2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/jiffies.h>
23 #include <linux/bootmem.h>
24 #include <linux/memblock.h>
25 #include <linux/compiler.h>
26 #include <linux/kernel.h>
27 #include <linux/kmemcheck.h>
28 #include <linux/kasan.h>
29 #include <linux/module.h>
30 #include <linux/suspend.h>
31 #include <linux/pagevec.h>
32 #include <linux/blkdev.h>
33 #include <linux/slab.h>
34 #include <linux/ratelimit.h>
35 #include <linux/oom.h>
36 #include <linux/notifier.h>
37 #include <linux/topology.h>
38 #include <linux/sysctl.h>
39 #include <linux/cpu.h>
40 #include <linux/cpuset.h>
41 #include <linux/memory_hotplug.h>
42 #include <linux/nodemask.h>
43 #include <linux/vmalloc.h>
44 #include <linux/vmstat.h>
45 #include <linux/mempolicy.h>
46 #include <linux/memremap.h>
47 #include <linux/stop_machine.h>
48 #include <linux/sort.h>
49 #include <linux/pfn.h>
50 #include <linux/backing-dev.h>
51 #include <linux/fault-inject.h>
52 #include <linux/page-isolation.h>
53 #include <linux/page_ext.h>
54 #include <linux/debugobjects.h>
55 #include <linux/kmemleak.h>
56 #include <linux/compaction.h>
57 #include <trace/events/kmem.h>
58 #include <linux/prefetch.h>
59 #include <linux/mm_inline.h>
60 #include <linux/migrate.h>
61 #include <linux/page_ext.h>
62 #include <linux/hugetlb.h>
63 #include <linux/sched/rt.h>
64 #include <linux/page_owner.h>
65 #include <linux/kthread.h>
67 #include <asm/sections.h>
68 #include <asm/tlbflush.h>
69 #include <asm/div64.h>
72 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
73 static DEFINE_MUTEX(pcp_batch_high_lock
);
74 #define MIN_PERCPU_PAGELIST_FRACTION (8)
76 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
77 DEFINE_PER_CPU(int, numa_node
);
78 EXPORT_PER_CPU_SYMBOL(numa_node
);
81 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
83 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
84 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
85 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
86 * defined in <linux/topology.h>.
88 DEFINE_PER_CPU(int, _numa_mem_
); /* Kernel "local memory" node */
89 EXPORT_PER_CPU_SYMBOL(_numa_mem_
);
90 int _node_numa_mem_
[MAX_NUMNODES
];
94 * Array of node states.
96 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
97 [N_POSSIBLE
] = NODE_MASK_ALL
,
98 [N_ONLINE
] = { { [0] = 1UL } },
100 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
101 #ifdef CONFIG_HIGHMEM
102 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
104 #ifdef CONFIG_MOVABLE_NODE
105 [N_MEMORY
] = { { [0] = 1UL } },
107 [N_CPU
] = { { [0] = 1UL } },
110 EXPORT_SYMBOL(node_states
);
112 /* Protect totalram_pages and zone->managed_pages */
113 static DEFINE_SPINLOCK(managed_page_count_lock
);
115 unsigned long totalram_pages __read_mostly
;
116 unsigned long totalreserve_pages __read_mostly
;
117 unsigned long totalcma_pages __read_mostly
;
119 int percpu_pagelist_fraction
;
120 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
123 * A cached value of the page's pageblock's migratetype, used when the page is
124 * put on a pcplist. Used to avoid the pageblock migratetype lookup when
125 * freeing from pcplists in most cases, at the cost of possibly becoming stale.
126 * Also the migratetype set in the page does not necessarily match the pcplist
127 * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
128 * other index - this ensures that it will be put on the correct CMA freelist.
130 static inline int get_pcppage_migratetype(struct page
*page
)
135 static inline void set_pcppage_migratetype(struct page
*page
, int migratetype
)
137 page
->index
= migratetype
;
140 #ifdef CONFIG_PM_SLEEP
142 * The following functions are used by the suspend/hibernate code to temporarily
143 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
144 * while devices are suspended. To avoid races with the suspend/hibernate code,
145 * they should always be called with pm_mutex held (gfp_allowed_mask also should
146 * only be modified with pm_mutex held, unless the suspend/hibernate code is
147 * guaranteed not to run in parallel with that modification).
150 static gfp_t saved_gfp_mask
;
152 void pm_restore_gfp_mask(void)
154 WARN_ON(!mutex_is_locked(&pm_mutex
));
155 if (saved_gfp_mask
) {
156 gfp_allowed_mask
= saved_gfp_mask
;
161 void pm_restrict_gfp_mask(void)
163 WARN_ON(!mutex_is_locked(&pm_mutex
));
164 WARN_ON(saved_gfp_mask
);
165 saved_gfp_mask
= gfp_allowed_mask
;
166 gfp_allowed_mask
&= ~(__GFP_IO
| __GFP_FS
);
169 bool pm_suspended_storage(void)
171 if ((gfp_allowed_mask
& (__GFP_IO
| __GFP_FS
)) == (__GFP_IO
| __GFP_FS
))
175 #endif /* CONFIG_PM_SLEEP */
177 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
178 unsigned int pageblock_order __read_mostly
;
181 static void __free_pages_ok(struct page
*page
, unsigned int order
);
184 * results with 256, 32 in the lowmem_reserve sysctl:
185 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
186 * 1G machine -> (16M dma, 784M normal, 224M high)
187 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
188 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
189 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
191 * TBD: should special case ZONE_DMA32 machines here - in those we normally
192 * don't need any ZONE_NORMAL reservation
194 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
195 #ifdef CONFIG_ZONE_DMA
198 #ifdef CONFIG_ZONE_DMA32
201 #ifdef CONFIG_HIGHMEM
207 EXPORT_SYMBOL(totalram_pages
);
209 static char * const zone_names
[MAX_NR_ZONES
] = {
210 #ifdef CONFIG_ZONE_DMA
213 #ifdef CONFIG_ZONE_DMA32
217 #ifdef CONFIG_HIGHMEM
221 #ifdef CONFIG_ZONE_DEVICE
226 char * const migratetype_names
[MIGRATE_TYPES
] = {
234 #ifdef CONFIG_MEMORY_ISOLATION
239 compound_page_dtor
* const compound_page_dtors
[] = {
242 #ifdef CONFIG_HUGETLB_PAGE
245 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
250 int min_free_kbytes
= 1024;
251 int user_min_free_kbytes
= -1;
252 int watermark_scale_factor
= 10;
254 static unsigned long __meminitdata nr_kernel_pages
;
255 static unsigned long __meminitdata nr_all_pages
;
256 static unsigned long __meminitdata dma_reserve
;
258 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
259 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
260 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
261 static unsigned long __initdata required_kernelcore
;
262 static unsigned long __initdata required_movablecore
;
263 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
264 static bool mirrored_kernelcore
;
266 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
268 EXPORT_SYMBOL(movable_zone
);
269 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
272 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
273 int nr_online_nodes __read_mostly
= 1;
274 EXPORT_SYMBOL(nr_node_ids
);
275 EXPORT_SYMBOL(nr_online_nodes
);
278 int page_group_by_mobility_disabled __read_mostly
;
280 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
281 static inline void reset_deferred_meminit(pg_data_t
*pgdat
)
283 pgdat
->first_deferred_pfn
= ULONG_MAX
;
286 /* Returns true if the struct page for the pfn is uninitialised */
287 static inline bool __meminit
early_page_uninitialised(unsigned long pfn
)
289 if (pfn
>= NODE_DATA(early_pfn_to_nid(pfn
))->first_deferred_pfn
)
295 static inline bool early_page_nid_uninitialised(unsigned long pfn
, int nid
)
297 if (pfn
>= NODE_DATA(nid
)->first_deferred_pfn
)
304 * Returns false when the remaining initialisation should be deferred until
305 * later in the boot cycle when it can be parallelised.
307 static inline bool update_defer_init(pg_data_t
*pgdat
,
308 unsigned long pfn
, unsigned long zone_end
,
309 unsigned long *nr_initialised
)
311 /* Always populate low zones for address-contrained allocations */
312 if (zone_end
< pgdat_end_pfn(pgdat
))
315 /* Initialise at least 2G of the highest zone */
317 if (*nr_initialised
> (2UL << (30 - PAGE_SHIFT
)) &&
318 (pfn
& (PAGES_PER_SECTION
- 1)) == 0) {
319 pgdat
->first_deferred_pfn
= pfn
;
326 static inline void reset_deferred_meminit(pg_data_t
*pgdat
)
330 static inline bool early_page_uninitialised(unsigned long pfn
)
335 static inline bool early_page_nid_uninitialised(unsigned long pfn
, int nid
)
340 static inline bool update_defer_init(pg_data_t
*pgdat
,
341 unsigned long pfn
, unsigned long zone_end
,
342 unsigned long *nr_initialised
)
349 void set_pageblock_migratetype(struct page
*page
, int migratetype
)
351 if (unlikely(page_group_by_mobility_disabled
&&
352 migratetype
< MIGRATE_PCPTYPES
))
353 migratetype
= MIGRATE_UNMOVABLE
;
355 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
356 PB_migrate
, PB_migrate_end
);
359 #ifdef CONFIG_DEBUG_VM
360 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
364 unsigned long pfn
= page_to_pfn(page
);
365 unsigned long sp
, start_pfn
;
368 seq
= zone_span_seqbegin(zone
);
369 start_pfn
= zone
->zone_start_pfn
;
370 sp
= zone
->spanned_pages
;
371 if (!zone_spans_pfn(zone
, pfn
))
373 } while (zone_span_seqretry(zone
, seq
));
376 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
377 pfn
, zone_to_nid(zone
), zone
->name
,
378 start_pfn
, start_pfn
+ sp
);
383 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
385 if (!pfn_valid_within(page_to_pfn(page
)))
387 if (zone
!= page_zone(page
))
393 * Temporary debugging check for pages not lying within a given zone.
395 static int bad_range(struct zone
*zone
, struct page
*page
)
397 if (page_outside_zone_boundaries(zone
, page
))
399 if (!page_is_consistent(zone
, page
))
405 static inline int bad_range(struct zone
*zone
, struct page
*page
)
411 static void bad_page(struct page
*page
, const char *reason
,
412 unsigned long bad_flags
)
414 static unsigned long resume
;
415 static unsigned long nr_shown
;
416 static unsigned long nr_unshown
;
418 /* Don't complain about poisoned pages */
419 if (PageHWPoison(page
)) {
420 page_mapcount_reset(page
); /* remove PageBuddy */
425 * Allow a burst of 60 reports, then keep quiet for that minute;
426 * or allow a steady drip of one report per second.
428 if (nr_shown
== 60) {
429 if (time_before(jiffies
, resume
)) {
435 "BUG: Bad page state: %lu messages suppressed\n",
442 resume
= jiffies
+ 60 * HZ
;
444 pr_alert("BUG: Bad page state in process %s pfn:%05lx\n",
445 current
->comm
, page_to_pfn(page
));
446 __dump_page(page
, reason
);
447 bad_flags
&= page
->flags
;
449 pr_alert("bad because of flags: %#lx(%pGp)\n",
450 bad_flags
, &bad_flags
);
451 dump_page_owner(page
);
456 /* Leave bad fields for debug, except PageBuddy could make trouble */
457 page_mapcount_reset(page
); /* remove PageBuddy */
458 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
462 * Higher-order pages are called "compound pages". They are structured thusly:
464 * The first PAGE_SIZE page is called the "head page" and have PG_head set.
466 * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
467 * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
469 * The first tail page's ->compound_dtor holds the offset in array of compound
470 * page destructors. See compound_page_dtors.
472 * The first tail page's ->compound_order holds the order of allocation.
473 * This usage means that zero-order pages may not be compound.
476 void free_compound_page(struct page
*page
)
478 __free_pages_ok(page
, compound_order(page
));
481 void prep_compound_page(struct page
*page
, unsigned int order
)
484 int nr_pages
= 1 << order
;
486 set_compound_page_dtor(page
, COMPOUND_PAGE_DTOR
);
487 set_compound_order(page
, order
);
489 for (i
= 1; i
< nr_pages
; i
++) {
490 struct page
*p
= page
+ i
;
491 set_page_count(p
, 0);
492 p
->mapping
= TAIL_MAPPING
;
493 set_compound_head(p
, page
);
495 atomic_set(compound_mapcount_ptr(page
), -1);
498 #ifdef CONFIG_DEBUG_PAGEALLOC
499 unsigned int _debug_guardpage_minorder
;
500 bool _debug_pagealloc_enabled __read_mostly
501 = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT
);
502 EXPORT_SYMBOL(_debug_pagealloc_enabled
);
503 bool _debug_guardpage_enabled __read_mostly
;
505 static int __init
early_debug_pagealloc(char *buf
)
510 if (strcmp(buf
, "on") == 0)
511 _debug_pagealloc_enabled
= true;
513 if (strcmp(buf
, "off") == 0)
514 _debug_pagealloc_enabled
= false;
518 early_param("debug_pagealloc", early_debug_pagealloc
);
520 static bool need_debug_guardpage(void)
522 /* If we don't use debug_pagealloc, we don't need guard page */
523 if (!debug_pagealloc_enabled())
529 static void init_debug_guardpage(void)
531 if (!debug_pagealloc_enabled())
534 _debug_guardpage_enabled
= true;
537 struct page_ext_operations debug_guardpage_ops
= {
538 .need
= need_debug_guardpage
,
539 .init
= init_debug_guardpage
,
542 static int __init
debug_guardpage_minorder_setup(char *buf
)
546 if (kstrtoul(buf
, 10, &res
) < 0 || res
> MAX_ORDER
/ 2) {
547 pr_err("Bad debug_guardpage_minorder value\n");
550 _debug_guardpage_minorder
= res
;
551 pr_info("Setting debug_guardpage_minorder to %lu\n", res
);
554 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup
);
556 static inline void set_page_guard(struct zone
*zone
, struct page
*page
,
557 unsigned int order
, int migratetype
)
559 struct page_ext
*page_ext
;
561 if (!debug_guardpage_enabled())
564 page_ext
= lookup_page_ext(page
);
565 __set_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
567 INIT_LIST_HEAD(&page
->lru
);
568 set_page_private(page
, order
);
569 /* Guard pages are not available for any usage */
570 __mod_zone_freepage_state(zone
, -(1 << order
), migratetype
);
573 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
574 unsigned int order
, int migratetype
)
576 struct page_ext
*page_ext
;
578 if (!debug_guardpage_enabled())
581 page_ext
= lookup_page_ext(page
);
582 __clear_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
584 set_page_private(page
, 0);
585 if (!is_migrate_isolate(migratetype
))
586 __mod_zone_freepage_state(zone
, (1 << order
), migratetype
);
589 struct page_ext_operations debug_guardpage_ops
= { NULL
, };
590 static inline void set_page_guard(struct zone
*zone
, struct page
*page
,
591 unsigned int order
, int migratetype
) {}
592 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
593 unsigned int order
, int migratetype
) {}
596 static inline void set_page_order(struct page
*page
, unsigned int order
)
598 set_page_private(page
, order
);
599 __SetPageBuddy(page
);
602 static inline void rmv_page_order(struct page
*page
)
604 __ClearPageBuddy(page
);
605 set_page_private(page
, 0);
609 * This function checks whether a page is free && is the buddy
610 * we can do coalesce a page and its buddy if
611 * (a) the buddy is not in a hole &&
612 * (b) the buddy is in the buddy system &&
613 * (c) a page and its buddy have the same order &&
614 * (d) a page and its buddy are in the same zone.
616 * For recording whether a page is in the buddy system, we set ->_mapcount
617 * PAGE_BUDDY_MAPCOUNT_VALUE.
618 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
619 * serialized by zone->lock.
621 * For recording page's order, we use page_private(page).
623 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
626 if (!pfn_valid_within(page_to_pfn(buddy
)))
629 if (page_is_guard(buddy
) && page_order(buddy
) == order
) {
630 if (page_zone_id(page
) != page_zone_id(buddy
))
633 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
638 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
640 * zone check is done late to avoid uselessly
641 * calculating zone/node ids for pages that could
644 if (page_zone_id(page
) != page_zone_id(buddy
))
647 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
655 * Freeing function for a buddy system allocator.
657 * The concept of a buddy system is to maintain direct-mapped table
658 * (containing bit values) for memory blocks of various "orders".
659 * The bottom level table contains the map for the smallest allocatable
660 * units of memory (here, pages), and each level above it describes
661 * pairs of units from the levels below, hence, "buddies".
662 * At a high level, all that happens here is marking the table entry
663 * at the bottom level available, and propagating the changes upward
664 * as necessary, plus some accounting needed to play nicely with other
665 * parts of the VM system.
666 * At each level, we keep a list of pages, which are heads of continuous
667 * free pages of length of (1 << order) and marked with _mapcount
668 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
670 * So when we are allocating or freeing one, we can derive the state of the
671 * other. That is, if we allocate a small block, and both were
672 * free, the remainder of the region must be split into blocks.
673 * If a block is freed, and its buddy is also free, then this
674 * triggers coalescing into a block of larger size.
679 static inline void __free_one_page(struct page
*page
,
681 struct zone
*zone
, unsigned int order
,
684 unsigned long page_idx
;
685 unsigned long combined_idx
;
686 unsigned long uninitialized_var(buddy_idx
);
688 unsigned int max_order
= MAX_ORDER
;
690 VM_BUG_ON(!zone_is_initialized(zone
));
691 VM_BUG_ON_PAGE(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
, page
);
693 VM_BUG_ON(migratetype
== -1);
694 if (is_migrate_isolate(migratetype
)) {
696 * We restrict max order of merging to prevent merge
697 * between freepages on isolate pageblock and normal
698 * pageblock. Without this, pageblock isolation
699 * could cause incorrect freepage accounting.
701 max_order
= min_t(unsigned int, MAX_ORDER
, pageblock_order
+ 1);
703 __mod_zone_freepage_state(zone
, 1 << order
, migratetype
);
706 page_idx
= pfn
& ((1 << max_order
) - 1);
708 VM_BUG_ON_PAGE(page_idx
& ((1 << order
) - 1), page
);
709 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
711 while (order
< max_order
- 1) {
712 buddy_idx
= __find_buddy_index(page_idx
, order
);
713 buddy
= page
+ (buddy_idx
- page_idx
);
714 if (!page_is_buddy(page
, buddy
, order
))
717 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
718 * merge with it and move up one order.
720 if (page_is_guard(buddy
)) {
721 clear_page_guard(zone
, buddy
, order
, migratetype
);
723 list_del(&buddy
->lru
);
724 zone
->free_area
[order
].nr_free
--;
725 rmv_page_order(buddy
);
727 combined_idx
= buddy_idx
& page_idx
;
728 page
= page
+ (combined_idx
- page_idx
);
729 page_idx
= combined_idx
;
732 set_page_order(page
, order
);
735 * If this is not the largest possible page, check if the buddy
736 * of the next-highest order is free. If it is, it's possible
737 * that pages are being freed that will coalesce soon. In case,
738 * that is happening, add the free page to the tail of the list
739 * so it's less likely to be used soon and more likely to be merged
740 * as a higher order page
742 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
743 struct page
*higher_page
, *higher_buddy
;
744 combined_idx
= buddy_idx
& page_idx
;
745 higher_page
= page
+ (combined_idx
- page_idx
);
746 buddy_idx
= __find_buddy_index(combined_idx
, order
+ 1);
747 higher_buddy
= higher_page
+ (buddy_idx
- combined_idx
);
748 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
749 list_add_tail(&page
->lru
,
750 &zone
->free_area
[order
].free_list
[migratetype
]);
755 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
757 zone
->free_area
[order
].nr_free
++;
760 static inline int free_pages_check(struct page
*page
)
762 const char *bad_reason
= NULL
;
763 unsigned long bad_flags
= 0;
765 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
766 bad_reason
= "nonzero mapcount";
767 if (unlikely(page
->mapping
!= NULL
))
768 bad_reason
= "non-NULL mapping";
769 if (unlikely(page_ref_count(page
) != 0))
770 bad_reason
= "nonzero _count";
771 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
)) {
772 bad_reason
= "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
773 bad_flags
= PAGE_FLAGS_CHECK_AT_FREE
;
776 if (unlikely(page
->mem_cgroup
))
777 bad_reason
= "page still charged to cgroup";
779 if (unlikely(bad_reason
)) {
780 bad_page(page
, bad_reason
, bad_flags
);
783 page_cpupid_reset_last(page
);
784 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
785 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
790 * Frees a number of pages from the PCP lists
791 * Assumes all pages on list are in same zone, and of same order.
792 * count is the number of pages to free.
794 * If the zone was previously in an "all pages pinned" state then look to
795 * see if this freeing clears that state.
797 * And clear the zone's pages_scanned counter, to hold off the "all pages are
798 * pinned" detection logic.
800 static void free_pcppages_bulk(struct zone
*zone
, int count
,
801 struct per_cpu_pages
*pcp
)
806 unsigned long nr_scanned
;
808 spin_lock(&zone
->lock
);
809 nr_scanned
= zone_page_state(zone
, NR_PAGES_SCANNED
);
811 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, -nr_scanned
);
815 struct list_head
*list
;
818 * Remove pages from lists in a round-robin fashion. A
819 * batch_free count is maintained that is incremented when an
820 * empty list is encountered. This is so more pages are freed
821 * off fuller lists instead of spinning excessively around empty
826 if (++migratetype
== MIGRATE_PCPTYPES
)
828 list
= &pcp
->lists
[migratetype
];
829 } while (list_empty(list
));
831 /* This is the only non-empty list. Free them all. */
832 if (batch_free
== MIGRATE_PCPTYPES
)
833 batch_free
= to_free
;
836 int mt
; /* migratetype of the to-be-freed page */
838 page
= list_last_entry(list
, struct page
, lru
);
839 /* must delete as __free_one_page list manipulates */
840 list_del(&page
->lru
);
842 mt
= get_pcppage_migratetype(page
);
843 /* MIGRATE_ISOLATE page should not go to pcplists */
844 VM_BUG_ON_PAGE(is_migrate_isolate(mt
), page
);
845 /* Pageblock could have been isolated meanwhile */
846 if (unlikely(has_isolate_pageblock(zone
)))
847 mt
= get_pageblock_migratetype(page
);
849 __free_one_page(page
, page_to_pfn(page
), zone
, 0, mt
);
850 trace_mm_page_pcpu_drain(page
, 0, mt
);
851 } while (--to_free
&& --batch_free
&& !list_empty(list
));
853 spin_unlock(&zone
->lock
);
856 static void free_one_page(struct zone
*zone
,
857 struct page
*page
, unsigned long pfn
,
861 unsigned long nr_scanned
;
862 spin_lock(&zone
->lock
);
863 nr_scanned
= zone_page_state(zone
, NR_PAGES_SCANNED
);
865 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, -nr_scanned
);
867 if (unlikely(has_isolate_pageblock(zone
) ||
868 is_migrate_isolate(migratetype
))) {
869 migratetype
= get_pfnblock_migratetype(page
, pfn
);
871 __free_one_page(page
, pfn
, zone
, order
, migratetype
);
872 spin_unlock(&zone
->lock
);
875 static int free_tail_pages_check(struct page
*head_page
, struct page
*page
)
880 * We rely page->lru.next never has bit 0 set, unless the page
881 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
883 BUILD_BUG_ON((unsigned long)LIST_POISON1
& 1);
885 if (!IS_ENABLED(CONFIG_DEBUG_VM
)) {
889 switch (page
- head_page
) {
891 /* the first tail page: ->mapping is compound_mapcount() */
892 if (unlikely(compound_mapcount(page
))) {
893 bad_page(page
, "nonzero compound_mapcount", 0);
899 * the second tail page: ->mapping is
900 * page_deferred_list().next -- ignore value.
