2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/jiffies.h>
23 #include <linux/bootmem.h>
24 #include <linux/memblock.h>
25 #include <linux/compiler.h>
26 #include <linux/kernel.h>
27 #include <linux/kmemcheck.h>
28 #include <linux/kasan.h>
29 #include <linux/module.h>
30 #include <linux/suspend.h>
31 #include <linux/pagevec.h>
32 #include <linux/blkdev.h>
33 #include <linux/slab.h>
34 #include <linux/ratelimit.h>
35 #include <linux/oom.h>
36 #include <linux/notifier.h>
37 #include <linux/topology.h>
38 #include <linux/sysctl.h>
39 #include <linux/cpu.h>
40 #include <linux/cpuset.h>
41 #include <linux/memory_hotplug.h>
42 #include <linux/nodemask.h>
43 #include <linux/vmalloc.h>
44 #include <linux/vmstat.h>
45 #include <linux/mempolicy.h>
46 #include <linux/memremap.h>
47 #include <linux/stop_machine.h>
48 #include <linux/sort.h>
49 #include <linux/pfn.h>
50 #include <linux/backing-dev.h>
51 #include <linux/fault-inject.h>
52 #include <linux/page-isolation.h>
53 #include <linux/page_ext.h>
54 #include <linux/debugobjects.h>
55 #include <linux/kmemleak.h>
56 #include <linux/compaction.h>
57 #include <trace/events/kmem.h>
58 #include <trace/events/oom.h>
59 #include <linux/prefetch.h>
60 #include <linux/mm_inline.h>
61 #include <linux/migrate.h>
62 #include <linux/hugetlb.h>
63 #include <linux/sched/rt.h>
64 #include <linux/sched/mm.h>
65 #include <linux/page_owner.h>
66 #include <linux/kthread.h>
67 #include <linux/memcontrol.h>
68 #include <linux/ftrace.h>
70 #include <asm/sections.h>
71 #include <asm/tlbflush.h>
72 #include <asm/div64.h>
75 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
76 static DEFINE_MUTEX(pcp_batch_high_lock
);
77 #define MIN_PERCPU_PAGELIST_FRACTION (8)
79 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
80 DEFINE_PER_CPU(int, numa_node
);
81 EXPORT_PER_CPU_SYMBOL(numa_node
);
84 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
86 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
87 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
88 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
89 * defined in <linux/topology.h>.
91 DEFINE_PER_CPU(int, _numa_mem_
); /* Kernel "local memory" node */
92 EXPORT_PER_CPU_SYMBOL(_numa_mem_
);
93 int _node_numa_mem_
[MAX_NUMNODES
];
96 /* work_structs for global per-cpu drains */
97 DEFINE_MUTEX(pcpu_drain_mutex
);
98 DEFINE_PER_CPU(struct work_struct
, pcpu_drain
);
100 #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY
101 volatile unsigned long latent_entropy __latent_entropy
;
102 EXPORT_SYMBOL(latent_entropy
);
106 * Array of node states.
108 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
109 [N_POSSIBLE
] = NODE_MASK_ALL
,
110 [N_ONLINE
] = { { [0] = 1UL } },
112 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
113 #ifdef CONFIG_HIGHMEM
114 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
116 #ifdef CONFIG_MOVABLE_NODE
117 [N_MEMORY
] = { { [0] = 1UL } },
119 [N_CPU
] = { { [0] = 1UL } },
122 EXPORT_SYMBOL(node_states
);
124 /* Protect totalram_pages and zone->managed_pages */
125 static DEFINE_SPINLOCK(managed_page_count_lock
);
127 unsigned long totalram_pages __read_mostly
;
128 unsigned long totalreserve_pages __read_mostly
;
129 unsigned long totalcma_pages __read_mostly
;
131 int percpu_pagelist_fraction
;
132 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
135 * A cached value of the page's pageblock's migratetype, used when the page is
136 * put on a pcplist. Used to avoid the pageblock migratetype lookup when
137 * freeing from pcplists in most cases, at the cost of possibly becoming stale.
138 * Also the migratetype set in the page does not necessarily match the pcplist
139 * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
140 * other index - this ensures that it will be put on the correct CMA freelist.
142 static inline int get_pcppage_migratetype(struct page
*page
)
147 static inline void set_pcppage_migratetype(struct page
*page
, int migratetype
)
149 page
->index
= migratetype
;
152 #ifdef CONFIG_PM_SLEEP
154 * The following functions are used by the suspend/hibernate code to temporarily
155 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
156 * while devices are suspended. To avoid races with the suspend/hibernate code,
157 * they should always be called with pm_mutex held (gfp_allowed_mask also should
158 * only be modified with pm_mutex held, unless the suspend/hibernate code is
159 * guaranteed not to run in parallel with that modification).
