not use atomic_set().
3482 * This would break get_page_unless_zero() users.
3484 atomic_add(size
- 1, &page
->_count
);
3486 /* reset page count bias and offset to start of new frag */
3487 nc
->pfmemalloc
= page_is_pfmemalloc(page
);
3488 nc
->pagecnt_bias
= size
;
3492 offset
= nc
->offset
- fragsz
;
3493 if (unlikely(offset
< 0)) {
3494 page
= virt_to_page(nc
->va
);
3496 if (!atomic_sub_and_test(nc
->pagecnt_bias
, &page
->_count
))
3499 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3500 /* if size can vary use size else just use PAGE_SIZE */
3503 /* OK, page count is 0, we can safely set it */
3504 atomic_set(&page
->_count
, size
);
3506 /* reset page count bias and offset to start of new frag */
3507 nc
->pagecnt_bias
= size
;
3508 offset
= size
- fragsz
;
3512 nc
->offset
= offset
;
3514 return nc
->va
+ offset
;
3516 EXPORT_SYMBOL(__alloc_page_frag
);
3519 * Frees a page fragment allocated out of either a compound or order 0 page.
3521 void __free_page_frag(void *addr
)
3523 struct page
*page
= virt_to_head_page(addr
);
3525 if (unlikely(put_page_testzero(page
)))
3526 __free_pages_ok(page
, compound_order(page
));
3528 EXPORT_SYMBOL(__free_page_frag
);
3531 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
3532 * of the current memory cgroup if __GFP_ACCOUNT is set, other than that it is
3533 * equivalent to alloc_pages.
3535 * It should be used when the caller would like to use kmalloc, but since the
3536 * allocation is large, it has to fall back to the page allocator.
3538 struct page
*alloc_kmem_pages(gfp_t gfp_mask
, unsigned int order
)
3542 page
= alloc_pages(gfp_mask
, order
);
3543 if (page
&& memcg_kmem_charge(page
, gfp_mask
, order
) != 0) {
3544 __free_pages(page
, order
);
3550 struct page
*alloc_kmem_pages_node(int nid
, gfp_t gfp_mask
, unsigned int order
)
3554 page
= alloc_pages_node(nid
, gfp_mask
, order
);
3555 if (page
&& memcg_kmem_charge(page
, gfp_mask
, order
) != 0) {
3556 __free_pages(page
, order
);
3563 * __free_kmem_pages and free_kmem_pages will free pages allocated with
3566 void __free_kmem_pages(struct page
*page
, unsigned int order
)
3568 memcg_kmem_uncharge(page
, order
);
3569 __free_pages(page
, order
);
3572 void free_kmem_pages(unsigned long addr
, unsigned int order
)
3575 VM_BUG_ON(!virt_addr_valid((void *)addr
));
3576 __free_kmem_pages(virt_to_page((void *)addr
), order
);
3580 static void *make_alloc_exact(unsigned long addr
, unsigned int order
,
3584 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
3585 unsigned long used
= addr
+ PAGE_ALIGN(size
);
3587 split_page(virt_to_page((void *)addr
), order
);
3588 while (used
< alloc_end
) {
3593 return (void *)addr
;
3597 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3598 * @size: the number of bytes to allocate
3599 * @gfp_mask: GFP flags for the allocation
3601 * This function is similar to alloc_pages(), except that it allocates the
3602 * minimum number of pages to satisfy the request. alloc_pages() can only
3603 * allocate memory in power-of-two pages.
3605 * This function is also limited by MAX_ORDER.
3607 * Memory allocated by this function must be released by free_pages_exact().
3609 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
3611 unsigned int order
= get_order(size
);
3614 addr
= __get_free_pages(gfp_mask
, order
);
3615 return make_alloc_exact(addr
, order
, size
);
3617 EXPORT_SYMBOL(alloc_pages_exact
);
3620 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3622 * @nid: the preferred node ID where memory should be allocated
3623 * @size: the number of bytes to allocate
3624 * @gfp_mask: GFP flags for the allocation
3626 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3629 void * __meminit
alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
3631 unsigned int order
= get_order(size
);
3632 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
3635 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
3639 * free_pages_exact - release memory allocated via alloc_pages_exact()
3640 * @virt: the value returned by alloc_pages_exact.
3641 * @size: size of allocation, same value as passed to alloc_pages_exact().
3643 * Release the memory allocated by a previous call to alloc_pages_exact.
3645 void free_pages_exact(void *virt
, size_t size
)
3647 unsigned long addr
= (unsigned long)virt
;
3648 unsigned long end
= addr
+ PAGE_ALIGN(size
);
3650 while (addr
< end
) {
3655 EXPORT_SYMBOL(free_pages_exact
);
3658 * nr_free_zone_pages - count number of pages beyond high watermark
3659 * @offset: The zone index of the highest zone
3661 * nr_free_zone_pages() counts the number of counts pages which are beyond the
3662 * high watermark within all zones at or below a given zone index. For each
3663 * zone, the number of pages is calculated as:
3664 * managed_pages - high_pages
3666 static unsigned long nr_free_zone_pages(int offset
)
3671 /* Just pick one node, since fallback list is circular */
3672 unsigned long sum
= 0;
3674 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
3676 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
3677 unsigned long size
= zone
->managed_pages
;
3678 unsigned long high
= high_wmark_pages(zone
);
3687 * nr_free_buffer_pages - count number of pages beyond high watermark
3689 * nr_free_buffer_pages() counts the number of pages which are beyond the high
3690 * watermark within ZONE_DMA and ZONE_NORMAL.
3692 unsigned long nr_free_buffer_pages(void)
3694 return nr_free_zone_pages(gfp_zone(GFP_USER
));
3696 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
3699 * nr_free_pagecache_pages - count number of pages beyond high watermark
3701 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3702 * high watermark within all zones.
3704 unsigned long nr_free_pagecache_pages(void)
3706 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
3709 static inline void show_node(struct zone
*zone
)
3711 if (IS_ENABLED(CONFIG_NUMA
))
3712 printk("Node %d ", zone_to_nid(zone
));
3715 void si_meminfo(struct sysinfo
*val
)
3717 val
->totalram
= totalram_pages
;
3718 val
->sharedram
= global_page_state(NR_SHMEM
);
3719 val
->freeram
= global_page_state(NR_FREE_PAGES
);
3720 val
->bufferram
= nr_blockdev_pages();
3721 val
->totalhigh
= totalhigh_pages
;
3722 val
->freehigh
= nr_free_highpages();
3723 val
->mem_unit
= PAGE_SIZE
;
3726 EXPORT_SYMBOL(si_meminfo
);
3729 void si_meminfo_node(struct sysinfo
*val
, int nid
)
3731 int zone_type
; /* needs to be signed */
3732 unsigned long managed_pages
= 0;
3733 pg_data_t
*pgdat
= NODE_DATA(nid
);
3735 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++)
3736 managed_pages
+= pgdat
->node_zones
[zone_type
].managed_pages
;
3737 val
->totalram
= managed_pages
;
3738 val
->sharedram
= node_page_state(nid
, NR_SHMEM
);
3739 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
3740 #ifdef CONFIG_HIGHMEM
3741 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].managed_pages
;
3742 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
3748 val
->mem_unit
= PAGE_SIZE
;
3753 * Determine whether the node should be displayed or not, depending on whether
3754 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
3756 bool skip_free_areas_node(unsigned int flags
, int nid
)
3759 unsigned int cpuset_mems_cookie
;
3761 if (!(flags
& SHOW_MEM_FILTER_NODES
))
3765 cpuset_mems_cookie
= read_mems_allowed_begin();
3766 ret
= !node_isset(nid
, cpuset_current_mems_allowed
);
3767 } while (read_mems_allowed_retry(cpuset_mems_cookie
));
3772 #define K(x) ((x) << (PAGE_SHIFT-10))
3774 static void show_migration_types(unsigned char type
)
3776 static const char types
[MIGRATE_TYPES
] = {
3777 [MIGRATE_UNMOVABLE
] = 'U',
3778 [MIGRATE_MOVABLE
] = 'M',
3779 [MIGRATE_RECLAIMABLE
] = 'E',
3780 [MIGRATE_HIGHATOMIC
] = 'H',
3782 [MIGRATE_CMA
] = 'C',
3784 #ifdef CONFIG_MEMORY_ISOLATION
3785 [MIGRATE_ISOLATE
] = 'I',
3788 char tmp
[MIGRATE_TYPES
+ 1];
3792 for (i
= 0; i
< MIGRATE_TYPES
; i
++) {
3793 if (type
& (1 << i
))
3798 printk("(%s) ", tmp
);
3802 * Show free area list (used inside shift_scroll-lock stuff)
3803 * We also calculate the percentage fragmentation. We do this by counting the
3804 * memory on each free list with the exception of the first item on the list.
3807 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
3810 void show_free_areas(unsigned int filter
)
3812 unsigned long free_pcp
= 0;
3816 for_each_populated_zone(zone
) {
3817 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3820 for_each_online_cpu(cpu
)
3821 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
3824 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3825 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3826 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
3827 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3828 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3829 " free:%lu free_pcp:%lu free_cma:%lu\n",
3830 global_page_state(NR_ACTIVE_ANON
),
3831 global_page_state(NR_INACTIVE_ANON
),
3832 global_page_state(NR_ISOLATED_ANON
),
3833 global_page_state(NR_ACTIVE_FILE
),
3834 global_page_state(NR_INACTIVE_FILE
),
3835 global_page_state(NR_ISOLATED_FILE
),
3836 global_page_state(NR_UNEVICTABLE
),
3837 global_page_state(NR_FILE_DIRTY
),
3838 global_page_state(NR_WRITEBACK
),
3839 global_page_state(NR_UNSTABLE_NFS
),
3840 global_page_state(NR_SLAB_RECLAIMABLE
),
3841 global_page_state(NR_SLAB_UNRECLAIMABLE
),
3842 global_page_state(NR_FILE_MAPPED
),
3843 global_page_state(NR_SHMEM
),
3844 global_page_state(NR_PAGETABLE
),
3845 global_page_state(NR_BOUNCE
),
3846 global_page_state(NR_FREE_PAGES
),
3848 global_page_state(NR_FREE_CMA_PAGES
));
3850 for_each_populated_zone(zone
) {
3853 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3857 for_each_online_cpu(cpu
)
3858 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
3866 " active_anon:%lukB"
3867 " inactive_anon:%lukB"
3868 " active_file:%lukB"
3869 " inactive_file:%lukB"
3870 " unevictable:%lukB"
3871 " isolated(anon):%lukB"
3872 " isolated(file):%lukB"
3880 " slab_reclaimable:%lukB"
3881 " slab_unreclaimable:%lukB"
3882 " kernel_stack:%lukB"
3889 " writeback_tmp:%lukB"
3890 " pages_scanned:%lu"
3891 " all_unreclaimable? %s"
3894 K(zone_page_state(zone
, NR_FREE_PAGES
)),
3895 K(min_wmark_pages(zone
)),
3896 K(low_wmark_pages(zone
)),
3897 K(high_wmark_pages(zone
)),
3898 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
3899 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
3900 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
3901 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
3902 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
3903 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
3904 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
3905 K(zone
->present_pages
),
3906 K(zone
->managed_pages
),
3907 K(zone_page_state(zone
, NR_MLOCK
)),
3908 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
3909 K(zone_page_state(zone
, NR_WRITEBACK
)),
3910 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
3911 K(zone_page_state(zone
, NR_SHMEM
)),
3912 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
3913 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
3914 zone_page_state(zone
, NR_KERNEL_STACK
) *
3916 K(zone_page_state(zone
, NR_PAGETABLE
)),
3917 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
3918 K(zone_page_state(zone
, NR_BOUNCE
)),
3920 K(this_cpu_read(zone
->pageset
->pcp
.count
)),
3921 K(zone_page_state(zone
, NR_FREE_CMA_PAGES
)),
3922 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
3923 K(zone_page_state(zone
, NR_PAGES_SCANNED
)),
3924 (!zone_reclaimable(zone
) ? "yes" : "no")
3926 printk("lowmem_reserve[]:");
3927 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3928 printk(" %ld", zone
->lowmem_reserve
[i
]);
3932 for_each_populated_zone(zone
) {
3934 unsigned long nr
[MAX_ORDER
], flags
, total
= 0;
3935 unsigned char types
[MAX_ORDER
];
3937 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3940 printk("%s: ", zone
->name
);
3942 spin_lock_irqsave(&zone
->lock
, flags
);
3943 for (order
= 0; order
< MAX_ORDER
; order
++) {
3944 struct free_area
*area
= &zone
->free_area
[order
];
3947 nr
[order
] = area
->nr_free
;
3948 total
+= nr
[order
] << order
;
3951 for (type
= 0; type
< MIGRATE_TYPES
; type
++) {
3952 if (!list_empty(&area
->free_list
[type
]))
3953 types
[order
] |= 1 << type
;
3956 spin_unlock_irqrestore(&zone
->lock
, flags
);
3957 for (order
= 0; order
< MAX_ORDER
; order
++) {
3958 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
3960 show_migration_types(types
[order
]);
3962 printk("= %lukB\n", K(total
));
3965 hugetlb_show_meminfo();
3967 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
3969 show_swap_cache_info();
3972 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
3974 zoneref
->zone
= zone
;
3975 zoneref
->zone_idx
= zone_idx(zone
);
3979 * Builds allocation fallback zone lists.