904 if (page
->mapping
!= TAIL_MAPPING
) {
905 bad_page(page
, "corrupted mapping in tail page", 0);
910 if (unlikely(!PageTail(page
))) {
911 bad_page(page
, "PageTail not set", 0);
914 if (unlikely(compound_head(page
) != head_page
)) {
915 bad_page(page
, "compound_head not consistent", 0);
920 page
->mapping
= NULL
;
921 clear_compound_head(page
);
925 static void __meminit
__init_single_page(struct page
*page
, unsigned long pfn
,
926 unsigned long zone
, int nid
)
928 set_page_links(page
, zone
, nid
, pfn
);
929 init_page_count(page
);
930 page_mapcount_reset(page
);
931 page_cpupid_reset_last(page
);
933 INIT_LIST_HEAD(&page
->lru
);
934 #ifdef WANT_PAGE_VIRTUAL
935 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
936 if (!is_highmem_idx(zone
))
937 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
941 static void __meminit
__init_single_pfn(unsigned long pfn
, unsigned long zone
,
944 return __init_single_page(pfn_to_page(pfn
), pfn
, zone
, nid
);
947 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
948 static void init_reserved_page(unsigned long pfn
)
953 if (!early_page_uninitialised(pfn
))
956 nid
= early_pfn_to_nid(pfn
);
957 pgdat
= NODE_DATA(nid
);
959 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
960 struct zone
*zone
= &pgdat
->node_zones
[zid
];
962 if (pfn
>= zone
->zone_start_pfn
&& pfn
< zone_end_pfn(zone
))
965 __init_single_pfn(pfn
, zid
, nid
);
968 static inline void init_reserved_page(unsigned long pfn
)
971 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
974 * Initialised pages do not have PageReserved set. This function is
975 * called for each range allocated by the bootmem allocator and
976 * marks the pages PageReserved. The remaining valid pages are later
977 * sent to the buddy page allocator.
979 void __meminit
reserve_bootmem_region(unsigned long start
, unsigned long end
)
981 unsigned long start_pfn
= PFN_DOWN(start
);
982 unsigned long end_pfn
= PFN_UP(end
);
984 for (; start_pfn
< end_pfn
; start_pfn
++) {
985 if (pfn_valid(start_pfn
)) {
986 struct page
*page
= pfn_to_page(start_pfn
);
988 init_reserved_page(start_pfn
);
990 /* Avoid false-positive PageTail() */
991 INIT_LIST_HEAD(&page
->lru
);
993 SetPageReserved(page
);
998 static bool free_pages_prepare(struct page
*page
, unsigned int order
)
1000 bool compound
= PageCompound(page
);
1003 VM_BUG_ON_PAGE(PageTail(page
), page
);
1004 VM_BUG_ON_PAGE(compound
&& compound_order(page
) != order
, page
);
1006 trace_mm_page_free(page
, order
);
1007 kmemcheck_free_shadow(page
, order
);
1008 kasan_free_pages(page
, order
);
1011 page
->mapping
= NULL
;
1012 bad
+= free_pages_check(page
);
1013 for (i
= 1; i
< (1 << order
); i
++) {
1015 bad
+= free_tail_pages_check(page
, page
+ i
);
1016 bad
+= free_pages_check(page
+ i
);
1021 reset_page_owner(page
, order
);
1023 if (!PageHighMem(page
)) {
1024 debug_check_no_locks_freed(page_address(page
),
1025 PAGE_SIZE
<< order
);
1026 debug_check_no_obj_freed(page_address(page
),
1027 PAGE_SIZE
<< order
);
1029 arch_free_page(page
, order
);
1030 kernel_poison_pages(page
, 1 << order
, 0);
1031 kernel_map_pages(page
, 1 << order
, 0);
1036 static void __free_pages_ok(struct page
*page
, unsigned int order
)
1038 unsigned long flags
;
1040 unsigned long pfn
= page_to_pfn(page
);
1042 if (!free_pages_prepare(page
, order
))
1045 migratetype
= get_pfnblock_migratetype(page
, pfn
);
1046 local_irq_save(flags
);
1047 __count_vm_events(PGFREE
, 1 << order
);
1048 free_one_page(page_zone(page
), page
, pfn
, order
, migratetype
);
1049 local_irq_restore(flags
);
1052 static void __init
__free_pages_boot_core(struct page
*page
,
1053 unsigned long pfn
, unsigned int order
)
1055 unsigned int nr_pages
= 1 << order
;
1056 struct page
*p
= page
;
1060 for (loop
= 0; loop
< (nr_pages
- 1); loop
++, p
++) {
1062 __ClearPageReserved(p
);
1063 set_page_count(p
, 0);
1065 __ClearPageReserved(p
);
1066 set_page_count(p
, 0);
1068 page_zone(page
)->managed_pages
+= nr_pages
;
1069 set_page_refcounted(page
);
1070 __free_pages(page
, order
);
1073 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1074 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1076 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata
;
1078 int __meminit
early_pfn_to_nid(unsigned long pfn
)
1080 static DEFINE_SPINLOCK(early_pfn_lock
);
1083 spin_lock(&early_pfn_lock
);
1084 nid
= __early_pfn_to_nid(pfn
, &early_pfnnid_cache
);
1087 spin_unlock(&early_pfn_lock
);
1093 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1094 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1095 struct mminit_pfnnid_cache
*state
)
1099 nid
= __early_pfn_to_nid(pfn
, state
);
1100 if (nid
>= 0 && nid
!= node
)
1105 /* Only safe to use early in boot when initialisation is single-threaded */
1106 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1108 return meminit_pfn_in_nid(pfn
, node
, &early_pfnnid_cache
);
1113 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1117 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1118 struct mminit_pfnnid_cache
*state
)
1125 void __init
__free_pages_bootmem(struct page
*page
, unsigned long pfn
,
1128 if (early_page_uninitialised(pfn
))
1130 return __free_pages_boot_core(page
, pfn
, order
);
1134 * Check that the whole (or subset of) a pageblock given by the interval of
1135 * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1136 * with the migration of free compaction scanner. The scanners then need to
1137 * use only pfn_valid_within() check for arches that allow holes within
1140 * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1142 * It's possible on some configurations to have a setup like node0 node1 node0
1143 * i.e. it's possible that all pages within a zones range of pages do not
1144 * belong to a single zone. We assume that a border between node0 and node1
1145 * can occur within a single pageblock, but not a node0 node1 node0
1146 * interleaving within a single pageblock. It is therefore sufficient to check
1147 * the first and last page of a pageblock and avoid checking each individual
1148 * page in a pageblock.
1150 struct page
*__pageblock_pfn_to_page(unsigned long start_pfn
,
1151 unsigned long end_pfn
, struct zone
*zone
)
1153 struct page
*start_page
;
1154 struct page
*end_page
;
1156 /* end_pfn is one past the range we are checking */
1159 if (!pfn_valid(start_pfn
) || !pfn_valid(end_pfn
))
1162 start_page
= pfn_to_page(start_pfn
);
1164 if (page_zone(start_page
) != zone
)
1167 end_page
= pfn_to_page(end_pfn
);
1169 /* This gives a shorter code than deriving page_zone(end_page) */
1170 if (page_zone_id(start_page
) != page_zone_id(end_page
))
1176 void set_zone_contiguous(struct zone
*zone
)
1178 unsigned long block_start_pfn
= zone
->zone_start_pfn
;
1179 unsigned long block_end_pfn
;
1181 block_end_pfn
= ALIGN(block_start_pfn
+ 1, pageblock_nr_pages
);
1182 for (; block_start_pfn
< zone_end_pfn(zone
);
1183 block_start_pfn
= block_end_pfn
,
1184 block_end_pfn
+= pageblock_nr_pages
) {
1186 block_end_pfn
= min(block_end_pfn
, zone_end_pfn(zone
));
1188 if (!__pageblock_pfn_to_page(block_start_pfn
,
1189 block_end_pfn
, zone
))
1193 /* We confirm that there is no hole */
1194 zone
->contiguous
= true;
1197 void clear_zone_contiguous(struct zone
*zone
)
1199 zone
->contiguous
= false;
1202 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1203 static void __init
deferred_free_range(struct page
*page
,
1204 unsigned long pfn
, int nr_pages
)
1211 /* Free a large naturally-aligned chunk if possible */
1212 if (nr_pages
== MAX_ORDER_NR_PAGES
&&
1213 (pfn
& (MAX_ORDER_NR_PAGES
-1)) == 0) {
1214 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
1215 __free_pages_boot_core(page
, pfn
, MAX_ORDER
-1);
1219 for (i
= 0; i
< nr_pages
; i
++, page
++, pfn
++)
1220 __free_pages_boot_core(page
, pfn
, 0);
1223 /* Completion tracking for deferred_init_memmap() threads */
1224 static atomic_t pgdat_init_n_undone __initdata
;
1225 static __initdata
DECLARE_COMPLETION(pgdat_init_all_done_comp
);
1227 static inline void __init
pgdat_init_report_one_done(void)
1229 if (atomic_dec_and_test(&pgdat_init_n_undone
))
1230 complete(&pgdat_init_all_done_comp
);
1233 /* Initialise remaining memory on a node */
1234 static int __init
deferred_init_memmap(void *data
)
1236 pg_data_t
*pgdat
= data
;
1237 int nid
= pgdat
->node_id
;
1238 struct mminit_pfnnid_cache nid_init_state
= { };
1239 unsigned long start
= jiffies
;
1240 unsigned long nr_pages
= 0;
1241 unsigned long walk_start
, walk_end
;
1244 unsigned long first_init_pfn
= pgdat
->first_deferred_pfn
;
1245 const struct cpumask
*cpumask
= cpumask_of_node(pgdat
->node_id
);
1247 if (first_init_pfn
== ULONG_MAX
) {
1248 pgdat_init_report_one_done();
1252 /* Bind memory initialisation thread to a local node if possible */
1253 if (!cpumask_empty(cpumask
))
1254 set_cpus_allowed_ptr(current
, cpumask
);
1256 /* Sanity check boundaries */
1257 BUG_ON(pgdat
->first_deferred_pfn
< pgdat
->node_start_pfn
);
1258 BUG_ON(pgdat
->first_deferred_pfn
> pgdat_end_pfn(pgdat
));
1259 pgdat
->first_deferred_pfn
= ULONG_MAX
;
1261 /* Only the highest zone is deferred so find it */
1262 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
1263 zone
= pgdat
->node_zones
+ zid
;
1264 if (first_init_pfn
< zone_end_pfn(zone
))
1268 for_each_mem_pfn_range(i
, nid
, &walk_start
, &walk_end
, NULL
) {
1269 unsigned long pfn
, end_pfn
;
1270 struct page
*page
= NULL
;
1271 struct page
*free_base_page
= NULL
;
1272 unsigned long free_base_pfn
= 0;
1275 end_pfn
= min(walk_end
, zone_end_pfn(zone
));
1276 pfn
= first_init_pfn
;
1277 if (pfn
< walk_start
)
1279 if (pfn
< zone
->zone_start_pfn
)
1280 pfn
= zone
->zone_start_pfn
;
1282 for (; pfn
< end_pfn
; pfn
++) {
1283 if (!pfn_valid_within(pfn
))
1287 * Ensure pfn_valid is checked every
1288 * MAX_ORDER_NR_PAGES for memory holes
1290 if ((pfn
& (MAX_ORDER_NR_PAGES
- 1)) == 0) {
1291 if (!pfn_valid(pfn
)) {
1297 if (!meminit_pfn_in_nid(pfn
, nid
, &nid_init_state
)) {
1302 /* Minimise pfn page lookups and scheduler checks */
1303 if (page
&& (pfn
& (MAX_ORDER_NR_PAGES
- 1)) != 0) {
1306 nr_pages
+= nr_to_free
;
1307 deferred_free_range(free_base_page
,
1308 free_base_pfn
, nr_to_free
);
1309 free_base_page
= NULL
;
1310 free_base_pfn
= nr_to_free
= 0;
1312 page
= pfn_to_page(pfn
);
1317 VM_BUG_ON(page_zone(page
) != zone
);
1321 __init_single_page(page
, pfn
, zid
, nid
);
1322 if (!free_base_page
) {
1323 free_base_page
= page
;
1324 free_base_pfn
= pfn
;
1329 /* Where possible, batch up pages for a single free */
1332 /* Free the current block of pages to allocator */
1333 nr_pages
+= nr_to_free
;
1334 deferred_free_range(free_base_page
, free_base_pfn
,
1336 free_base_page
= NULL
;
1337 free_base_pfn
= nr_to_free
= 0;
1340 first_init_pfn
= max(end_pfn
, first_init_pfn
);
1343 /* Sanity check that the next zone really is unpopulated */
1344 WARN_ON(++zid
< MAX_NR_ZONES
&& populated_zone(++zone
));
1346 pr_info("node %d initialised, %lu pages in %ums\n", nid
, nr_pages
,
1347 jiffies_to_msecs(jiffies
- start
));
1349 pgdat_init_report_one_done();
1352 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1354 void __init
page_alloc_init_late(void)
1358 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1361 /* There will be num_node_state(N_MEMORY) threads */
1362 atomic_set(&pgdat_init_n_undone
, num_node_state(N_MEMORY
));
1363 for_each_node_state(nid
, N_MEMORY
) {
1364 kthread_run(deferred_init_memmap
, NODE_DATA(nid
), "pgdatinit%d", nid
);
1367 /* Block until all are initialised */
1368 wait_for_completion(&pgdat_init_all_done_comp
);
1370 /* Reinit limits that are based on free pages after the kernel is up */
1371 files_maxfiles_init();
1374 for_each_populated_zone(zone
)
1375 set_zone_contiguous(zone
);
1379 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1380 void __init
init_cma_reserved_pageblock(struct page
*page
)
1382 unsigned i
= pageblock_nr_pages
;
1383 struct page
*p
= page
;
1386 __ClearPageReserved(p
);
1387 set_page_count(p
, 0);
1390 set_pageblock_migratetype(page
, MIGRATE_CMA
);
1392 if (pageblock_order
>= MAX_ORDER
) {
1393 i
= pageblock_nr_pages
;
1396 set_page_refcounted(p
);
1397 __free_pages(p
, MAX_ORDER
- 1);
1398 p
+= MAX_ORDER_NR_PAGES
;
1399 } while (i
-= MAX_ORDER_NR_PAGES
);
1401 set_page_refcounted(page
);
1402 __free_pages(page
, pageblock_order
);
1405 adjust_managed_page_count(page
, pageblock_nr_pages
);
1410 * The order of subdivision here is critical for the IO subsystem.
1411 * Please do not alter this order without good reasons and regression
1412 * testing. Specifically, as large blocks of memory are subdivided,
1413 * the order in which smaller blocks are delivered depends on the order
1414 * they're subdivided in this function. This is the primary factor
1415 * influencing the order in which pages are delivered to the IO
1416 * subsystem according to empirical testing, and this is also justified
1417 * by considering the behavior of a buddy system containing a single
1418 * large block of memory acted on by a series of small allocations.
1419 * This behavior is a critical factor in sglist merging's success.
1423 static inline void expand(struct zone
*zone
, struct page
*page
,
1424 int low
, int high
, struct free_area
*area
,
1427 unsigned long size
= 1 << high
;
1429 while (high
> low
) {
1433 VM_BUG_ON_PAGE(bad_range(zone
, &page
[size
]), &page
[size
]);
1435 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC
) &&
1436 debug_guardpage_enabled() &&
1437 high
< debug_guardpage_minorder()) {
1439 * Mark as guard pages (or page), that will allow to
1440 * merge back to allocator when buddy will be freed.
1441 * Corresponding page table entries will not be touched,
1442 * pages will stay not present in virtual address space
1444 set_page_guard(zone
, &page
[size
], high
, migratetype
);
1447 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
1449 set_page_order(&page
[size
], high
);
1454 * This page is about to be returned from the page allocator
1456 static inline int check_new_page(struct page
*page
)
1458 const char *bad_reason
= NULL
;
1459 unsigned long bad_flags
= 0;
1461 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
1462 bad_reason
= "nonzero mapcount";
1463 if (unlikely(page
->mapping
!= NULL
))
1464 bad_reason
= "non-NULL mapping";
1465 if (unlikely(page_ref_count(page
) != 0))
1466 bad_reason
= "nonzero _count";
1467 if (unlikely(page
->flags
& __PG_HWPOISON
)) {
1468 bad_reason
= "HWPoisoned (hardware-corrupted)";
1469 bad_flags
= __PG_HWPOISON
;
1471 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)) {
1472 bad_reason
= "PAGE_FLAGS_CHECK_AT_PREP flag set";
1473 bad_flags
= PAGE_FLAGS_CHECK_AT_PREP
;
1476 if (unlikely(page
->mem_cgroup
))
1477 bad_reason
= "page still charged to cgroup";
1479 if (unlikely(bad_reason
)) {
1480 bad_page(page
, bad_reason
, bad_flags
);
1486 static inline bool free_pages_prezeroed(bool poisoned
)
1488 return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO
) &&
1489 page_poisoning_enabled() && poisoned
;
1492 static int prep_new_page(struct page
*page
, unsigned int order
, gfp_t gfp_flags
,
1496 bool poisoned
= true;
1498 for (i
= 0; i
< (1 << order
); i
++) {
1499 struct page
*p
= page
+ i
;
1500 if (unlikely(check_new_page(p
)))
1503 poisoned
&= page_is_poisoned(p
);
1506 set_page_private(page
, 0);
1507 set_page_refcounted(page
);
1509 arch_alloc_page(page
, order
);
1510 kernel_map_pages(page
, 1 << order
, 1);
1511 kernel_poison_pages(page
, 1 << order
, 1);
1512 kasan_alloc_pages(page
, order
);
1514 if (!free_pages_prezeroed(poisoned
) && (gfp_flags
& __GFP_ZERO
))
1515 for (i
= 0; i
< (1 << order
); i
++)
1516 clear_highpage(page
+ i
);
1518 if (order
&& (gfp_flags
& __GFP_COMP
))
1519 prep_compound_page(page
, order
);
1521 set_page_owner(page
, order
, gfp_flags
);
1524 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1525 * allocate the page. The expectation is that the caller is taking
1526 * steps that will free more memory. The caller should avoid the page
1527 * being used for !PFMEMALLOC purposes.
1529 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
1530 set_page_pfmemalloc(page
);
1532 clear_page_pfmemalloc(page
);
1538 * Go through the free lists for the given migratetype and remove
1539 * the smallest available page from the freelists
1542 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
1545 unsigned int current_order
;
1546 struct free_area
*area
;
1549 /* Find a page of the appropriate size in the preferred list */
1550 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
1551 area
= &(zone
->free_area
[current_order
]);
1552 page
= list_first_entry_or_null(&area
->free_list
[migratetype
],
1556 list_del(&page
->lru
);
1557 rmv_page_order(page
);
1559 expand(zone
, page
, order
, current_order
, area
, migratetype
);
1560 set_pcppage_migratetype(page
, migratetype
);
1569 * This array describes the order lists are fallen back to when
1570 * the free lists for the desirable migrate type are depleted
1572 static int fallbacks
[MIGRATE_TYPES
][4] = {
1573 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1574 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1575 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_TYPES
},
1577 [MIGRATE_CMA
] = { MIGRATE_TYPES
}, /* Never used */
1579 #ifdef CONFIG_MEMORY_ISOLATION
1580 [MIGRATE_ISOLATE
] = { MIGRATE_TYPES
}, /* Never used */
1585 static struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1588 return __rmqueue_smallest(zone
, order
, MIGRATE_CMA
);
1591 static inline struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1592 unsigned int order
) { return NULL
; }
1596 * Move the free pages in a range to the free lists of the requested type.
1597 * Note that start_page and end_pages are not aligned on a pageblock
1598 * boundary. If alignment is required, use move_freepages_block()
1600 int move_freepages(struct zone
*zone
,
1601 struct page
*start_page
, struct page
*end_page
,
1606 int pages_moved
= 0;
1608 #ifndef CONFIG_HOLES_IN_ZONE
1610 * page_zone is not safe to call in this context when
1611 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1612 * anyway as we check zone boundaries in move_freepages_block().
1613 * Remove at a later date when no bug reports exist related to
1614 * grouping pages by mobility
1616 VM_BUG_ON(page_zone(start_page
) != page_zone(end_page
));
1619 for (page
= start_page
; page
<= end_page
;) {
1620 /* Make sure we are not inadvertently changing nodes */
1621 VM_BUG_ON_PAGE(page_to_nid(page
) != zone_to_nid(zone
), page
);
1623 if (!pfn_valid_within(page_to_pfn(page
))) {
1628 if (!PageBuddy(page
)) {
1633 order
= page_order(page
);
1634 list_move(&page
->lru
,
1635 &zone
->free_area
[order
].free_list
[migratetype
]);
1637 pages_moved
+= 1 << order
;
1643 int move_freepages_block(struct zone
*zone
, struct page
*page
,
1646 unsigned long start_pfn
, end_pfn
;
1647 struct page
*start_page
, *end_page
;
1649 start_pfn
= page_to_pfn(page
);
1650 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
1651 start_page
= pfn_to_page(start_pfn
);
1652 end_page
= start_page
+ pageblock_nr_pages
- 1;
1653 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
1655 /* Do not cross zone boundaries */
1656 if (!zone_spans_pfn(zone
, start_pfn
))
1658 if (!zone_spans_pfn(zone
, end_pfn
))
1661 return move_freepages(zone
, start_page
, end_page
, migratetype
);
1664 static void change_pageblock_range(struct page
*pageblock_page
,
1665 int start_order
, int migratetype
)
1667 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
1669 while (nr_pageblocks
--) {
1670 set_pageblock_migratetype(pageblock_page
, migratetype
);
1671 pageblock_page
+= pageblock_nr_pages
;
1676 * When we are falling back to another migratetype during allocation, try to
1677 * steal extra free pages from the same pageblocks to satisfy further
1678 * allocations, instead of polluting multiple pageblocks.
1680 * If we are stealing a relatively large buddy page, it is likely there will
1681 * be more free pages in the pageblock, so try to steal them all. For
1682 * reclaimable and unmovable allocations, we steal regardless of page size,
1683 * as fragmentation caused by those allocations polluting movable pageblocks
1684 * is worse than movable allocations stealing from unmovable and reclaimable
1687 static bool can_steal_fallback(unsigned int order
, int start_mt
)
1690 * Leaving this order check is intended, although there is
1691 * relaxed order check in next check. The reason is that
1692 * we can actually steal whole pageblock if this condition met,
1693 * but, below check doesn't guarantee it and that is just heuristic
1694 * so could be changed anytime.
1696 if (order
>= pageblock_order
)
1699 if (order
>= pageblock_order
/ 2 ||
1700 start_mt
== MIGRATE_RECLAIMABLE
||
1701 start_mt
== MIGRATE_UNMOVABLE
||
1702 page_group_by_mobility_disabled
)
1709 * This function implements actual steal behaviour. If order is large enough,
1710 * we can steal whole pageblock. If not, we first move freepages in this
1711 * pageblock and check whether half of pages are moved or not. If half of
1712 * pages are moved, we can change migratetype of pageblock and permanently
1713 * use it's pages as requested migratetype in the future.
1715 static void steal_suitable_fallback(struct zone
*zone
, struct page
*page
,
1718 unsigned int current_order
= page_order(page
);
1721 /* Take ownership for orders >= pageblock_order */
1722 if (current_order
>= pageblock_order
) {
1723 change_pageblock_range(page
, current_order
, start_type
);
1727 pages
= move_freepages_block(zone
, page
, start_type
);
1729 /* Claim the whole block if over half of it is free */
1730 if (pages
>= (1 << (pageblock_order
-1)) ||
1731 page_group_by_mobility_disabled
)
1732 set_pageblock_migratetype(page
, start_type
);
1736 * Check whether there is a suitable fallback freepage with requested order.
1737 * If only_stealable is true, this function returns fallback_mt only if
1738 * we can steal other freepages all together. This would help to reduce
1739 * fragmentation due to mixed migratetype pages in one pageblock.
1741 int find_suitable_fallback(struct free_area
*area
, unsigned int order
,
1742 int migratetype
, bool only_stealable
, bool *can_steal
)
1747 if (area
->nr_free
== 0)
1752 fallback_mt
= fallbacks
[migratetype
][i
];
1753 if (fallback_mt
== MIGRATE_TYPES
)
1756 if (list_empty(&area
->free_list
[fallback_mt
]))
1759 if (can_steal_fallback(order
, migratetype
))
1762 if (!only_stealable
)
1773 * Reserve a pageblock for exclusive use of high-order atomic allocations if
1774 * there are no empty page blocks that contain a page with a suitable order
1776 static void reserve_highatomic_pageblock(struct page
*page
, struct zone
*zone
,
1777 unsigned int alloc_order
)
1780 unsigned long max_managed
, flags
;
1783 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
1784 * Check is race-prone but harmless.