162 static gfp_t saved_gfp_mask
;
164 void pm_restore_gfp_mask(void)
166 WARN_ON(!mutex_is_locked(&pm_mutex
));
167 if (saved_gfp_mask
) {
168 gfp_allowed_mask
= saved_gfp_mask
;
173 void pm_restrict_gfp_mask(void)
175 WARN_ON(!mutex_is_locked(&pm_mutex
));
176 WARN_ON(saved_gfp_mask
);
177 saved_gfp_mask
= gfp_allowed_mask
;
178 gfp_allowed_mask
&= ~(__GFP_IO
| __GFP_FS
);
181 bool pm_suspended_storage(void)
183 if ((gfp_allowed_mask
& (__GFP_IO
| __GFP_FS
)) == (__GFP_IO
| __GFP_FS
))
187 #endif /* CONFIG_PM_SLEEP */
189 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
190 unsigned int pageblock_order __read_mostly
;
193 static void __free_pages_ok(struct page
*page
, unsigned int order
);
196 * results with 256, 32 in the lowmem_reserve sysctl:
197 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
198 * 1G machine -> (16M dma, 784M normal, 224M high)
199 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
200 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
201 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
203 * TBD: should special case ZONE_DMA32 machines here - in those we normally
204 * don't need any ZONE_NORMAL reservation
206 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
207 #ifdef CONFIG_ZONE_DMA
210 #ifdef CONFIG_ZONE_DMA32
213 #ifdef CONFIG_HIGHMEM
219 EXPORT_SYMBOL(totalram_pages
);
221 static char * const zone_names
[MAX_NR_ZONES
] = {
222 #ifdef CONFIG_ZONE_DMA
225 #ifdef CONFIG_ZONE_DMA32
229 #ifdef CONFIG_HIGHMEM
233 #ifdef CONFIG_ZONE_DEVICE
238 char * const migratetype_names
[MIGRATE_TYPES
] = {
246 #ifdef CONFIG_MEMORY_ISOLATION
251 compound_page_dtor
* const compound_page_dtors
[] = {
254 #ifdef CONFIG_HUGETLB_PAGE
257 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
262 int min_free_kbytes
= 1024;
263 int user_min_free_kbytes
= -1;
264 int watermark_scale_factor
= 10;
266 static unsigned long __meminitdata nr_kernel_pages
;
267 static unsigned long __meminitdata nr_all_pages
;
268 static unsigned long __meminitdata dma_reserve
;
270 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
271 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
272 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
273 static unsigned long __initdata required_kernelcore
;
274 static unsigned long __initdata required_movablecore
;
275 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
276 static bool mirrored_kernelcore
;
278 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
280 EXPORT_SYMBOL(movable_zone
);
281 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
284 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
285 int nr_online_nodes __read_mostly
= 1;
286 EXPORT_SYMBOL(nr_node_ids
);
287 EXPORT_SYMBOL(nr_online_nodes
);
290 int page_group_by_mobility_disabled __read_mostly
;
292 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
293 static inline void reset_deferred_meminit(pg_data_t
*pgdat
)
295 pgdat
->first_deferred_pfn
= ULONG_MAX
;
298 /* Returns true if the struct page for the pfn is uninitialised */
299 static inline bool __meminit
early_page_uninitialised(unsigned long pfn
)
301 int nid
= early_pfn_to_nid(pfn
);
303 if (node_online(nid
) && pfn
>= NODE_DATA(nid
)->first_deferred_pfn
)
310 * Returns false when the remaining initialisation should be deferred until
311 * later in the boot cycle when it can be parallelised.
313 static inline bool update_defer_init(pg_data_t
*pgdat
,
314 unsigned long pfn
, unsigned long zone_end
,
315 unsigned long *nr_initialised
)
317 unsigned long max_initialise
;
319 /* Always populate low zones for address-contrained allocations */
320 if (zone_end
< pgdat_end_pfn(pgdat
))
323 * Initialise at least 2G of a node but also take into account that
324 * two large system hashes that can take up 1GB for 0.25TB/node.
326 max_initialise
= max(2UL << (30 - PAGE_SHIFT
),
327 (pgdat
->node_spanned_pages
>> 8));
330 if ((*nr_initialised
> max_initialise
) &&
331 (pfn
& (PAGES_PER_SECTION
- 1)) == 0) {
332 pgdat
->first_deferred_pfn
= pfn
;
339 static inline void reset_deferred_meminit(pg_data_t
*pgdat
)
343 static inline bool early_page_uninitialised(unsigned long pfn
)
348 static inline bool update_defer_init(pg_data_t
*pgdat
,
349 unsigned long pfn
, unsigned long zone_end
,
350 unsigned long *nr_initialised
)
356 /* Return a pointer to the bitmap storing bits affecting a block of pages */
357 static inline unsigned long *get_pageblock_bitmap(struct page
*page
,
360 #ifdef CONFIG_SPARSEMEM
361 return __pfn_to_section(pfn
)->pageblock_flags
;
363 return page_zone(page
)->pageblock_flags
;
364 #endif /* CONFIG_SPARSEMEM */
367 static inline int pfn_to_bitidx(struct page
*page
, unsigned long pfn
)
369 #ifdef CONFIG_SPARSEMEM
370 pfn
&= (PAGES_PER_SECTION
-1);
371 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
373 pfn
= pfn
- round_down(page_zone(page
)->zone_start_pfn
, pageblock_nr_pages
);
374 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
375 #endif /* CONFIG_SPARSEMEM */
379 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
380 * @page: The page within the block of interest
381 * @pfn: The target page frame number
382 * @end_bitidx: The last bit of interest to retrieve
383 * @mask: mask of bits that the caller is interested in
385 * Return: pageblock_bits flags
387 static __always_inline
unsigned long __get_pfnblock_flags_mask(struct page
*page
,
389 unsigned long end_bitidx
,
392 unsigned long *bitmap
;
393 unsigned long bitidx
, word_bitidx
;
396 bitmap
= get_pageblock_bitmap(page
, pfn
);
397 bitidx
= pfn_to_bitidx(page
, pfn
);
398 word_bitidx
= bitidx
/ BITS_PER_LONG
;
399 bitidx
&= (BITS_PER_LONG
-1);
401 word
= bitmap
[word_bitidx
];
402 bitidx
+= end_bitidx
;
403 return (word
>> (BITS_PER_LONG
- bitidx
- 1)) & mask
;
406 unsigned long get_pfnblock_flags_mask(struct page
*page
, unsigned long pfn
,
407 unsigned long end_bitidx
,
410 return __get_pfnblock_flags_mask(page
, pfn
, end_bitidx
, mask
);
413 static __always_inline
int get_pfnblock_migratetype(struct page
*page
, unsigned long pfn
)
415 return __get_pfnblock_flags_mask(page
, pfn
, PB_migrate_end
, MIGRATETYPE_MASK
);
419 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
420 * @page: The page within the block of interest
421 * @flags: The flags to set
422 * @pfn: The target page frame number
423 * @end_bitidx: The last bit of interest
424 * @mask: mask of bits that the caller is interested in
426 void set_pfnblock_flags_mask(struct page
*page
, unsigned long flags
,
428 unsigned long end_bitidx
,
431 unsigned long *bitmap
;
432 unsigned long bitidx
, word_bitidx
;
433 unsigned long old_word
, word
;
435 BUILD_BUG_ON(NR_PAGEBLOCK_BITS
!= 4);
437 bitmap
= get_pageblock_bitmap(page
, pfn
);
438 bitidx
= pfn_to_bitidx(page
, pfn
);
439 word_bitidx
= bitidx
/ BITS_PER_LONG
;
440 bitidx
&= (BITS_PER_LONG
-1);
442 VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page
), pfn
), page
);
444 bitidx
+= end_bitidx
;
445 mask
<<= (BITS_PER_LONG
- bitidx
- 1);
446 flags
<<= (BITS_PER_LONG
- bitidx
- 1);
448 word
= READ_ONCE(bitmap
[word_bitidx
]);
450 old_word
= cmpxchg(&bitmap
[word_bitidx
], word
, (word
& ~mask
) | flags
);
451 if (word
== old_word
)
457 void set_pageblock_migratetype(struct page
*page
, int migratetype
)
459 if (unlikely(page_group_by_mobility_disabled
&&
460 migratetype
< MIGRATE_PCPTYPES
))
461 migratetype
= MIGRATE_UNMOVABLE
;
463 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
464 PB_migrate
, PB_migrate_end
);
467 #ifdef CONFIG_DEBUG_VM
468 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
472 unsigned long pfn
= page_to_pfn(page
);
473 unsigned long sp
, start_pfn
;
476 seq
= zone_span_seqbegin(zone
);
477 start_pfn
= zone
->zone_start_pfn
;
478 sp
= zone
->spanned_pages
;
479 if (!zone_spans_pfn(zone
, pfn
))
481 } while (zone_span_seqretry(zone
, seq
));
484 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
485 pfn
, zone_to_nid(zone
), zone
->name
,
486 start_pfn
, start_pfn
+ sp
);
491 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
493 if (!pfn_valid_within(page_to_pfn(page
)))
495 if (zone
!= page_zone(page
))
501 * Temporary debugging check for pages not lying within a given zone.