3981 * Add all populated zones of a node to the zonelist.
3983 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
3987 enum zone_type zone_type
= MAX_NR_ZONES
;
3991 zone
= pgdat
->node_zones
+ zone_type
;
3992 if (populated_zone(zone
)) {
3993 zoneref_set_zone(zone
,
3994 &zonelist
->_zonerefs
[nr_zones
++]);
3995 check_highest_zone(zone_type
);
3997 } while (zone_type
);
4005 * 0 = automatic detection of better ordering.
4006 * 1 = order by ([node] distance, -zonetype)
4007 * 2 = order by (-zonetype, [node] distance)
4009 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
4010 * the same zonelist. So only NUMA can configure this param.
4012 #define ZONELIST_ORDER_DEFAULT 0
4013 #define ZONELIST_ORDER_NODE 1
4014 #define ZONELIST_ORDER_ZONE 2
4016 /* zonelist order in the kernel.
4017 * set_zonelist_order() will set this to NODE or ZONE.
4019 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4020 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
4024 /* The value user specified ....changed by config */
4025 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4026 /* string for sysctl */
4027 #define NUMA_ZONELIST_ORDER_LEN 16
4028 char numa_zonelist_order
[16] = "default";
4031 * interface for configure zonelist ordering.
4032 * command line option "numa_zonelist_order"
4033 * = "[dD]efault - default, automatic configuration.
4034 * = "[nN]ode - order by node locality, then by zone within node
4035 * = "[zZ]one - order by zone, then by locality within zone
4038 static int __parse_numa_zonelist_order(char *s
)
4040 if (*s
== 'd' || *s
== 'D') {
4041 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4042 } else if (*s
== 'n' || *s
== 'N') {
4043 user_zonelist_order
= ZONELIST_ORDER_NODE
;
4044 } else if (*s
== 'z' || *s
== 'Z') {
4045 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
4048 "Ignoring invalid numa_zonelist_order value: "
4055 static __init
int setup_numa_zonelist_order(char *s
)
4062 ret
= __parse_numa_zonelist_order(s
);
4064 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
4068 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
4071 * sysctl handler for numa_zonelist_order
4073 int numa_zonelist_order_handler(struct ctl_table
*table
, int write
,
4074 void __user
*buffer
, size_t *length
,
4077 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
4079 static DEFINE_MUTEX(zl_order_mutex
);
4081 mutex_lock(&zl_order_mutex
);
4083 if (strlen((char *)table
->data
) >= NUMA_ZONELIST_ORDER_LEN
) {
4087 strcpy(saved_string
, (char *)table
->data
);
4089 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
4093 int oldval
= user_zonelist_order
;
4095 ret
= __parse_numa_zonelist_order((char *)table
->data
);
4098 * bogus value. restore saved string
4100 strncpy((char *)table
->data
, saved_string
,
4101 NUMA_ZONELIST_ORDER_LEN
);
4102 user_zonelist_order
= oldval
;
4103 } else if (oldval
!= user_zonelist_order
) {
4104 mutex_lock(&zonelists_mutex
);
4105 build_all_zonelists(NULL
, NULL
);
4106 mutex_unlock(&zonelists_mutex
);
4110 mutex_unlock(&zl_order_mutex
);
4115 #define MAX_NODE_LOAD (nr_online_nodes)
4116 static int node_load
[MAX_NUMNODES
];
4119 * find_next_best_node - find the next node that should appear in a given node's fallback list
4120 * @node: node whose fallback list we're appending
4121 * @used_node_mask: nodemask_t of already used nodes
4123 * We use a number of factors to determine which is the next node that should
4124 * appear on a given node's fallback list. The node should not have appeared
4125 * already in @node's fallback list, and it should be the next closest node
4126 * according to the distance array (which contains arbitrary distance values
4127 * from each node to each node in the system), and should also prefer nodes
4128 * with no CPUs, since presumably they'll have very little allocation pressure
4129 * on them otherwise.
4130 * It returns -1 if no node is found.
4132 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
4135 int min_val
= INT_MAX
;
4136 int best_node
= NUMA_NO_NODE
;
4137 const struct cpumask
*tmp
= cpumask_of_node(0);
4139 /* Use the local node if we haven't already */
4140 if (!node_isset(node
, *used_node_mask
)) {
4141 node_set(node
, *used_node_mask
);
4145 for_each_node_state(n
, N_MEMORY
) {
4147 /* Don't want a node to appear more than once */
4148 if (node_isset(n
, *used_node_mask
))
4151 /* Use the distance array to find the distance */
4152 val
= node_distance(node
, n
);
4154 /* Penalize nodes under us ("prefer the next node") */
4157 /* Give preference to headless and unused nodes */
4158 tmp
= cpumask_of_node(n
);
4159 if (!cpumask_empty(tmp
))
4160 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
4162 /* Slight preference for less loaded node */
4163 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
4164 val
+= node_load
[n
];
4166 if (val
< min_val
) {
4173 node_set(best_node
, *used_node_mask
);
4180 * Build zonelists ordered by node and zones within node.
4181 * This results in maximum locality--normal zone overflows into local
4182 * DMA zone, if any--but risks exhausting DMA zone.
4184 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
4187 struct zonelist
*zonelist
;
4189 zonelist
= &pgdat
->node_zonelists
[0];
4190 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
4192 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4193 zonelist
->_zonerefs
[j
].zone
= NULL
;
4194 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4198 * Build gfp_thisnode zonelists
4200 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
4203 struct zonelist
*zonelist
;
4205 zonelist
= &pgdat
->node_zonelists
[1];
4206 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4207 zonelist
->_zonerefs
[j
].zone
= NULL
;
4208 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4212 * Build zonelists ordered by zone and nodes within zones.
4213 * This results in conserving DMA zone[s] until all Normal memory is
4214 * exhausted, but results in overflowing to remote node while memory
4215 * may still exist in local DMA zone.
4217 static int node_order
[MAX_NUMNODES
];
4219 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
4222 int zone_type
; /* needs to be signed */
4224 struct zonelist
*zonelist
;
4226 zonelist
= &pgdat
->node_zonelists
[0];
4228 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
4229 for (j
= 0; j
< nr_nodes
; j
++) {
4230 node
= node_order
[j
];
4231 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
4232 if (populated_zone(z
)) {
4234 &zonelist
->_zonerefs
[pos
++]);
4235 check_highest_zone(zone_type
);
4239 zonelist
->_zonerefs
[pos
].zone
= NULL
;
4240 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
4243 #if defined(CONFIG_64BIT)
4245 * Devices that require DMA32/DMA are relatively rare and do not justify a
4246 * penalty to every machine in case the specialised case applies. Default
4247 * to Node-ordering on 64-bit NUMA machines
4249 static int default_zonelist_order(void)
4251 return ZONELIST_ORDER_NODE
;
4255 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4256 * by the kernel. If processes running on node 0 deplete the low memory zone
4257 * then reclaim will occur more frequency increasing stalls and potentially
4258 * be easier to OOM if a large percentage of the zone is under writeback or
4259 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4260 * Hence, default to zone ordering on 32-bit.
4262 static int default_zonelist_order(void)
4264 return ZONELIST_ORDER_ZONE
;
4266 #endif /* CONFIG_64BIT */
4268 static void set_zonelist_order(void)
4270 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
4271 current_zonelist_order
= default_zonelist_order();
4273 current_zonelist_order
= user_zonelist_order
;
4276 static void build_zonelists(pg_data_t
*pgdat
)
4279 nodemask_t used_mask
;
4280 int local_node
, prev_node
;
4281 struct zonelist
*zonelist
;
4282 unsigned int order
= current_zonelist_order
;
4284 /* initialize zonelists */
4285 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
4286 zonelist
= pgdat
->node_zonelists
+ i
;
4287 zonelist
->_zonerefs
[0].zone
= NULL
;
4288 zonelist
->_zonerefs
[0].zone_idx
= 0;
4291 /* NUMA-aware ordering of nodes */
4292 local_node
= pgdat
->node_id
;
4293 load
= nr_online_nodes
;
4294 prev_node
= local_node
;
4295 nodes_clear(used_mask
);
4297 memset(node_order
, 0, sizeof(node_order
));
4300 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
4302 * We don't want to pressure a particular node.
4303 * So adding penalty to the first node in same
4304 * distance group to make it round-robin.
4306 if (node_distance(local_node
, node
) !=
4307 node_distance(local_node
, prev_node
))
4308 node_load
[node
] = load
;
4312 if (order
== ZONELIST_ORDER_NODE
)
4313 build_zonelists_in_node_order(pgdat
, node
);
4315 node_order
[i
++] = node
; /* remember order */
4318 if (order
== ZONELIST_ORDER_ZONE
) {
4319 /* calculate node order -- i.e., DMA last! */
4320 build_zonelists_in_zone_order(pgdat
, i
);
4323 build_thisnode_zonelists(pgdat
);
4326 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4328 * Return node id of node used for "local" allocations.
4329 * I.e., first node id of first zone in arg node's generic zonelist.
4330 * Used for initializing percpu 'numa_mem', which is used primarily
4331 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4333 int local_memory_node(int node
)
4337 (void)first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
4338 gfp_zone(GFP_KERNEL
),
4345 #else /* CONFIG_NUMA */
4347 static void set_zonelist_order(void)
4349 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
4352 static void build_zonelists(pg_data_t
*pgdat
)
4354 int node
, local_node
;
4356 struct zonelist
*zonelist
;
4358 local_node
= pgdat
->node_id
;
4360 zonelist
= &pgdat
->node_zonelists
[0];
4361 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4364 * Now we build the zonelist so that it contains the zones
4365 * of all the other nodes.
4366 * We don't want to pressure a particular node, so when
4367 * building the zones for node N, we make sure that the
4368 * zones coming right after the local ones are those from
4369 * node N+1 (modulo N)
4371 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
4372 if (!node_online(node
))
4374 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4376 for (node
= 0; node
< local_node
; node
++) {
4377 if (!node_online(node
))
4379 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4382 zonelist
->_zonerefs
[j
].zone
= NULL
;
4383 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4386 #endif /* CONFIG_NUMA */
4389 * Boot pageset table. One per cpu which is going to be used for all
4390 * zones and all nodes. The parameters will be set in such a way
4391 * that an item put on a list will immediately be handed over to
4392 * the buddy list. This is safe since pageset manipulation is done
4393 * with interrupts disabled.
4395 * The boot_pagesets must be kept even after bootup is complete for
4396 * unused processors and/or zones. They do play a role for bootstrapping
4397 * hotplugged processors.
4399 * zoneinfo_show() and maybe other functions do
4400 * not check if the processor is online before following the pageset pointer.
4401 * Other parts of the kernel may not check if the zone is available.
4403 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
4404 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
4405 static void setup_zone_pageset(struct zone
*zone
);
4408 * Global mutex to protect against size modification of zonelists
4409 * as well as to serialize pageset setup for the new populated zone.
4411 DEFINE_MUTEX(zonelists_mutex
);
4413 /* return values int ....just for stop_machine() */
4414 static int __build_all_zonelists(void *data
)
4418 pg_data_t
*self
= data
;
4421 memset(node_load
, 0, sizeof(node_load
));
4424 if (self
&& !node_online(self
->node_id
)) {
4425 build_zonelists(self
);
4428 for_each_online_node(nid
) {
4429 pg_data_t
*pgdat
= NODE_DATA(nid
);
4431 build_zonelists(pgdat
);
4435 * Initialize the boot_pagesets that are going to be used
4436 * for bootstrapping processors. The real pagesets for
4437 * each zone will be allocated later when the per cpu
4438 * allocator is available.
4440 * boot_pagesets are used also for bootstrapping offline
4441 * cpus if the system is already booted because the pagesets
4442 * are needed to initialize allocators on a specific cpu too.
4443 * F.e. the percpu allocator needs the page allocator which
4444 * needs the percpu allocator in order to allocate its pagesets
4445 * (a chicken-egg dilemma).