1786 max_managed
= (zone
->managed_pages
/ 100) + pageblock_nr_pages
;
1787 if (zone
->nr_reserved_highatomic
>= max_managed
)
1790 spin_lock_irqsave(&zone
->lock
, flags
);
1792 /* Recheck the nr_reserved_highatomic limit under the lock */
1793 if (zone
->nr_reserved_highatomic
>= max_managed
)
1797 mt
= get_pageblock_migratetype(page
);
1798 if (mt
!= MIGRATE_HIGHATOMIC
&&
1799 !is_migrate_isolate(mt
) && !is_migrate_cma(mt
)) {
1800 zone
->nr_reserved_highatomic
+= pageblock_nr_pages
;
1801 set_pageblock_migratetype(page
, MIGRATE_HIGHATOMIC
);
1802 move_freepages_block(zone
, page
, MIGRATE_HIGHATOMIC
);
1806 spin_unlock_irqrestore(&zone
->lock
, flags
);
1810 * Used when an allocation is about to fail under memory pressure. This
1811 * potentially hurts the reliability of high-order allocations when under
1812 * intense memory pressure but failed atomic allocations should be easier
1813 * to recover from than an OOM.
1815 static void unreserve_highatomic_pageblock(const struct alloc_context
*ac
)
1817 struct zonelist
*zonelist
= ac
->zonelist
;
1818 unsigned long flags
;
1824 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
, ac
->high_zoneidx
,
1826 /* Preserve at least one pageblock */
1827 if (zone
->nr_reserved_highatomic
<= pageblock_nr_pages
)
1830 spin_lock_irqsave(&zone
->lock
, flags
);
1831 for (order
= 0; order
< MAX_ORDER
; order
++) {
1832 struct free_area
*area
= &(zone
->free_area
[order
]);
1834 page
= list_first_entry_or_null(
1835 &area
->free_list
[MIGRATE_HIGHATOMIC
],
1841 * It should never happen but changes to locking could
1842 * inadvertently allow a per-cpu drain to add pages
1843 * to MIGRATE_HIGHATOMIC while unreserving so be safe
1844 * and watch for underflows.
1846 zone
->nr_reserved_highatomic
-= min(pageblock_nr_pages
,
1847 zone
->nr_reserved_highatomic
);
1850 * Convert to ac->migratetype and avoid the normal
1851 * pageblock stealing heuristics. Minimally, the caller
1852 * is doing the work and needs the pages. More
1853 * importantly, if the block was always converted to
1854 * MIGRATE_UNMOVABLE or another type then the number
1855 * of pageblocks that cannot be completely freed
1858 set_pageblock_migratetype(page
, ac
->migratetype
);
1859 move_freepages_block(zone
, page
, ac
->migratetype
);
1860 spin_unlock_irqrestore(&zone
->lock
, flags
);
1863 spin_unlock_irqrestore(&zone
->lock
, flags
);
1867 /* Remove an element from the buddy allocator from the fallback list */
1868 static inline struct page
*
1869 __rmqueue_fallback(struct zone
*zone
, unsigned int order
, int start_migratetype
)
1871 struct free_area
*area
;
1872 unsigned int current_order
;
1877 /* Find the largest possible block of pages in the other list */
1878 for (current_order
= MAX_ORDER
-1;
1879 current_order
>= order
&& current_order
<= MAX_ORDER
-1;
1881 area
= &(zone
->free_area
[current_order
]);
1882 fallback_mt
= find_suitable_fallback(area
, current_order
,
1883 start_migratetype
, false, &can_steal
);
1884 if (fallback_mt
== -1)
1887 page
= list_first_entry(&area
->free_list
[fallback_mt
],
1890 steal_suitable_fallback(zone
, page
, start_migratetype
);
1892 /* Remove the page from the freelists */
1894 list_del(&page
->lru
);
1895 rmv_page_order(page
);
1897 expand(zone
, page
, order
, current_order
, area
,
1900 * The pcppage_migratetype may differ from pageblock's
1901 * migratetype depending on the decisions in
1902 * find_suitable_fallback(). This is OK as long as it does not
1903 * differ for MIGRATE_CMA pageblocks. Those can be used as
1904 * fallback only via special __rmqueue_cma_fallback() function
1906 set_pcppage_migratetype(page
, start_migratetype
);
1908 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
1909 start_migratetype
, fallback_mt
);
1918 * Do the hard work of removing an element from the buddy allocator.
1919 * Call me with the zone->lock already held.
1921 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
1926 page
= __rmqueue_smallest(zone
, order
, migratetype
);
1927 if (unlikely(!page
)) {
1928 if (migratetype
== MIGRATE_MOVABLE
)
1929 page
= __rmqueue_cma_fallback(zone
, order
);
1932 page
= __rmqueue_fallback(zone
, order
, migratetype
);
1935 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
1940 * Obtain a specified number of elements from the buddy allocator, all under
1941 * a single hold of the lock, for efficiency. Add them to the supplied list.
1942 * Returns the number of new pages which were placed at *list.
1944 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
1945 unsigned long count
, struct list_head
*list
,
1946 int migratetype
, bool cold
)
1950 spin_lock(&zone
->lock
);
1951 for (i
= 0; i
< count
; ++i
) {
1952 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
1953 if (unlikely(page
== NULL
))
1957 * Split buddy pages returned by expand() are received here
1958 * in physical page order. The page is added to the callers and
1959 * list and the list head then moves forward. From the callers
1960 * perspective, the linked list is ordered by page number in
1961 * some conditions. This is useful for IO devices that can
1962 * merge IO requests if the physical pages are ordered
1966 list_add(&page
->lru
, list
);
1968 list_add_tail(&page
->lru
, list
);
1970 if (is_migrate_cma(get_pcppage_migratetype(page
)))
1971 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
,
1974 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
1975 spin_unlock(&zone
->lock
);
1981 * Called from the vmstat counter updater to drain pagesets of this
1982 * currently executing processor on remote nodes after they have
1985 * Note that this function must be called with the thread pinned to
1986 * a single processor.
1988 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
1990 unsigned long flags
;
1991 int to_drain
, batch
;
1993 local_irq_save(flags
);
1994 batch
= READ_ONCE(pcp
->batch
);
1995 to_drain
= min(pcp
->count
, batch
);
1997 free_pcppages_bulk(zone
, to_drain
, pcp
);
1998 pcp
->count
-= to_drain
;
2000 local_irq_restore(flags
);
2005 * Drain pcplists of the indicated processor and zone.
2007 * The processor must either be the current processor and the
2008 * thread pinned to the current processor or a processor that
2011 static void drain_pages_zone(unsigned int cpu
, struct zone
*zone
)
2013 unsigned long flags
;
2014 struct per_cpu_pageset
*pset
;
2015 struct per_cpu_pages
*pcp
;
2017 local_irq_save(flags
);
2018 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
2022 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
2025 local_irq_restore(flags
);
2029 * Drain pcplists of all zones on the indicated processor.
2031 * The processor must either be the current processor and the
2032 * thread pinned to the current processor or a processor that
2035 static void drain_pages(unsigned int cpu
)
2039 for_each_populated_zone(zone
) {
2040 drain_pages_zone(cpu
, zone
);
2045 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
2047 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2048 * the single zone's pages.
2050 void drain_local_pages(struct zone
*zone
)
2052 int cpu
= smp_processor_id();
2055 drain_pages_zone(cpu
, zone
);
2061 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2063 * When zone parameter is non-NULL, spill just the single zone's pages.
2065 * Note that this code is protected against sending an IPI to an offline
2066 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
2067 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
2068 * nothing keeps CPUs from showing up after we populated the cpumask and
2069 * before the call to on_each_cpu_mask().
2071 void drain_all_pages(struct zone
*zone
)
2076 * Allocate in the BSS so we wont require allocation in
2077 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2079 static cpumask_t cpus_with_pcps
;
2082 * We don't care about racing with CPU hotplug event
2083 * as offline notification will cause the notified
2084 * cpu to drain that CPU pcps and on_each_cpu_mask
2085 * disables preemption as part of its processing
2087 for_each_online_cpu(cpu
) {
2088 struct per_cpu_pageset
*pcp
;
2090 bool has_pcps
= false;
2093 pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
2097 for_each_populated_zone(z
) {
2098 pcp
= per_cpu_ptr(z
->pageset
, cpu
);
2099 if (pcp
->pcp
.count
) {
2107 cpumask_set_cpu(cpu
, &cpus_with_pcps
);
2109 cpumask_clear_cpu(cpu
, &cpus_with_pcps
);
2111 on_each_cpu_mask(&cpus_with_pcps
, (smp_call_func_t
) drain_local_pages
,
2115 #ifdef CONFIG_HIBERNATION
2117 void mark_free_pages(struct zone
*zone
)
2119 unsigned long pfn
, max_zone_pfn
;
2120 unsigned long flags
;
2121 unsigned int order
, t
;
2124 if (zone_is_empty(zone
))
2127 spin_lock_irqsave(&zone
->lock
, flags
);
2129 max_zone_pfn
= zone_end_pfn(zone
);
2130 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
2131 if (pfn_valid(pfn
)) {
2132 page
= pfn_to_page(pfn
);
2133 if (!swsusp_page_is_forbidden(page
))
2134 swsusp_unset_page_free(page
);
2137 for_each_migratetype_order(order
, t
) {
2138 list_for_each_entry(page
,
2139 &zone
->free_area
[order
].free_list
[t
], lru
) {
2142 pfn
= page_to_pfn(page
);
2143 for (i
= 0; i
< (1UL << order
); i
++)
2144 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
2147 spin_unlock_irqrestore(&zone
->lock
, flags
);
2149 #endif /* CONFIG_PM */
2152 * Free a 0-order page
2153 * cold == true ? free a cold page : free a hot page
2155 void free_hot_cold_page(struct page
*page
, bool cold
)
2157 struct zone
*zone
= page_zone(page
);
2158 struct per_cpu_pages
*pcp
;
2159 unsigned long flags
;
2160 unsigned long pfn
= page_to_pfn(page
);
2163 if (!free_pages_prepare(page
, 0))
2166 migratetype
= get_pfnblock_migratetype(page
, pfn
);
2167 set_pcppage_migratetype(page
, migratetype
);
2168 local_irq_save(flags
);
2169 __count_vm_event(PGFREE
);
2172 * We only track unmovable, reclaimable and movable on pcp lists.
2173 * Free ISOLATE pages back to the allocator because they are being
2174 * offlined but treat RESERVE as movable pages so we can get those
2175 * areas back if necessary. Otherwise, we may have to free
2176 * excessively into the page allocator
2178 if (migratetype
>= MIGRATE_PCPTYPES
) {
2179 if (unlikely(is_migrate_isolate(migratetype
))) {
2180 free_one_page(zone
, page
, pfn
, 0, migratetype
);
2183 migratetype
= MIGRATE_MOVABLE
;
2186 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2188 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
2190 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
2192 if (pcp
->count
>= pcp
->high
) {
2193 unsigned long batch
= READ_ONCE(pcp
->batch
);
2194 free_pcppages_bulk(zone
, batch
, pcp
);
2195 pcp
->count
-= batch
;
2199 local_irq_restore(flags
);
2203 * Free a list of 0-order pages
2205 void free_hot_cold_page_list(struct list_head
*list
, bool cold
)
2207 struct page
*page
, *next
;
2209 list_for_each_entry_safe(page
, next
, list
, lru
) {
2210 trace_mm_page_free_batched(page
, cold
);
2211 free_hot_cold_page(page
, cold
);
2216 * split_page takes a non-compound higher-order page, and splits it into
2217 * n (1<<order) sub-pages: page[0..n]
2218 * Each sub-page must be freed individually.
2220 * Note: this is probably too low level an operation for use in drivers.
2221 * Please consult with lkml before using this in your driver.
2223 void split_page(struct page
*page
, unsigned int order
)
2228 VM_BUG_ON_PAGE(PageCompound(page
), page
);
2229 VM_BUG_ON_PAGE(!page_count(page
), page
);
2231 #ifdef CONFIG_KMEMCHECK
2233 * Split shadow pages too, because free(page[0]) would
2234 * otherwise free the whole shadow.
2236 if (kmemcheck_page_is_tracked(page
))
2237 split_page(virt_to_page(page
[0].shadow
), order
);
2240 gfp_mask
= get_page_owner_gfp(page
);
2241 set_page_owner(page
, 0, gfp_mask
);
2242 for (i
= 1; i
< (1 << order
); i
++) {
2243 set_page_refcounted(page
+ i
);
2244 set_page_owner(page
+ i
, 0, gfp_mask
);
2247 EXPORT_SYMBOL_GPL(split_page
);
2249 int __isolate_free_page(struct page
*page
, unsigned int order
)
2251 unsigned long watermark
;
2255 BUG_ON(!PageBuddy(page
));
2257 zone
= page_zone(page
);
2258 mt
= get_pageblock_migratetype(page
);
2260 if (!is_migrate_isolate(mt
)) {
2261 /* Obey watermarks as if the page was being allocated */
2262 watermark
= low_wmark_pages(zone
) + (1 << order
);
2263 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
2266 __mod_zone_freepage_state(zone
, -(1UL << order
), mt
);
2269 /* Remove page from free list */
2270 list_del(&page
->lru
);
2271 zone
->free_area
[order
].nr_free
--;
2272 rmv_page_order(page
);
2274 set_page_owner(page
, order
, __GFP_MOVABLE
);
2276 /* Set the pageblock if the isolated page is at least a pageblock */
2277 if (order
>= pageblock_order
- 1) {
2278 struct page
*endpage
= page
+ (1 << order
) - 1;
2279 for (; page
< endpage
; page
+= pageblock_nr_pages
) {
2280 int mt
= get_pageblock_migratetype(page
);
2281 if (!is_migrate_isolate(mt
) && !is_migrate_cma(mt
))
2282 set_pageblock_migratetype(page
,
2288 return 1UL << order
;
2292 * Similar to split_page except the page is already free. As this is only
2293 * being used for migration, the migratetype of the block also changes.
2294 * As this is called with interrupts disabled, the caller is responsible
2295 * for calling arch_alloc_page() and kernel_map_page() after interrupts
2298 * Note: this is probably too low level an operation for use in drivers.
2299 * Please consult with lkml before using this in your driver.
2301 int split_free_page(struct page
*page
)
2306 order
= page_order(page
);
2308 nr_pages
= __isolate_free_page(page
, order
);
2312 /* Split into individual pages */
2313 set_page_refcounted(page
);
2314 split_page(page
, order
);
2319 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2322 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
2323 struct zone
*zone
, unsigned int order
,
2324 gfp_t gfp_flags
, int alloc_flags
, int migratetype
)
2326 unsigned long flags
;
2328 bool cold
= ((gfp_flags
& __GFP_COLD
) != 0);
2330 if (likely(order
== 0)) {
2331 struct per_cpu_pages
*pcp
;
2332 struct list_head
*list
;
2334 local_irq_save(flags
);
2335 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2336 list
= &pcp
->lists
[migratetype
];
2337 if (list_empty(list
)) {
2338 pcp
->count
+= rmqueue_bulk(zone
, 0,
2341 if (unlikely(list_empty(list
)))
2346 page
= list_last_entry(list
, struct page
, lru
);
2348 page
= list_first_entry(list
, struct page
, lru
);
2350 list_del(&page
->lru
);
2354 * We most definitely don't want callers attempting to
2355 * allocate greater than order-1 page units with __GFP_NOFAIL.
2357 WARN_ON_ONCE((gfp_flags
& __GFP_NOFAIL
) && (order
> 1));
2358 spin_lock_irqsave(&zone
->lock
, flags
);
2361 if (alloc_flags
& ALLOC_HARDER
) {
2362 page
= __rmqueue_smallest(zone
, order
, MIGRATE_HIGHATOMIC
);
2364 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
2367 page
= __rmqueue(zone
, order
, migratetype
);
2368 spin_unlock(&zone
->lock
);
2371 __mod_zone_freepage_state(zone
, -(1 << order
),
2372 get_pcppage_migratetype(page
));
2375 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
, -(1 << order
));
2376 if (atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]) <= 0 &&
2377 !test_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
))
2378 set_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
);
2380 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
2381 zone_statistics(preferred_zone
, zone
, gfp_flags
);
2382 local_irq_restore(flags
);
2384 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
2388 local_irq_restore(flags
);
2392 #ifdef CONFIG_FAIL_PAGE_ALLOC
2395 struct fault_attr attr
;
2397 bool ignore_gfp_highmem
;
2398 bool ignore_gfp_reclaim
;
2400 } fail_page_alloc
= {
2401 .attr
= FAULT_ATTR_INITIALIZER
,
2402 .ignore_gfp_reclaim
= true,
2403 .ignore_gfp_highmem
= true,
2407 static int __init
setup_fail_page_alloc(char *str
)
2409 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
2411 __setup("fail_page_alloc=", setup_fail_page_alloc
);
2413 static bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2415 if (order
< fail_page_alloc
.min_order
)
2417 if (gfp_mask
& __GFP_NOFAIL
)
2419 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
2421 if (fail_page_alloc
.ignore_gfp_reclaim
&&
2422 (gfp_mask
& __GFP_DIRECT_RECLAIM
))
2425 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
2428 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2430 static int __init
fail_page_alloc_debugfs(void)
2432 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
2435 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
2436 &fail_page_alloc
.attr
);
2438 return PTR_ERR(dir
);
2440 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
2441 &fail_page_alloc
.ignore_gfp_reclaim
))
2443 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
2444 &fail_page_alloc
.ignore_gfp_highmem
))
2446 if (!debugfs_create_u32("min-order", mode
, dir
,
2447 &fail_page_alloc
.min_order
))
2452 debugfs_remove_recursive(dir
);
2457 late_initcall(fail_page_alloc_debugfs
);
2459 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2461 #else /* CONFIG_FAIL_PAGE_ALLOC */
2463 static inline bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2468 #endif /* CONFIG_FAIL_PAGE_ALLOC */
2471 * Return true if free base pages are above 'mark'. For high-order checks it
2472 * will return true of the order-0 watermark is reached and there is at least
2473 * one free page of a suitable size. Checking now avoids taking the zone lock
2474 * to check in the allocation paths if no pages are free.
2476 static bool __zone_watermark_ok(struct zone
*z
, unsigned int order
,
2477 unsigned long mark
, int classzone_idx
, int alloc_flags
,
2482 const int alloc_harder
= (alloc_flags
& ALLOC_HARDER
);
2484 /* free_pages may go negative - that's OK */
2485 free_pages
-= (1 << order
) - 1;
2487 if (alloc_flags
& ALLOC_HIGH
)
2491 * If the caller does not have rights to ALLOC_HARDER then subtract
2492 * the high-atomic reserves. This will over-estimate the size of the
2493 * atomic reserve but it avoids a search.
2495 if (likely(!alloc_harder
))
2496 free_pages
-= z
->nr_reserved_highatomic
;
2501 /* If allocation can't use CMA areas don't use free CMA pages */
2502 if (!(alloc_flags
& ALLOC_CMA
))
2503 free_pages
-= zone_page_state(z
, NR_FREE_CMA_PAGES
);
2507 * Check watermarks for an order-0 allocation request. If these
2508 * are not met, then a high-order request also cannot go ahead
2509 * even if a suitable page happened to be free.
2511 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
2514 /* If this is an order-0 request then the watermark is fine */
2518 /* For a high-order request, check at least one suitable page is free */
2519 for (o
= order
; o
< MAX_ORDER
; o
++) {
2520 struct free_area
*area
= &z
->free_area
[o
];
2529 for (mt
= 0; mt
< MIGRATE_PCPTYPES
; mt
++) {
2530 if (!list_empty(&area
->free_list
[mt
]))
2535 if ((alloc_flags
& ALLOC_CMA
) &&
2536 !list_empty(&area
->free_list
[MIGRATE_CMA
])) {
2544 bool zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
2545 int classzone_idx
, int alloc_flags
)
2547 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
2548 zone_page_state(z
, NR_FREE_PAGES
));
2551 bool zone_watermark_ok_safe(struct zone
*z
, unsigned int order
,
2552 unsigned long mark
, int classzone_idx
)
2554 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
2556 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
2557 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
2559 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, 0,
2564 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
2566 return local_zone
->node
== zone
->node
;
2569 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2571 return node_distance(zone_to_nid(local_zone
), zone_to_nid(zone
)) <
2574 #else /* CONFIG_NUMA */
2575 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
2580 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2584 #endif /* CONFIG_NUMA */
2586 static void reset_alloc_batches(struct zone
*preferred_zone
)
2588 struct zone
*zone
= preferred_zone
->zone_pgdat
->node_zones
;
2591 mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
2592 high_wmark_pages(zone
) - low_wmark_pages(zone
) -
2593 atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]));
2594 clear_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
);
2595 } while (zone
++ != preferred_zone
);
2599 * get_page_from_freelist goes through the zonelist trying to allocate
2602 static struct page
*
2603 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
, int alloc_flags
,
2604 const struct alloc_context
*ac
)
2606 struct zonelist
*zonelist
= ac
->zonelist
;
2608 struct page
*page
= NULL
;
2610 int nr_fair_skipped
= 0;
2611 bool zonelist_rescan
;
2614 zonelist_rescan
= false;
2617 * Scan zonelist, looking for a zone with enough free.
2618 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2620 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
, ac
->high_zoneidx
,
2624 if (cpusets_enabled() &&
2625 (alloc_flags
& ALLOC_CPUSET
) &&
2626 !cpuset_zone_allowed(zone
, gfp_mask
))
2629 * Distribute pages in proportion to the individual
2630 * zone size to ensure fair page aging. The zone a
2631 * page was allocated in should have no effect on the
2632 * time the page has in memory before being reclaimed.
2634 if (alloc_flags
& ALLOC_FAIR
) {
2635 if (!zone_local(ac
->preferred_zone
, zone
))
2637 if (test_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
)) {
2643 * When allocating a page cache page for writing, we
2644 * want to get it from a zone that is within its dirty
2645 * limit, such that no single zone holds more than its
2646 * proportional share of globally allowed dirty pages.
2647 * The dirty limits take into account the zone's
2648 * lowmem reserves and high watermark so that kswapd
2649 * should be able to balance it without having to
2650 * write pages from its LRU list.
2652 * This may look like it could increase pressure on
2653 * lower zones by failing allocations in higher zones
2654 * before they are full. But the pages that do spill
2655 * over are limited as the lower zones are protected
2656 * by this very same mechanism. It should not become
2657 * a practical burden to them.
2659 * XXX: For now, allow allocations to potentially
2660 * exceed the per-zone dirty limit in the slowpath
2661 * (spread_dirty_pages unset) before going into reclaim,
2662 * which is important when on a NUMA setup the allowed
2663 * zones are together not big enough to reach the
2664 * global limit. The proper fix for these situations
2665 * will require awareness of zones in the
2666 * dirty-throttling and the flusher threads.
2668 if (ac
->spread_dirty_pages
&& !zone_dirty_ok(zone
))
2671 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
2672 if (!zone_watermark_ok(zone
, order
, mark
,
2673 ac
->classzone_idx
, alloc_flags
)) {
2676 /* Checked here to keep the fast path fast */
2677 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
2678 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
2681 if (zone_reclaim_mode
== 0 ||
2682 !zone_allows_reclaim(ac
->preferred_zone
, zone
))
2685 ret
= zone_reclaim(zone
, gfp_mask
, order
);
2687 case ZONE_RECLAIM_NOSCAN
:
2690 case ZONE_RECLAIM_FULL
:
2691 /* scanned but unreclaimable */
2694 /* did we reclaim enough */
2695 if (zone_watermark_ok(zone
, order
, mark
,
2696 ac
->classzone_idx
, alloc_flags
))
2704 page
= buffered_rmqueue(ac
->preferred_zone
, zone
, order
,
2705 gfp_mask
, alloc_flags
, ac
->migratetype
);
2707 if (prep_new_page(page
, order
, gfp_mask
, alloc_flags
))
2711 * If this is a high-order atomic allocation then check
2712 * if the pageblock should be reserved for the future
2714 if (unlikely(order
&& (alloc_flags
& ALLOC_HARDER
)))
2715 reserve_highatomic_pageblock(page
, zone
, order
);
2722 * The first pass makes sure allocations are spread fairly within the
2723 * local node. However, the local node might have free pages left
2724 * after the fairness batches are exhausted, and remote zones haven't
2725 * even been considered yet. Try once more without fairness, and
2726 * include remote zones now, before entering the slowpath and waking
2727 * kswapd: prefer spilling to a remote zone over swapping locally.
2729 if (alloc_flags
& ALLOC_FAIR
) {
2730 alloc_flags
&= ~ALLOC_FAIR
;
2731 if (nr_fair_skipped
) {
2732 zonelist_rescan
= true;
2733 reset_alloc_batches(ac
->preferred_zone
);
2735 if (nr_online_nodes
> 1)
2736 zonelist_rescan
= true;
2739 if (zonelist_rescan
)
2746 * Large machines with many possible nodes should not always dump per-node
2747 * meminfo in irq context.