503 static int bad_range(struct zone
*zone
, struct page
*page
)
505 if (page_outside_zone_boundaries(zone
, page
))
507 if (!page_is_consistent(zone
, page
))
513 static inline int bad_range(struct zone
*zone
, struct page
*page
)
519 static void bad_page(struct page
*page
, const char *reason
,
520 unsigned long bad_flags
)
522 static unsigned long resume
;
523 static unsigned long nr_shown
;
524 static unsigned long nr_unshown
;
527 * Allow a burst of 60 reports, then keep quiet for that minute;
528 * or allow a steady drip of one report per second.
530 if (nr_shown
== 60) {
531 if (time_before(jiffies
, resume
)) {
537 "BUG: Bad page state: %lu messages suppressed\n",
544 resume
= jiffies
+ 60 * HZ
;
546 pr_alert("BUG: Bad page state in process %s pfn:%05lx\n",
547 current
->comm
, page_to_pfn(page
));
548 __dump_page(page
, reason
);
549 bad_flags
&= page
->flags
;
551 pr_alert("bad because of flags: %#lx(%pGp)\n",
552 bad_flags
, &bad_flags
);
553 dump_page_owner(page
);
558 /* Leave bad fields for debug, except PageBuddy could make trouble */
559 page_mapcount_reset(page
); /* remove PageBuddy */
560 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
564 * Higher-order pages are called "compound pages". They are structured thusly:
566 * The first PAGE_SIZE page is called the "head page" and have PG_head set.
568 * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
569 * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
571 * The first tail page's ->compound_dtor holds the offset in array of compound
572 * page destructors. See compound_page_dtors.
574 * The first tail page's ->compound_order holds the order of allocation.
575 * This usage means that zero-order pages may not be compound.
578 void free_compound_page(struct page
*page
)
580 __free_pages_ok(page
, compound_order(page
));
583 void prep_compound_page(struct page
*page
, unsigned int order
)
586 int nr_pages
= 1 << order
;
588 set_compound_page_dtor(page
, COMPOUND_PAGE_DTOR
);
589 set_compound_order(page
, order
);
591 for (i
= 1; i
< nr_pages
; i
++) {
592 struct page
*p
= page
+ i
;
593 set_page_count(p
, 0);
594 p
->mapping
= TAIL_MAPPING
;
595 set_compound_head(p
, page
);
597 atomic_set(compound_mapcount_ptr(page
), -1);
600 #ifdef CONFIG_DEBUG_PAGEALLOC
601 unsigned int _debug_guardpage_minorder
;
602 bool _debug_pagealloc_enabled __read_mostly
603 = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT
);
604 EXPORT_SYMBOL(_debug_pagealloc_enabled
);
605 bool _debug_guardpage_enabled __read_mostly
;
607 static int __init
early_debug_pagealloc(char *buf
)
611 return kstrtobool(buf
, &_debug_pagealloc_enabled
);
613 early_param("debug_pagealloc", early_debug_pagealloc
);
615 static bool need_debug_guardpage(void)
617 /* If we don't use debug_pagealloc, we don't need guard page */
618 if (!debug_pagealloc_enabled())
621 if (!debug_guardpage_minorder())
627 static void init_debug_guardpage(void)
629 if (!debug_pagealloc_enabled())
632 if (!debug_guardpage_minorder())
635 _debug_guardpage_enabled
= true;
638 struct page_ext_operations debug_guardpage_ops
= {
639 .need
= need_debug_guardpage
,
640 .init
= init_debug_guardpage
,
643 static int __init
debug_guardpage_minorder_setup(char *buf
)
647 if (kstrtoul(buf
, 10, &res
) < 0 || res
> MAX_ORDER
/ 2) {
648 pr_err("Bad debug_guardpage_minorder value\n");
651 _debug_guardpage_minorder
= res
;
652 pr_info("Setting debug_guardpage_minorder to %lu\n", res
);
655 early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup
);
657 static inline bool set_page_guard(struct zone
*zone
, struct page
*page
,
658 unsigned int order
, int migratetype
)
660 struct page_ext
*page_ext
;
662 if (!debug_guardpage_enabled())
665 if (order
>= debug_guardpage_minorder())
668 page_ext
= lookup_page_ext(page
);
669 if (unlikely(!page_ext
))
672 __set_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
674 INIT_LIST_HEAD(&page
->lru
);
675 set_page_private(page
, order
);
676 /* Guard pages are not available for any usage */
677 __mod_zone_freepage_state(zone
, -(1 << order
), migratetype
);
682 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
683 unsigned int order
, int migratetype
)
685 struct page_ext
*page_ext
;
687 if (!debug_guardpage_enabled())
690 page_ext
= lookup_page_ext(page
);
691 if (unlikely(!page_ext
))
694 __clear_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
696 set_page_private(page
, 0);
697 if (!is_migrate_isolate(migratetype
))
698 __mod_zone_freepage_state(zone
, (1 << order
), migratetype
);
701 struct page_ext_operations debug_guardpage_ops
;
702 static inline bool set_page_guard(struct zone
*zone
, struct page
*page
,
703 unsigned int order
, int migratetype
) { return false; }
704 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
705 unsigned int order
, int migratetype
) {}
708 static inline void set_page_order(struct page
*page
, unsigned int order
)
710 set_page_private(page
, order
);
711 __SetPageBuddy(page
);
714 static inline void rmv_page_order(struct page
*page
)
716 __ClearPageBuddy(page
);
717 set_page_private(page
, 0);
721 * This function checks whether a page is free && is the buddy
722 * we can do coalesce a page and its buddy if
723 * (a) the buddy is not in a hole (check before calling!) &&
724 * (b) the buddy is in the buddy system &&
725 * (c) a page and its buddy have the same order &&
726 * (d) a page and its buddy are in the same zone.