4447 for_each_possible_cpu(cpu
) {
4448 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
4450 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4452 * We now know the "local memory node" for each node--
4453 * i.e., the node of the first zone in the generic zonelist.
4454 * Set up numa_mem percpu variable for on-line cpus. During
4455 * boot, only the boot cpu should be on-line; we'll init the
4456 * secondary cpus' numa_mem as they come on-line. During
4457 * node/memory hotplug, we'll fixup all on-line cpus.
4459 if (cpu_online(cpu
))
4460 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
4467 static noinline
void __init
4468 build_all_zonelists_init(void)
4470 __build_all_zonelists(NULL
);
4471 mminit_verify_zonelist();
4472 cpuset_init_current_mems_allowed();
4476 * Called with zonelists_mutex held always
4477 * unless system_state == SYSTEM_BOOTING.
4479 * __ref due to (1) call of __meminit annotated setup_zone_pageset
4480 * [we're only called with non-NULL zone through __meminit paths] and
4481 * (2) call of __init annotated helper build_all_zonelists_init
4482 * [protected by SYSTEM_BOOTING].
4484 void __ref
build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
)
4486 set_zonelist_order();
4488 if (system_state
== SYSTEM_BOOTING
) {
4489 build_all_zonelists_init();
4491 #ifdef CONFIG_MEMORY_HOTPLUG
4493 setup_zone_pageset(zone
);
4495 /* we have to stop all cpus to guarantee there is no user
4497 stop_machine(__build_all_zonelists
, pgdat
, NULL
);
4498 /* cpuset refresh routine should be here */
4500 vm_total_pages
= nr_free_pagecache_pages();
4502 * Disable grouping by mobility if the number of pages in the
4503 * system is too low to allow the mechanism to work. It would be
4504 * more accurate, but expensive to check per-zone. This check is
4505 * made on memory-hotadd so a system can start with mobility
4506 * disabled and enable it later
4508 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
4509 page_group_by_mobility_disabled
= 1;
4511 page_group_by_mobility_disabled
= 0;
4513 pr_info("Built %i zonelists in %s order, mobility grouping %s. "
4514 "Total pages: %ld\n",
4516 zonelist_order_name
[current_zonelist_order
],
4517 page_group_by_mobility_disabled
? "off" : "on",
4520 pr_info("Policy zone: %s\n", zone_names
[policy_zone
]);
4525 * Helper functions to size the waitqueue hash table.
4526 * Essentially these want to choose hash table sizes sufficiently
4527 * large so that collisions trying to wait on pages are rare.
4528 * But in fact, the number of active page waitqueues on typical
4529 * systems is ridiculously low, less than 200. So this is even
4530 * conservative, even though it seems large.
4532 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
4533 * waitqueues, i.e. the size of the waitq table given the number of pages.
4535 #define PAGES_PER_WAITQUEUE 256
4537 #ifndef CONFIG_MEMORY_HOTPLUG
4538 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
4540 unsigned long size
= 1;
4542 pages
/= PAGES_PER_WAITQUEUE
;
4544 while (size
< pages
)
4548 * Once we have dozens or even hundreds of threads sleeping
4549 * on IO we've got bigger problems than wait queue collision.
4550 * Limit the size of the wait table to a reasonable size.
4552 size
= min(size
, 4096UL);
4554 return max(size
, 4UL);
4558 * A zone's size might be changed by hot-add, so it is not possible to determine
4559 * a suitable size for its wait_table. So we use the maximum size now.
4561 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
4563 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
4564 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4565 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
4567 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4568 * or more by the traditional way. (See above). It equals:
4570 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
4571 * ia64(16K page size) : = ( 8G + 4M)byte.
4572 * powerpc (64K page size) : = (32G +16M)byte.
4574 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
4581 * This is an integer logarithm so that shifts can be used later
4582 * to extract the more random high bits from the multiplicative
4583 * hash function before the remainder is taken.
4585 static inline unsigned long wait_table_bits(unsigned long size
)
4591 * Initially all pages are reserved - free ones are freed
4592 * up by free_all_bootmem() once the early boot process is
4593 * done. Non-atomic initialization, single-pass.
4595 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
4596 unsigned long start_pfn
, enum memmap_context context
)
4598 struct vmem_altmap
*altmap
= to_vmem_altmap(__pfn_to_phys(start_pfn
));
4599 unsigned long end_pfn
= start_pfn
+ size
;
4600 pg_data_t
*pgdat
= NODE_DATA(nid
);
4602 unsigned long nr_initialised
= 0;
4603 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4604 struct memblock_region
*r
= NULL
, *tmp
;
4607 if (highest_memmap_pfn
< end_pfn
- 1)
4608 highest_memmap_pfn
= end_pfn
- 1;
4611 * Honor reservation requested by the driver for this ZONE_DEVICE
4614 if (altmap
&& start_pfn
== altmap
->base_pfn
)
4615 start_pfn
+= altmap
->reserve
;
4617 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
4619 * There can be holes in boot-time mem_map[]s handed to this
4620 * function. They do not exist on hotplugged memory.
4622 if (context
!= MEMMAP_EARLY
)
4625 if (!early_pfn_valid(pfn
))
4627 if (!early_pfn_in_nid(pfn
, nid
))
4629 if (!update_defer_init(pgdat
, pfn
, end_pfn
, &nr_initialised
))
4632 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4634 * If not mirrored_kernelcore and ZONE_MOVABLE exists, range
4635 * from zone_movable_pfn[nid] to end of each node should be
4636 * ZONE_MOVABLE not ZONE_NORMAL. skip it.
4638 if (!mirrored_kernelcore
&& zone_movable_pfn
[nid
])
4639 if (zone
== ZONE_NORMAL
&& pfn
>= zone_movable_pfn
[nid
])
4643 * Check given memblock attribute by firmware which can affect
4644 * kernel memory layout. If zone==ZONE_MOVABLE but memory is
4645 * mirrored, it's an overlapped memmap init. skip it.
4647 if (mirrored_kernelcore
&& zone
== ZONE_MOVABLE
) {
4648 if (!r
|| pfn
>= memblock_region_memory_end_pfn(r
)) {
4649 for_each_memblock(memory
, tmp
)
4650 if (pfn
< memblock_region_memory_end_pfn(tmp
))
4654 if (pfn
>= memblock_region_memory_base_pfn(r
) &&
4655 memblock_is_mirror(r
)) {
4656 /* already initialized as NORMAL */
4657 pfn
= memblock_region_memory_end_pfn(r
);
4665 * Mark the block movable so that blocks are reserved for
4666 * movable at startup. This will force kernel allocations
4667 * to reserve their blocks rather than leaking throughout
4668 * the address space during boot when many long-lived
4669 * kernel allocations are made.
4671 * bitmap is created for zone's valid pfn range. but memmap
4672 * can be created for invalid pages (for alignment)
4673 * check here not to call set_pageblock_migratetype() against
4676 if (!(pfn
& (pageblock_nr_pages
- 1))) {
4677 struct page
*page
= pfn_to_page(pfn
);
4679 __init_single_page(page
, pfn
, zone
, nid
);
4680 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4682 __init_single_pfn(pfn
, zone
, nid
);
4687 static void __meminit
zone_init_free_lists(struct zone
*zone
)
4689 unsigned int order
, t
;
4690 for_each_migratetype_order(order
, t
) {
4691 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
4692 zone
->free_area
[order
].nr_free
= 0;
4696 #ifndef __HAVE_ARCH_MEMMAP_INIT
4697 #define memmap_init(size, nid, zone, start_pfn) \
4698 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4701 static int zone_batchsize(struct zone
*zone
)
4707 * The per-cpu-pages pools are set to around 1000th of the
4708 * size of the zone. But no more than 1/2 of a meg.
4710 * OK, so we don't know how big the cache is. So guess.
4712 batch
= zone
->managed_pages
/ 1024;
4713 if (batch
* PAGE_SIZE
> 512 * 1024)
4714 batch
= (512 * 1024) / PAGE_SIZE
;
4715 batch
/= 4; /* We effectively *= 4 below */
4720 * Clamp the batch to a 2^n - 1 value. Having a power
4721 * of 2 value was found to be more likely to have
4722 * suboptimal cache aliasing properties in some cases.
4724 * For example if 2 tasks are alternately allocating
4725 * batches of pages, one task can end up with a lot
4726 * of pages of one half of the possible page colors
4727 * and the other with pages of the other colors.
4729 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
4734 /* The deferral and batching of frees should be suppressed under NOMMU
4737 * The problem is that NOMMU needs to be able to allocate large chunks
4738 * of contiguous memory as there's no hardware page translation to
4739 * assemble apparent contiguous memory from discontiguous pages.
4741 * Queueing large contiguous runs of pages for batching, however,
4742 * causes the pages to actually be freed in smaller chunks. As there
4743 * can be a significant delay between the individual batches being
4744 * recycled, this leads to the once large chunks of space being
4745 * fragmented and becoming unavailable for high-order allocations.
4752 * pcp->high and pcp->batch values are related and dependent on one another:
4753 * ->batch must never be higher then ->high.
4754 * The following function updates them in a safe manner without read side
4757 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4758 * those fields changing asynchronously (acording the the above rule).
4760 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4761 * outside of boot time (or some other assurance that no concurrent updaters
4764 static void pageset_update(struct per_cpu_pages
*pcp
, unsigned long high
,
4765 unsigned long batch
)
4767 /* start with a fail safe value for batch */
4771 /* Update high, then batch, in order */
4778 /* a companion to pageset_set_high() */
4779 static void pageset_set_batch(struct per_cpu_pageset
*p
, unsigned long batch
)
4781 pageset_update(&p
->pcp
, 6 * batch
, max(1UL, 1 * batch
));
4784 static void pageset_init(struct per_cpu_pageset
*p
)
4786 struct per_cpu_pages
*pcp
;
4789 memset(p
, 0, sizeof(*p
));
4793 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
4794 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
4797 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
4800 pageset_set_batch(p
, batch
);
4804 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
4805 * to the value high for the pageset p.
4807 static void pageset_set_high(struct per_cpu_pageset
*p
,
4810 unsigned long batch
= max(1UL, high
/ 4);
4811 if ((high
/ 4) > (PAGE_SHIFT
* 8))
4812 batch
= PAGE_SHIFT
* 8;
4814 pageset_update(&p
->pcp
, high
, batch
);
4817 static void pageset_set_high_and_batch(struct zone
*zone
,
4818 struct per_cpu_pageset
*pcp
)
4820 if (percpu_pagelist_fraction
)
4821 pageset_set_high(pcp
,
4822 (zone
->managed_pages
/
4823 percpu_pagelist_fraction
));
4825 pageset_set_batch(pcp
, zone_batchsize(zone
));
4828 static void __meminit
zone_pageset_init(struct zone
*zone
, int cpu
)
4830 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
4833 pageset_set_high_and_batch(zone
, pcp
);
4836 static void __meminit
setup_zone_pageset(struct zone
*zone
)
4839 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
4840 for_each_possible_cpu(cpu
)
4841 zone_pageset_init(zone
, cpu
);
4845 * Allocate per cpu pagesets and initialize them.
4846 * Before this call only boot pagesets were available.
4848 void __init
setup_per_cpu_pageset(void)
4852 for_each_populated_zone(zone
)
4853 setup_zone_pageset(zone
);
4856 static noinline __init_refok
4857 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
4863 * The per-page waitqueue mechanism uses hashed waitqueues
4866 zone
->wait_table_hash_nr_entries
=
4867 wait_table_hash_nr_entries(zone_size_pages
);
4868 zone
->wait_table_bits
=
4869 wait_table_bits(zone
->wait_table_hash_nr_entries
);
4870 alloc_size
= zone
->wait_table_hash_nr_entries
4871 * sizeof(wait_queue_head_t
);
4873 if (!slab_is_available()) {
4874 zone
->wait_table
= (wait_queue_head_t
*)
4875 memblock_virt_alloc_node_nopanic(
4876 alloc_size
, zone
->zone_pgdat
->node_id
);
4879 * This case means that a zone whose size was 0 gets new memory
4880 * via memory hot-add.
4881 * But it may be the case that a new node was hot-added. In
4882 * this case vmalloc() will not be able to use this new node's
4883 * memory - this wait_table must be initialized to use this new
4884 * node itself as well.
4885 * To use this new node's memory, further consideration will be
4888 zone
->wait_table
= vmalloc(alloc_size
);
4890 if (!zone
->wait_table
)
4893 for (i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
4894 init_waitqueue_head(zone
->wait_table
+ i
);
4899 static __meminit
void zone_pcp_init(struct zone
*zone
)
4902 * per cpu subsystem is not up at this point. The following code
4903 * relies on the ability of the linker to provide the
4904 * offset of a (static) per cpu variable into the per cpu area.