2749 static inline bool should_suppress_show_mem(void)
2754 ret
= in_interrupt();
2759 static DEFINE_RATELIMIT_STATE(nopage_rs
,
2760 DEFAULT_RATELIMIT_INTERVAL
,
2761 DEFAULT_RATELIMIT_BURST
);
2763 void warn_alloc_failed(gfp_t gfp_mask
, unsigned int order
, const char *fmt
, ...)
2765 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
2767 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
) ||
2768 debug_guardpage_minorder() > 0)
2772 * This documents exceptions given to allocations in certain
2773 * contexts that are allowed to allocate outside current's set
2776 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2777 if (test_thread_flag(TIF_MEMDIE
) ||
2778 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
2779 filter
&= ~SHOW_MEM_FILTER_NODES
;
2780 if (in_interrupt() || !(gfp_mask
& __GFP_DIRECT_RECLAIM
))
2781 filter
&= ~SHOW_MEM_FILTER_NODES
;
2784 struct va_format vaf
;
2787 va_start(args
, fmt
);
2792 pr_warn("%pV", &vaf
);
2797 pr_warn("%s: page allocation failure: order:%u, mode:%#x(%pGg)\n",
2798 current
->comm
, order
, gfp_mask
, &gfp_mask
);
2800 if (!should_suppress_show_mem())
2804 static inline struct page
*
2805 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
2806 const struct alloc_context
*ac
, unsigned long *did_some_progress
)
2808 struct oom_control oc
= {
2809 .zonelist
= ac
->zonelist
,
2810 .nodemask
= ac
->nodemask
,
2811 .gfp_mask
= gfp_mask
,
2816 *did_some_progress
= 0;
2819 * Acquire the oom lock. If that fails, somebody else is
2820 * making progress for us.
2822 if (!mutex_trylock(&oom_lock
)) {
2823 *did_some_progress
= 1;
2824 schedule_timeout_uninterruptible(1);
2829 * Go through the zonelist yet one more time, keep very high watermark
2830 * here, this is only to catch a parallel oom killing, we must fail if
2831 * we're still under heavy pressure.
2833 page
= get_page_from_freelist(gfp_mask
| __GFP_HARDWALL
, order
,
2834 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
, ac
);
2838 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2839 /* Coredumps can quickly deplete all memory reserves */
2840 if (current
->flags
& PF_DUMPCORE
)
2842 /* The OOM killer will not help higher order allocs */
2843 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2845 /* The OOM killer does not needlessly kill tasks for lowmem */
2846 if (ac
->high_zoneidx
< ZONE_NORMAL
)
2848 /* The OOM killer does not compensate for IO-less reclaim */
2849 if (!(gfp_mask
& __GFP_FS
)) {
2851 * XXX: Page reclaim didn't yield anything,
2852 * and the OOM killer can't be invoked, but
2853 * keep looping as per tradition.
2855 *did_some_progress
= 1;
2858 if (pm_suspended_storage())
2860 /* The OOM killer may not free memory on a specific node */
2861 if (gfp_mask
& __GFP_THISNODE
)
2864 /* Exhausted what can be done so it's blamo time */
2865 if (out_of_memory(&oc
) || WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
)) {
2866 *did_some_progress
= 1;
2868 if (gfp_mask
& __GFP_NOFAIL
) {
2869 page
= get_page_from_freelist(gfp_mask
, order
,
2870 ALLOC_NO_WATERMARKS
|ALLOC_CPUSET
, ac
);
2872 * fallback to ignore cpuset restriction if our nodes
2876 page
= get_page_from_freelist(gfp_mask
, order
,
2877 ALLOC_NO_WATERMARKS
, ac
);
2881 mutex_unlock(&oom_lock
);
2885 #ifdef CONFIG_COMPACTION
2886 /* Try memory compaction for high-order allocations before reclaim */
2887 static struct page
*
2888 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2889 int alloc_flags
, const struct alloc_context
*ac
,
2890 enum migrate_mode mode
, int *contended_compaction
,
2891 bool *deferred_compaction
)
2893 unsigned long compact_result
;
2899 current
->flags
|= PF_MEMALLOC
;
2900 compact_result
= try_to_compact_pages(gfp_mask
, order
, alloc_flags
, ac
,
2901 mode
, contended_compaction
);
2902 current
->flags
&= ~PF_MEMALLOC
;
2904 switch (compact_result
) {
2905 case COMPACT_DEFERRED
:
2906 *deferred_compaction
= true;
2908 case COMPACT_SKIPPED
:
2915 * At least in one zone compaction wasn't deferred or skipped, so let's
2916 * count a compaction stall
2918 count_vm_event(COMPACTSTALL
);
2920 page
= get_page_from_freelist(gfp_mask
, order
,
2921 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
2924 struct zone
*zone
= page_zone(page
);
2926 zone
->compact_blockskip_flush
= false;
2927 compaction_defer_reset(zone
, order
, true);
2928 count_vm_event(COMPACTSUCCESS
);
2933 * It's bad if compaction run occurs and fails. The most likely reason
2934 * is that pages exist, but not enough to satisfy watermarks.
2936 count_vm_event(COMPACTFAIL
);
2943 static inline struct page
*
2944 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2945 int alloc_flags
, const struct alloc_context
*ac
,
2946 enum migrate_mode mode
, int *contended_compaction
,
2947 bool *deferred_compaction
)
2951 #endif /* CONFIG_COMPACTION */
2953 /* Perform direct synchronous page reclaim */
2955 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
,
2956 const struct alloc_context
*ac
)
2958 struct reclaim_state reclaim_state
;
2963 /* We now go into synchronous reclaim */
2964 cpuset_memory_pressure_bump();
2965 current
->flags
|= PF_MEMALLOC
;
2966 lockdep_set_current_reclaim_state(gfp_mask
);
2967 reclaim_state
.reclaimed_slab
= 0;
2968 current
->reclaim_state
= &reclaim_state
;
2970 progress
= try_to_free_pages(ac
->zonelist
, order
, gfp_mask
,
2973 current
->reclaim_state
= NULL
;
2974 lockdep_clear_current_reclaim_state();
2975 current
->flags
&= ~PF_MEMALLOC
;
2982 /* The really slow allocator path where we enter direct reclaim */
2983 static inline struct page
*
2984 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
2985 int alloc_flags
, const struct alloc_context
*ac
,
2986 unsigned long *did_some_progress
)
2988 struct page
*page
= NULL
;
2989 bool drained
= false;
2991 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, ac
);
2992 if (unlikely(!(*did_some_progress
)))
2996 page
= get_page_from_freelist(gfp_mask
, order
,
2997 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
3000 * If an allocation failed after direct reclaim, it could be because
3001 * pages are pinned on the per-cpu lists or in high alloc reserves.
3002 * Shrink them them and try again
3004 if (!page
&& !drained
) {
3005 unreserve_highatomic_pageblock(ac
);
3006 drain_all_pages(NULL
);
3014 static void wake_all_kswapds(unsigned int order
, const struct alloc_context
*ac
)
3019 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
,
3020 ac
->high_zoneidx
, ac
->nodemask
)
3021 wakeup_kswapd(zone
, order
, zone_idx(ac
->preferred_zone
));
3025 gfp_to_alloc_flags(gfp_t gfp_mask
)
3027 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
3029 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
3030 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
3033 * The caller may dip into page reserves a bit more if the caller
3034 * cannot run direct reclaim, or if the caller has realtime scheduling
3035 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
3036 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
3038 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
3040 if (gfp_mask
& __GFP_ATOMIC
) {
3042 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
3043 * if it can't schedule.
3045 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
3046 alloc_flags
|= ALLOC_HARDER
;
3048 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
3049 * comment for __cpuset_node_allowed().
3051 alloc_flags
&= ~ALLOC_CPUSET
;
3052 } else if (unlikely(rt_task(current
)) && !in_interrupt())
3053 alloc_flags
|= ALLOC_HARDER
;
3055 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
3056 if (gfp_mask
& __GFP_MEMALLOC
)
3057 alloc_flags
|= ALLOC_NO_WATERMARKS
;
3058 else if (in_serving_softirq() && (current
->flags
& PF_MEMALLOC
))
3059 alloc_flags
|= ALLOC_NO_WATERMARKS
;
3060 else if (!in_interrupt() &&
3061 ((current
->flags
& PF_MEMALLOC
) ||
3062 unlikely(test_thread_flag(TIF_MEMDIE
))))
3063 alloc_flags
|= ALLOC_NO_WATERMARKS
;
3066 if (gfpflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
3067 alloc_flags
|= ALLOC_CMA
;
3072 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask
)
3074 return !!(gfp_to_alloc_flags(gfp_mask
) & ALLOC_NO_WATERMARKS
);
3077 static inline bool is_thp_gfp_mask(gfp_t gfp_mask
)
3079 return (gfp_mask
& (GFP_TRANSHUGE
| __GFP_KSWAPD_RECLAIM
)) == GFP_TRANSHUGE
;
3082 static inline struct page
*
3083 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
3084 struct alloc_context
*ac
)
3086 bool can_direct_reclaim
= gfp_mask
& __GFP_DIRECT_RECLAIM
;
3087 struct page
*page
= NULL
;
3089 unsigned long pages_reclaimed
= 0;
3090 unsigned long did_some_progress
;
3091 enum migrate_mode migration_mode
= MIGRATE_ASYNC
;
3092 bool deferred_compaction
= false;
3093 int contended_compaction
= COMPACT_CONTENDED_NONE
;
3096 * In the slowpath, we sanity check order to avoid ever trying to
3097 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3098 * be using allocators in order of preference for an area that is
3101 if (order
>= MAX_ORDER
) {
3102 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
3107 * We also sanity check to catch abuse of atomic reserves being used by
3108 * callers that are not in atomic context.
3110 if (WARN_ON_ONCE((gfp_mask
& (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)) ==
3111 (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)))
3112 gfp_mask
&= ~__GFP_ATOMIC
;
3115 if (gfp_mask
& __GFP_KSWAPD_RECLAIM
)
3116 wake_all_kswapds(order
, ac
);
3119 * OK, we're below the kswapd watermark and have kicked background
3120 * reclaim. Now things get more complex, so set up alloc_flags according
3121 * to how we want to proceed.
3123 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
3126 * Find the true preferred zone if the allocation is unconstrained by
3129 if (!(alloc_flags
& ALLOC_CPUSET
) && !ac
->nodemask
) {
3130 struct zoneref
*preferred_zoneref
;
3131 preferred_zoneref
= first_zones_zonelist(ac
->zonelist
,
3132 ac
->high_zoneidx
, NULL
, &ac
->preferred_zone
);
3133 ac
->classzone_idx
= zonelist_zone_idx(preferred_zoneref
);
3136 /* This is the last chance, in general, before the goto nopage. */
3137 page
= get_page_from_freelist(gfp_mask
, order
,
3138 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
3142 /* Allocate without watermarks if the context allows */
3143 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
3145 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3146 * the allocation is high priority and these type of
3147 * allocations are system rather than user orientated
3149 ac
->zonelist
= node_zonelist(numa_node_id(), gfp_mask
);
3150 page
= get_page_from_freelist(gfp_mask
, order
,
3151 ALLOC_NO_WATERMARKS
, ac
);
3156 /* Caller is not willing to reclaim, we can't balance anything */
3157 if (!can_direct_reclaim
) {
3159 * All existing users of the __GFP_NOFAIL are blockable, so warn
3160 * of any new users that actually allow this type of allocation
3163 WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
);
3167 /* Avoid recursion of direct reclaim */
3168 if (current
->flags
& PF_MEMALLOC
) {
3170 * __GFP_NOFAIL request from this context is rather bizarre
3171 * because we cannot reclaim anything and only can loop waiting
3172 * for somebody to do a work for us.
3174 if (WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
)) {
3181 /* Avoid allocations with no watermarks from looping endlessly */
3182 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
3186 * Try direct compaction. The first pass is asynchronous. Subsequent
3187 * attempts after direct reclaim are synchronous
3189 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
, ac
,
3191 &contended_compaction
,
3192 &deferred_compaction
);
3196 /* Checks for THP-specific high-order allocations */
3197 if (is_thp_gfp_mask(gfp_mask
)) {
3199 * If compaction is deferred for high-order allocations, it is
3200 * because sync compaction recently failed. If this is the case
3201 * and the caller requested a THP allocation, we do not want
3202 * to heavily disrupt the system, so we fail the allocation
3203 * instead of entering direct reclaim.
3205 if (deferred_compaction
)
3209 * In all zones where compaction was attempted (and not
3210 * deferred or skipped), lock contention has been detected.
3211 * For THP allocation we do not want to disrupt the others
3212 * so we fallback to base pages instead.
3214 if (contended_compaction
== COMPACT_CONTENDED_LOCK
)
3218 * If compaction was aborted due to need_resched(), we do not
3219 * want to further increase allocation latency, unless it is
3220 * khugepaged trying to collapse.
3222 if (contended_compaction
== COMPACT_CONTENDED_SCHED
3223 && !(current
->flags
& PF_KTHREAD
))
3228 * It can become very expensive to allocate transparent hugepages at
3229 * fault, so use asynchronous memory compaction for THP unless it is
3230 * khugepaged trying to collapse.
3232 if (!is_thp_gfp_mask(gfp_mask
) || (current
->flags
& PF_KTHREAD
))
3233 migration_mode
= MIGRATE_SYNC_LIGHT
;
3235 /* Try direct reclaim and then allocating */
3236 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
, alloc_flags
, ac
,
3237 &did_some_progress
);
3241 /* Do not loop if specifically requested */
3242 if (gfp_mask
& __GFP_NORETRY
)
3245 /* Keep reclaiming pages as long as there is reasonable progress */
3246 pages_reclaimed
+= did_some_progress
;
3247 if ((did_some_progress
&& order
<= PAGE_ALLOC_COSTLY_ORDER
) ||
3248 ((gfp_mask
& __GFP_REPEAT
) && pages_reclaimed
< (1 << order
))) {
3249 /* Wait for some write requests to complete then retry */
3250 wait_iff_congested(ac
->preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
3254 /* Reclaim has failed us, start killing things */
3255 page
= __alloc_pages_may_oom(gfp_mask
, order
, ac
, &did_some_progress
);
3259 /* Retry as long as the OOM killer is making progress */
3260 if (did_some_progress
)
3265 * High-order allocations do not necessarily loop after
3266 * direct reclaim and reclaim/compaction depends on compaction
3267 * being called after reclaim so call directly if necessary
3269 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
,
3271 &contended_compaction
,
3272 &deferred_compaction
);
3276 warn_alloc_failed(gfp_mask
, order
, NULL
);
3282 * This is the 'heart' of the zoned buddy allocator.
3285 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
3286 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
3288 struct zoneref
*preferred_zoneref
;
3289 struct page
*page
= NULL
;
3290 unsigned int cpuset_mems_cookie
;
3291 int alloc_flags
= ALLOC_WMARK_LOW
|ALLOC_CPUSET
|ALLOC_FAIR
;
3292 gfp_t alloc_mask
; /* The gfp_t that was actually used for allocation */
3293 struct alloc_context ac
= {
3294 .high_zoneidx
= gfp_zone(gfp_mask
),
3295 .nodemask
= nodemask
,
3296 .migratetype
= gfpflags_to_migratetype(gfp_mask
),
3299 gfp_mask
&= gfp_allowed_mask
;
3301 lockdep_trace_alloc(gfp_mask
);
3303 might_sleep_if(gfp_mask
& __GFP_DIRECT_RECLAIM
);
3305 if (should_fail_alloc_page(gfp_mask
, order
))
3309 * Check the zones suitable for the gfp_mask contain at least one
3310 * valid zone. It's possible to have an empty zonelist as a result
3311 * of __GFP_THISNODE and a memoryless node
3313 if (unlikely(!zonelist
->_zonerefs
->zone
))
3316 if (IS_ENABLED(CONFIG_CMA
) && ac
.migratetype
== MIGRATE_MOVABLE
)
3317 alloc_flags
|= ALLOC_CMA
;
3320 cpuset_mems_cookie
= read_mems_allowed_begin();
3322 /* We set it here, as __alloc_pages_slowpath might have changed it */
3323 ac
.zonelist
= zonelist
;
3325 /* Dirty zone balancing only done in the fast path */
3326 ac
.spread_dirty_pages
= (gfp_mask
& __GFP_WRITE
);
3328 /* The preferred zone is used for statistics later */
3329 preferred_zoneref
= first_zones_zonelist(ac
.zonelist
, ac
.high_zoneidx
,
3330 ac
.nodemask
? : &cpuset_current_mems_allowed
,
3331 &ac
.preferred_zone
);
3332 if (!ac
.preferred_zone
)
3334 ac
.classzone_idx
= zonelist_zone_idx(preferred_zoneref
);
3336 /* First allocation attempt */
3337 alloc_mask
= gfp_mask
|__GFP_HARDWALL
;
3338 page
= get_page_from_freelist(alloc_mask
, order
, alloc_flags
, &ac
);
3339 if (unlikely(!page
)) {
3341 * Runtime PM, block IO and its error handling path
3342 * can deadlock because I/O on the device might not
3345 alloc_mask
= memalloc_noio_flags(gfp_mask
);
3346 ac
.spread_dirty_pages
= false;
3348 page
= __alloc_pages_slowpath(alloc_mask
, order
, &ac
);
3351 if (kmemcheck_enabled
&& page
)
3352 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
3354 trace_mm_page_alloc(page
, order
, alloc_mask
, ac
.migratetype
);
3358 * When updating a task's mems_allowed, it is possible to race with
3359 * parallel threads in such a way that an allocation can fail while
3360 * the mask is being updated. If a page allocation is about to fail,
3361 * check if the cpuset changed during allocation and if so, retry.
3363 if (unlikely(!page
&& read_mems_allowed_retry(cpuset_mems_cookie
)))
3368 EXPORT_SYMBOL(__alloc_pages_nodemask
);
3371 * Common helper functions.
3373 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
3378 * __get_free_pages() returns a 32-bit address, which cannot represent
3381 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
3383 page
= alloc_pages(gfp_mask
, order
);
3386 return (unsigned long) page_address(page
);
3388 EXPORT_SYMBOL(__get_free_pages
);
3390 unsigned long get_zeroed_page(gfp_t gfp_mask
)
3392 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
3394 EXPORT_SYMBOL(get_zeroed_page
);
3396 void __free_pages(struct page
*page
, unsigned int order
)
3398 if (put_page_testzero(page
)) {
3400 free_hot_cold_page(page
, false);
3402 __free_pages_ok(page
, order
);
3406 EXPORT_SYMBOL(__free_pages
);
3408 void free_pages(unsigned long addr
, unsigned int order
)
3411 VM_BUG_ON(!virt_addr_valid((void *)addr
));
3412 __free_pages(virt_to_page((void *)addr
), order
);
3416 EXPORT_SYMBOL(free_pages
);
3420 * An arbitrary-length arbitrary-offset area of memory which resides
3421 * within a 0 or higher order page. Multiple fragments within that page
3422 * are individually refcounted, in the page's reference counter.
3424 * The page_frag functions below provide a simple allocation framework for
3425 * page fragments. This is used by the network stack and network device
3426 * drivers to provide a backing region of memory for use as either an
3427 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3429 static struct page
*__page_frag_refill(struct page_frag_cache
*nc
,
3432 struct page
*page
= NULL
;
3433 gfp_t gfp
= gfp_mask
;
3435 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3436 gfp_mask
|= __GFP_COMP
| __GFP_NOWARN
| __GFP_NORETRY
|
3438 page
= alloc_pages_node(NUMA_NO_NODE
, gfp_mask
,
3439 PAGE_FRAG_CACHE_MAX_ORDER
);
3440 nc
->size
= page
? PAGE_FRAG_CACHE_MAX_SIZE
: PAGE_SIZE
;
3442 if (unlikely(!page
))
3443 page
= alloc_pages_node(NUMA_NO_NODE
, gfp
, 0);
3445 nc
->va
= page
? page_address(page
) : NULL
;
3450 void *__alloc_page_frag(struct page_frag_cache
*nc
,
3451 unsigned int fragsz
, gfp_t gfp_mask
)
3453 unsigned int size
= PAGE_SIZE
;
3457 if (unlikely(!nc
->va
)) {
3459 page
= __page_frag_refill(nc
, gfp_mask
);
3463 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3464 /* if size can vary use size else just use PAGE_SIZE */
3467 /* Even if we own the page, we do not use atomic_set().
3468 * This would break get_page_unless_zero() users.
3470 page_ref_add(page
, size
- 1);
3472 /* reset page count bias and offset to start of new frag */
3473 nc
->pfmemalloc
= page_is_pfmemalloc(page
);
3474 nc
->pagecnt_bias
= size
;
3478 offset
= nc
->offset
- fragsz
;
3479 if (unlikely(offset
< 0)) {
3480 page
= virt_to_page(nc
->va
);
3482 if (!page_ref_sub_and_test(page
, nc
->pagecnt_bias
))
3485 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3486 /* if size can vary use size else just use PAGE_SIZE */
3489 /* OK, page count is 0, we can safely set it */
3490 set_page_count(page
, size
);
3492 /* reset page count bias and offset to start of new frag */
3493 nc
->pagecnt_bias
= size
;
3494 offset
= size
- fragsz
;
3498 nc
->offset
= offset
;
3500 return nc
->va
+ offset
;
3502 EXPORT_SYMBOL(__alloc_page_frag
);
3505 * Frees a page fragment allocated out of either a compound or order 0 page.
3507 void __free_page_frag(void *addr
)
3509 struct page
*page
= virt_to_head_page(addr
);
3511 if (unlikely(put_page_testzero(page
)))
3512 __free_pages_ok(page
, compound_order(page
));
3514 EXPORT_SYMBOL(__free_page_frag
);
3517 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
3518 * of the current memory cgroup if __GFP_ACCOUNT is set, other than that it is
3519 * equivalent to alloc_pages.
3521 * It should be used when the caller would like to use kmalloc, but since the
3522 * allocation is large, it has to fall back to the page allocator.
3524 struct page
*alloc_kmem_pages(gfp_t gfp_mask
, unsigned int order
)
3528 page
= alloc_pages(gfp_mask
, order
);
3529 if (page
&& memcg_kmem_charge(page
, gfp_mask
, order
) != 0) {
3530 __free_pages(page
, order
);
3536 struct page
*alloc_kmem_pages_node(int nid
, gfp_t gfp_mask
, unsigned int order
)
3540 page
= alloc_pages_node(nid
, gfp_mask
, order
);
3541 if (page
&& memcg_kmem_charge(page
, gfp_mask
, order
) != 0) {
3542 __free_pages(page
, order
);
3549 * __free_kmem_pages and free_kmem_pages will free pages allocated with
3552 void __free_kmem_pages(struct page
*page
, unsigned int order
)
3554 memcg_kmem_uncharge(page
, order
);
3555 __free_pages(page
, order
);
3558 void free_kmem_pages(unsigned long addr
, unsigned int order
)
3561 VM_BUG_ON(!virt_addr_valid((void *)addr
));
3562 __free_kmem_pages(virt_to_page((void *)addr
), order
);
3566 static void *make_alloc_exact(unsigned long addr
, unsigned int order
,
3570 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
3571 unsigned long used
= addr
+ PAGE_ALIGN(size
);
3573 split_page(virt_to_page((void *)addr
), order
);
3574 while (used
< alloc_end
) {
3579 return (void *)addr
;
3583 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3584 * @size: the number of bytes to allocate
3585 * @gfp_mask: GFP flags for the allocation
3587 * This function is similar to alloc_pages(), except that it allocates the
3588 * minimum number of pages to satisfy the request. alloc_pages() can only
3589 * allocate memory in power-of-two pages.
3591 * This function is also limited by MAX_ORDER.
3593 * Memory allocated by this function must be released by free_pages_exact().
3595 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
3597 unsigned int order
= get_order(size
);
3600 addr
= __get_free_pages(gfp_mask
, order
);
3601 return make_alloc_exact(addr
, order
, size
);
3603 EXPORT_SYMBOL(alloc_pages_exact
);
3606 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3608 * @nid: the preferred node ID where memory should be allocated
3609 * @size: the number of bytes to allocate
3610 * @gfp_mask: GFP flags for the allocation
3612 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3615 void * __meminit
alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
3617 unsigned int order
= get_order(size
);
3618 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
3621 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
3625 * free_pages_exact - release memory allocated via alloc_pages_exact()
3626 * @virt: the value returned by alloc_pages_exact.