728 * For recording whether a page is in the buddy system, we set ->_mapcount
729 * PAGE_BUDDY_MAPCOUNT_VALUE.
730 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
731 * serialized by zone->lock.
733 * For recording page's order, we use page_private(page).
735 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
738 if (page_is_guard(buddy
) && page_order(buddy
) == order
) {
739 if (page_zone_id(page
) != page_zone_id(buddy
))
742 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
747 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
749 * zone check is done late to avoid uselessly
750 * calculating zone/node ids for pages that could
753 if (page_zone_id(page
) != page_zone_id(buddy
))
756 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
764 * Freeing function for a buddy system allocator.
766 * The concept of a buddy system is to maintain direct-mapped table
767 * (containing bit values) for memory blocks of various "orders".
768 * The bottom level table contains the map for the smallest allocatable
769 * units of memory (here, pages), and each level above it describes
770 * pairs of units from the levels below, hence, "buddies".
771 * At a high level, all that happens here is marking the table entry
772 * at the bottom level available, and propagating the changes upward
773 * as necessary, plus some accounting needed to play nicely with other
774 * parts of the VM system.
775 * At each level, we keep a list of pages, which are heads of continuous
776 * free pages of length of (1 << order) and marked with _mapcount
777 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
779 * So when we are allocating or freeing one, we can derive the state of the
780 * other. That is, if we allocate a small block, and both were
781 * free, the remainder of the region must be split into blocks.
782 * If a block is freed, and its buddy is also free, then this
783 * triggers coalescing into a block of larger size.
788 static inline void __free_one_page(struct page
*page
,
790 struct zone
*zone
, unsigned int order
,
793 unsigned long combined_pfn
;
794 unsigned long uninitialized_var(buddy_pfn
);
796 unsigned int max_order
;
798 max_order
= min_t(unsigned int, MAX_ORDER
, pageblock_order
+ 1);
800 VM_BUG_ON(!zone_is_initialized(zone
));
801 VM_BUG_ON_PAGE(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
, page
);
803 VM_BUG_ON(migratetype
== -1);
804 if (likely(!is_migrate_isolate(migratetype
)))
805 __mod_zone_freepage_state(zone
, 1 << order
, migratetype
);
807 VM_BUG_ON_PAGE(pfn
& ((1 << order
) - 1), page
);
808 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
811 while (order
< max_order
- 1) {
812 buddy_pfn
= __find_buddy_pfn(pfn
, order
);
813 buddy
= page
+ (buddy_pfn
- pfn
);
815 if (!pfn_valid_within(buddy_pfn
))
817 if (!page_is_buddy(page
, buddy
, order
))
820 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
821 * merge with it and move up one order.
823 if (page_is_guard(buddy
)) {
824 clear_page_guard(zone
, buddy
, order
, migratetype
);
826 list_del(&buddy
->lru
);
827 zone
->free_area
[order
].nr_free
--;
828 rmv_page_order(buddy
);
830 combined_pfn
= buddy_pfn
& pfn
;
831 page
= page
+ (combined_pfn
- pfn
);
835 if (max_order
< MAX_ORDER
) {
836 /* If we are here, it means order is >= pageblock_order.
837 * We want to prevent merge between freepages on isolate
838 * pageblock and normal pageblock. Without this, pageblock
839 * isolation could cause incorrect freepage or CMA accounting.
841 * We don't want to hit this code for the more frequent
844 if (unlikely(has_isolate_pageblock(zone
))) {
847 buddy_pfn
= __find_buddy_pfn(pfn
, order
);
848 buddy
= page
+ (buddy_pfn
- pfn
);
849 buddy_mt
= get_pageblock_migratetype(buddy
);
851 if (migratetype
!= buddy_mt
852 && (is_migrate_isolate(migratetype
) ||
853 is_migrate_isolate(buddy_mt
)))
857 goto continue_merging
;
861 set_page_order(page
, order
);
864 * If this is not the largest possible page, check if the buddy
865 * of the next-highest order is free. If it is, it's possible
866 * that pages are being freed that will coalesce soon. In case,
867 * that is happening, add the free page to the tail of the list
868 * so it's less likely to be used soon and more likely to be merged
869 * as a higher order page
871 if ((order
< MAX_ORDER
-2) && pfn_valid_within(buddy_pfn
)) {
872 struct page
*higher_page
, *higher_buddy
;
873 combined_pfn
= buddy_pfn
& pfn
;
874 higher_page
= page
+ (combined_pfn
- pfn
);
875 buddy_pfn
= __find_buddy_pfn(combined_pfn
, order
+ 1);
876 higher_buddy
= higher_page
+ (buddy_pfn
- combined_pfn
);
877 if (pfn_valid_within(buddy_pfn
) &&
878 page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
879 list_add_tail(&page
->lru
,
880 &zone
->free_area
[order
].free_list
[migratetype
]);
885 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
887 zone
->free_area
[order
].nr_free
++;
891 * A bad page could be due to a number of fields. Instead of multiple branches,
892 * try and check multiple fields with one check. The caller must do a detailed
893 * check if necessary.