4906 zone
->pageset
= &boot_pageset
;
4908 if (populated_zone(zone
))
4909 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
4910 zone
->name
, zone
->present_pages
,
4911 zone_batchsize(zone
));
4914 int __meminit
init_currently_empty_zone(struct zone
*zone
,
4915 unsigned long zone_start_pfn
,
4918 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
4920 ret
= zone_wait_table_init(zone
, size
);
4923 pgdat
->nr_zones
= zone_idx(zone
) + 1;
4925 zone
->zone_start_pfn
= zone_start_pfn
;
4927 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
4928 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4930 (unsigned long)zone_idx(zone
),
4931 zone_start_pfn
, (zone_start_pfn
+ size
));
4933 zone_init_free_lists(zone
);
4938 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4939 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4942 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4944 int __meminit
__early_pfn_to_nid(unsigned long pfn
,
4945 struct mminit_pfnnid_cache
*state
)
4947 unsigned long start_pfn
, end_pfn
;
4950 if (state
->last_start
<= pfn
&& pfn
< state
->last_end
)
4951 return state
->last_nid
;
4953 nid
= memblock_search_pfn_nid(pfn
, &start_pfn
, &end_pfn
);
4955 state
->last_start
= start_pfn
;
4956 state
->last_end
= end_pfn
;
4957 state
->last_nid
= nid
;
4962 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4965 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
4966 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4967 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
4969 * If an architecture guarantees that all ranges registered contain no holes
4970 * and may be freed, this this function may be used instead of calling
4971 * memblock_free_early_nid() manually.
4973 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
4975 unsigned long start_pfn
, end_pfn
;
4978 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
4979 start_pfn
= min(start_pfn
, max_low_pfn
);
4980 end_pfn
= min(end_pfn
, max_low_pfn
);
4982 if (start_pfn
< end_pfn
)
4983 memblock_free_early_nid(PFN_PHYS(start_pfn
),
4984 (end_pfn
- start_pfn
) << PAGE_SHIFT
,
4990 * sparse_memory_present_with_active_regions - Call memory_present for each active range
4991 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4993 * If an architecture guarantees that all ranges registered contain no holes and may
4994 * be freed, this function may be used instead of calling memory_present() manually.
4996 void __init
sparse_memory_present_with_active_regions(int nid
)
4998 unsigned long start_pfn
, end_pfn
;
5001 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
5002 memory_present(this_nid
, start_pfn
, end_pfn
);
5006 * get_pfn_range_for_nid - Return the start and end page frames for a node
5007 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
5008 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
5009 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
5011 * It returns the start and end page frame of a node based on information
5012 * provided by memblock_set_node(). If called for a node
5013 * with no available memory, a warning is printed and the start and end
5016 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
5017 unsigned long *start_pfn
, unsigned long *end_pfn
)
5019 unsigned long this_start_pfn
, this_end_pfn
;
5025 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
5026 *start_pfn
= min(*start_pfn
, this_start_pfn
);
5027 *end_pfn
= max(*end_pfn
, this_end_pfn
);
5030 if (*start_pfn
== -1UL)
5035 * This finds a zone that can be used for ZONE_MOVABLE pages. The
5036 * assumption is made that zones within a node are ordered in monotonic
5037 * increasing memory addresses so that the "highest" populated zone is used
5039 static void __init
find_usable_zone_for_movable(void)
5042 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
5043 if (zone_index
== ZONE_MOVABLE
)
5046 if (arch_zone_highest_possible_pfn
[zone_index
] >
5047 arch_zone_lowest_possible_pfn
[zone_index
])
5051 VM_BUG_ON(zone_index
== -1);
5052 movable_zone
= zone_index
;
5056 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
5057 * because it is sized independent of architecture. Unlike the other zones,
5058 * the starting point for ZONE_MOVABLE is not fixed. It may be different
5059 * in each node depending on the size of each node and how evenly kernelcore
5060 * is distributed. This helper function adjusts the zone ranges
5061 * provided by the architecture for a given node by using the end of the
5062 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5063 * zones within a node are in order of monotonic increases memory addresses
5065 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
5066 unsigned long zone_type
,
5067 unsigned long node_start_pfn
,
5068 unsigned long node_end_pfn
,
5069 unsigned long *zone_start_pfn
,
5070 unsigned long *zone_end_pfn
)
5072 /* Only adjust if ZONE_MOVABLE is on this node */
5073 if (zone_movable_pfn
[nid
]) {
5074 /* Size ZONE_MOVABLE */
5075 if (zone_type
== ZONE_MOVABLE
) {
5076 *zone_start_pfn
= zone_movable_pfn
[nid
];
5077 *zone_end_pfn
= min(node_end_pfn
,
5078 arch_zone_highest_possible_pfn
[movable_zone
]);
5080 /* Check if this whole range is within ZONE_MOVABLE */
5081 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
5082 *zone_start_pfn
= *zone_end_pfn
;
5087 * Return the number of pages a zone spans in a node, including holes
5088 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5090 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5091 unsigned long zone_type
,
5092 unsigned long node_start_pfn
,
5093 unsigned long node_end_pfn
,
5094 unsigned long *zone_start_pfn
,
5095 unsigned long *zone_end_pfn
,
5096 unsigned long *ignored
)
5098 /* When hotadd a new node from cpu_up(), the node should be empty */
5099 if (!node_start_pfn
&& !node_end_pfn
)
5102 /* Get the start and end of the zone */
5103 *zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
5104 *zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
5105 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5106 node_start_pfn
, node_end_pfn
,
5107 zone_start_pfn
, zone_end_pfn
);
5109 /* Check that this node has pages within the zone's required range */
5110 if (*zone_end_pfn
< node_start_pfn
|| *zone_start_pfn
> node_end_pfn
)
5113 /* Move the zone boundaries inside the node if necessary */
5114 *zone_end_pfn
= min(*zone_end_pfn
, node_end_pfn
);
5115 *zone_start_pfn
= max(*zone_start_pfn
, node_start_pfn
);
5117 /* Return the spanned pages */
5118 return *zone_end_pfn
- *zone_start_pfn
;
5122 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
5123 * then all holes in the requested range will be accounted for.
5125 unsigned long __meminit
__absent_pages_in_range(int nid
,
5126 unsigned long range_start_pfn
,
5127 unsigned long range_end_pfn
)
5129 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
5130 unsigned long start_pfn
, end_pfn
;
5133 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5134 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
5135 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
5136 nr_absent
-= end_pfn
- start_pfn
;
5142 * absent_pages_in_range - Return number of page frames in holes within a range
5143 * @start_pfn: The start PFN to start searching for holes
5144 * @end_pfn: The end PFN to stop searching for holes
5146 * It returns the number of pages frames in memory holes within a range.
5148 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
5149 unsigned long end_pfn
)
5151 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
5154 /* Return the number of page frames in holes in a zone on a node */
5155 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5156 unsigned long zone_type
,
5157 unsigned long node_start_pfn
,
5158 unsigned long node_end_pfn
,
5159 unsigned long *ignored
)
5161 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
5162 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
5163 unsigned long zone_start_pfn
, zone_end_pfn
;
5164 unsigned long nr_absent
;
5166 /* When hotadd a new node from cpu_up(), the node should be empty */
5167 if (!node_start_pfn
&& !node_end_pfn
)
5170 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
5171 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
5173 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5174 node_start_pfn
, node_end_pfn
,
5175 &zone_start_pfn
, &zone_end_pfn
);
5176 nr_absent
= __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
5179 * ZONE_MOVABLE handling.
5180 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
5183 if (zone_movable_pfn
[nid
]) {
5184 if (mirrored_kernelcore
) {
5185 unsigned long start_pfn
, end_pfn
;
5186 struct memblock_region
*r
;
5188 for_each_memblock(memory
, r
) {
5189 start_pfn
= clamp(memblock_region_memory_base_pfn(r
),
5190 zone_start_pfn
, zone_end_pfn
);
5191 end_pfn
= clamp(memblock_region_memory_end_pfn(r
),
5192 zone_start_pfn
, zone_end_pfn
);
5194 if (zone_type
== ZONE_MOVABLE
&&
5195 memblock_is_mirror(r
))
5196 nr_absent
+= end_pfn
- start_pfn
;
5198 if (zone_type
== ZONE_NORMAL
&&
5199 !memblock_is_mirror(r
))
5200 nr_absent
+= end_pfn
- start_pfn
;
5203 if (zone_type
== ZONE_NORMAL
)
5204 nr_absent
+= node_end_pfn
- zone_movable_pfn
[nid
];
5211 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5212 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5213 unsigned long zone_type
,
5214 unsigned long node_start_pfn
,
5215 unsigned long node_end_pfn
,
5216 unsigned long *zone_start_pfn
,
5217 unsigned long *zone_end_pfn
,
5218 unsigned long *zones_size
)
5222 *zone_start_pfn
= node_start_pfn
;
5223 for (zone
= 0; zone
< zone_type
; zone
++)
5224 *zone_start_pfn
+= zones_size
[zone
];
5226 *zone_end_pfn
= *zone_start_pfn
+ zones_size
[zone_type
];
5228 return zones_size
[zone_type
];
5231 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5232 unsigned long zone_type
,
5233 unsigned long node_start_pfn
,
5234 unsigned long node_end_pfn
,
5235 unsigned long *zholes_size
)
5240 return zholes_size
[zone_type
];
5243 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5245 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
5246 unsigned long node_start_pfn
,
5247 unsigned long node_end_pfn
,
5248 unsigned long *zones_size
,
5249 unsigned long *zholes_size
)
5251 unsigned long realtotalpages
= 0, totalpages
= 0;
5254 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5255 struct zone
*zone
= pgdat
->node_zones
+ i
;
5256 unsigned long zone_start_pfn
, zone_end_pfn
;
5257 unsigned long size
, real_size
;
5259 size
= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
5265 real_size
= size
- zone_absent_pages_in_node(pgdat
->node_id
, i
,
5266 node_start_pfn
, node_end_pfn
,
5269 zone
->zone_start_pfn
= zone_start_pfn
;
5271 zone
->zone_start_pfn
= 0;
5272 zone
->spanned_pages
= size
;
5273 zone
->present_pages
= real_size
;
5276 realtotalpages
+= real_size
;
5279 pgdat
->node_spanned_pages
= totalpages
;
5280 pgdat
->node_present_pages
= realtotalpages
;
5281 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
5285 #ifndef CONFIG_SPARSEMEM
5287 * Calculate the size of the zone->blockflags rounded to an unsigned long
5288 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5289 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
5290 * round what is now in bits to nearest long in bits, then return it in
5293 static unsigned long __init
usemap_size(unsigned long zone_start_pfn
, unsigned long zonesize
)
5295 unsigned long usemapsize
;
5297 zonesize
+= zone_start_pfn
& (pageblock_nr_pages
-1);
5298 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
5299 usemapsize
= usemapsize
>> pageblock_order
;
5300 usemapsize
*= NR_PAGEBLOCK_BITS
;
5301 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
5303 return usemapsize
/ 8;
5306 static void __init
setup_usemap(struct pglist_data
*pgdat
,
5308 unsigned long zone_start_pfn
,
5309 unsigned long zonesize
)
5311 unsigned long usemapsize
= usemap_size(zone_start_pfn
, zonesize
);
5312 zone
->pageblock_flags
= NULL
;
5314 zone
->pageblock_flags
=
5315 memblock_virt_alloc_node_nopanic(usemapsize
,
5319 static inline void setup_usemap(struct pglist_data
*pgdat
, struct zone
*zone
,
5320 unsigned long zone_start_pfn
, unsigned long zonesize
) {}
5321 #endif /* CONFIG_SPARSEMEM */
5323 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
5325 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
5326 void __paginginit
set_pageblock_order(void)
5330 /* Check that pageblock_nr_pages has not already been setup */
5331 if (pageblock_order
)
5334 if (HPAGE_SHIFT
> PAGE_SHIFT
)
5335 order
= HUGETLB_PAGE_ORDER
;
5337 order
= MAX_ORDER
- 1;
5340 * Assume the largest contiguous order of interest is a huge page.
5341 * This value may be variable depending on boot parameters on IA64 and
5344 pageblock_order
= order
;
5346 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5349 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
5350 * is unused as pageblock_order is set at compile-time. See
5351 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5354 void __paginginit
set_pageblock_order(void)
5358 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5360 static unsigned long __paginginit
calc_memmap_size(unsigned long spanned_pages
,
5361 unsigned long present_pages
)
5363 unsigned long pages
= spanned_pages
;
5366 * Provide a more accurate estimation if there are holes within
5367 * the zone and SPARSEMEM is in use. If there are holes within the
5368 * zone, each populated memory region may cost us one or two extra
5369 * memmap pages due to alignment because memmap pages for each
5370 * populated regions may not naturally algined on page boundary.