3627 * @size: size of allocation, same value as passed to alloc_pages_exact().
3629 * Release the memory allocated by a previous call to alloc_pages_exact.
3631 void free_pages_exact(void *virt
, size_t size
)
3633 unsigned long addr
= (unsigned long)virt
;
3634 unsigned long end
= addr
+ PAGE_ALIGN(size
);
3636 while (addr
< end
) {
3641 EXPORT_SYMBOL(free_pages_exact
);
3644 * nr_free_zone_pages - count number of pages beyond high watermark
3645 * @offset: The zone index of the highest zone
3647 * nr_free_zone_pages() counts the number of counts pages which are beyond the
3648 * high watermark within all zones at or below a given zone index. For each
3649 * zone, the number of pages is calculated as:
3650 * managed_pages - high_pages
3652 static unsigned long nr_free_zone_pages(int offset
)
3657 /* Just pick one node, since fallback list is circular */
3658 unsigned long sum
= 0;
3660 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
3662 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
3663 unsigned long size
= zone
->managed_pages
;
3664 unsigned long high
= high_wmark_pages(zone
);
3673 * nr_free_buffer_pages - count number of pages beyond high watermark
3675 * nr_free_buffer_pages() counts the number of pages which are beyond the high
3676 * watermark within ZONE_DMA and ZONE_NORMAL.
3678 unsigned long nr_free_buffer_pages(void)
3680 return nr_free_zone_pages(gfp_zone(GFP_USER
));
3682 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
3685 * nr_free_pagecache_pages - count number of pages beyond high watermark
3687 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3688 * high watermark within all zones.
3690 unsigned long nr_free_pagecache_pages(void)
3692 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
3695 static inline void show_node(struct zone
*zone
)
3697 if (IS_ENABLED(CONFIG_NUMA
))
3698 printk("Node %d ", zone_to_nid(zone
));
3701 long si_mem_available(void)
3704 unsigned long pagecache
;
3705 unsigned long wmark_low
= 0;
3706 unsigned long pages
[NR_LRU_LISTS
];
3710 for (lru
= LRU_BASE
; lru
< NR_LRU_LISTS
; lru
++)
3711 pages
[lru
] = global_page_state(NR_LRU_BASE
+ lru
);
3714 wmark_low
+= zone
->watermark
[WMARK_LOW
];
3717 * Estimate the amount of memory available for userspace allocations,
3718 * without causing swapping.
3720 available
= global_page_state(NR_FREE_PAGES
) - totalreserve_pages
;
3723 * Not all the page cache can be freed, otherwise the system will
3724 * start swapping. Assume at least half of the page cache, or the
3725 * low watermark worth of cache, needs to stay.
3727 pagecache
= pages
[LRU_ACTIVE_FILE
] + pages
[LRU_INACTIVE_FILE
];
3728 pagecache
-= min(pagecache
/ 2, wmark_low
);
3729 available
+= pagecache
;
3732 * Part of the reclaimable slab consists of items that are in use,
3733 * and cannot be freed. Cap this estimate at the low watermark.
3735 available
+= global_page_state(NR_SLAB_RECLAIMABLE
) -
3736 min(global_page_state(NR_SLAB_RECLAIMABLE
) / 2, wmark_low
);
3742 EXPORT_SYMBOL_GPL(si_mem_available
);
3744 void si_meminfo(struct sysinfo
*val
)
3746 val
->totalram
= totalram_pages
;
3747 val
->sharedram
= global_page_state(NR_SHMEM
);
3748 val
->freeram
= global_page_state(NR_FREE_PAGES
);
3749 val
->bufferram
= nr_blockdev_pages();
3750 val
->totalhigh
= totalhigh_pages
;
3751 val
->freehigh
= nr_free_highpages();
3752 val
->mem_unit
= PAGE_SIZE
;
3755 EXPORT_SYMBOL(si_meminfo
);
3758 void si_meminfo_node(struct sysinfo
*val
, int nid
)
3760 int zone_type
; /* needs to be signed */
3761 unsigned long managed_pages
= 0;
3762 pg_data_t
*pgdat
= NODE_DATA(nid
);
3764 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++)
3765 managed_pages
+= pgdat
->node_zones
[zone_type
].managed_pages
;
3766 val
->totalram
= managed_pages
;
3767 val
->sharedram
= node_page_state(nid
, NR_SHMEM
);
3768 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
3769 #ifdef CONFIG_HIGHMEM
3770 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].managed_pages
;
3771 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
3777 val
->mem_unit
= PAGE_SIZE
;
3782 * Determine whether the node should be displayed or not, depending on whether
3783 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
3785 bool skip_free_areas_node(unsigned int flags
, int nid
)
3788 unsigned int cpuset_mems_cookie
;
3790 if (!(flags
& SHOW_MEM_FILTER_NODES
))
3794 cpuset_mems_cookie
= read_mems_allowed_begin();
3795 ret
= !node_isset(nid
, cpuset_current_mems_allowed
);
3796 } while (read_mems_allowed_retry(cpuset_mems_cookie
));
3801 #define K(x) ((x) << (PAGE_SHIFT-10))
3803 static void show_migration_types(unsigned char type
)
3805 static const char types
[MIGRATE_TYPES
] = {
3806 [MIGRATE_UNMOVABLE
] = 'U',
3807 [MIGRATE_MOVABLE
] = 'M',
3808 [MIGRATE_RECLAIMABLE
] = 'E',
3809 [MIGRATE_HIGHATOMIC
] = 'H',
3811 [MIGRATE_CMA
] = 'C',
3813 #ifdef CONFIG_MEMORY_ISOLATION
3814 [MIGRATE_ISOLATE
] = 'I',
3817 char tmp
[MIGRATE_TYPES
+ 1];
3821 for (i
= 0; i
< MIGRATE_TYPES
; i
++) {
3822 if (type
& (1 << i
))
3827 printk("(%s) ", tmp
);
3831 * Show free area list (used inside shift_scroll-lock stuff)
3832 * We also calculate the percentage fragmentation. We do this by counting the
3833 * memory on each free list with the exception of the first item on the list.
3836 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
3839 void show_free_areas(unsigned int filter
)
3841 unsigned long free_pcp
= 0;
3845 for_each_populated_zone(zone
) {
3846 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3849 for_each_online_cpu(cpu
)
3850 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
3853 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3854 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3855 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
3856 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3857 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3858 " free:%lu free_pcp:%lu free_cma:%lu\n",
3859 global_page_state(NR_ACTIVE_ANON
),
3860 global_page_state(NR_INACTIVE_ANON
),
3861 global_page_state(NR_ISOLATED_ANON
),
3862 global_page_state(NR_ACTIVE_FILE
),
3863 global_page_state(NR_INACTIVE_FILE
),
3864 global_page_state(NR_ISOLATED_FILE
),
3865 global_page_state(NR_UNEVICTABLE
),
3866 global_page_state(NR_FILE_DIRTY
),
3867 global_page_state(NR_WRITEBACK
),
3868 global_page_state(NR_UNSTABLE_NFS
),
3869 global_page_state(NR_SLAB_RECLAIMABLE
),
3870 global_page_state(NR_SLAB_UNRECLAIMABLE
),
3871 global_page_state(NR_FILE_MAPPED
),
3872 global_page_state(NR_SHMEM
),
3873 global_page_state(NR_PAGETABLE
),
3874 global_page_state(NR_BOUNCE
),
3875 global_page_state(NR_FREE_PAGES
),
3877 global_page_state(NR_FREE_CMA_PAGES
));
3879 for_each_populated_zone(zone
) {
3882 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3886 for_each_online_cpu(cpu
)
3887 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
3895 " active_anon:%lukB"
3896 " inactive_anon:%lukB"
3897 " active_file:%lukB"
3898 " inactive_file:%lukB"
3899 " unevictable:%lukB"
3900 " isolated(anon):%lukB"
3901 " isolated(file):%lukB"
3909 " slab_reclaimable:%lukB"
3910 " slab_unreclaimable:%lukB"
3911 " kernel_stack:%lukB"
3918 " writeback_tmp:%lukB"
3919 " pages_scanned:%lu"
3920 " all_unreclaimable? %s"
3923 K(zone_page_state(zone
, NR_FREE_PAGES
)),
3924 K(min_wmark_pages(zone
)),
3925 K(low_wmark_pages(zone
)),
3926 K(high_wmark_pages(zone
)),
3927 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
3928 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
3929 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
3930 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
3931 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
3932 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
3933 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
3934 K(zone
->present_pages
),
3935 K(zone
->managed_pages
),
3936 K(zone_page_state(zone
, NR_MLOCK
)),
3937 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
3938 K(zone_page_state(zone
, NR_WRITEBACK
)),
3939 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
3940 K(zone_page_state(zone
, NR_SHMEM
)),
3941 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
3942 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
3943 zone_page_state(zone
, NR_KERNEL_STACK
) *
3945 K(zone_page_state(zone
, NR_PAGETABLE
)),
3946 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
3947 K(zone_page_state(zone
, NR_BOUNCE
)),
3949 K(this_cpu_read(zone
->pageset
->pcp
.count
)),
3950 K(zone_page_state(zone
, NR_FREE_CMA_PAGES
)),
3951 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
3952 K(zone_page_state(zone
, NR_PAGES_SCANNED
)),
3953 (!zone_reclaimable(zone
) ? "yes" : "no")
3955 printk("lowmem_reserve[]:");
3956 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3957 printk(" %ld", zone
->lowmem_reserve
[i
]);
3961 for_each_populated_zone(zone
) {
3963 unsigned long nr
[MAX_ORDER
], flags
, total
= 0;
3964 unsigned char types
[MAX_ORDER
];
3966 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3969 printk("%s: ", zone
->name
);
3971 spin_lock_irqsave(&zone
->lock
, flags
);
3972 for (order
= 0; order
< MAX_ORDER
; order
++) {
3973 struct free_area
*area
= &zone
->free_area
[order
];
3976 nr
[order
] = area
->nr_free
;
3977 total
+= nr
[order
] << order
;
3980 for (type
= 0; type
< MIGRATE_TYPES
; type
++) {
3981 if (!list_empty(&area
->free_list
[type
]))
3982 types
[order
] |= 1 << type
;
3985 spin_unlock_irqrestore(&zone
->lock
, flags
);
3986 for (order
= 0; order
< MAX_ORDER
; order
++) {
3987 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
3989 show_migration_types(types
[order
]);
3991 printk("= %lukB\n", K(total
));
3994 hugetlb_show_meminfo();
3996 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
3998 show_swap_cache_info();
4001 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
4003 zoneref
->zone
= zone
;
4004 zoneref
->zone_idx
= zone_idx(zone
);
4008 * Builds allocation fallback zone lists.
4010 * Add all populated zones of a node to the zonelist.
4012 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
4016 enum zone_type zone_type
= MAX_NR_ZONES
;
4020 zone
= pgdat
->node_zones
+ zone_type
;
4021 if (populated_zone(zone
)) {
4022 zoneref_set_zone(zone
,
4023 &zonelist
->_zonerefs
[nr_zones
++]);
4024 check_highest_zone(zone_type
);
4026 } while (zone_type
);
4034 * 0 = automatic detection of better ordering.
4035 * 1 = order by ([node] distance, -zonetype)
4036 * 2 = order by (-zonetype, [node] distance)
4038 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
4039 * the same zonelist. So only NUMA can configure this param.
4041 #define ZONELIST_ORDER_DEFAULT 0
4042 #define ZONELIST_ORDER_NODE 1
4043 #define ZONELIST_ORDER_ZONE 2
4045 /* zonelist order in the kernel.
4046 * set_zonelist_order() will set this to NODE or ZONE.
4048 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4049 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
4053 /* The value user specified ....changed by config */
4054 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4055 /* string for sysctl */
4056 #define NUMA_ZONELIST_ORDER_LEN 16
4057 char numa_zonelist_order
[16] = "default";
4060 * interface for configure zonelist ordering.
4061 * command line option "numa_zonelist_order"
4062 * = "[dD]efault - default, automatic configuration.
4063 * = "[nN]ode - order by node locality, then by zone within node
4064 * = "[zZ]one - order by zone, then by locality within zone
4067 static int __parse_numa_zonelist_order(char *s
)
4069 if (*s
== 'd' || *s
== 'D') {
4070 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4071 } else if (*s
== 'n' || *s
== 'N') {
4072 user_zonelist_order
= ZONELIST_ORDER_NODE
;
4073 } else if (*s
== 'z' || *s
== 'Z') {
4074 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
4076 pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s
);
4082 static __init
int setup_numa_zonelist_order(char *s
)
4089 ret
= __parse_numa_zonelist_order(s
);
4091 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
4095 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
4098 * sysctl handler for numa_zonelist_order
4100 int numa_zonelist_order_handler(struct ctl_table
*table
, int write
,
4101 void __user
*buffer
, size_t *length
,
4104 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
4106 static DEFINE_MUTEX(zl_order_mutex
);
4108 mutex_lock(&zl_order_mutex
);
4110 if (strlen((char *)table
->data
) >= NUMA_ZONELIST_ORDER_LEN
) {
4114 strcpy(saved_string
, (char *)table
->data
);
4116 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
4120 int oldval
= user_zonelist_order
;
4122 ret
= __parse_numa_zonelist_order((char *)table
->data
);
4125 * bogus value. restore saved string
4127 strncpy((char *)table
->data
, saved_string
,
4128 NUMA_ZONELIST_ORDER_LEN
);
4129 user_zonelist_order
= oldval
;
4130 } else if (oldval
!= user_zonelist_order
) {
4131 mutex_lock(&zonelists_mutex
);
4132 build_all_zonelists(NULL
, NULL
);
4133 mutex_unlock(&zonelists_mutex
);
4137 mutex_unlock(&zl_order_mutex
);
4142 #define MAX_NODE_LOAD (nr_online_nodes)
4143 static int node_load
[MAX_NUMNODES
];
4146 * find_next_best_node - find the next node that should appear in a given node's fallback list
4147 * @node: node whose fallback list we're appending
4148 * @used_node_mask: nodemask_t of already used nodes
4150 * We use a number of factors to determine which is the next node that should
4151 * appear on a given node's fallback list. The node should not have appeared
4152 * already in @node's fallback list, and it should be the next closest node
4153 * according to the distance array (which contains arbitrary distance values
4154 * from each node to each node in the system), and should also prefer nodes
4155 * with no CPUs, since presumably they'll have very little allocation pressure
4156 * on them otherwise.
4157 * It returns -1 if no node is found.
4159 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
4162 int min_val
= INT_MAX
;
4163 int best_node
= NUMA_NO_NODE
;
4164 const struct cpumask
*tmp
= cpumask_of_node(0);
4166 /* Use the local node if we haven't already */
4167 if (!node_isset(node
, *used_node_mask
)) {
4168 node_set(node
, *used_node_mask
);
4172 for_each_node_state(n
, N_MEMORY
) {
4174 /* Don't want a node to appear more than once */
4175 if (node_isset(n
, *used_node_mask
))
4178 /* Use the distance array to find the distance */
4179 val
= node_distance(node
, n
);
4181 /* Penalize nodes under us ("prefer the next node") */
4184 /* Give preference to headless and unused nodes */
4185 tmp
= cpumask_of_node(n
);
4186 if (!cpumask_empty(tmp
))
4187 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
4189 /* Slight preference for less loaded node */
4190 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
4191 val
+= node_load
[n
];
4193 if (val
< min_val
) {
4200 node_set(best_node
, *used_node_mask
);
4207 * Build zonelists ordered by node and zones within node.
4208 * This results in maximum locality--normal zone overflows into local
4209 * DMA zone, if any--but risks exhausting DMA zone.
4211 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
4214 struct zonelist
*zonelist
;
4216 zonelist
= &pgdat
->node_zonelists
[0];
4217 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
4219 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4220 zonelist
->_zonerefs
[j
].zone
= NULL
;
4221 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4225 * Build gfp_thisnode zonelists
4227 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
4230 struct zonelist
*zonelist
;
4232 zonelist
= &pgdat
->node_zonelists
[1];
4233 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4234 zonelist
->_zonerefs
[j
].zone
= NULL
;
4235 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4239 * Build zonelists ordered by zone and nodes within zones.
4240 * This results in conserving DMA zone[s] until all Normal memory is
4241 * exhausted, but results in overflowing to remote node while memory
4242 * may still exist in local DMA zone.
4244 static int node_order
[MAX_NUMNODES
];
4246 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
4249 int zone_type
; /* needs to be signed */
4251 struct zonelist
*zonelist
;
4253 zonelist
= &pgdat
->node_zonelists
[0];
4255 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
4256 for (j
= 0; j
< nr_nodes
; j
++) {
4257 node
= node_order
[j
];
4258 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
4259 if (populated_zone(z
)) {
4261 &zonelist
->_zonerefs
[pos
++]);
4262 check_highest_zone(zone_type
);
4266 zonelist
->_zonerefs
[pos
].zone
= NULL
;
4267 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
4270 #if defined(CONFIG_64BIT)
4272 * Devices that require DMA32/DMA are relatively rare and do not justify a
4273 * penalty to every machine in case the specialised case applies. Default
4274 * to Node-ordering on 64-bit NUMA machines
4276 static int default_zonelist_order(void)
4278 return ZONELIST_ORDER_NODE
;
4282 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4283 * by the kernel. If processes running on node 0 deplete the low memory zone
4284 * then reclaim will occur more frequency increasing stalls and potentially
4285 * be easier to OOM if a large percentage of the zone is under writeback or
4286 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4287 * Hence, default to zone ordering on 32-bit.
4289 static int default_zonelist_order(void)
4291 return ZONELIST_ORDER_ZONE
;
4293 #endif /* CONFIG_64BIT */
4295 static void set_zonelist_order(void)
4297 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
4298 current_zonelist_order
= default_zonelist_order();
4300 current_zonelist_order
= user_zonelist_order
;
4303 static void build_zonelists(pg_data_t
*pgdat
)
4306 nodemask_t used_mask
;
4307 int local_node
, prev_node
;
4308 struct zonelist
*zonelist
;
4309 unsigned int order
= current_zonelist_order
;
4311 /* initialize zonelists */
4312 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
4313 zonelist
= pgdat
->node_zonelists
+ i
;
4314 zonelist
->_zonerefs
[0].zone
= NULL
;
4315 zonelist
->_zonerefs
[0].zone_idx
= 0;
4318 /* NUMA-aware ordering of nodes */
4319 local_node
= pgdat
->node_id
;
4320 load
= nr_online_nodes
;
4321 prev_node
= local_node
;
4322 nodes_clear(used_mask
);
4324 memset(node_order
, 0, sizeof(node_order
));
4327 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
4329 * We don't want to pressure a particular node.
4330 * So adding penalty to the first node in same
4331 * distance group to make it round-robin.
4333 if (node_distance(local_node
, node
) !=
4334 node_distance(local_node
, prev_node
))
4335 node_load
[node
] = load
;
4339 if (order
== ZONELIST_ORDER_NODE
)
4340 build_zonelists_in_node_order(pgdat
, node
);
4342 node_order
[i
++] = node
; /* remember order */
4345 if (order
== ZONELIST_ORDER_ZONE
) {
4346 /* calculate node order -- i.e., DMA last! */
4347 build_zonelists_in_zone_order(pgdat
, i
);
4350 build_thisnode_zonelists(pgdat
);
4353 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4355 * Return node id of node used for "local" allocations.
4356 * I.e., first node id of first zone in arg node's generic zonelist.
4357 * Used for initializing percpu 'numa_mem', which is used primarily
4358 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4360 int local_memory_node(int node
)
4364 (void)first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
4365 gfp_zone(GFP_KERNEL
),
4372 #else /* CONFIG_NUMA */
4374 static void set_zonelist_order(void)
4376 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
4379 static void build_zonelists(pg_data_t
*pgdat
)
4381 int node
, local_node
;
4383 struct zonelist
*zonelist
;
4385 local_node
= pgdat
->node_id
;
4387 zonelist
= &pgdat
->node_zonelists
[0];
4388 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4391 * Now we build the zonelist so that it contains the zones
4392 * of all the other nodes.
4393 * We don't want to pressure a particular node, so when
4394 * building the zones for node N, we make sure that the
4395 * zones coming right after the local ones are those from
4396 * node N+1 (modulo N)
4398 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
4399 if (!node_online(node
))
4401 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4403 for (node
= 0; node
< local_node
; node
++) {
4404 if (!node_online(node
))
4406 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4409 zonelist
->_zonerefs
[j
].zone
= NULL
;
4410 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4413 #endif /* CONFIG_NUMA */
4416 * Boot pageset table. One per cpu which is going to be used for all
4417 * zones and all nodes. The parameters will be set in such a way
4418 * that an item put on a list will immediately be handed over to
4419 * the buddy list. This is safe since pageset manipulation is done
4420 * with interrupts disabled.
4422 * The boot_pagesets must be kept even after bootup is complete for
4423 * unused processors and/or zones. They do play a role for bootstrapping
4424 * hotplugged processors.
4426 * zoneinfo_show() and maybe other functions do
4427 * not check if the processor is online before following the pageset pointer.
4428 * Other parts of the kernel may not check if the zone is available.
4430 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
4431 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
4432 static void setup_zone_pageset(struct zone
*zone
);
4435 * Global mutex to protect against size modification of zonelists
4436 * as well as to serialize pageset setup for the new populated zone.
4438 DEFINE_MUTEX(zonelists_mutex
);
4440 /* return values int ....just for stop_machine() */
4441 static int __build_all_zonelists(void *data
)
4445 pg_data_t
*self
= data
;
4448 memset(node_load
, 0, sizeof(node_load
));
4451 if (self
&& !node_online(self
->node_id
)) {
4452 build_zonelists(self
);
4455 for_each_online_node(nid
) {
4456 pg_data_t
*pgdat
= NODE_DATA(nid
);
4458 build_zonelists(pgdat
);
4462 * Initialize the boot_pagesets that are going to be used
4463 * for bootstrapping processors. The real pagesets for
4464 * each zone will be allocated later when the per cpu
4465 * allocator is available.
4467 * boot_pagesets are used also for bootstrapping offline
4468 * cpus if the system is already booted because the pagesets
4469 * are needed to initialize allocators on a specific cpu too.
4470 * F.e. the percpu allocator needs the page allocator which
4471 * needs the percpu allocator in order to allocate its pagesets
4472 * (a chicken-egg dilemma).
4474 for_each_possible_cpu(cpu
) {
4475 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
4477 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4479 * We now know the "local memory node" for each node--
4480 * i.e., the node of the first zone in the generic zonelist.
4481 * Set up numa_mem percpu variable for on-line cpus. During
4482 * boot, only the boot cpu should be on-line; we'll init the
4483 * secondary cpus' numa_mem as they come on-line. During
4484 * node/memory hotplug, we'll fixup all on-line cpus.
4486 if (cpu_online(cpu
))
4487 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
4494 static noinline
void __init
4495 build_all_zonelists_init(void)
4497 __build_all_zonelists(NULL
);
4498 mminit_verify_zonelist();
4499 cpuset_init_current_mems_allowed();
4503 * Called with zonelists_mutex held always
4504 * unless system_state == SYSTEM_BOOTING.
4506 * __ref due to (1) call of __meminit annotated setup_zone_pageset
4507 * [we're only called with non-NULL zone through __meminit paths] and
4508 * (2) call of __init annotated helper build_all_zonelists_init
4509 * [protected by SYSTEM_BOOTING].
4511 void __ref
build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
)
4513 set_zonelist_order();
4515 if (system_state
== SYSTEM_BOOTING
) {
4516 build_all_zonelists_init();
4518 #ifdef CONFIG_MEMORY_HOTPLUG
4520 setup_zone_pageset(zone
);
4522 /* we have to stop all cpus to guarantee there is no user
4524 stop_machine(__build_all_zonelists
, pgdat
, NULL
);
4525 /* cpuset refresh routine should be here */
4527 vm_total_pages
= nr_free_pagecache_pages();
4529 * Disable grouping by mobility if the number of pages in the
4530 * system is too low to allow the mechanism to work. It would be
4531 * more accurate, but expensive to check per-zone. This check is
4532 * made on memory-hotadd so a system can start with mobility
4533 * disabled and enable it later
4535 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
4536 page_group_by_mobility_disabled
= 1;
4538 page_group_by_mobility_disabled
= 0;
4540 pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n",
4542 zonelist_order_name
[current_zonelist_order
],
4543 page_group_by_mobility_disabled
? "off" : "on",
4546 pr_info("Policy zone: %s\n", zone_names
[policy_zone
]);
4551 * Helper functions to size the waitqueue hash table.