895 static inline bool page_expected_state(struct page
*page
,
896 unsigned long check_flags
)
898 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
901 if (unlikely((unsigned long)page
->mapping
|
902 page_ref_count(page
) |
904 (unsigned long)page
->mem_cgroup
|
906 (page
->flags
& check_flags
)))
912 static void free_pages_check_bad(struct page
*page
)
914 const char *bad_reason
;
915 unsigned long bad_flags
;
920 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
921 bad_reason
= "nonzero mapcount";
922 if (unlikely(page
->mapping
!= NULL
))
923 bad_reason
= "non-NULL mapping";
924 if (unlikely(page_ref_count(page
) != 0))
925 bad_reason
= "nonzero _refcount";
926 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
)) {
927 bad_reason
= "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
928 bad_flags
= PAGE_FLAGS_CHECK_AT_FREE
;
931 if (unlikely(page
->mem_cgroup
))
932 bad_reason
= "page still charged to cgroup";
934 bad_page(page
, bad_reason
, bad_flags
);
937 static inline int free_pages_check(struct page
*page
)
939 if (likely(page_expected_state(page
, PAGE_FLAGS_CHECK_AT_FREE
)))
942 /* Something has gone sideways, find it */
943 free_pages_check_bad(page
);
947 static int free_tail_pages_check(struct page
*head_page
, struct page
*page
)
952 * We rely page->lru.next never has bit 0 set, unless the page
953 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
955 BUILD_BUG_ON((unsigned long)LIST_POISON1
& 1);
957 if (!IS_ENABLED(CONFIG_DEBUG_VM
)) {
961 switch (page
- head_page
) {
963 /* the first tail page: ->mapping is compound_mapcount() */
964 if (unlikely(compound_mapcount(page
))) {
965 bad_page(page
, "nonzero compound_mapcount", 0);
971 * the second tail page: ->mapping is
972 * page_deferred_list().next -- ignore value.
976 if (page
->mapping
!= TAIL_MAPPING
) {
977 bad_page(page
, "corrupted mapping in tail page", 0);
982 if (unlikely(!PageTail(page
))) {
983 bad_page(page
, "PageTail not set", 0);
986 if (unlikely(compound_head(page
) != head_page
)) {
987 bad_page(page
, "compound_head not consistent", 0);
992 page
->mapping
= NULL
;
993 clear_compound_head(page
);
997 static __always_inline
bool free_pages_prepare(struct page
*page
,
998 unsigned int order
, bool check_free
)
1002 VM_BUG_ON_PAGE(PageTail(page
), page
);
1004 trace_mm_page_free(page
, order
);
1005 kmemcheck_free_shadow(page
, order
);
1008 * Check tail pages before head page information is cleared to
1009 * avoid checking PageCompound for order-0 pages.
1011 if (unlikely(order
)) {
1012 bool compound
= PageCompound(page
);
1015 VM_BUG_ON_PAGE(compound
&& compound_order(page
) != order
, page
);
1018 ClearPageDoubleMap(page
);
1019 for (i
= 1; i
< (1 << order
); i
++) {
1021 bad
+= free_tail_pages_check(page
, page
+ i
);
1022 if (unlikely(free_pages_check(page
+ i
))) {
1026 (page
+ i
)->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
1029 if (PageMappingFlags(page
))
1030 page
->mapping
= NULL
;
1031 if (memcg_kmem_enabled() && PageKmemcg(page
))
1032 memcg_kmem_uncharge(page
, order
);
1034 bad
+= free_pages_check(page
);
1038 page_cpupid_reset_last(page
);
1039 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
1040 reset_page_owner(page
, order
);
1042 if (!PageHighMem(page
)) {
1043 debug_check_no_locks_freed(page_address(page
),
1044 PAGE_SIZE
<< order
);
1045 debug_check_no_obj_freed(page_address(page
),
1046 PAGE_SIZE
<< order
);
1048 arch_free_page(page
, order
);
1049 kernel_poison_pages(page
, 1 << order
, 0);
1050 kernel_map_pages(page
, 1 << order
, 0);
1051 kasan_free_pages(page
, order
);
1056 #ifdef CONFIG_DEBUG_VM
1057 static inline bool free_pcp_prepare(struct page
*page
)
1059 return free_pages_prepare(page
, 0, true);
1062 static inline bool bulkfree_pcp_prepare(struct page
*page
)
1067 static bool free_pcp_prepare(struct page
*page
)
1069 return free_pages_prepare(page
, 0, false);
1072 static bool bulkfree_pcp_prepare(struct page
*page
)
1074 return free_pages_check(page
);
1076 #endif /* CONFIG_DEBUG_VM */
1079 * Frees a number of pages from the PCP lists
1080 * Assumes all pages on list are in same zone, and of same order.
1081 * count is the number of pages to free.
1083 * If the zone was previously in an "all pages pinned" state then look to
1084 * see if this freeing clears that state.
1086 * And clear the zone's pages_scanned counter, to hold off the "all pages are
1087 * pinned" detection logic.