5371 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5373 if (spanned_pages
> present_pages
+ (present_pages
>> 4) &&
5374 IS_ENABLED(CONFIG_SPARSEMEM
))
5375 pages
= present_pages
;
5377 return PAGE_ALIGN(pages
* sizeof(struct page
)) >> PAGE_SHIFT
;
5381 * Set up the zone data structures:
5382 * - mark all pages reserved
5383 * - mark all memory queues empty
5384 * - clear the memory bitmaps
5386 * NOTE: pgdat should get zeroed by caller.
5388 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
)
5391 int nid
= pgdat
->node_id
;
5394 pgdat_resize_init(pgdat
);
5395 #ifdef CONFIG_NUMA_BALANCING
5396 spin_lock_init(&pgdat
->numabalancing_migrate_lock
);
5397 pgdat
->numabalancing_migrate_nr_pages
= 0;
5398 pgdat
->numabalancing_migrate_next_window
= jiffies
;
5400 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5401 spin_lock_init(&pgdat
->split_queue_lock
);
5402 INIT_LIST_HEAD(&pgdat
->split_queue
);
5403 pgdat
->split_queue_len
= 0;
5405 init_waitqueue_head(&pgdat
->kswapd_wait
);
5406 init_waitqueue_head(&pgdat
->pfmemalloc_wait
);
5407 pgdat_page_ext_init(pgdat
);
5409 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5410 struct zone
*zone
= pgdat
->node_zones
+ j
;
5411 unsigned long size
, realsize
, freesize
, memmap_pages
;
5412 unsigned long zone_start_pfn
= zone
->zone_start_pfn
;
5414 size
= zone
->spanned_pages
;
5415 realsize
= freesize
= zone
->present_pages
;
5418 * Adjust freesize so that it accounts for how much memory
5419 * is used by this zone for memmap. This affects the watermark
5420 * and per-cpu initialisations
5422 memmap_pages
= calc_memmap_size(size
, realsize
);
5423 if (!is_highmem_idx(j
)) {
5424 if (freesize
>= memmap_pages
) {
5425 freesize
-= memmap_pages
;
5428 " %s zone: %lu pages used for memmap\n",
5429 zone_names
[j
], memmap_pages
);
5432 " %s zone: %lu pages exceeds freesize %lu\n",
5433 zone_names
[j
], memmap_pages
, freesize
);
5436 /* Account for reserved pages */
5437 if (j
== 0 && freesize
> dma_reserve
) {
5438 freesize
-= dma_reserve
;
5439 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
5440 zone_names
[0], dma_reserve
);
5443 if (!is_highmem_idx(j
))
5444 nr_kernel_pages
+= freesize
;
5445 /* Charge for highmem memmap if there are enough kernel pages */
5446 else if (nr_kernel_pages
> memmap_pages
* 2)
5447 nr_kernel_pages
-= memmap_pages
;
5448 nr_all_pages
+= freesize
;
5451 * Set an approximate value for lowmem here, it will be adjusted
5452 * when the bootmem allocator frees pages into the buddy system.
5453 * And all highmem pages will be managed by the buddy system.
5455 zone
->managed_pages
= is_highmem_idx(j
) ? realsize
: freesize
;
5458 zone
->min_unmapped_pages
= (freesize
*sysctl_min_unmapped_ratio
)
5460 zone
->min_slab_pages
= (freesize
* sysctl_min_slab_ratio
) / 100;
5462 zone
->name
= zone_names
[j
];
5463 spin_lock_init(&zone
->lock
);
5464 spin_lock_init(&zone
->lru_lock
);
5465 zone_seqlock_init(zone
);
5466 zone
->zone_pgdat
= pgdat
;
5467 zone_pcp_init(zone
);
5469 /* For bootup, initialized properly in watermark setup */
5470 mod_zone_page_state(zone
, NR_ALLOC_BATCH
, zone
->managed_pages
);
5472 lruvec_init(&zone
->lruvec
);
5476 set_pageblock_order();
5477 setup_usemap(pgdat
, zone
, zone_start_pfn
, size
);
5478 ret
= init_currently_empty_zone(zone
, zone_start_pfn
, size
);
5480 memmap_init(size
, nid
, j
, zone_start_pfn
);
5484 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
5486 unsigned long __maybe_unused start
= 0;
5487 unsigned long __maybe_unused offset
= 0;
5489 /* Skip empty nodes */
5490 if (!pgdat
->node_spanned_pages
)
5493 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5494 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
5495 offset
= pgdat
->node_start_pfn
- start
;
5496 /* ia64 gets its own node_mem_map, before this, without bootmem */
5497 if (!pgdat
->node_mem_map
) {
5498 unsigned long size
, end
;
5502 * The zone's endpoints aren't required to be MAX_ORDER
5503 * aligned but the node_mem_map endpoints must be in order
5504 * for the buddy allocator to function correctly.
5506 end
= pgdat_end_pfn(pgdat
);
5507 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
5508 size
= (end
- start
) * sizeof(struct page
);
5509 map
= alloc_remap(pgdat
->node_id
, size
);
5511 map
= memblock_virt_alloc_node_nopanic(size
,
5513 pgdat
->node_mem_map
= map
+ offset
;
5515 #ifndef CONFIG_NEED_MULTIPLE_NODES
5517 * With no DISCONTIG, the global mem_map is just set as node 0's
5519 if (pgdat
== NODE_DATA(0)) {
5520 mem_map
= NODE_DATA(0)->node_mem_map
;
5521 #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
5522 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
5524 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5527 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
5530 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
5531 unsigned long node_start_pfn
, unsigned long *zholes_size
)
5533 pg_data_t
*pgdat
= NODE_DATA(nid
);
5534 unsigned long start_pfn
= 0;
5535 unsigned long end_pfn
= 0;
5537 /* pg_data_t should be reset to zero when it's allocated */
5538 WARN_ON(pgdat
->nr_zones
|| pgdat
->classzone_idx
);
5540 reset_deferred_meminit(pgdat
);
5541 pgdat
->node_id
= nid
;
5542 pgdat
->node_start_pfn
= node_start_pfn
;
5543 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5544 get_pfn_range_for_nid(nid
, &start_pfn
, &end_pfn
);
5545 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid
,
5546 (u64
)start_pfn
<< PAGE_SHIFT
,
5547 end_pfn
? ((u64
)end_pfn
<< PAGE_SHIFT
) - 1 : 0);
5549 start_pfn
= node_start_pfn
;
5551 calculate_node_totalpages(pgdat
, start_pfn
, end_pfn
,
5552 zones_size
, zholes_size
);
5554 alloc_node_mem_map(pgdat
);
5555 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5556 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5557 nid
, (unsigned long)pgdat
,
5558 (unsigned long)pgdat
->node_mem_map
);
5561 free_area_init_core(pgdat
);
5564 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5566 #if MAX_NUMNODES > 1
5568 * Figure out the number of possible node ids.
5570 void __init
setup_nr_node_ids(void)
5572 unsigned int highest
;
5574 highest
= find_last_bit(node_possible_map
.bits
, MAX_NUMNODES
);
5575 nr_node_ids
= highest
+ 1;
5580 * node_map_pfn_alignment - determine the maximum internode alignment
5582 * This function should be called after node map is populated and sorted.
5583 * It calculates the maximum power of two alignment which can distinguish
5586 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5587 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
5588 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
5589 * shifted, 1GiB is enough and this function will indicate so.
5591 * This is used to test whether pfn -> nid mapping of the chosen memory
5592 * model has fine enough granularity to avoid incorrect mapping for the
5593 * populated node map.
5595 * Returns the determined alignment in pfn's. 0 if there is no alignment
5596 * requirement (single node).
5598 unsigned long __init
node_map_pfn_alignment(void)
5600 unsigned long accl_mask
= 0, last_end
= 0;
5601 unsigned long start
, end
, mask
;
5605 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
5606 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
5613 * Start with a mask granular enough to pin-point to the
5614 * start pfn and tick off bits one-by-one until it becomes
5615 * too coarse to separate the current node from the last.
5617 mask
= ~((1 << __ffs(start
)) - 1);
5618 while (mask
&& last_end
<= (start
& (mask
<< 1)))
5621 /* accumulate all internode masks */
5625 /* convert mask to number of pages */
5626 return ~accl_mask
+ 1;
5629 /* Find the lowest pfn for a node */
5630 static unsigned long __init
find_min_pfn_for_node(int nid
)
5632 unsigned long min_pfn
= ULONG_MAX
;
5633 unsigned long start_pfn
;
5636 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
5637 min_pfn
= min(min_pfn
, start_pfn
);
5639 if (min_pfn
== ULONG_MAX
) {
5641 "Could not find start_pfn for node %d\n", nid
);
5649 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5651 * It returns the minimum PFN based on information provided via
5652 * memblock_set_node().
5654 unsigned long __init
find_min_pfn_with_active_regions(void)
5656 return find_min_pfn_for_node(MAX_NUMNODES
);
5660 * early_calculate_totalpages()
5661 * Sum pages in active regions for movable zone.
5662 * Populate N_MEMORY for calculating usable_nodes.
5664 static unsigned long __init
early_calculate_totalpages(void)
5666 unsigned long totalpages
= 0;
5667 unsigned long start_pfn
, end_pfn
;
5670 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
5671 unsigned long pages
= end_pfn
- start_pfn
;
5673 totalpages
+= pages
;
5675 node_set_state(nid
, N_MEMORY
);
5681 * Find the PFN the Movable zone begins in each node. Kernel memory
5682 * is spread evenly between nodes as long as the nodes have enough
5683 * memory. When they don't, some nodes will have more kernelcore than
5686 static void __init
find_zone_movable_pfns_for_nodes(void)
5689 unsigned long usable_startpfn
;
5690 unsigned long kernelcore_node
, kernelcore_remaining
;
5691 /* save the state before borrow the nodemask */
5692 nodemask_t saved_node_state
= node_states
[N_MEMORY
];
5693 unsigned long totalpages
= early_calculate_totalpages();
5694 int usable_nodes
= nodes_weight(node_states
[N_MEMORY
]);
5695 struct memblock_region
*r
;
5697 /* Need to find movable_zone earlier when movable_node is specified. */
5698 find_usable_zone_for_movable();
5701 * If movable_node is specified, ignore kernelcore and movablecore
5704 if (movable_node_is_enabled()) {
5705 for_each_memblock(memory
, r
) {
5706 if (!memblock_is_hotpluggable(r
))
5711 usable_startpfn
= PFN_DOWN(r
->base
);
5712 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
5713 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
5721 * If kernelcore=mirror is specified, ignore movablecore option
5723 if (mirrored_kernelcore
) {
5724 bool mem_below_4gb_not_mirrored
= false;
5726 for_each_memblock(memory
, r
) {
5727 if (memblock_is_mirror(r
))
5732 usable_startpfn
= memblock_region_memory_base_pfn(r
);
5734 if (usable_startpfn
< 0x100000) {
5735 mem_below_4gb_not_mirrored
= true;
5739 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
5740 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
5744 if (mem_below_4gb_not_mirrored
)
5745 pr_warn("This configuration results in unmirrored kernel memory.");
5751 * If movablecore=nn[KMG] was specified, calculate what size of
5752 * kernelcore that corresponds so that memory usable for
5753 * any allocation type is evenly spread. If both kernelcore
5754 * and movablecore are specified, then the value of kernelcore
5755 * will be used for required_kernelcore if it's greater than
5756 * what movablecore would have allowed.
5758 if (required_movablecore
) {
5759 unsigned long corepages
;
5762 * Round-up so that ZONE_MOVABLE is at least as large as what
5763 * was requested by the user
5765 required_movablecore
=
5766 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
5767 required_movablecore
= min(totalpages
, required_movablecore
);
5768 corepages
= totalpages
- required_movablecore
;
5770 required_kernelcore
= max(required_kernelcore
, corepages
);
5774 * If kernelcore was not specified or kernelcore size is larger
5775 * than totalpages, there is no ZONE_MOVABLE.