4552 * Essentially these want to choose hash table sizes sufficiently
4553 * large so that collisions trying to wait on pages are rare.
4554 * But in fact, the number of active page waitqueues on typical
4555 * systems is ridiculously low, less than 200. So this is even
4556 * conservative, even though it seems large.
4558 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
4559 * waitqueues, i.e. the size of the waitq table given the number of pages.
4561 #define PAGES_PER_WAITQUEUE 256
4563 #ifndef CONFIG_MEMORY_HOTPLUG
4564 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
4566 unsigned long size
= 1;
4568 pages
/= PAGES_PER_WAITQUEUE
;
4570 while (size
< pages
)
4574 * Once we have dozens or even hundreds of threads sleeping
4575 * on IO we've got bigger problems than wait queue collision.
4576 * Limit the size of the wait table to a reasonable size.
4578 size
= min(size
, 4096UL);
4580 return max(size
, 4UL);
4584 * A zone's size might be changed by hot-add, so it is not possible to determine
4585 * a suitable size for its wait_table. So we use the maximum size now.
4587 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
4589 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
4590 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4591 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
4593 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4594 * or more by the traditional way. (See above). It equals:
4596 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
4597 * ia64(16K page size) : = ( 8G + 4M)byte.
4598 * powerpc (64K page size) : = (32G +16M)byte.
4600 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
4607 * This is an integer logarithm so that shifts can be used later
4608 * to extract the more random high bits from the multiplicative
4609 * hash function before the remainder is taken.
4611 static inline unsigned long wait_table_bits(unsigned long size
)
4617 * Initially all pages are reserved - free ones are freed
4618 * up by free_all_bootmem() once the early boot process is
4619 * done. Non-atomic initialization, single-pass.
4621 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
4622 unsigned long start_pfn
, enum memmap_context context
)
4624 struct vmem_altmap
*altmap
= to_vmem_altmap(__pfn_to_phys(start_pfn
));
4625 unsigned long end_pfn
= start_pfn
+ size
;
4626 pg_data_t
*pgdat
= NODE_DATA(nid
);
4628 unsigned long nr_initialised
= 0;
4629 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4630 struct memblock_region
*r
= NULL
, *tmp
;
4633 if (highest_memmap_pfn
< end_pfn
- 1)
4634 highest_memmap_pfn
= end_pfn
- 1;
4637 * Honor reservation requested by the driver for this ZONE_DEVICE
4640 if (altmap
&& start_pfn
== altmap
->base_pfn
)
4641 start_pfn
+= altmap
->reserve
;
4643 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
4645 * There can be holes in boot-time mem_map[]s handed to this
4646 * function. They do not exist on hotplugged memory.
4648 if (context
!= MEMMAP_EARLY
)
4651 if (!early_pfn_valid(pfn
))
4653 if (!early_pfn_in_nid(pfn
, nid
))
4655 if (!update_defer_init(pgdat
, pfn
, end_pfn
, &nr_initialised
))
4658 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4660 * If not mirrored_kernelcore and ZONE_MOVABLE exists, range
4661 * from zone_movable_pfn[nid] to end of each node should be
4662 * ZONE_MOVABLE not ZONE_NORMAL. skip it.
4664 if (!mirrored_kernelcore
&& zone_movable_pfn
[nid
])
4665 if (zone
== ZONE_NORMAL
&& pfn
>= zone_movable_pfn
[nid
])
4669 * Check given memblock attribute by firmware which can affect
4670 * kernel memory layout. If zone==ZONE_MOVABLE but memory is
4671 * mirrored, it's an overlapped memmap init. skip it.
4673 if (mirrored_kernelcore
&& zone
== ZONE_MOVABLE
) {
4674 if (!r
|| pfn
>= memblock_region_memory_end_pfn(r
)) {
4675 for_each_memblock(memory
, tmp
)
4676 if (pfn
< memblock_region_memory_end_pfn(tmp
))
4680 if (pfn
>= memblock_region_memory_base_pfn(r
) &&
4681 memblock_is_mirror(r
)) {
4682 /* already initialized as NORMAL */
4683 pfn
= memblock_region_memory_end_pfn(r
);
4691 * Mark the block movable so that blocks are reserved for
4692 * movable at startup. This will force kernel allocations
4693 * to reserve their blocks rather than leaking throughout
4694 * the address space during boot when many long-lived
4695 * kernel allocations are made.
4697 * bitmap is created for zone's valid pfn range. but memmap
4698 * can be created for invalid pages (for alignment)
4699 * check here not to call set_pageblock_migratetype() against
4702 if (!(pfn
& (pageblock_nr_pages
- 1))) {
4703 struct page
*page
= pfn_to_page(pfn
);
4705 __init_single_page(page
, pfn
, zone
, nid
);
4706 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4708 __init_single_pfn(pfn
, zone
, nid
);
4713 static void __meminit
zone_init_free_lists(struct zone
*zone
)
4715 unsigned int order
, t
;
4716 for_each_migratetype_order(order
, t
) {
4717 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
4718 zone
->free_area
[order
].nr_free
= 0;
4722 #ifndef __HAVE_ARCH_MEMMAP_INIT
4723 #define memmap_init(size, nid, zone, start_pfn) \
4724 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4727 static int zone_batchsize(struct zone
*zone
)
4733 * The per-cpu-pages pools are set to around 1000th of the
4734 * size of the zone. But no more than 1/2 of a meg.
4736 * OK, so we don't know how big the cache is. So guess.
4738 batch
= zone
->managed_pages
/ 1024;
4739 if (batch
* PAGE_SIZE
> 512 * 1024)
4740 batch
= (512 * 1024) / PAGE_SIZE
;
4741 batch
/= 4; /* We effectively *= 4 below */
4746 * Clamp the batch to a 2^n - 1 value. Having a power
4747 * of 2 value was found to be more likely to have
4748 * suboptimal cache aliasing properties in some cases.
4750 * For example if 2 tasks are alternately allocating
4751 * batches of pages, one task can end up with a lot
4752 * of pages of one half of the possible page colors
4753 * and the other with pages of the other colors.
4755 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
4760 /* The deferral and batching of frees should be suppressed under NOMMU
4763 * The problem is that NOMMU needs to be able to allocate large chunks
4764 * of contiguous memory as there's no hardware page translation to
4765 * assemble apparent contiguous memory from discontiguous pages.
4767 * Queueing large contiguous runs of pages for batching, however,
4768 * causes the pages to actually be freed in smaller chunks. As there
4769 * can be a significant delay between the individual batches being
4770 * recycled, this leads to the once large chunks of space being
4771 * fragmented and becoming unavailable for high-order allocations.
4778 * pcp->high and pcp->batch values are related and dependent on one another:
4779 * ->batch must never be higher then ->high.
4780 * The following function updates them in a safe manner without read side
4783 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4784 * those fields changing asynchronously (acording the the above rule).
4786 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4787 * outside of boot time (or some other assurance that no concurrent updaters
4790 static void pageset_update(struct per_cpu_pages
*pcp
, unsigned long high
,
4791 unsigned long batch
)
4793 /* start with a fail safe value for batch */
4797 /* Update high, then batch, in order */
4804 /* a companion to pageset_set_high() */
4805 static void pageset_set_batch(struct per_cpu_pageset
*p
, unsigned long batch
)
4807 pageset_update(&p
->pcp
, 6 * batch
, max(1UL, 1 * batch
));
4810 static void pageset_init(struct per_cpu_pageset
*p
)
4812 struct per_cpu_pages
*pcp
;
4815 memset(p
, 0, sizeof(*p
));
4819 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
4820 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
4823 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
4826 pageset_set_batch(p
, batch
);
4830 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
4831 * to the value high for the pageset p.
4833 static void pageset_set_high(struct per_cpu_pageset
*p
,
4836 unsigned long batch
= max(1UL, high
/ 4);
4837 if ((high
/ 4) > (PAGE_SHIFT
* 8))
4838 batch
= PAGE_SHIFT
* 8;
4840 pageset_update(&p
->pcp
, high
, batch
);
4843 static void pageset_set_high_and_batch(struct zone
*zone
,
4844 struct per_cpu_pageset
*pcp
)
4846 if (percpu_pagelist_fraction
)
4847 pageset_set_high(pcp
,
4848 (zone
->managed_pages
/
4849 percpu_pagelist_fraction
));
4851 pageset_set_batch(pcp
, zone_batchsize(zone
));
4854 static void __meminit
zone_pageset_init(struct zone
*zone
, int cpu
)
4856 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
4859 pageset_set_high_and_batch(zone
, pcp
);
4862 static void __meminit
setup_zone_pageset(struct zone
*zone
)
4865 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
4866 for_each_possible_cpu(cpu
)
4867 zone_pageset_init(zone
, cpu
);
4871 * Allocate per cpu pagesets and initialize them.
4872 * Before this call only boot pagesets were available.
4874 void __init
setup_per_cpu_pageset(void)
4878 for_each_populated_zone(zone
)
4879 setup_zone_pageset(zone
);
4882 static noinline __init_refok
4883 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
4889 * The per-page waitqueue mechanism uses hashed waitqueues
4892 zone
->wait_table_hash_nr_entries
=
4893 wait_table_hash_nr_entries(zone_size_pages
);
4894 zone
->wait_table_bits
=
4895 wait_table_bits(zone
->wait_table_hash_nr_entries
);
4896 alloc_size
= zone
->wait_table_hash_nr_entries
4897 * sizeof(wait_queue_head_t
);
4899 if (!slab_is_available()) {
4900 zone
->wait_table
= (wait_queue_head_t
*)
4901 memblock_virt_alloc_node_nopanic(
4902 alloc_size
, zone
->zone_pgdat
->node_id
);
4905 * This case means that a zone whose size was 0 gets new memory
4906 * via memory hot-add.
4907 * But it may be the case that a new node was hot-added. In
4908 * this case vmalloc() will not be able to use this new node's
4909 * memory - this wait_table must be initialized to use this new
4910 * node itself as well.
4911 * To use this new node's memory, further consideration will be
4914 zone
->wait_table
= vmalloc(alloc_size
);
4916 if (!zone
->wait_table
)
4919 for (i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
4920 init_waitqueue_head(zone
->wait_table
+ i
);
4925 static __meminit
void zone_pcp_init(struct zone
*zone
)
4928 * per cpu subsystem is not up at this point. The following code
4929 * relies on the ability of the linker to provide the
4930 * offset of a (static) per cpu variable into the per cpu area.
4932 zone
->pageset
= &boot_pageset
;
4934 if (populated_zone(zone
))
4935 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
4936 zone
->name
, zone
->present_pages
,
4937 zone_batchsize(zone
));
4940 int __meminit
init_currently_empty_zone(struct zone
*zone
,
4941 unsigned long zone_start_pfn
,
4944 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
4946 ret
= zone_wait_table_init(zone
, size
);
4949 pgdat
->nr_zones
= zone_idx(zone
) + 1;
4951 zone
->zone_start_pfn
= zone_start_pfn
;
4953 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
4954 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4956 (unsigned long)zone_idx(zone
),
4957 zone_start_pfn
, (zone_start_pfn
+ size
));
4959 zone_init_free_lists(zone
);
4964 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4965 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4968 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4970 int __meminit
__early_pfn_to_nid(unsigned long pfn
,
4971 struct mminit_pfnnid_cache
*state
)
4973 unsigned long start_pfn
, end_pfn
;
4976 if (state
->last_start
<= pfn
&& pfn
< state
->last_end
)
4977 return state
->last_nid
;
4979 nid
= memblock_search_pfn_nid(pfn
, &start_pfn
, &end_pfn
);
4981 state
->last_start
= start_pfn
;
4982 state
->last_end
= end_pfn
;
4983 state
->last_nid
= nid
;
4988 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4991 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
4992 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4993 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
4995 * If an architecture guarantees that all ranges registered contain no holes
4996 * and may be freed, this this function may be used instead of calling
4997 * memblock_free_early_nid() manually.
4999 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
5001 unsigned long start_pfn
, end_pfn
;
5004 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
5005 start_pfn
= min(start_pfn
, max_low_pfn
);
5006 end_pfn
= min(end_pfn
, max_low_pfn
);
5008 if (start_pfn
< end_pfn
)
5009 memblock_free_early_nid(PFN_PHYS(start_pfn
),
5010 (end_pfn
- start_pfn
) << PAGE_SHIFT
,
5016 * sparse_memory_present_with_active_regions - Call memory_present for each active range
5017 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
5019 * If an architecture guarantees that all ranges registered contain no holes and may
5020 * be freed, this function may be used instead of calling memory_present() manually.
5022 void __init
sparse_memory_present_with_active_regions(int nid
)
5024 unsigned long start_pfn
, end_pfn
;
5027 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
5028 memory_present(this_nid
, start_pfn
, end_pfn
);
5032 * get_pfn_range_for_nid - Return the start and end page frames for a node
5033 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
5034 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
5035 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
5037 * It returns the start and end page frame of a node based on information
5038 * provided by memblock_set_node(). If called for a node
5039 * with no available memory, a warning is printed and the start and end
5042 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
5043 unsigned long *start_pfn
, unsigned long *end_pfn
)
5045 unsigned long this_start_pfn
, this_end_pfn
;
5051 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
5052 *start_pfn
= min(*start_pfn
, this_start_pfn
);
5053 *end_pfn
= max(*end_pfn
, this_end_pfn
);
5056 if (*start_pfn
== -1UL)
5061 * This finds a zone that can be used for ZONE_MOVABLE pages. The
5062 * assumption is made that zones within a node are ordered in monotonic
5063 * increasing memory addresses so that the "highest" populated zone is used
5065 static void __init
find_usable_zone_for_movable(void)
5068 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
5069 if (zone_index
== ZONE_MOVABLE
)
5072 if (arch_zone_highest_possible_pfn
[zone_index
] >
5073 arch_zone_lowest_possible_pfn
[zone_index
])
5077 VM_BUG_ON(zone_index
== -1);
5078 movable_zone
= zone_index
;
5082 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
5083 * because it is sized independent of architecture. Unlike the other zones,
5084 * the starting point for ZONE_MOVABLE is not fixed. It may be different
5085 * in each node depending on the size of each node and how evenly kernelcore
5086 * is distributed. This helper function adjusts the zone ranges
5087 * provided by the architecture for a given node by using the end of the
5088 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5089 * zones within a node are in order of monotonic increases memory addresses
5091 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
5092 unsigned long zone_type
,
5093 unsigned long node_start_pfn
,
5094 unsigned long node_end_pfn
,
5095 unsigned long *zone_start_pfn
,
5096 unsigned long *zone_end_pfn
)
5098 /* Only adjust if ZONE_MOVABLE is on this node */
5099 if (zone_movable_pfn
[nid
]) {
5100 /* Size ZONE_MOVABLE */
5101 if (zone_type
== ZONE_MOVABLE
) {
5102 *zone_start_pfn
= zone_movable_pfn
[nid
];
5103 *zone_end_pfn
= min(node_end_pfn
,
5104 arch_zone_highest_possible_pfn
[movable_zone
]);
5106 /* Check if this whole range is within ZONE_MOVABLE */
5107 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
5108 *zone_start_pfn
= *zone_end_pfn
;
5113 * Return the number of pages a zone spans in a node, including holes
5114 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5116 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5117 unsigned long zone_type
,
5118 unsigned long node_start_pfn
,
5119 unsigned long node_end_pfn
,
5120 unsigned long *zone_start_pfn
,
5121 unsigned long *zone_end_pfn
,
5122 unsigned long *ignored
)
5124 /* When hotadd a new node from cpu_up(), the node should be empty */
5125 if (!node_start_pfn
&& !node_end_pfn
)
5128 /* Get the start and end of the zone */
5129 *zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
5130 *zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
5131 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5132 node_start_pfn
, node_end_pfn
,
5133 zone_start_pfn
, zone_end_pfn
);
5135 /* Check that this node has pages within the zone's required range */
5136 if (*zone_end_pfn
< node_start_pfn
|| *zone_start_pfn
> node_end_pfn
)
5139 /* Move the zone boundaries inside the node if necessary */
5140 *zone_end_pfn
= min(*zone_end_pfn
, node_end_pfn
);
5141 *zone_start_pfn
= max(*zone_start_pfn
, node_start_pfn
);
5143 /* Return the spanned pages */
5144 return *zone_end_pfn
- *zone_start_pfn
;
5148 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
5149 * then all holes in the requested range will be accounted for.
5151 unsigned long __meminit
__absent_pages_in_range(int nid
,
5152 unsigned long range_start_pfn
,
5153 unsigned long range_end_pfn
)
5155 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
5156 unsigned long start_pfn
, end_pfn
;
5159 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5160 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
5161 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
5162 nr_absent
-= end_pfn
- start_pfn
;
5168 * absent_pages_in_range - Return number of page frames in holes within a range
5169 * @start_pfn: The start PFN to start searching for holes
5170 * @end_pfn: The end PFN to stop searching for holes
5172 * It returns the number of pages frames in memory holes within a range.
5174 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
5175 unsigned long end_pfn
)
5177 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
5180 /* Return the number of page frames in holes in a zone on a node */
5181 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5182 unsigned long zone_type
,
5183 unsigned long node_start_pfn
,
5184 unsigned long node_end_pfn
,
5185 unsigned long *ignored
)
5187 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
5188 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
5189 unsigned long zone_start_pfn
, zone_end_pfn
;
5190 unsigned long nr_absent
;
5192 /* When hotadd a new node from cpu_up(), the node should be empty */
5193 if (!node_start_pfn
&& !node_end_pfn
)
5196 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
5197 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
5199 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5200 node_start_pfn
, node_end_pfn
,
5201 &zone_start_pfn
, &zone_end_pfn
);
5202 nr_absent
= __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
5205 * ZONE_MOVABLE handling.
5206 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
5209 if (zone_movable_pfn
[nid
]) {
5210 if (mirrored_kernelcore
) {
5211 unsigned long start_pfn
, end_pfn
;
5212 struct memblock_region
*r
;
5214 for_each_memblock(memory
, r
) {
5215 start_pfn
= clamp(memblock_region_memory_base_pfn(r
),
5216 zone_start_pfn
, zone_end_pfn
);
5217 end_pfn
= clamp(memblock_region_memory_end_pfn(r
),
5218 zone_start_pfn
, zone_end_pfn
);
5220 if (zone_type
== ZONE_MOVABLE
&&
5221 memblock_is_mirror(r
))
5222 nr_absent
+= end_pfn
- start_pfn
;
5224 if (zone_type
== ZONE_NORMAL
&&
5225 !memblock_is_mirror(r
))
5226 nr_absent
+= end_pfn
- start_pfn
;
5229 if (zone_type
== ZONE_NORMAL
)
5230 nr_absent
+= node_end_pfn
- zone_movable_pfn
[nid
];
5237 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5238 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5239 unsigned long zone_type
,
5240 unsigned long node_start_pfn
,
5241 unsigned long node_end_pfn
,
5242 unsigned long *zone_start_pfn
,
5243 unsigned long *zone_end_pfn
,
5244 unsigned long *zones_size
)
5248 *zone_start_pfn
= node_start_pfn
;
5249 for (zone
= 0; zone
< zone_type
; zone
++)
5250 *zone_start_pfn
+= zones_size
[zone
];
5252 *zone_end_pfn
= *zone_start_pfn
+ zones_size
[zone_type
];
5254 return zones_size
[zone_type
];
5257 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5258 unsigned long zone_type
,
5259 unsigned long node_start_pfn
,
5260 unsigned long node_end_pfn
,
5261 unsigned long *zholes_size
)
5266 return zholes_size
[zone_type
];
5269 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5271 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
5272 unsigned long node_start_pfn
,
5273 unsigned long node_end_pfn
,
5274 unsigned long *zones_size
,
5275 unsigned long *zholes_size
)
5277 unsigned long realtotalpages
= 0, totalpages
= 0;
5280 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5281 struct zone
*zone
= pgdat
->node_zones
+ i
;
5282 unsigned long zone_start_pfn
, zone_end_pfn
;
5283 unsigned long size
, real_size
;
5285 size
= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
5291 real_size
= size
- zone_absent_pages_in_node(pgdat
->node_id
, i
,
5292 node_start_pfn
, node_end_pfn
,
5295 zone
->zone_start_pfn
= zone_start_pfn
;
5297 zone
->zone_start_pfn
= 0;
5298 zone
->spanned_pages
= size
;
5299 zone
->present_pages
= real_size
;
5302 realtotalpages
+= real_size
;
5305 pgdat
->node_spanned_pages
= totalpages
;
5306 pgdat
->node_present_pages
= realtotalpages
;
5307 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
5311 #ifndef CONFIG_SPARSEMEM
5313 * Calculate the size of the zone->blockflags rounded to an unsigned long
5314 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5315 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
5316 * round what is now in bits to nearest long in bits, then return it in
5319 static unsigned long __init
usemap_size(unsigned long zone_start_pfn
, unsigned long zonesize
)
5321 unsigned long usemapsize
;
5323 zonesize
+= zone_start_pfn
& (pageblock_nr_pages
-1);
5324 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
5325 usemapsize
= usemapsize
>> pageblock_order
;
5326 usemapsize
*= NR_PAGEBLOCK_BITS
;
5327 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
5329 return usemapsize
/ 8;
5332 static void __init
setup_usemap(struct pglist_data
*pgdat
,
5334 unsigned long zone_start_pfn
,
5335 unsigned long zonesize
)
5337 unsigned long usemapsize
= usemap_size(zone_start_pfn
, zonesize
);
5338 zone
->pageblock_flags
= NULL
;
5340 zone
->pageblock_flags
=
5341 memblock_virt_alloc_node_nopanic(usemapsize
,
5345 static inline void setup_usemap(struct pglist_data
*pgdat
, struct zone
*zone
,
5346 unsigned long zone_start_pfn
, unsigned long zonesize
) {}
5347 #endif /* CONFIG_SPARSEMEM */
5349 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
5351 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
5352 void __paginginit
set_pageblock_order(void)
5356 /* Check that pageblock_nr_pages has not already been setup */
5357 if (pageblock_order
)
5360 if (HPAGE_SHIFT
> PAGE_SHIFT
)
5361 order
= HUGETLB_PAGE_ORDER
;
5363 order
= MAX_ORDER
- 1;
5366 * Assume the largest contiguous order of interest is a huge page.
5367 * This value may be variable depending on boot parameters on IA64 and
5370 pageblock_order
= order
;
5372 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5375 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
5376 * is unused as pageblock_order is set at compile-time. See
5377 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5380 void __paginginit
set_pageblock_order(void)
5384 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5386 static unsigned long __paginginit
calc_memmap_size(unsigned long spanned_pages
,
5387 unsigned long present_pages
)
5389 unsigned long pages
= spanned_pages
;
5392 * Provide a more accurate estimation if there are holes within
5393 * the zone and SPARSEMEM is in use. If there are holes within the
5394 * zone, each populated memory region may cost us one or two extra
5395 * memmap pages due to alignment because memmap pages for each
5396 * populated regions may not naturally algined on page boundary.
5397 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5399 if (spanned_pages
> present_pages
+ (present_pages
>> 4) &&
5400 IS_ENABLED(CONFIG_SPARSEMEM
))
5401 pages
= present_pages
;
5403 return PAGE_ALIGN(pages
* sizeof(struct page
)) >> PAGE_SHIFT
;
5407 * Set up the zone data structures:
5408 * - mark all pages reserved
5409 * - mark all memory queues empty
5410 * - clear the memory bitmaps
5412 * NOTE: pgdat should get zeroed by caller.