1089 static void free_pcppages_bulk(struct zone
*zone
, int count
,
1090 struct per_cpu_pages
*pcp
)
1092 int migratetype
= 0;
1094 bool isolated_pageblocks
;
1096 spin_lock(&zone
->lock
);
1097 isolated_pageblocks
= has_isolate_pageblock(zone
);
1101 struct list_head
*list
;
1104 * Remove pages from lists in a round-robin fashion. A
1105 * batch_free count is maintained that is incremented when an
1106 * empty list is encountered. This is so more pages are freed
1107 * off fuller lists instead of spinning excessively around empty
1112 if (++migratetype
== MIGRATE_PCPTYPES
)
1114 list
= &pcp
->lists
[migratetype
];
1115 } while (list_empty(list
));
1117 /* This is the only non-empty list. Free them all. */
1118 if (batch_free
== MIGRATE_PCPTYPES
)
1122 int mt
; /* migratetype of the to-be-freed page */
1124 page
= list_last_entry(list
, struct page
, lru
);
1125 /* must delete as __free_one_page list manipulates */
1126 list_del(&page
->lru
);
1128 mt
= get_pcppage_migratetype(page
);
1129 /* MIGRATE_ISOLATE page should not go to pcplists */
1130 VM_BUG_ON_PAGE(is_migrate_isolate(mt
), page
);
1131 /* Pageblock could have been isolated meanwhile */
1132 if (unlikely(isolated_pageblocks
))
1133 mt
= get_pageblock_migratetype(page
);
1135 if (bulkfree_pcp_prepare(page
))
1138 __free_one_page(page
, page_to_pfn(page
), zone
, 0, mt
);
1139 trace_mm_page_pcpu_drain(page
, 0, mt
);
1140 } while (--count
&& --batch_free
&& !list_empty(list
));
1142 spin_unlock(&zone
->lock
);
1145 static void free_one_page(struct zone
*zone
,
1146 struct page
*page
, unsigned long pfn
,
1150 spin_lock(&zone
->lock
);
1151 if (unlikely(has_isolate_pageblock(zone
) ||
1152 is_migrate_isolate(migratetype
))) {
1153 migratetype
= get_pfnblock_migratetype(page
, pfn
);
1155 __free_one_page(page
, pfn
, zone
, order
, migratetype
);
1156 spin_unlock(&zone
->lock
);
1159 static void __meminit
__init_single_page(struct page
*page
, unsigned long pfn
,
1160 unsigned long zone
, int nid
)
1162 set_page_links(page
, zone
, nid
, pfn
);
1163 init_page_count(page
);
1164 page_mapcount_reset(page
);
1165 page_cpupid_reset_last(page
);
1167 INIT_LIST_HEAD(&page
->lru
);
1168 #ifdef WANT_PAGE_VIRTUAL
1169 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1170 if (!is_highmem_idx(zone
))
1171 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
1175 static void __meminit
__init_single_pfn(unsigned long pfn
, unsigned long zone
,
1178 return __init_single_page(pfn_to_page(pfn
), pfn
, zone
, nid
);
1181 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1182 static void init_reserved_page(unsigned long pfn
)
1187 if (!early_page_uninitialised(pfn
))
1190 nid
= early_pfn_to_nid(pfn
);
1191 pgdat
= NODE_DATA(nid
);
1193 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
1194 struct zone
*zone
= &pgdat
->node_zones
[zid
];
1196 if (pfn
>= zone
->zone_start_pfn
&& pfn
< zone_end_pfn(zone
))
1199 __init_single_pfn(pfn
, zid
, nid
);
1202 static inline void init_reserved_page(unsigned long pfn
)
1205 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1208 * Initialised pages do not have PageReserved set. This function is
1209 * called for each range allocated by the bootmem allocator and
1210 * marks the pages PageReserved. The remaining valid pages are later
1211 * sent to the buddy page allocator.
1213 void __meminit
reserve_bootmem_region(phys_addr_t start
, phys_addr_t end
)
1215 unsigned long start_pfn
= PFN_DOWN(start
);
1216 unsigned long end_pfn
= PFN_UP(end
);
1218 for (; start_pfn
< end_pfn
; start_pfn
++) {
1219 if (pfn_valid(start_pfn
)) {
1220 struct page
*page
= pfn_to_page(start_pfn
);
1222 init_reserved_page(start_pfn
);
1224 /* Avoid false-positive PageTail() */
1225 INIT_LIST_HEAD(&page
->lru
);
1227 SetPageReserved(page
);
1232 static void __free_pages_ok(struct page
*page
, unsigned int order
)
1234 unsigned long flags
;
1236 unsigned long pfn
= page_to_pfn(page
);
1238 if (!free_pages_prepare(page
, order
, true))
1241 migratetype
= get_pfnblock_migratetype(page
, pfn
);
1242 local_irq_save(flags
);
1243 __count_vm_events(PGFREE
, 1 << order
);
1244 free_one_page(page_zone(page
), page
, pfn
, order
, migratetype
);
1245 local_irq_restore(flags
);
1248 static void __init
__free_pages_boot_core(struct page
*page
, unsigned int order
)
1250 unsigned int nr_pages
= 1 << order
;
1251 struct page
*p
= page
;
1255 for (loop
= 0; loop
< (nr_pages
- 1); loop
++, p
++) {
1257 __ClearPageReserved(p
);
1258 set_page_count(p
, 0);
1260 __ClearPageReserved(p
);
1261 set_page_count(p
, 0);
1263 page_zone(page
)->managed_pages
+= nr_pages
;
1264 set_page_refcounted(page
);
1265 __free_pages(page
, order
);
1268 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1269 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1271 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata
;
1273 int __meminit
early_pfn_to_nid(unsigned long pfn
)
1275 static DEFINE_SPINLOCK(early_pfn_lock
);
1278 spin_lock(&early_pfn_lock
);
1279 nid
= __early_pfn_to_nid(pfn
, &early_pfnnid_cache
);
1281 nid
= first_online_node
;
1282 spin_unlock(&early_pfn_lock
);
1288 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1289 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1290 struct mminit_pfnnid_cache
*state
)
1294 nid
= __early_pfn_to_nid(pfn
, state
);
1295 if (nid
>= 0 && nid
!= node
)
1300 /* Only safe to use early in boot when initialisation is single-threaded */
1301 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1303 return meminit_pfn_in_nid(pfn
, node
, &early_pfnnid_cache
);
1308 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1312 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1313 struct mminit_pfnnid_cache
*state
)
1320 void __init
__free_pages_bootmem(struct page
*page
, unsigned long pfn
,
1323 if (early_page_uninitialised(pfn
))
1325 return __free_pages_boot_core(page
, order
);
1329 * Check that the whole (or subset of) a pageblock given by the interval of
1330 * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1331 * with the migration of free compaction scanner. The scanners then need to
1332 * use only pfn_valid_within() check for arches that allow holes within
1335 * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1337 * It's possible on some configurations to have a setup like node0 node1 node0
1338 * i.e. it's possible that all pages within a zones range of pages do not
1339 * belong to a single zone. We assume that a border between node0 and node1
1340 * can occur within a single pageblock, but not a node0 node1 node0
1341 * interleaving within a single pageblock. It is therefore sufficient to check
1342 * the first and last page of a pageblock and avoid checking each individual
1343 * page in a pageblock.