5777 if (!required_kernelcore
|| required_kernelcore
>= totalpages
)
5780 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5781 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
5784 /* Spread kernelcore memory as evenly as possible throughout nodes */
5785 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5786 for_each_node_state(nid
, N_MEMORY
) {
5787 unsigned long start_pfn
, end_pfn
;
5790 * Recalculate kernelcore_node if the division per node
5791 * now exceeds what is necessary to satisfy the requested
5792 * amount of memory for the kernel
5794 if (required_kernelcore
< kernelcore_node
)
5795 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5798 * As the map is walked, we track how much memory is usable
5799 * by the kernel using kernelcore_remaining. When it is
5800 * 0, the rest of the node is usable by ZONE_MOVABLE
5802 kernelcore_remaining
= kernelcore_node
;
5804 /* Go through each range of PFNs within this node */
5805 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5806 unsigned long size_pages
;
5808 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
5809 if (start_pfn
>= end_pfn
)
5812 /* Account for what is only usable for kernelcore */
5813 if (start_pfn
< usable_startpfn
) {
5814 unsigned long kernel_pages
;
5815 kernel_pages
= min(end_pfn
, usable_startpfn
)
5818 kernelcore_remaining
-= min(kernel_pages
,
5819 kernelcore_remaining
);
5820 required_kernelcore
-= min(kernel_pages
,
5821 required_kernelcore
);
5823 /* Continue if range is now fully accounted */
5824 if (end_pfn
<= usable_startpfn
) {
5827 * Push zone_movable_pfn to the end so
5828 * that if we have to rebalance
5829 * kernelcore across nodes, we will
5830 * not double account here
5832 zone_movable_pfn
[nid
] = end_pfn
;
5835 start_pfn
= usable_startpfn
;
5839 * The usable PFN range for ZONE_MOVABLE is from
5840 * start_pfn->end_pfn. Calculate size_pages as the
5841 * number of pages used as kernelcore
5843 size_pages
= end_pfn
- start_pfn
;
5844 if (size_pages
> kernelcore_remaining
)
5845 size_pages
= kernelcore_remaining
;
5846 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
5849 * Some kernelcore has been met, update counts and
5850 * break if the kernelcore for this node has been
5853 required_kernelcore
-= min(required_kernelcore
,
5855 kernelcore_remaining
-= size_pages
;
5856 if (!kernelcore_remaining
)
5862 * If there is still required_kernelcore, we do another pass with one
5863 * less node in the count. This will push zone_movable_pfn[nid] further
5864 * along on the nodes that still have memory until kernelcore is
5868 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
5872 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5873 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
5874 zone_movable_pfn
[nid
] =
5875 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
5878 /* restore the node_state */
5879 node_states
[N_MEMORY
] = saved_node_state
;
5882 /* Any regular or high memory on that node ? */
5883 static void check_for_memory(pg_data_t
*pgdat
, int nid
)
5885 enum zone_type zone_type
;
5887 if (N_MEMORY
== N_NORMAL_MEMORY
)
5890 for (zone_type
= 0; zone_type
<= ZONE_MOVABLE
- 1; zone_type
++) {
5891 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
5892 if (populated_zone(zone
)) {
5893 node_set_state(nid
, N_HIGH_MEMORY
);
5894 if (N_NORMAL_MEMORY
!= N_HIGH_MEMORY
&&
5895 zone_type
<= ZONE_NORMAL
)
5896 node_set_state(nid
, N_NORMAL_MEMORY
);
5903 * free_area_init_nodes - Initialise all pg_data_t and zone data
5904 * @max_zone_pfn: an array of max PFNs for each zone
5906 * This will call free_area_init_node() for each active node in the system.
5907 * Using the page ranges provided by memblock_set_node(), the size of each
5908 * zone in each node and their holes is calculated. If the maximum PFN
5909 * between two adjacent zones match, it is assumed that the zone is empty.
5910 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5911 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5912 * starts where the previous one ended. For example, ZONE_DMA32 starts
5913 * at arch_max_dma_pfn.
5915 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
5917 unsigned long start_pfn
, end_pfn
;
5920 /* Record where the zone boundaries are */
5921 memset(arch_zone_lowest_possible_pfn
, 0,
5922 sizeof(arch_zone_lowest_possible_pfn
));
5923 memset(arch_zone_highest_possible_pfn
, 0,
5924 sizeof(arch_zone_highest_possible_pfn
));
5925 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
5926 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
5927 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
5928 if (i
== ZONE_MOVABLE
)
5930 arch_zone_lowest_possible_pfn
[i
] =
5931 arch_zone_highest_possible_pfn
[i
-1];
5932 arch_zone_highest_possible_pfn
[i
] =
5933 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
5935 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
5936 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
5938 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5939 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
5940 find_zone_movable_pfns_for_nodes();
5942 /* Print out the zone ranges */
5943 pr_info("Zone ranges:\n");
5944 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5945 if (i
== ZONE_MOVABLE
)
5947 pr_info(" %-8s ", zone_names
[i
]);
5948 if (arch_zone_lowest_possible_pfn
[i
] ==
5949 arch_zone_highest_possible_pfn
[i
])
5952 pr_cont("[mem %#018Lx-%#018Lx]\n",
5953 (u64
)arch_zone_lowest_possible_pfn
[i
]
5955 ((u64
)arch_zone_highest_possible_pfn
[i
]
5956 << PAGE_SHIFT
) - 1);
5959 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5960 pr_info("Movable zone start for each node\n");
5961 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
5962 if (zone_movable_pfn
[i
])
5963 pr_info(" Node %d: %#018Lx\n", i
,
5964 (u64
)zone_movable_pfn
[i
] << PAGE_SHIFT
);
5967 /* Print out the early node map */
5968 pr_info("Early memory node ranges\n");
5969 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
5970 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid
,
5971 (u64
)start_pfn
<< PAGE_SHIFT
,
5972 ((u64
)end_pfn
<< PAGE_SHIFT
) - 1);
5974 /* Initialise every node */
5975 mminit_verify_pageflags_layout();
5976 setup_nr_node_ids();
5977 for_each_online_node(nid
) {
5978 pg_data_t
*pgdat
= NODE_DATA(nid
);
5979 free_area_init_node(nid
, NULL
,
5980 find_min_pfn_for_node(nid
), NULL
);
5982 /* Any memory on that node */
5983 if (pgdat
->node_present_pages
)
5984 node_set_state(nid
, N_MEMORY
);
5985 check_for_memory(pgdat
, nid
);
5989 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
5991 unsigned long long coremem
;
5995 coremem
= memparse(p
, &p
);
5996 *core
= coremem
>> PAGE_SHIFT
;
5998 /* Paranoid check that UL is enough for the coremem value */
5999 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
6005 * kernelcore=size sets the amount of memory for use for allocations that
6006 * cannot be reclaimed or migrated.
6008 static int __init
cmdline_parse_kernelcore(char *p
)
6010 /* parse kernelcore=mirror */
6011 if (parse_option_str(p
, "mirror")) {
6012 mirrored_kernelcore
= true;
6016 return cmdline_parse_core(p
, &required_kernelcore
);
6020 * movablecore=size sets the amount of memory for use for allocations that
6021 * can be reclaimed or migrated.
6023 static int __init
cmdline_parse_movablecore(char *p
)
6025 return cmdline_parse_core(p
, &required_movablecore
);
6028 early_param("kernelcore", cmdline_parse_kernelcore
);
6029 early_param("movablecore", cmdline_parse_movablecore
);
6031 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6033 void adjust_managed_page_count(struct page
*page
, long count
)
6035 spin_lock(&managed_page_count_lock
);
6036 page_zone(page
)->managed_pages
+= count
;
6037 totalram_pages
+= count
;
6038 #ifdef CONFIG_HIGHMEM
6039 if (PageHighMem(page
))
6040 totalhigh_pages
+= count
;
6042 spin_unlock(&managed_page_count_lock
);
6044 EXPORT_SYMBOL(adjust_managed_page_count
);
6046 unsigned long free_reserved_area(void *start
, void *end
, int poison
, char *s
)
6049 unsigned long pages
= 0;
6051 start
= (void *)PAGE_ALIGN((unsigned long)start
);
6052 end
= (void *)((unsigned long)end
& PAGE_MASK
);
6053 for (pos
= start
; pos
< end
; pos
+= PAGE_SIZE
, pages
++) {
6054 if ((unsigned int)poison
<= 0xFF)
6055 memset(pos
, poison
, PAGE_SIZE
);
6056 free_reserved_page(virt_to_page(pos
));
6060 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
6061 s
, pages
<< (PAGE_SHIFT
- 10), start
, end
);
6065 EXPORT_SYMBOL(free_reserved_area
);
6067 #ifdef CONFIG_HIGHMEM
6068 void free_highmem_page(struct page
*page
)
6070 __free_reserved_page(page
);
6072 page_zone(page
)->managed_pages
++;
6078 void __init
mem_init_print_info(const char *str
)
6080 unsigned long physpages
, codesize
, datasize
, rosize
, bss_size
;
6081 unsigned long init_code_size
, init_data_size
;
6083 physpages
= get_num_physpages();
6084 codesize
= _etext
- _stext
;
6085 datasize
= _edata
- _sdata
;
6086 rosize
= __end_rodata
- __start_rodata
;
6087 bss_size
= __bss_stop
- __bss_start
;
6088 init_data_size
= __init_end
- __init_begin
;
6089 init_code_size
= _einittext
- _sinittext
;
6092 * Detect special cases and adjust section sizes accordingly:
6093 * 1) .init.* may be embedded into .data sections
6094 * 2) .init.text.* may be out of [__init_begin, __init_end],
6095 * please refer to arch/tile/kernel/vmlinux.lds.S.
6096 * 3) .rodata.* may be embedded into .text or .data sections.
6098 #define adj_init_size(start, end, size, pos, adj) \
6100 if (start <= pos && pos < end && size > adj) \
6104 adj_init_size(__init_begin
, __init_end
, init_data_size
,
6105 _sinittext
, init_code_size
);
6106 adj_init_size(_stext
, _etext
, codesize
, _sinittext
, init_code_size
);
6107 adj_init_size(_sdata
, _edata
, datasize
, __init_begin
, init_data_size
);
6108 adj_init_size(_stext
, _etext
, codesize
, __start_rodata
, rosize
);
6109 adj_init_size(_sdata
, _edata
, datasize
, __start_rodata
, rosize
);
6111 #undef adj_init_size
6113 pr_info("Memory: %luK/%luK available "
6114 "(%luK kernel code, %luK rwdata, %luK rodata, "
6115 "%luK init, %luK bss, %luK reserved, %luK cma-reserved"
6116 #ifdef CONFIG_HIGHMEM
6120 nr_free_pages() << (PAGE_SHIFT
-10), physpages
<< (PAGE_SHIFT
-10),
6121 codesize
>> 10, datasize
>> 10, rosize
>> 10,
6122 (init_data_size
+ init_code_size
) >> 10, bss_size
>> 10,
6123 (physpages
- totalram_pages
- totalcma_pages
) << (PAGE_SHIFT
-10),
6124 totalcma_pages
<< (PAGE_SHIFT
-10),
6125 #ifdef CONFIG_HIGHMEM
6126 totalhigh_pages
<< (PAGE_SHIFT
-10),
6128 str
? ", " : "", str
? str
: "");
6132 * set_dma_reserve - set the specified number of pages reserved in the first zone
6133 * @new_dma_reserve: The number of pages to mark reserved
6135 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
6136 * In the DMA zone, a significant percentage may be consumed by kernel image
6137 * and other unfreeable allocations which can skew the watermarks badly. This
6138 * function may optionally be used to account for unfreeable pages in the
6139 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6140 * smaller per-cpu batchsize.
6142 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
6144 dma_reserve
= new_dma_reserve
;
6147 void __init
free_area_init(unsigned long *zones_size
)
6149 free_area_init_node(0, zones_size
,
6150 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
6153 static int page_alloc_cpu_notify(struct notifier_block
*self
,
6154 unsigned long action
, void *hcpu
)
6156 int cpu
= (unsigned long)hcpu
;
6158 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
6159 lru_add_drain_cpu(cpu
);
6163 * Spill the event counters of the dead processor
6164 * into the current processors event counters.
6165 * This artificially elevates the count of the current
6168 vm_events_fold_cpu(cpu
);
6171 * Zero the differential counters of the dead processor
6172 * so that the vm statistics are consistent.
6174 * This is only okay since the processor is dead and cannot
6175 * race with what we are doing.
6177 cpu_vm_stats_fold(cpu
);
6182 void __init
page_alloc_init(void)
6184 hotcpu_notifier(page_alloc_cpu_notify
, 0);
6188 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
6189 * or min_free_kbytes changes.
6191 static void calculate_totalreserve_pages(void)
6193 struct pglist_data
*pgdat
;
6194 unsigned long reserve_pages
= 0;
6195 enum zone_type i
, j
;
6197 for_each_online_pgdat(pgdat
) {
6198 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
6199 struct zone
*zone
= pgdat
->node_zones
+ i
;
6202 /* Find valid and maximum lowmem_reserve in the zone */
6203 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
6204 if (zone
->lowmem_reserve
[j
] > max
)
6205 max
= zone
->lowmem_reserve
[j
];
6208 /* we treat the high watermark as reserved pages. */
6209 max
+= high_wmark_pages(zone
);
6211 if (max
> zone
->managed_pages
)
6212 max
= zone
->managed_pages
;
6214 zone
->totalreserve_pages
= max
;
6216 reserve_pages
+= max
;
6219 totalreserve_pages
= reserve_pages
;
6223 * setup_per_zone_lowmem_reserve - called whenever
6224 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
6225 * has a correct pages reserved value, so an adequate number of
6226 * pages are left in the zone after a successful __alloc_pages().