5414 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
)
5417 int nid
= pgdat
->node_id
;
5420 pgdat_resize_init(pgdat
);
5421 #ifdef CONFIG_NUMA_BALANCING
5422 spin_lock_init(&pgdat
->numabalancing_migrate_lock
);
5423 pgdat
->numabalancing_migrate_nr_pages
= 0;
5424 pgdat
->numabalancing_migrate_next_window
= jiffies
;
5426 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5427 spin_lock_init(&pgdat
->split_queue_lock
);
5428 INIT_LIST_HEAD(&pgdat
->split_queue
);
5429 pgdat
->split_queue_len
= 0;
5431 init_waitqueue_head(&pgdat
->kswapd_wait
);
5432 init_waitqueue_head(&pgdat
->pfmemalloc_wait
);
5433 #ifdef CONFIG_COMPACTION
5434 init_waitqueue_head(&pgdat
->kcompactd_wait
);
5436 pgdat_page_ext_init(pgdat
);
5438 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5439 struct zone
*zone
= pgdat
->node_zones
+ j
;
5440 unsigned long size
, realsize
, freesize
, memmap_pages
;
5441 unsigned long zone_start_pfn
= zone
->zone_start_pfn
;
5443 size
= zone
->spanned_pages
;
5444 realsize
= freesize
= zone
->present_pages
;
5447 * Adjust freesize so that it accounts for how much memory
5448 * is used by this zone for memmap. This affects the watermark
5449 * and per-cpu initialisations
5451 memmap_pages
= calc_memmap_size(size
, realsize
);
5452 if (!is_highmem_idx(j
)) {
5453 if (freesize
>= memmap_pages
) {
5454 freesize
-= memmap_pages
;
5457 " %s zone: %lu pages used for memmap\n",
5458 zone_names
[j
], memmap_pages
);
5460 pr_warn(" %s zone: %lu pages exceeds freesize %lu\n",
5461 zone_names
[j
], memmap_pages
, freesize
);
5464 /* Account for reserved pages */
5465 if (j
== 0 && freesize
> dma_reserve
) {
5466 freesize
-= dma_reserve
;
5467 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
5468 zone_names
[0], dma_reserve
);
5471 if (!is_highmem_idx(j
))
5472 nr_kernel_pages
+= freesize
;
5473 /* Charge for highmem memmap if there are enough kernel pages */
5474 else if (nr_kernel_pages
> memmap_pages
* 2)
5475 nr_kernel_pages
-= memmap_pages
;
5476 nr_all_pages
+= freesize
;
5479 * Set an approximate value for lowmem here, it will be adjusted
5480 * when the bootmem allocator frees pages into the buddy system.
5481 * And all highmem pages will be managed by the buddy system.
5483 zone
->managed_pages
= is_highmem_idx(j
) ? realsize
: freesize
;
5486 zone
->min_unmapped_pages
= (freesize
*sysctl_min_unmapped_ratio
)
5488 zone
->min_slab_pages
= (freesize
* sysctl_min_slab_ratio
) / 100;
5490 zone
->name
= zone_names
[j
];
5491 spin_lock_init(&zone
->lock
);
5492 spin_lock_init(&zone
->lru_lock
);
5493 zone_seqlock_init(zone
);
5494 zone
->zone_pgdat
= pgdat
;
5495 zone_pcp_init(zone
);
5497 /* For bootup, initialized properly in watermark setup */
5498 mod_zone_page_state(zone
, NR_ALLOC_BATCH
, zone
->managed_pages
);
5500 lruvec_init(&zone
->lruvec
);
5504 set_pageblock_order();
5505 setup_usemap(pgdat
, zone
, zone_start_pfn
, size
);
5506 ret
= init_currently_empty_zone(zone
, zone_start_pfn
, size
);
5508 memmap_init(size
, nid
, j
, zone_start_pfn
);
5512 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
5514 unsigned long __maybe_unused start
= 0;
5515 unsigned long __maybe_unused offset
= 0;
5517 /* Skip empty nodes */
5518 if (!pgdat
->node_spanned_pages
)
5521 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5522 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
5523 offset
= pgdat
->node_start_pfn
- start
;
5524 /* ia64 gets its own node_mem_map, before this, without bootmem */
5525 if (!pgdat
->node_mem_map
) {
5526 unsigned long size
, end
;
5530 * The zone's endpoints aren't required to be MAX_ORDER
5531 * aligned but the node_mem_map endpoints must be in order
5532 * for the buddy allocator to function correctly.
5534 end
= pgdat_end_pfn(pgdat
);
5535 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
5536 size
= (end
- start
) * sizeof(struct page
);
5537 map
= alloc_remap(pgdat
->node_id
, size
);
5539 map
= memblock_virt_alloc_node_nopanic(size
,
5541 pgdat
->node_mem_map
= map
+ offset
;
5543 #ifndef CONFIG_NEED_MULTIPLE_NODES
5545 * With no DISCONTIG, the global mem_map is just set as node 0's
5547 if (pgdat
== NODE_DATA(0)) {
5548 mem_map
= NODE_DATA(0)->node_mem_map
;
5549 #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
5550 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
5552 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5555 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
5558 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
5559 unsigned long node_start_pfn
, unsigned long *zholes_size
)
5561 pg_data_t
*pgdat
= NODE_DATA(nid
);
5562 unsigned long start_pfn
= 0;
5563 unsigned long end_pfn
= 0;
5565 /* pg_data_t should be reset to zero when it's allocated */
5566 WARN_ON(pgdat
->nr_zones
|| pgdat
->classzone_idx
);
5568 reset_deferred_meminit(pgdat
);
5569 pgdat
->node_id
= nid
;
5570 pgdat
->node_start_pfn
= node_start_pfn
;
5571 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5572 get_pfn_range_for_nid(nid
, &start_pfn
, &end_pfn
);
5573 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid
,
5574 (u64
)start_pfn
<< PAGE_SHIFT
,
5575 end_pfn
? ((u64
)end_pfn
<< PAGE_SHIFT
) - 1 : 0);
5577 start_pfn
= node_start_pfn
;
5579 calculate_node_totalpages(pgdat
, start_pfn
, end_pfn
,
5580 zones_size
, zholes_size
);
5582 alloc_node_mem_map(pgdat
);
5583 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5584 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5585 nid
, (unsigned long)pgdat
,
5586 (unsigned long)pgdat
->node_mem_map
);
5589 free_area_init_core(pgdat
);
5592 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5594 #if MAX_NUMNODES > 1
5596 * Figure out the number of possible node ids.
5598 void __init
setup_nr_node_ids(void)
5600 unsigned int highest
;
5602 highest
= find_last_bit(node_possible_map
.bits
, MAX_NUMNODES
);
5603 nr_node_ids
= highest
+ 1;
5608 * node_map_pfn_alignment - determine the maximum internode alignment
5610 * This function should be called after node map is populated and sorted.
5611 * It calculates the maximum power of two alignment which can distinguish
5614 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5615 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
5616 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
5617 * shifted, 1GiB is enough and this function will indicate so.
5619 * This is used to test whether pfn -> nid mapping of the chosen memory
5620 * model has fine enough granularity to avoid incorrect mapping for the
5621 * populated node map.
5623 * Returns the determined alignment in pfn's. 0 if there is no alignment
5624 * requirement (single node).
5626 unsigned long __init
node_map_pfn_alignment(void)
5628 unsigned long accl_mask
= 0, last_end
= 0;
5629 unsigned long start
, end
, mask
;
5633 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
5634 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
5641 * Start with a mask granular enough to pin-point to the
5642 * start pfn and tick off bits one-by-one until it becomes
5643 * too coarse to separate the current node from the last.
5645 mask
= ~((1 << __ffs(start
)) - 1);
5646 while (mask
&& last_end
<= (start
& (mask
<< 1)))
5649 /* accumulate all internode masks */
5653 /* convert mask to number of pages */
5654 return ~accl_mask
+ 1;
5657 /* Find the lowest pfn for a node */
5658 static unsigned long __init
find_min_pfn_for_node(int nid
)
5660 unsigned long min_pfn
= ULONG_MAX
;
5661 unsigned long start_pfn
;
5664 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
5665 min_pfn
= min(min_pfn
, start_pfn
);
5667 if (min_pfn
== ULONG_MAX
) {
5668 pr_warn("Could not find start_pfn for node %d\n", nid
);
5676 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5678 * It returns the minimum PFN based on information provided via
5679 * memblock_set_node().
5681 unsigned long __init
find_min_pfn_with_active_regions(void)
5683 return find_min_pfn_for_node(MAX_NUMNODES
);
5687 * early_calculate_totalpages()
5688 * Sum pages in active regions for movable zone.
5689 * Populate N_MEMORY for calculating usable_nodes.
5691 static unsigned long __init
early_calculate_totalpages(void)
5693 unsigned long totalpages
= 0;
5694 unsigned long start_pfn
, end_pfn
;
5697 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
5698 unsigned long pages
= end_pfn
- start_pfn
;
5700 totalpages
+= pages
;
5702 node_set_state(nid
, N_MEMORY
);
5708 * Find the PFN the Movable zone begins in each node. Kernel memory
5709 * is spread evenly between nodes as long as the nodes have enough
5710 * memory. When they don't, some nodes will have more kernelcore than
5713 static void __init
find_zone_movable_pfns_for_nodes(void)
5716 unsigned long usable_startpfn
;
5717 unsigned long kernelcore_node
, kernelcore_remaining
;
5718 /* save the state before borrow the nodemask */
5719 nodemask_t saved_node_state
= node_states
[N_MEMORY
];
5720 unsigned long totalpages
= early_calculate_totalpages();
5721 int usable_nodes
= nodes_weight(node_states
[N_MEMORY
]);
5722 struct memblock_region
*r
;
5724 /* Need to find movable_zone earlier when movable_node is specified. */
5725 find_usable_zone_for_movable();
5728 * If movable_node is specified, ignore kernelcore and movablecore
5731 if (movable_node_is_enabled()) {
5732 for_each_memblock(memory
, r
) {
5733 if (!memblock_is_hotpluggable(r
))
5738 usable_startpfn
= PFN_DOWN(r
->base
);
5739 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
5740 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
5748 * If kernelcore=mirror is specified, ignore movablecore option
5750 if (mirrored_kernelcore
) {
5751 bool mem_below_4gb_not_mirrored
= false;
5753 for_each_memblock(memory
, r
) {
5754 if (memblock_is_mirror(r
))
5759 usable_startpfn
= memblock_region_memory_base_pfn(r
);
5761 if (usable_startpfn
< 0x100000) {
5762 mem_below_4gb_not_mirrored
= true;
5766 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
5767 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
5771 if (mem_below_4gb_not_mirrored
)
5772 pr_warn("This configuration results in unmirrored kernel memory.");
5778 * If movablecore=nn[KMG] was specified, calculate what size of
5779 * kernelcore that corresponds so that memory usable for
5780 * any allocation type is evenly spread. If both kernelcore
5781 * and movablecore are specified, then the value of kernelcore
5782 * will be used for required_kernelcore if it's greater than
5783 * what movablecore would have allowed.
5785 if (required_movablecore
) {
5786 unsigned long corepages
;
5789 * Round-up so that ZONE_MOVABLE is at least as large as what
5790 * was requested by the user
5792 required_movablecore
=
5793 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
5794 required_movablecore
= min(totalpages
, required_movablecore
);
5795 corepages
= totalpages
- required_movablecore
;
5797 required_kernelcore
= max(required_kernelcore
, corepages
);
5801 * If kernelcore was not specified or kernelcore size is larger
5802 * than totalpages, there is no ZONE_MOVABLE.
5804 if (!required_kernelcore
|| required_kernelcore
>= totalpages
)
5807 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5808 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
5811 /* Spread kernelcore memory as evenly as possible throughout nodes */
5812 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5813 for_each_node_state(nid
, N_MEMORY
) {
5814 unsigned long start_pfn
, end_pfn
;
5817 * Recalculate kernelcore_node if the division per node
5818 * now exceeds what is necessary to satisfy the requested
5819 * amount of memory for the kernel
5821 if (required_kernelcore
< kernelcore_node
)
5822 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5825 * As the map is walked, we track how much memory is usable
5826 * by the kernel using kernelcore_remaining. When it is
5827 * 0, the rest of the node is usable by ZONE_MOVABLE
5829 kernelcore_remaining
= kernelcore_node
;
5831 /* Go through each range of PFNs within this node */
5832 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5833 unsigned long size_pages
;
5835 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
5836 if (start_pfn
>= end_pfn
)
5839 /* Account for what is only usable for kernelcore */
5840 if (start_pfn
< usable_startpfn
) {
5841 unsigned long kernel_pages
;
5842 kernel_pages
= min(end_pfn
, usable_startpfn
)
5845 kernelcore_remaining
-= min(kernel_pages
,
5846 kernelcore_remaining
);
5847 required_kernelcore
-= min(kernel_pages
,
5848 required_kernelcore
);
5850 /* Continue if range is now fully accounted */
5851 if (end_pfn
<= usable_startpfn
) {
5854 * Push zone_movable_pfn to the end so
5855 * that if we have to rebalance
5856 * kernelcore across nodes, we will
5857 * not double account here
5859 zone_movable_pfn
[nid
] = end_pfn
;
5862 start_pfn
= usable_startpfn
;
5866 * The usable PFN range for ZONE_MOVABLE is from
5867 * start_pfn->end_pfn. Calculate size_pages as the
5868 * number of pages used as kernelcore
5870 size_pages
= end_pfn
- start_pfn
;
5871 if (size_pages
> kernelcore_remaining
)
5872 size_pages
= kernelcore_remaining
;
5873 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
5876 * Some kernelcore has been met, update counts and
5877 * break if the kernelcore for this node has been
5880 required_kernelcore
-= min(required_kernelcore
,
5882 kernelcore_remaining
-= size_pages
;
5883 if (!kernelcore_remaining
)
5889 * If there is still required_kernelcore, we do another pass with one
5890 * less node in the count. This will push zone_movable_pfn[nid] further
5891 * along on the nodes that still have memory until kernelcore is
5895 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
5899 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5900 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
5901 zone_movable_pfn
[nid
] =
5902 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
5905 /* restore the node_state */
5906 node_states
[N_MEMORY
] = saved_node_state
;
5909 /* Any regular or high memory on that node ? */
5910 static void check_for_memory(pg_data_t
*pgdat
, int nid
)
5912 enum zone_type zone_type
;
5914 if (N_MEMORY
== N_NORMAL_MEMORY
)
5917 for (zone_type
= 0; zone_type
<= ZONE_MOVABLE
- 1; zone_type
++) {
5918 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
5919 if (populated_zone(zone
)) {
5920 node_set_state(nid
, N_HIGH_MEMORY
);
5921 if (N_NORMAL_MEMORY
!= N_HIGH_MEMORY
&&
5922 zone_type
<= ZONE_NORMAL
)
5923 node_set_state(nid
, N_NORMAL_MEMORY
);
5930 * free_area_init_nodes - Initialise all pg_data_t and zone data
5931 * @max_zone_pfn: an array of max PFNs for each zone
5933 * This will call free_area_init_node() for each active node in the system.
5934 * Using the page ranges provided by memblock_set_node(), the size of each
5935 * zone in each node and their holes is calculated. If the maximum PFN
5936 * between two adjacent zones match, it is assumed that the zone is empty.
5937 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5938 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5939 * starts where the previous one ended. For example, ZONE_DMA32 starts
5940 * at arch_max_dma_pfn.
5942 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
5944 unsigned long start_pfn
, end_pfn
;
5947 /* Record where the zone boundaries are */
5948 memset(arch_zone_lowest_possible_pfn
, 0,
5949 sizeof(arch_zone_lowest_possible_pfn
));
5950 memset(arch_zone_highest_possible_pfn
, 0,
5951 sizeof(arch_zone_highest_possible_pfn
));
5952 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
5953 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
5954 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
5955 if (i
== ZONE_MOVABLE
)
5957 arch_zone_lowest_possible_pfn
[i
] =
5958 arch_zone_highest_possible_pfn
[i
-1];
5959 arch_zone_highest_possible_pfn
[i
] =
5960 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
5962 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
5963 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
5965 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5966 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
5967 find_zone_movable_pfns_for_nodes();
5969 /* Print out the zone ranges */
5970 pr_info("Zone ranges:\n");
5971 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5972 if (i
== ZONE_MOVABLE
)
5974 pr_info(" %-8s ", zone_names
[i
]);
5975 if (arch_zone_lowest_possible_pfn
[i
] ==
5976 arch_zone_highest_possible_pfn
[i
])
5979 pr_cont("[mem %#018Lx-%#018Lx]\n",
5980 (u64
)arch_zone_lowest_possible_pfn
[i
]
5982 ((u64
)arch_zone_highest_possible_pfn
[i
]
5983 << PAGE_SHIFT
) - 1);
5986 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5987 pr_info("Movable zone start for each node\n");
5988 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
5989 if (zone_movable_pfn
[i
])
5990 pr_info(" Node %d: %#018Lx\n", i
,
5991 (u64
)zone_movable_pfn
[i
] << PAGE_SHIFT
);
5994 /* Print out the early node map */
5995 pr_info("Early memory node ranges\n");
5996 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
5997 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid
,
5998 (u64
)start_pfn
<< PAGE_SHIFT
,
5999 ((u64
)end_pfn
<< PAGE_SHIFT
) - 1);
6001 /* Initialise every node */
6002 mminit_verify_pageflags_layout();
6003 setup_nr_node_ids();
6004 for_each_online_node(nid
) {
6005 pg_data_t
*pgdat
= NODE_DATA(nid
);
6006 free_area_init_node(nid
, NULL
,
6007 find_min_pfn_for_node(nid
), NULL
);
6009 /* Any memory on that node */
6010 if (pgdat
->node_present_pages
)
6011 node_set_state(nid
, N_MEMORY
);
6012 check_for_memory(pgdat
, nid
);
6016 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
6018 unsigned long long coremem
;
6022 coremem
= memparse(p
, &p
);
6023 *core
= coremem
>> PAGE_SHIFT
;
6025 /* Paranoid check that UL is enough for the coremem value */
6026 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
6032 * kernelcore=size sets the amount of memory for use for allocations that
6033 * cannot be reclaimed or migrated.
6035 static int __init
cmdline_parse_kernelcore(char *p
)
6037 /* parse kernelcore=mirror */
6038 if (parse_option_str(p
, "mirror")) {
6039 mirrored_kernelcore
= true;
6043 return cmdline_parse_core(p
, &required_kernelcore
);
6047 * movablecore=size sets the amount of memory for use for allocations that
6048 * can be reclaimed or migrated.
6050 static int __init
cmdline_parse_movablecore(char *p
)
6052 return cmdline_parse_core(p
, &required_movablecore
);
6055 early_param("kernelcore", cmdline_parse_kernelcore
);
6056 early_param("movablecore", cmdline_parse_movablecore
);
6058 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6060 void adjust_managed_page_count(struct page
*page
, long count
)
6062 spin_lock(&managed_page_count_lock
);
6063 page_zone(page
)->managed_pages
+= count
;
6064 totalram_pages
+= count
;
6065 #ifdef CONFIG_HIGHMEM
6066 if (PageHighMem(page
))
6067 totalhigh_pages
+= count
;
6069 spin_unlock(&managed_page_count_lock
);
6071 EXPORT_SYMBOL(adjust_managed_page_count
);
6073 unsigned long free_reserved_area(void *start
, void *end
, int poison
, char *s
)
6076 unsigned long pages
= 0;
6078 start
= (void *)PAGE_ALIGN((unsigned long)start
);
6079 end
= (void *)((unsigned long)end
& PAGE_MASK
);
6080 for (pos
= start
; pos
< end
; pos
+= PAGE_SIZE
, pages
++) {
6081 if ((unsigned int)poison
<= 0xFF)
6082 memset(pos
, poison
, PAGE_SIZE
);
6083 free_reserved_page(virt_to_page(pos
));
6087 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
6088 s
, pages
<< (PAGE_SHIFT
- 10), start
, end
);
6092 EXPORT_SYMBOL(free_reserved_area
);
6094 #ifdef CONFIG_HIGHMEM
6095 void free_highmem_page(struct page
*page
)
6097 __free_reserved_page(page
);
6099 page_zone(page
)->managed_pages
++;
6105 void __init
mem_init_print_info(const char *str
)
6107 unsigned long physpages
, codesize
, datasize
, rosize
, bss_size
;
6108 unsigned long init_code_size
, init_data_size
;
6110 physpages
= get_num_physpages();
6111 codesize
= _etext
- _stext
;
6112 datasize
= _edata
- _sdata
;
6113 rosize
= __end_rodata
- __start_rodata
;
6114 bss_size
= __bss_stop
- __bss_start
;
6115 init_data_size
= __init_end
- __init_begin
;
6116 init_code_size
= _einittext
- _sinittext
;
6119 * Detect special cases and adjust section sizes accordingly:
6120 * 1) .init.* may be embedded into .data sections
6121 * 2) .init.text.* may be out of [__init_begin, __init_end],
6122 * please refer to arch/tile/kernel/vmlinux.lds.S.
6123 * 3) .rodata.* may be embedded into .text or .data sections.
6125 #define adj_init_size(start, end, size, pos, adj) \
6127 if (start <= pos && pos < end && size > adj) \
6131 adj_init_size(__init_begin
, __init_end
, init_data_size
,
6132 _sinittext
, init_code_size
);
6133 adj_init_size(_stext
, _etext
, codesize
, _sinittext
, init_code_size
);
6134 adj_init_size(_sdata
, _edata
, datasize
, __init_begin
, init_data_size
);
6135 adj_init_size(_stext
, _etext
, codesize
, __start_rodata
, rosize
);
6136 adj_init_size(_sdata
, _edata
, datasize
, __start_rodata
, rosize
);
6138 #undef adj_init_size
6140 pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
6141 #ifdef CONFIG_HIGHMEM
6145 nr_free_pages() << (PAGE_SHIFT
- 10),
6146 physpages
<< (PAGE_SHIFT
- 10),
6147 codesize
>> 10, datasize
>> 10, rosize
>> 10,
6148 (init_data_size
+ init_code_size
) >> 10, bss_size
>> 10,
6149 (physpages
- totalram_pages
- totalcma_pages
) << (PAGE_SHIFT
- 10),
6150 totalcma_pages
<< (PAGE_SHIFT
- 10),
6151 #ifdef CONFIG_HIGHMEM
6152 totalhigh_pages
<< (PAGE_SHIFT
- 10),
6154 str
? ", " : "", str
? str
: "");
6158 * set_dma_reserve - set the specified number of pages reserved in the first zone
6159 * @new_dma_reserve: The number of pages to mark reserved
6161 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
6162 * In the DMA zone, a significant percentage may be consumed by kernel image
6163 * and other unfreeable allocations which can skew the watermarks badly. This
6164 * function may optionally be used to account for unfreeable pages in the
6165 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6166 * smaller per-cpu batchsize.
6168 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
6170 dma_reserve
= new_dma_reserve
;
6173 void __init
free_area_init(unsigned long *zones_size
)
6175 free_area_init_node(0, zones_size
,
6176 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
6179 static int page_alloc_cpu_notify(struct notifier_block
*self
,
6180 unsigned long action
, void *hcpu
)
6182 int cpu
= (unsigned long)hcpu
;
6184 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
6185 lru_add_drain_cpu(cpu
);
6189 * Spill the event counters of the dead processor
6190 * into the current processors event counters.
6191 * This artificially elevates the count of the current
6194 vm_events_fold_cpu(cpu
);
6197 * Zero the differential counters of the dead processor
6198 * so that the vm statistics are consistent.
6200 * This is only okay since the processor is dead and cannot
6201 * race with what we are doing.
6203 cpu_vm_stats_fold(cpu
);
6208 void __init
page_alloc_init(void)
6210 hotcpu_notifier(page_alloc_cpu_notify
, 0);
6214 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
6215 * or min_free_kbytes changes.
6217 static void calculate_totalreserve_pages(void)
6219 struct pglist_data
*pgdat
;
6220 unsigned long reserve_pages
= 0;
6221 enum zone_type i
, j
;
6223 for_each_online_pgdat(pgdat
) {
6224 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
6225 struct zone
*zone
= pgdat
->node_zones
+ i
;
6228 /* Find valid and maximum lowmem_reserve in the zone */
6229 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
6230 if (zone
->lowmem_reserve
[j
] > max
)
6231 max
= zone
->lowmem_reserve
[j
];
6234 /* we treat the high watermark as reserved pages. */
6235 max
+= high_wmark_pages(zone
);
6237 if (max
> zone
->managed_pages
)
6238 max
= zone
->managed_pages
;
6240 zone
->totalreserve_pages
= max
;
6242 reserve_pages
+= max
;
6245 totalreserve_pages
= reserve_pages
;
6249 * setup_per_zone_lowmem_reserve - called whenever
6250 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
6251 * has a correct pages reserved value, so an adequate number of
6252 * pages are left in the zone after a successful __alloc_pages().