1345 struct page
*__pageblock_pfn_to_page(unsigned long start_pfn
,
1346 unsigned long end_pfn
, struct zone
*zone
)
1348 struct page
*start_page
;
1349 struct page
*end_page
;
1351 /* end_pfn is one past the range we are checking */
1354 if (!pfn_valid(start_pfn
) || !pfn_valid(end_pfn
))
1357 start_page
= pfn_to_page(start_pfn
);
1359 if (page_zone(start_page
) != zone
)
1362 end_page
= pfn_to_page(end_pfn
);
1364 /* This gives a shorter code than deriving page_zone(end_page) */
1365 if (page_zone_id(start_page
) != page_zone_id(end_page
))
1371 void set_zone_contiguous(struct zone
*zone
)
1373 unsigned long block_start_pfn
= zone
->zone_start_pfn
;
1374 unsigned long block_end_pfn
;
1376 block_end_pfn
= ALIGN(block_start_pfn
+ 1, pageblock_nr_pages
);
1377 for (; block_start_pfn
< zone_end_pfn(zone
);
1378 block_start_pfn
= block_end_pfn
,
1379 block_end_pfn
+= pageblock_nr_pages
) {
1381 block_end_pfn
= min(block_end_pfn
, zone_end_pfn(zone
));
1383 if (!__pageblock_pfn_to_page(block_start_pfn
,
1384 block_end_pfn
, zone
))
1388 /* We confirm that there is no hole */
1389 zone
->contiguous
= true;
1392 void clear_zone_contiguous(struct zone
*zone
)
1394 zone
->contiguous
= false;
1397 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1398 static void __init
deferred_free_range(struct page
*page
,
1399 unsigned long pfn
, int nr_pages
)
1406 /* Free a large naturally-aligned chunk if possible */
1407 if (nr_pages
== pageblock_nr_pages
&&
1408 (pfn
& (pageblock_nr_pages
- 1)) == 0) {
1409 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
1410 __free_pages_boot_core(page
, pageblock_order
);
1414 for (i
= 0; i
< nr_pages
; i
++, page
++, pfn
++) {
1415 if ((pfn
& (pageblock_nr_pages
- 1)) == 0)
1416 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
1417 __free_pages_boot_core(page
, 0);
1421 /* Completion tracking for deferred_init_memmap() threads */
1422 static atomic_t pgdat_init_n_undone __initdata
;
1423 static __initdata
DECLARE_COMPLETION(pgdat_init_all_done_comp
);
1425 static inline void __init
pgdat_init_report_one_done(void)
1427 if (atomic_dec_and_test(&pgdat_init_n_undone
))
1428 complete(&pgdat_init_all_done_comp
);
1431 /* Initialise remaining memory on a node */
1432 static int __init
deferred_init_memmap(void *data
)
1434 pg_data_t
*pgdat
= data
;
1435 int nid
= pgdat
->node_id
;
1436 struct mminit_pfnnid_cache nid_init_state
= { };
1437 unsigned long start
= jiffies
;
1438 unsigned long nr_pages
= 0;
1439 unsigned long walk_start
, walk_end
;
1442 unsigned long first_init_pfn
= pgdat
->first_deferred_pfn
;
1443 const struct cpumask
*cpumask
= cpumask_of_node(pgdat
->node_id
);
1445 if (first_init_pfn
== ULONG_MAX
) {
1446 pgdat_init_report_one_done();
1450 /* Bind memory initialisation thread to a local node if possible */
1451 if (!cpumask_empty(cpumask
))
1452 set_cpus_allowed_ptr(current
, cpumask
);
1454 /* Sanity check boundaries */
1455 BUG_ON(pgdat
->first_deferred_pfn
< pgdat
->node_start_pfn
);
1456 BUG_ON(pgdat
->first_deferred_pfn
> pgdat_end_pfn(pgdat
));
1457 pgdat
->first_deferred_pfn
= ULONG_MAX
;
1459 /* Only the highest zone is deferred so find it */
1460 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
1461 zone
= pgdat
->node_zones
+ zid
;
1462 if (first_init_pfn
< zone_end_pfn(zone
))
1466 for_each_mem_pfn_range(i
, nid
, &walk_start
, &walk_end
, NULL
) {
1467 unsigned long pfn
, end_pfn
;
1468 struct page
*page
= NULL
;
1469 struct page
*free_base_page
= NULL
;
1470 unsigned long free_base_pfn
= 0;
1473 end_pfn
= min(walk_end
, zone_end_pfn(zone
));
1474 pfn
= first_init_pfn
;
1475 if (pfn
< walk_start
)
1477 if (pfn
< zone
->zone_start_pfn
)
1478 pfn
= zone
->zone_start_pfn
;
1480 for (; pfn
< end_pfn
; pfn
++) {
1481 if (!pfn_valid_within(pfn
))
1485 * Ensure pfn_valid is checked every
1486 * pageblock_nr_pages for memory holes
1488 if ((pfn
& (pageblock_nr_pages
- 1)) == 0) {
1489 if (!pfn_valid(pfn
)) {
1495 if (!meminit_pfn_in_nid(pfn
, nid
, &nid_init_state
)) {
1500 /* Minimise pfn page lookups and scheduler checks */
1501 if (page
&& (pfn
& (pageblock_nr_pages
- 1)) != 0) {
1504 nr_pages
+= nr_to_free
;
1505 deferred_free_range(free_base_page
,
1506 free_base_pfn
, nr_to_free
);
1507 free_base_page
= NULL
;
1508 free_base_pfn
= nr_to_free
= 0;
1510 page
= pfn_to_page(pfn
);
1515 VM_BUG_ON(page_zone(page
) != zone
);
1519 __init_single_page(page
, pfn
, zid
, nid
);
1520 if (!free_base_page
) {
1521 free_base_page
= page
;
1522 free_base_pfn
= pfn
;
1527 /* Where possible, batch up pages for a single free */
1530 /* Free the current block of pages to allocator */
1531 nr_pages
+= nr_to_free
;
1532 deferred_free_range(free_base_page
, free_base_pfn
,
1534 free_base_page
= NULL
;
1535 free_base_pfn
= nr_to_free
= 0;
1537 /* Free the last block of pages to allocator */
1538 nr_pages
+= nr_to_free
;
1539 deferred_free_range(free_base_page
, free_base_pfn
, nr_to_free
);
1541 first_init_pfn
= max(end_pfn
, first_init_pfn
);
1544 /* Sanity check that the next zone really is unpopulated */
1545 WARN_ON(++zid
< MAX_NR_ZONES
&& populated_zone(++zone
));
1547 pr_info("node %d initialised, %lu pages in %ums\n", nid
, nr_pages
,
1548 jiffies_to_msecs(jiffies
- start
));
1550 pgdat_init_report_one_done();
1553 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1555 void __init
page_alloc_init_late(void)
1559 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1562 /* There will be num_node_state(N_MEMORY) threads */
1563 atomic_set(&pgdat_init_n_undone
, num_node_state(N_MEMORY
));
1564 for_each_node_state(nid
, N_MEMORY
) {
1565 kthread_run(deferred_init_memmap
, NODE_DATA(nid
), "pgdatinit%d", nid
);
1568 /* Block until all are initialised */
1569 wait_for_completion(&pgdat_init_all_done_comp
);
1571 /* Reinit limits that are based on free pages after the kernel is up */
1572 files_maxfiles_init();
1575 for_each_populated_zone(zone
)
1576 set_zone_contiguous(zone
);
1580 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1581 void __init
init_cma_reserved_pageblock(struct page
*page
)
1583 unsigned i
= pageblock_nr_pages
;
1584 struct page
*p
= page
;
1587 __ClearPageReserved(p
);
1588 set_page_count(p
, 0);
1591 set_pageblock_migratetype(page
, MIGRATE_CMA
);
1593 if (pageblock_order
>= MAX_ORDER
) {
1594 i
= pageblock_nr_pages
;
1597 set_page_refcounted(p
);
1598 __free_pages(p
, MAX_ORDER
- 1);
1599 p
+= MAX_ORDER_NR_PAGES
;
1600 } while (i
-= MAX_ORDER_NR_PAGES
);
1602 set_page_refcounted(page
);
1603 __free_pages(page
, pageblock_order
);
1606 adjust_managed_page_count(page
, pageblock_nr_pages
);
1611 * The order of subdivision here is critical for the IO subsystem.