6228 static void setup_per_zone_lowmem_reserve(void)
6230 struct pglist_data
*pgdat
;
6231 enum zone_type j
, idx
;
6233 for_each_online_pgdat(pgdat
) {
6234 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
6235 struct zone
*zone
= pgdat
->node_zones
+ j
;
6236 unsigned long managed_pages
= zone
->managed_pages
;
6238 zone
->lowmem_reserve
[j
] = 0;
6242 struct zone
*lower_zone
;
6246 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
6247 sysctl_lowmem_reserve_ratio
[idx
] = 1;
6249 lower_zone
= pgdat
->node_zones
+ idx
;
6250 lower_zone
->lowmem_reserve
[j
] = managed_pages
/
6251 sysctl_lowmem_reserve_ratio
[idx
];
6252 managed_pages
+= lower_zone
->managed_pages
;
6257 /* update totalreserve_pages */
6258 calculate_totalreserve_pages();
6261 static void __setup_per_zone_wmarks(void)
6263 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
6264 unsigned long lowmem_pages
= 0;
6266 unsigned long flags
;
6268 /* Calculate total number of !ZONE_HIGHMEM pages */
6269 for_each_zone(zone
) {
6270 if (!is_highmem(zone
))
6271 lowmem_pages
+= zone
->managed_pages
;
6274 for_each_zone(zone
) {
6277 spin_lock_irqsave(&zone
->lock
, flags
);
6278 tmp
= (u64
)pages_min
* zone
->managed_pages
;
6279 do_div(tmp
, lowmem_pages
);
6280 if (is_highmem(zone
)) {
6282 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6283 * need highmem pages, so cap pages_min to a small
6286 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
6287 * deltas control asynch page reclaim, and so should
6288 * not be capped for highmem.
6290 unsigned long min_pages
;
6292 min_pages
= zone
->managed_pages
/ 1024;
6293 min_pages
= clamp(min_pages
, SWAP_CLUSTER_MAX
, 128UL);
6294 zone
->watermark
[WMARK_MIN
] = min_pages
;
6297 * If it's a lowmem zone, reserve a number of pages
6298 * proportionate to the zone's size.
6300 zone
->watermark
[WMARK_MIN
] = tmp
;
6303 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + (tmp
>> 2);
6304 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + (tmp
>> 1);
6306 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
6307 high_wmark_pages(zone
) - low_wmark_pages(zone
) -
6308 atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]));
6310 spin_unlock_irqrestore(&zone
->lock
, flags
);
6313 /* update totalreserve_pages */
6314 calculate_totalreserve_pages();
6318 * setup_per_zone_wmarks - called when min_free_kbytes changes
6319 * or when memory is hot-{added|removed}
6321 * Ensures that the watermark[min,low,high] values for each zone are set
6322 * correctly with respect to min_free_kbytes.
6324 void setup_per_zone_wmarks(void)
6326 mutex_lock(&zonelists_mutex
);
6327 __setup_per_zone_wmarks();
6328 mutex_unlock(&zonelists_mutex
);
6332 * The inactive anon list should be small enough that the VM never has to
6333 * do too much work, but large enough that each inactive page has a chance
6334 * to be referenced again before it is swapped out.
6336 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
6337 * INACTIVE_ANON pages on this zone's LRU, maintained by the
6338 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
6339 * the anonymous pages are kept on the inactive list.
6342 * memory ratio inactive anon
6343 * -------------------------------------
6352 static void __meminit
calculate_zone_inactive_ratio(struct zone
*zone
)
6354 unsigned int gb
, ratio
;
6356 /* Zone size in gigabytes */
6357 gb
= zone
->managed_pages
>> (30 - PAGE_SHIFT
);
6359 ratio
= int_sqrt(10 * gb
);
6363 zone
->inactive_ratio
= ratio
;
6366 static void __meminit
setup_per_zone_inactive_ratio(void)
6371 calculate_zone_inactive_ratio(zone
);
6375 * Initialise min_free_kbytes.
6377 * For small machines we want it small (128k min). For large machines
6378 * we want it large (64MB max). But it is not linear, because network
6379 * bandwidth does not increase linearly with machine size. We use
6381 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
6382 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6398 int __meminit
init_per_zone_wmark_min(void)
6400 unsigned long lowmem_kbytes
;
6401 int new_min_free_kbytes
;
6403 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
6404 new_min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
6406 if (new_min_free_kbytes
> user_min_free_kbytes
) {
6407 min_free_kbytes
= new_min_free_kbytes
;
6408 if (min_free_kbytes
< 128)
6409 min_free_kbytes
= 128;
6410 if (min_free_kbytes
> 65536)
6411 min_free_kbytes
= 65536;
6413 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6414 new_min_free_kbytes
, user_min_free_kbytes
);
6416 setup_per_zone_wmarks();
6417 refresh_zone_stat_thresholds();
6418 setup_per_zone_lowmem_reserve();
6419 setup_per_zone_inactive_ratio();
6422 module_init(init_per_zone_wmark_min
)
6425 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
6426 * that we can call two helper functions whenever min_free_kbytes
6429 int min_free_kbytes_sysctl_handler(struct ctl_table
*table
, int write
,
6430 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6434 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6439 user_min_free_kbytes
= min_free_kbytes
;
6440 setup_per_zone_wmarks();
6446 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6447 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6452 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6457 zone
->min_unmapped_pages
= (zone
->managed_pages
*
6458 sysctl_min_unmapped_ratio
) / 100;
6462 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6463 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6468 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6473 zone
->min_slab_pages
= (zone
->managed_pages
*
6474 sysctl_min_slab_ratio
) / 100;
6480 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6481 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6482 * whenever sysctl_lowmem_reserve_ratio changes.
6484 * The reserve ratio obviously has absolutely no relation with the
6485 * minimum watermarks. The lowmem reserve ratio can only make sense
6486 * if in function of the boot time zone sizes.
6488 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6489 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6491 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6492 setup_per_zone_lowmem_reserve();
6497 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
6498 * cpu. It is the fraction of total pages in each zone that a hot per cpu
6499 * pagelist can have before it gets flushed back to buddy allocator.
6501 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table
*table
, int write
,
6502 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6505 int old_percpu_pagelist_fraction
;
6508 mutex_lock(&pcp_batch_high_lock
);
6509 old_percpu_pagelist_fraction
= percpu_pagelist_fraction
;
6511 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6512 if (!write
|| ret
< 0)
6515 /* Sanity checking to avoid pcp imbalance */
6516 if (percpu_pagelist_fraction
&&
6517 percpu_pagelist_fraction
< MIN_PERCPU_PAGELIST_FRACTION
) {
6518 percpu_pagelist_fraction
= old_percpu_pagelist_fraction
;
6524 if (percpu_pagelist_fraction
== old_percpu_pagelist_fraction
)
6527 for_each_populated_zone(zone
) {
6530 for_each_possible_cpu(cpu
)
6531 pageset_set_high_and_batch(zone
,
6532 per_cpu_ptr(zone
->pageset
, cpu
));
6535 mutex_unlock(&pcp_batch_high_lock
);
6540 int hashdist
= HASHDIST_DEFAULT
;
6542 static int __init
set_hashdist(char *str
)
6546 hashdist
= simple_strtoul(str
, &str
, 0);
6549 __setup("hashdist=", set_hashdist
);
6553 * allocate a large system hash table from bootmem
6554 * - it is assumed that the hash table must contain an exact power-of-2
6555 * quantity of entries
6556 * - limit is the number of hash buckets, not the total allocation size
6558 void *__init
alloc_large_system_hash(const char *tablename
,
6559 unsigned long bucketsize
,
6560 unsigned long numentries
,
6563 unsigned int *_hash_shift
,
6564 unsigned int *_hash_mask
,
6565 unsigned long low_limit
,
6566 unsigned long high_limit
)
6568 unsigned long long max
= high_limit
;
6569 unsigned long log2qty
, size
;
6572 /* allow the kernel cmdline to have a say */
6574 /* round applicable memory size up to nearest megabyte */
6575 numentries
= nr_kernel_pages
;
6577 /* It isn't necessary when PAGE_SIZE >= 1MB */
6578 if (PAGE_SHIFT
< 20)
6579 numentries
= round_up(numentries
, (1<<20)/PAGE_SIZE
);
6581 /* limit to 1 bucket per 2^scale bytes of low memory */
6582 if (scale
> PAGE_SHIFT
)
6583 numentries
>>= (scale
- PAGE_SHIFT
);
6585 numentries
<<= (PAGE_SHIFT
- scale
);
6587 /* Make sure we've got at least a 0-order allocation.. */
6588 if (unlikely(flags
& HASH_SMALL
)) {
6589 /* Makes no sense without HASH_EARLY */
6590 WARN_ON(!(flags
& HASH_EARLY
));
6591 if (!(numentries
>> *_hash_shift
)) {
6592 numentries
= 1UL << *_hash_shift
;
6593 BUG_ON(!numentries
);
6595 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
6596 numentries
= PAGE_SIZE
/ bucketsize
;
6598 numentries
= roundup_pow_of_two(numentries
);
6600 /* limit allocation size to 1/16 total memory by default */
6602 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
6603 do_div(max
, bucketsize
);
6605 max
= min(max
, 0x80000000ULL
);
6607 if (numentries
< low_limit
)
6608 numentries
= low_limit
;
6609 if (numentries
> max
)
6612 log2qty
= ilog2(numentries
);
6615 size
= bucketsize
<< log2qty
;
6616 if (flags
& HASH_EARLY
)
6617 table
= memblock_virt_alloc_nopanic(size
, 0);
6619 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
6622 * If bucketsize is not a power-of-two, we may free
6623 * some pages at the end of hash table which
6624 * alloc_pages_exact() automatically does
6626 if (get_order(size
) < MAX_ORDER
) {
6627 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
6628 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
6631 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
6634 panic("Failed to allocate %s hash table\n", tablename
);
6636 printk(KERN_INFO
"%s hash table entries: %ld (order: %d, %lu bytes)\n",
6639 ilog2(size
) - PAGE_SHIFT
,
6643 *_hash_shift
= log2qty
;
6645 *_hash_mask
= (1 << log2qty
) - 1;
6650 /* Return a pointer to the bitmap storing bits affecting a block of pages */
6651 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
6654 #ifdef CONFIG_SPARSEMEM
6655 return __pfn_to_section(pfn
)->pageblock_flags
;
6657 return zone
->pageblock_flags
;
6658 #endif /* CONFIG_SPARSEMEM */
6661 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
6663 #ifdef CONFIG_SPARSEMEM
6664 pfn
&= (PAGES_PER_SECTION
-1);
6665 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
6667 pfn
= pfn
- round_down(zone
->zone_start_pfn
, pageblock_nr_pages
);
6668 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
6669 #endif /* CONFIG_SPARSEMEM */
6673 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
6674 * @page: The page within the block of interest
6675 * @pfn: The target page frame number
6676 * @end_bitidx: The last bit of interest to retrieve
6677 * @mask: mask of bits that the caller is interested in
6679 * Return: pageblock_bits flags
6681 unsigned long get_pfnblock_flags_mask(struct page
*page
, unsigned long pfn
,
6682 unsigned long end_bitidx
,
6686 unsigned long *bitmap
;
6687 unsigned long bitidx
, word_bitidx
;
6690 zone
= page_zone(page
);
6691 bitmap
= get_pageblock_bitmap(zone
, pfn
);
6692 bitidx
= pfn_to_bitidx(zone
, pfn
);
6693 word_bitidx
= bitidx
/ BITS_PER_LONG
;
6694 bitidx
&= (BITS_PER_LONG
-1);
6696 word
= bitmap
[word_bitidx
];
6697 bitidx
+= end_bitidx
;
6698 return (word
>> (BITS_PER_LONG
- bitidx
- 1)) & mask
;
6702 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
6703 * @page: The page within the block of interest
6704 * @flags: The flags to set
6705 * @pfn: The target page frame number
6706 * @end_bitidx: The last bit of interest
6707 * @mask: mask of bits that the caller is interested in
6709 void set_pfnblock_flags_mask(struct page
*page
, unsigned long flags
,
6711 unsigned long end_bitidx
,
6715 unsigned long *bitmap
;
6716 unsigned long bitidx
, word_bitidx
;
6717 unsigned long old_word
, word
;
6719 BUILD_BUG_ON(NR_PAGEBLOCK_BITS
!= 4);
6721 zone
= page_zone(page
);
6722 bitmap
= get_pageblock_bitmap(zone
, pfn
);
6723 bitidx
= pfn_to_bitidx(zone
, pfn
);
6724 word_bitidx
= bitidx
/ BITS_PER_LONG
;
6725 bitidx
&= (BITS_PER_LONG
-1);
6727 VM_BUG_ON_PAGE(!zone_spans_pfn(zone
, pfn
), page
);
6729 bitidx
+= end_bitidx
;
6730 mask
<<= (BITS_PER_LONG
- bitidx
- 1);
6731 flags
<<= (BITS_PER_LONG
- bitidx
- 1);
6733 word
= READ_ONCE(bitmap
[word_bitidx
]);
6735 old_word
= cmpxchg(&bitmap
[word_bitidx
], word
, (word
& ~mask
) | flags
);
6736 if (word
== old_word
)
6743 * This function checks whether pageblock includes unmovable pages or not.