6254 static void setup_per_zone_lowmem_reserve(void)
6256 struct pglist_data
*pgdat
;
6257 enum zone_type j
, idx
;
6259 for_each_online_pgdat(pgdat
) {
6260 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
6261 struct zone
*zone
= pgdat
->node_zones
+ j
;
6262 unsigned long managed_pages
= zone
->managed_pages
;
6264 zone
->lowmem_reserve
[j
] = 0;
6268 struct zone
*lower_zone
;
6272 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
6273 sysctl_lowmem_reserve_ratio
[idx
] = 1;
6275 lower_zone
= pgdat
->node_zones
+ idx
;
6276 lower_zone
->lowmem_reserve
[j
] = managed_pages
/
6277 sysctl_lowmem_reserve_ratio
[idx
];
6278 managed_pages
+= lower_zone
->managed_pages
;
6283 /* update totalreserve_pages */
6284 calculate_totalreserve_pages();
6287 static void __setup_per_zone_wmarks(void)
6289 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
6290 unsigned long lowmem_pages
= 0;
6292 unsigned long flags
;
6294 /* Calculate total number of !ZONE_HIGHMEM pages */
6295 for_each_zone(zone
) {
6296 if (!is_highmem(zone
))
6297 lowmem_pages
+= zone
->managed_pages
;
6300 for_each_zone(zone
) {
6303 spin_lock_irqsave(&zone
->lock
, flags
);
6304 tmp
= (u64
)pages_min
* zone
->managed_pages
;
6305 do_div(tmp
, lowmem_pages
);
6306 if (is_highmem(zone
)) {
6308 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6309 * need highmem pages, so cap pages_min to a small
6312 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
6313 * deltas control asynch page reclaim, and so should
6314 * not be capped for highmem.
6316 unsigned long min_pages
;
6318 min_pages
= zone
->managed_pages
/ 1024;
6319 min_pages
= clamp(min_pages
, SWAP_CLUSTER_MAX
, 128UL);
6320 zone
->watermark
[WMARK_MIN
] = min_pages
;
6323 * If it's a lowmem zone, reserve a number of pages
6324 * proportionate to the zone's size.
6326 zone
->watermark
[WMARK_MIN
] = tmp
;
6330 * Set the kswapd watermarks distance according to the
6331 * scale factor in proportion to available memory, but
6332 * ensure a minimum size on small systems.
6334 tmp
= max_t(u64
, tmp
>> 2,
6335 mult_frac(zone
->managed_pages
,
6336 watermark_scale_factor
, 10000));
6338 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + tmp
;
6339 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + tmp
* 2;
6341 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
6342 high_wmark_pages(zone
) - low_wmark_pages(zone
) -
6343 atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]));
6345 spin_unlock_irqrestore(&zone
->lock
, flags
);
6348 /* update totalreserve_pages */
6349 calculate_totalreserve_pages();
6353 * setup_per_zone_wmarks - called when min_free_kbytes changes
6354 * or when memory is hot-{added|removed}
6356 * Ensures that the watermark[min,low,high] values for each zone are set
6357 * correctly with respect to min_free_kbytes.
6359 void setup_per_zone_wmarks(void)
6361 mutex_lock(&zonelists_mutex
);
6362 __setup_per_zone_wmarks();
6363 mutex_unlock(&zonelists_mutex
);
6367 * The inactive anon list should be small enough that the VM never has to
6368 * do too much work, but large enough that each inactive page has a chance
6369 * to be referenced again before it is swapped out.
6371 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
6372 * INACTIVE_ANON pages on this zone's LRU, maintained by the
6373 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
6374 * the anonymous pages are kept on the inactive list.
6377 * memory ratio inactive anon
6378 * -------------------------------------
6387 static void __meminit
calculate_zone_inactive_ratio(struct zone
*zone
)
6389 unsigned int gb
, ratio
;
6391 /* Zone size in gigabytes */
6392 gb
= zone
->managed_pages
>> (30 - PAGE_SHIFT
);
6394 ratio
= int_sqrt(10 * gb
);
6398 zone
->inactive_ratio
= ratio
;
6401 static void __meminit
setup_per_zone_inactive_ratio(void)
6406 calculate_zone_inactive_ratio(zone
);
6410 * Initialise min_free_kbytes.
6412 * For small machines we want it small (128k min). For large machines
6413 * we want it large (64MB max). But it is not linear, because network
6414 * bandwidth does not increase linearly with machine size. We use
6416 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
6417 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6433 int __meminit
init_per_zone_wmark_min(void)
6435 unsigned long lowmem_kbytes
;
6436 int new_min_free_kbytes
;
6438 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
6439 new_min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
6441 if (new_min_free_kbytes
> user_min_free_kbytes
) {
6442 min_free_kbytes
= new_min_free_kbytes
;
6443 if (min_free_kbytes
< 128)
6444 min_free_kbytes
= 128;
6445 if (min_free_kbytes
> 65536)
6446 min_free_kbytes
= 65536;
6448 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6449 new_min_free_kbytes
, user_min_free_kbytes
);
6451 setup_per_zone_wmarks();
6452 refresh_zone_stat_thresholds();
6453 setup_per_zone_lowmem_reserve();
6454 setup_per_zone_inactive_ratio();
6457 module_init(init_per_zone_wmark_min
)
6460 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
6461 * that we can call two helper functions whenever min_free_kbytes
6464 int min_free_kbytes_sysctl_handler(struct ctl_table
*table
, int write
,
6465 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6469 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6474 user_min_free_kbytes
= min_free_kbytes
;
6475 setup_per_zone_wmarks();
6480 int watermark_scale_factor_sysctl_handler(struct ctl_table
*table
, int write
,
6481 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6485 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6490 setup_per_zone_wmarks();
6496 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6497 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6502 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6507 zone
->min_unmapped_pages
= (zone
->managed_pages
*
6508 sysctl_min_unmapped_ratio
) / 100;
6512 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6513 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6518 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6523 zone
->min_slab_pages
= (zone
->managed_pages
*
6524 sysctl_min_slab_ratio
) / 100;
6530 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6531 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6532 * whenever sysctl_lowmem_reserve_ratio changes.
6534 * The reserve ratio obviously has absolutely no relation with the
6535 * minimum watermarks. The lowmem reserve ratio can only make sense
6536 * if in function of the boot time zone sizes.
6538 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6539 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6541 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6542 setup_per_zone_lowmem_reserve();
6547 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
6548 * cpu. It is the fraction of total pages in each zone that a hot per cpu
6549 * pagelist can have before it gets flushed back to buddy allocator.
6551 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table
*table
, int write
,
6552 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6555 int old_percpu_pagelist_fraction
;
6558 mutex_lock(&pcp_batch_high_lock
);
6559 old_percpu_pagelist_fraction
= percpu_pagelist_fraction
;
6561 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6562 if (!write
|| ret
< 0)
6565 /* Sanity checking to avoid pcp imbalance */
6566 if (percpu_pagelist_fraction
&&
6567 percpu_pagelist_fraction
< MIN_PERCPU_PAGELIST_FRACTION
) {
6568 percpu_pagelist_fraction
= old_percpu_pagelist_fraction
;
6574 if (percpu_pagelist_fraction
== old_percpu_pagelist_fraction
)
6577 for_each_populated_zone(zone
) {
6580 for_each_possible_cpu(cpu
)
6581 pageset_set_high_and_batch(zone
,
6582 per_cpu_ptr(zone
->pageset
, cpu
));
6585 mutex_unlock(&pcp_batch_high_lock
);
6590 int hashdist
= HASHDIST_DEFAULT
;
6592 static int __init
set_hashdist(char *str
)
6596 hashdist
= simple_strtoul(str
, &str
, 0);
6599 __setup("hashdist=", set_hashdist
);
6603 * allocate a large system hash table from bootmem
6604 * - it is assumed that the hash table must contain an exact power-of-2
6605 * quantity of entries
6606 * - limit is the number of hash buckets, not the total allocation size
6608 void *__init
alloc_large_system_hash(const char *tablename
,
6609 unsigned long bucketsize
,
6610 unsigned long numentries
,
6613 unsigned int *_hash_shift
,
6614 unsigned int *_hash_mask
,
6615 unsigned long low_limit
,
6616 unsigned long high_limit
)
6618 unsigned long long max
= high_limit
;
6619 unsigned long log2qty
, size
;
6622 /* allow the kernel cmdline to have a say */
6624 /* round applicable memory size up to nearest megabyte */
6625 numentries
= nr_kernel_pages
;
6627 /* It isn't necessary when PAGE_SIZE >= 1MB */
6628 if (PAGE_SHIFT
< 20)
6629 numentries
= round_up(numentries
, (1<<20)/PAGE_SIZE
);
6631 /* limit to 1 bucket per 2^scale bytes of low memory */
6632 if (scale
> PAGE_SHIFT
)
6633 numentries
>>= (scale
- PAGE_SHIFT
);
6635 numentries
<<= (PAGE_SHIFT
- scale
);
6637 /* Make sure we've got at least a 0-order allocation.. */
6638 if (unlikely(flags
& HASH_SMALL
)) {
6639 /* Makes no sense without HASH_EARLY */
6640 WARN_ON(!(flags
& HASH_EARLY
));
6641 if (!(numentries
>> *_hash_shift
)) {
6642 numentries
= 1UL << *_hash_shift
;
6643 BUG_ON(!numentries
);
6645 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
6646 numentries
= PAGE_SIZE
/ bucketsize
;
6648 numentries
= roundup_pow_of_two(numentries
);
6650 /* limit allocation size to 1/16 total memory by default */
6652 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
6653 do_div(max
, bucketsize
);
6655 max
= min(max
, 0x80000000ULL
);
6657 if (numentries
< low_limit
)
6658 numentries
= low_limit
;
6659 if (numentries
> max
)
6662 log2qty
= ilog2(numentries
);
6665 size
= bucketsize
<< log2qty
;
6666 if (flags
& HASH_EARLY
)
6667 table
= memblock_virt_alloc_nopanic(size
, 0);
6669 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
6672 * If bucketsize is not a power-of-two, we may free
6673 * some pages at the end of hash table which
6674 * alloc_pages_exact() automatically does
6676 if (get_order(size
) < MAX_ORDER
) {
6677 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
6678 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
6681 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
6684 panic("Failed to allocate %s hash table\n", tablename
);
6686 pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n",
6687 tablename
, 1UL << log2qty
, ilog2(size
) - PAGE_SHIFT
, size
);
6690 *_hash_shift
= log2qty
;
6692 *_hash_mask
= (1 << log2qty
) - 1;
6697 /* Return a pointer to the bitmap storing bits affecting a block of pages */
6698 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
6701 #ifdef CONFIG_SPARSEMEM
6702 return __pfn_to_section(pfn
)->pageblock_flags
;
6704 return zone
->pageblock_flags
;
6705 #endif /* CONFIG_SPARSEMEM */
6708 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
6710 #ifdef CONFIG_SPARSEMEM
6711 pfn
&= (PAGES_PER_SECTION
-1);
6712 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
6714 pfn
= pfn
- round_down(zone
->zone_start_pfn
, pageblock_nr_pages
);
6715 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
6716 #endif /* CONFIG_SPARSEMEM */
6720 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
6721 * @page: The page within the block of interest
6722 * @pfn: The target page frame number
6723 * @end_bitidx: The last bit of interest to retrieve
6724 * @mask: mask of bits that the caller is interested in
6726 * Return: pageblock_bits flags
6728 unsigned long get_pfnblock_flags_mask(struct page
*page
, unsigned long pfn
,
6729 unsigned long end_bitidx
,
6733 unsigned long *bitmap
;
6734 unsigned long bitidx
, word_bitidx
;
6737 zone
= page_zone(page
);
6738 bitmap
= get_pageblock_bitmap(zone
, pfn
);
6739 bitidx
= pfn_to_bitidx(zone
, pfn
);
6740 word_bitidx
= bitidx
/ BITS_PER_LONG
;
6741 bitidx
&= (BITS_PER_LONG
-1);
6743 word
= bitmap
[word_bitidx
];
6744 bitidx
+= end_bitidx
;
6745 return (word
>> (BITS_PER_LONG
- bitidx
- 1)) & mask
;
6749 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
6750 * @page: The page within the block of interest
6751 * @flags: The flags to set
6752 * @pfn: The target page frame number
6753 * @end_bitidx: The last bit of interest
6754 * @mask: mask of bits that the caller is interested in
6756 void set_pfnblock_flags_mask(struct page
*page
, unsigned long flags
,
6758 unsigned long end_bitidx
,
6762 unsigned long *bitmap
;
6763 unsigned long bitidx
, word_bitidx
;
6764 unsigned long old_word
, word
;
6766 BUILD_BUG_ON(NR_PAGEBLOCK_BITS
!= 4);
6768 zone
= page_zone(page
);
6769 bitmap
= get_pageblock_bitmap(zone
, pfn
);
6770 bitidx
= pfn_to_bitidx(zone
, pfn
);
6771 word_bitidx
= bitidx
/ BITS_PER_LONG
;
6772 bitidx
&= (BITS_PER_LONG
-1);
6774 VM_BUG_ON_PAGE(!zone_spans_pfn(zone
, pfn
), page
);
6776 bitidx
+= end_bitidx
;
6777 mask
<<= (BITS_PER_LONG
- bitidx
- 1);
6778 flags
<<= (BITS_PER_LONG
- bitidx
- 1);
6780 word
= READ_ONCE(bitmap
[word_bitidx
]);
6782 old_word
= cmpxchg(&bitmap
[word_bitidx
], word
, (word
& ~mask
) | flags
);
6783 if (word
== old_word
)
6790 * This function checks whether pageblock includes unmovable pages or not.
6791 * If @count is not zero, it is okay to include less @count unmovable pages
6793 * PageLRU check without isolation or lru_lock could race so that
6794 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6795 * expect this function should be exact.
6797 bool has_unmovable_pages(struct zone
*zone
, struct page
*page
, int count
,
6798 bool skip_hwpoisoned_pages
)
6800 unsigned long pfn
, iter
, found
;
6804 * For avoiding noise data, lru_add_drain_all() should be called
6805 * If ZONE_MOVABLE, the zone never contains unmovable pages
6807 if (zone_idx(zone
) == ZONE_MOVABLE
)
6809 mt
= get_pageblock_migratetype(page
);
6810 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
6813 pfn
= page_to_pfn(page
);
6814 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
6815 unsigned long check
= pfn
+ iter
;
6817 if (!pfn_valid_within(check
))
6820 page
= pfn_to_page(check
);
6823 * Hugepages are not in LRU lists, but they're movable.
6824 * We need not scan over tail pages bacause we don't
6825 * handle each tail page individually in migration.
6827 if (PageHuge(page
)) {
6828 iter
= round_up(iter
+ 1, 1<<compound_order(page
)) - 1;
6833 * We can't use page_count without pin a page
6834 * because another CPU can free compound page.
6835 * This check already skips compound tails of THP
6836 * because their page->_count is zero at all time.
6838 if (!page_ref_count(page
)) {
6839 if (PageBuddy(page
))
6840 iter
+= (1 << page_order(page
)) - 1;
6845 * The HWPoisoned page may be not in buddy system, and
6846 * page_count() is not 0.
6848 if (skip_hwpoisoned_pages
&& PageHWPoison(page
))
6854 * If there are RECLAIMABLE pages, we need to check
6855 * it. But now, memory offline itself doesn't call
6856 * shrink_node_slabs() and it still to be fixed.
6859 * If the page is not RAM, page_count()should be 0.
6860 * we don't need more check. This is an _used_ not-movable page.
6862 * The problematic thing here is PG_reserved pages. PG_reserved
6863 * is set to both of a memory hole page and a _used_ kernel
6872 bool is_pageblock_removable_nolock(struct page
*page
)
6878 * We have to be careful here because we are iterating over memory
6879 * sections which are not zone aware so we might end up outside of
6880 * the zone but still within the section.
6881 * We have to take care about the node as well. If the node is offline
6882 * its NODE_DATA will be NULL - see page_zone.
6884 if (!node_online(page_to_nid(page
)))
6887 zone
= page_zone(page
);
6888 pfn
= page_to_pfn(page
);
6889 if (!zone_spans_pfn(zone
, pfn
))
6892 return !has_unmovable_pages(zone
, page
, 0, true);
6895 #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
6897 static unsigned long pfn_max_align_down(unsigned long pfn
)
6899 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
6900 pageblock_nr_pages
) - 1);
6903 static unsigned long pfn_max_align_up(unsigned long pfn
)
6905 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
6906 pageblock_nr_pages
));
6909 /* [start, end) must belong to a single zone. */
6910 static int __alloc_contig_migrate_range(struct compact_control
*cc
,
6911 unsigned long start
, unsigned long end
)
6913 /* This function is based on compact_zone() from compaction.c. */
6914 unsigned long nr_reclaimed
;
6915 unsigned long pfn
= start
;
6916 unsigned int tries
= 0;
6921 while (pfn
< end
|| !list_empty(&cc
->migratepages
)) {
6922 if (fatal_signal_pending(current
)) {
6927 if (list_empty(&cc
->migratepages
)) {
6928 cc
->nr_migratepages
= 0;
6929 pfn
= isolate_migratepages_range(cc
, pfn
, end
);
6935 } else if (++tries
== 5) {
6936 ret
= ret
< 0 ? ret
: -EBUSY
;
6940 nr_reclaimed
= reclaim_clean_pages_from_list(cc
->zone
,
6942 cc
->nr_migratepages
-= nr_reclaimed
;
6944 ret
= migrate_pages(&cc
->migratepages
, alloc_migrate_target
,
6945 NULL
, 0, cc
->mode
, MR_CMA
);
6948 putback_movable_pages(&cc
->migratepages
);
6955 * alloc_contig_range() -- tries to allocate given range of pages
6956 * @start: start PFN to allocate
6957 * @end: one-past-the-last PFN to allocate
6958 * @migratetype: migratetype of the underlaying pageblocks (either
6959 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6960 * in range must have the same migratetype and it must
6961 * be either of the two.
6963 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6964 * aligned, however it's the caller's responsibility to guarantee that
6965 * we are the only thread that changes migrate type of pageblocks the
6968 * The PFN range must belong to a single zone.
6970 * Returns zero on success or negative error code. On success all
6971 * pages which PFN is in [start, end) are allocated for the caller and
6972 * need to be freed with free_contig_range().
6974 int alloc_contig_range(unsigned long start
, unsigned long end
,
6975 unsigned migratetype
)
6977 unsigned long outer_start
, outer_end
;
6981 struct compact_control cc
= {
6982 .nr_migratepages
= 0,
6984 .zone
= page_zone(pfn_to_page(start
)),
6985 .mode
= MIGRATE_SYNC
,
6986 .ignore_skip_hint
= true,
6988 INIT_LIST_HEAD(&cc
.migratepages
);
6991 * What we do here is we mark all pageblocks in range as
6992 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6993 * have different sizes, and due to the way page allocator
6994 * work, we align the range to biggest of the two pages so
6995 * that page allocator won't try to merge buddies from
6996 * different pageblocks and change MIGRATE_ISOLATE to some
6997 * other migration type.
6999 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
7000 * migrate the pages from an unaligned range (ie. pages that
7001 * we are interested in). This will put all the pages in
7002 * range back to page allocator as MIGRATE_ISOLATE.
7004 * When this is done, we take the pages in range from page
7005 * allocator removing them from the buddy system. This way
7006 * page allocator will never consider using them.
7008 * This lets us mark the pageblocks back as
7009 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
7010 * aligned range but not in the unaligned, original range are
7011 * put back to page allocator so that buddy can use them.
7014 ret
= start_isolate_page_range(pfn_max_align_down(start
),
7015 pfn_max_align_up(end
), migratetype
,
7021 * In case of -EBUSY, we'd like to know which page causes problem.
7022 * So, just fall through. We will check it in test_pages_isolated().
7024 ret
= __alloc_contig_migrate_range(&cc
, start
, end
);
7025 if (ret
&& ret
!= -EBUSY
)
7029 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
7030 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
7031 * more, all pages in [start, end) are free in page allocator.
7032 * What we are going to do is to allocate all pages from
7033 * [start, end) (that is remove them from page allocator).
7035 * The only problem is that pages at the beginning and at the
7036 * end of interesting range may be not aligned with pages that
7037 * page allocator holds, ie. they can be part of higher order
7038 * pages. Because of this, we reserve the bigger range and
7039 * once this is done free the pages we are not interested in.
7041 * We don't have to hold zone->lock here because the pages are
7042 * isolated thus they won't get removed from buddy.
7045 lru_add_drain_all();
7046 drain_all_pages(cc
.zone
);
7049 outer_start
= start
;
7050 while (!PageBuddy(pfn_to_page(outer_start
))) {
7051 if (++order
>= MAX_ORDER
) {
7052 outer_start
= start
;
7055 outer_start
&= ~0UL << order
;
7058 if (outer_start
!= start
) {
7059 order
= page_order(pfn_to_page(outer_start
));
7062 * outer_start page could be small order buddy page and
7063 * it doesn't include start page. Adjust outer_start
7064 * in this case to report failed page properly
7065 * on tracepoint in test_pages_isolated()
7067 if (outer_start
+ (1UL << order
) <= start
)
7068 outer_start
= start
;
7071 /* Make sure the range is really isolated. */
7072 if (test_pages_isolated(outer_start
, end
, false)) {
7073 pr_info("%s: [%lx, %lx) PFNs busy\n",
7074 __func__
, outer_start
, end
);
7079 /* Grab isolated pages from freelists. */
7080 outer_end
= isolate_freepages_range(&cc
, outer_start
, end
);
7086 /* Free head and tail (if any) */
7087 if (start
!= outer_start
)
7088 free_contig_range(outer_start
, start
- outer_start
);
7089 if (end
!= outer_end
)
7090 free_contig_range(end
, outer_end
- end
);
7093 undo_isolate_page_range(pfn_max_align_down(start
),
7094 pfn_max_align_up(end
), migratetype
);
7098 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
7100 unsigned int count
= 0;
7102 for (; nr_pages
--; pfn
++) {
7103 struct page
*page
= pfn_to_page(pfn
);
7105 count
+= page_count(page
) != 1;
7108 WARN(count
!= 0, "%d pages are still in use!\n", count
);
7112 #ifdef CONFIG_MEMORY_HOTPLUG
7114 * The zone indicated has a new number of managed_pages; batch sizes and percpu
7115 * page high values need to be recalulated.
7117 void __meminit
zone_pcp_update(struct zone
*zone
)
7120 mutex_lock(&pcp_batch_high_lock
);
7121 for_each_possible_cpu(cpu
)
7122 pageset_set_high_and_batch(zone
,
7123 per_cpu_ptr(zone
->pageset
, cpu
));
7124 mutex_unlock(&pcp_batch_high_lock
);
7128 void zone_pcp_reset(struct zone
*zone
)
7130 unsigned long flags
;
7132 struct per_cpu_pageset
*pset
;
7134 /* avoid races with drain_pages() */
7135 local_irq_save(flags
);
7136 if (zone
->pageset
!= &boot_pageset
) {
7137 for_each_online_cpu(cpu
) {
7138 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
7139 drain_zonestat(zone
, pset
);
7141 free_percpu(zone
->pageset
);
7142 zone
->pageset
= &boot_pageset
;
7144 local_irq_restore(flags
);
7147 #ifdef CONFIG_MEMORY_HOTREMOVE
7149 * All pages in the range must be isolated before calling this.
7152 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
7156 unsigned int order
, i
;
7158 unsigned long flags
;
7159 /* find the first valid pfn */
7160 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
7165 zone
= page_zone(pfn_to_page(pfn
));
7166 spin_lock_irqsave(&zone
->lock
, flags
);
7168 while (pfn
< end_pfn
) {
7169 if (!pfn_valid(pfn
)) {
7173 page
= pfn_to_page(pfn
);
7175 * The HWPoisoned page may be not in buddy system, and
7176 * page_count() is not 0.
7178 if (unlikely(!PageBuddy(page
) && PageHWPoison(page
))) {
7180 SetPageReserved(page
);
7184 BUG_ON(page_count(page
));
7185 BUG_ON(!PageBuddy(page
));
7186 order
= page_order(page
);
7187 #ifdef CONFIG_DEBUG_VM
7188 pr_info("remove from free list %lx %d %lx\n",
7189 pfn
, 1 << order
, end_pfn
);
7191 list_del(&page
->lru
);
7192 rmv_page_order(page
);
7193 zone
->free_area
[order
].nr_free
--;
7194 for (i
= 0; i
< (1 << order
); i
++)
7195 SetPageReserved((page
+i
));
7196 pfn
+= (1 << order
);
7198 spin_unlock_irqrestore(&zone
->lock
, flags
);
7202 bool is_free_buddy_page(struct page
*page
)
7204 struct zone
*zone
= page_zone(page
);
7205 unsigned long pfn
= page_to_pfn(page
);
7206 unsigned long flags
;
7209 spin_lock_irqsave(&zone
->lock
, flags
);
7210 for (order
= 0; order
< MAX_ORDER
; order
++) {
7211 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
7213 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
7216 spin_unlock_irqrestore(&zone
->lock
, flags
);
7218 return order
< MAX_ORDER
;