6744 * If @count is not zero, it is okay to include less @count unmovable pages
6746 * PageLRU check without isolation or lru_lock could race so that
6747 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6748 * expect this function should be exact.
6750 bool has_unmovable_pages(struct zone
*zone
, struct page
*page
, int count
,
6751 bool skip_hwpoisoned_pages
)
6753 unsigned long pfn
, iter
, found
;
6757 * For avoiding noise data, lru_add_drain_all() should be called
6758 * If ZONE_MOVABLE, the zone never contains unmovable pages
6760 if (zone_idx(zone
) == ZONE_MOVABLE
)
6762 mt
= get_pageblock_migratetype(page
);
6763 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
6766 pfn
= page_to_pfn(page
);
6767 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
6768 unsigned long check
= pfn
+ iter
;
6770 if (!pfn_valid_within(check
))
6773 page
= pfn_to_page(check
);
6776 * Hugepages are not in LRU lists, but they're movable.
6777 * We need not scan over tail pages bacause we don't
6778 * handle each tail page individually in migration.
6780 if (PageHuge(page
)) {
6781 iter
= round_up(iter
+ 1, 1<<compound_order(page
)) - 1;
6786 * We can't use page_count without pin a page
6787 * because another CPU can free compound page.
6788 * This check already skips compound tails of THP
6789 * because their page->_count is zero at all time.
6791 if (!atomic_read(&page
->_count
)) {
6792 if (PageBuddy(page
))
6793 iter
+= (1 << page_order(page
)) - 1;
6798 * The HWPoisoned page may be not in buddy system, and
6799 * page_count() is not 0.
6801 if (skip_hwpoisoned_pages
&& PageHWPoison(page
))
6807 * If there are RECLAIMABLE pages, we need to check
6808 * it. But now, memory offline itself doesn't call
6809 * shrink_node_slabs() and it still to be fixed.
6812 * If the page is not RAM, page_count()should be 0.
6813 * we don't need more check. This is an _used_ not-movable page.
6815 * The problematic thing here is PG_reserved pages. PG_reserved
6816 * is set to both of a memory hole page and a _used_ kernel
6825 bool is_pageblock_removable_nolock(struct page
*page
)
6831 * We have to be careful here because we are iterating over memory
6832 * sections which are not zone aware so we might end up outside of
6833 * the zone but still within the section.
6834 * We have to take care about the node as well. If the node is offline
6835 * its NODE_DATA will be NULL - see page_zone.
6837 if (!node_online(page_to_nid(page
)))
6840 zone
= page_zone(page
);
6841 pfn
= page_to_pfn(page
);
6842 if (!zone_spans_pfn(zone
, pfn
))
6845 return !has_unmovable_pages(zone
, page
, 0, true);
6848 #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
6850 static unsigned long pfn_max_align_down(unsigned long pfn
)
6852 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
6853 pageblock_nr_pages
) - 1);
6856 static unsigned long pfn_max_align_up(unsigned long pfn
)
6858 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
6859 pageblock_nr_pages
));
6862 /* [start, end) must belong to a single zone. */
6863 static int __alloc_contig_migrate_range(struct compact_control
*cc
,
6864 unsigned long start
, unsigned long end
)
6866 /* This function is based on compact_zone() from compaction.c. */
6867 unsigned long nr_reclaimed
;
6868 unsigned long pfn
= start
;
6869 unsigned int tries
= 0;
6874 while (pfn
< end
|| !list_empty(&cc
->migratepages
)) {
6875 if (fatal_signal_pending(current
)) {
6880 if (list_empty(&cc
->migratepages
)) {
6881 cc
->nr_migratepages
= 0;
6882 pfn
= isolate_migratepages_range(cc
, pfn
, end
);
6888 } else if (++tries
== 5) {
6889 ret
= ret
< 0 ? ret
: -EBUSY
;
6893 nr_reclaimed
= reclaim_clean_pages_from_list(cc
->zone
,
6895 cc
->nr_migratepages
-= nr_reclaimed
;
6897 ret
= migrate_pages(&cc
->migratepages
, alloc_migrate_target
,
6898 NULL
, 0, cc
->mode
, MR_CMA
);
6901 putback_movable_pages(&cc
->migratepages
);
6908 * alloc_contig_range() -- tries to allocate given range of pages
6909 * @start: start PFN to allocate
6910 * @end: one-past-the-last PFN to allocate
6911 * @migratetype: migratetype of the underlaying pageblocks (either
6912 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6913 * in range must have the same migratetype and it must
6914 * be either of the two.
6916 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6917 * aligned, however it's the caller's responsibility to guarantee that
6918 * we are the only thread that changes migrate type of pageblocks the
6921 * The PFN range must belong to a single zone.
6923 * Returns zero on success or negative error code. On success all
6924 * pages which PFN is in [start, end) are allocated for the caller and
6925 * need to be freed with free_contig_range().
6927 int alloc_contig_range(unsigned long start
, unsigned long end
,
6928 unsigned migratetype
)
6930 unsigned long outer_start
, outer_end
;
6934 struct compact_control cc
= {
6935 .nr_migratepages
= 0,
6937 .zone
= page_zone(pfn_to_page(start
)),
6938 .mode
= MIGRATE_SYNC
,
6939 .ignore_skip_hint
= true,
6941 INIT_LIST_HEAD(&cc
.migratepages
);
6944 * What we do here is we mark all pageblocks in range as
6945 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6946 * have different sizes, and due to the way page allocator
6947 * work, we align the range to biggest of the two pages so
6948 * that page allocator won't try to merge buddies from
6949 * different pageblocks and change MIGRATE_ISOLATE to some
6950 * other migration type.
6952 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6953 * migrate the pages from an unaligned range (ie. pages that
6954 * we are interested in). This will put all the pages in
6955 * range back to page allocator as MIGRATE_ISOLATE.
6957 * When this is done, we take the pages in range from page
6958 * allocator removing them from the buddy system. This way
6959 * page allocator will never consider using them.
6961 * This lets us mark the pageblocks back as
6962 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6963 * aligned range but not in the unaligned, original range are
6964 * put back to page allocator so that buddy can use them.
6967 ret
= start_isolate_page_range(pfn_max_align_down(start
),
6968 pfn_max_align_up(end
), migratetype
,
6974 * In case of -EBUSY, we'd like to know which page causes problem.
6975 * So, just fall through. We will check it in test_pages_isolated().
6977 ret
= __alloc_contig_migrate_range(&cc
, start
, end
);
6978 if (ret
&& ret
!= -EBUSY
)
6982 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6983 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6984 * more, all pages in [start, end) are free in page allocator.
6985 * What we are going to do is to allocate all pages from
6986 * [start, end) (that is remove them from page allocator).
6988 * The only problem is that pages at the beginning and at the
6989 * end of interesting range may be not aligned with pages that
6990 * page allocator holds, ie. they can be part of higher order
6991 * pages. Because of this, we reserve the bigger range and
6992 * once this is done free the pages we are not interested in.
6994 * We don't have to hold zone->lock here because the pages are
6995 * isolated thus they won't get removed from buddy.
6998 lru_add_drain_all();
6999 drain_all_pages(cc
.zone
);
7002 outer_start
= start
;
7003 while (!PageBuddy(pfn_to_page(outer_start
))) {
7004 if (++order
>= MAX_ORDER
) {
7005 outer_start
= start
;
7008 outer_start
&= ~0UL << order
;
7011 if (outer_start
!= start
) {
7012 order
= page_order(pfn_to_page(outer_start
));
7015 * outer_start page could be small order buddy page and
7016 * it doesn't include start page. Adjust outer_start
7017 * in this case to report failed page properly
7018 * on tracepoint in test_pages_isolated()
7020 if (outer_start
+ (1UL << order
) <= start
)
7021 outer_start
= start
;
7024 /* Make sure the range is really isolated. */
7025 if (test_pages_isolated(outer_start
, end
, false)) {
7026 pr_info("%s: [%lx, %lx) PFNs busy\n",
7027 __func__
, outer_start
, end
);
7032 /* Grab isolated pages from freelists. */
7033 outer_end
= isolate_freepages_range(&cc
, outer_start
, end
);
7039 /* Free head and tail (if any) */
7040 if (start
!= outer_start
)
7041 free_contig_range(outer_start
, start
- outer_start
);
7042 if (end
!= outer_end
)
7043 free_contig_range(end
, outer_end
- end
);
7046 undo_isolate_page_range(pfn_max_align_down(start
),
7047 pfn_max_align_up(end
), migratetype
);
7051 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
7053 unsigned int count
= 0;
7055 for (; nr_pages
--; pfn
++) {
7056 struct page
*page
= pfn_to_page(pfn
);
7058 count
+= page_count(page
) != 1;
7061 WARN(count
!= 0, "%d pages are still in use!\n", count
);
7065 #ifdef CONFIG_MEMORY_HOTPLUG
7067 * The zone indicated has a new number of managed_pages; batch sizes and percpu
7068 * page high values need to be recalulated.
7070 void __meminit
zone_pcp_update(struct zone
*zone
)
7073 mutex_lock(&pcp_batch_high_lock
);
7074 for_each_possible_cpu(cpu
)
7075 pageset_set_high_and_batch(zone
,
7076 per_cpu_ptr(zone
->pageset
, cpu
));
7077 mutex_unlock(&pcp_batch_high_lock
);
7081 void zone_pcp_reset(struct zone
*zone
)
7083 unsigned long flags
;
7085 struct per_cpu_pageset
*pset
;
7087 /* avoid races with drain_pages() */
7088 local_irq_save(flags
);
7089 if (zone
->pageset
!= &boot_pageset
) {
7090 for_each_online_cpu(cpu
) {
7091 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
7092 drain_zonestat(zone
, pset
);
7094 free_percpu(zone
->pageset
);
7095 zone
->pageset
= &boot_pageset
;
7097 local_irq_restore(flags
);
7100 #ifdef CONFIG_MEMORY_HOTREMOVE
7102 * All pages in the range must be isolated before calling this.
7105 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
7109 unsigned int order
, i
;
7111 unsigned long flags
;
7112 /* find the first valid pfn */
7113 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
7118 zone
= page_zone(pfn_to_page(pfn
));
7119 spin_lock_irqsave(&zone
->lock
, flags
);
7121 while (pfn
< end_pfn
) {
7122 if (!pfn_valid(pfn
)) {
7126 page
= pfn_to_page(pfn
);
7128 * The HWPoisoned page may be not in buddy system, and
7129 * page_count() is not 0.
7131 if (unlikely(!PageBuddy(page
) && PageHWPoison(page
))) {
7133 SetPageReserved(page
);
7137 BUG_ON(page_count(page
));
7138 BUG_ON(!PageBuddy(page
));
7139 order
= page_order(page
);
7140 #ifdef CONFIG_DEBUG_VM
7141 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
7142 pfn
, 1 << order
, end_pfn
);
7144 list_del(&page
->lru
);
7145 rmv_page_order(page
);
7146 zone
->free_area
[order
].nr_free
--;
7147 for (i
= 0; i
< (1 << order
); i
++)
7148 SetPageReserved((page
+i
));
7149 pfn
+= (1 << order
);
7151 spin_unlock_irqrestore(&zone
->lock
, flags
);
7155 #ifdef CONFIG_MEMORY_FAILURE
7156 bool is_free_buddy_page(struct page
*page
)
7158 struct zone
*zone
= page_zone(page
);
7159 unsigned long pfn
= page_to_pfn(page
);
7160 unsigned long flags
;
7163 spin_lock_irqsave(&zone
->lock
, flags
);
7164 for (order
= 0; order
< MAX_ORDER
; order
++) {
7165 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
7167 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
7170 spin_unlock_irqrestore(&zone
->lock
, flags
);
7172 return order
< MAX_ORDER
;