1612 * Please do not alter this order without good reasons and regression
1613 * testing. Specifically, as large blocks of memory are subdivided,
1614 * the order in which smaller blocks are delivered depends on the order
1615 * they're subdivided in this function. This is the primary factor
1616 * influencing the order in which pages are delivered to the IO
1617 * subsystem according to empirical testing, and this is also justified
1618 * by considering the behavior of a buddy system containing a single
1619 * large block of memory acted on by a series of small allocations.
1620 * This behavior is a critical factor in sglist merging's success.
1624 static inline void expand(struct zone
*zone
, struct page
*page
,
1625 int low
, int high
, struct free_area
*area
,
1628 unsigned long size
= 1 << high
;
1630 while (high
> low
) {
1634 VM_BUG_ON_PAGE(bad_range(zone
, &page
[size
]), &page
[size
]);
1637 * Mark as guard pages (or page), that will allow to
1638 * merge back to allocator when buddy will be freed.
1639 * Corresponding page table entries will not be touched,
1640 * pages will stay not present in virtual address space
1642 if (set_page_guard(zone
, &page
[size
], high
, migratetype
))
1645 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
1647 set_page_order(&page
[size
], high
);
1651 static void check_new_page_bad(struct page
*page
)
1653 const char *bad_reason
= NULL
;
1654 unsigned long bad_flags
= 0;
1656 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
1657 bad_reason
= "nonzero mapcount";
1658 if (unlikely(page
->mapping
!= NULL
))
1659 bad_reason
= "non-NULL mapping";
1660 if (unlikely(page_ref_count(page
) != 0))
1661 bad_reason
= "nonzero _count";
1662 if (unlikely(page
->flags
& __PG_HWPOISON
)) {
1663 bad_reason
= "HWPoisoned (hardware-corrupted)";
1664 bad_flags
= __PG_HWPOISON
;
1665 /* Don't complain about hwpoisoned pages */
1666 page_mapcount_reset(page
); /* remove PageBuddy */
1669 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)) {
1670 bad_reason
= "PAGE_FLAGS_CHECK_AT_PREP flag set";
1671 bad_flags
= PAGE_FLAGS_CHECK_AT_PREP
;
1674 if (unlikely(page
->mem_cgroup
))
1675 bad_reason
= "page still charged to cgroup";
1677 bad_page(page
, bad_reason
, bad_flags
);
1681 * This page is about to be returned from the page allocator
1683 static inline int check_new_page(struct page
*page
)
1685 if (likely(page_expected_state(page
,
1686 PAGE_FLAGS_CHECK_AT_PREP
|__PG_HWPOISON
)))
1689 check_new_page_bad(page
);
1693 static inline bool free_pages_prezeroed(void)
1695 return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO
) &&
1696 page_poisoning_enabled();
1699 #ifdef CONFIG_DEBUG_VM
1700 static bool check_pcp_refill(struct page
*page
)
1705 static bool check_new_pcp(struct page
*page
)
1707 return check_new_page(page
);
1710 static bool check_pcp_refill(struct page
*page
)
1712 return check_new_page(page
);
1714 static bool check_new_pcp(struct page
*page
)
1718 #endif /* CONFIG_DEBUG_VM */
1720 static bool check_new_pages(struct page
*page
, unsigned int order
)
1723 for (i
= 0; i
< (1 << order
); i
++) {
1724 struct page
*p
= page
+ i
;
1726 if (unlikely(check_new_page(p
)))
1733 inline void post_alloc_hook(struct page
*page
, unsigned int order
,
1736 set_page_private(page
, 0);
1737 set_page_refcounted(page
);
1739 arch_alloc_page(page
, order
);
1740 kernel_map_pages(page
, 1 << order
, 1);
1741 kernel_poison_pages(page
, 1 << order
, 1);
1742 kasan_alloc_pages(page
, order
);
1743 set_page_owner(page
, order
, gfp_flags
);
1746 static void prep_new_page(struct page
*page
, unsigned int order
, gfp_t gfp_flags
,
1747 unsigned int alloc_flags
)
1751 post_alloc_hook(page
, order
, gfp_flags
);
1753 if (!free_pages_prezeroed() && (gfp_flags
& __GFP_ZERO
))
1754 for (i
= 0; i
< (1 << order
); i
++)
1755 clear_highpage(page
+ i
);
1757 if (order
&& (gfp_flags
& __GFP_COMP
))
1758 prep_compound_page(page
, order
);
1761 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1762 * allocate the page. The expectation is that the caller is taking
1763 * steps that will free more memory. The caller should avoid the page
1764 * being used for !PFMEMALLOC purposes.
1766 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
1767 set_page_pfmemalloc(page
);
1769 clear_page_pfmemalloc(page
);
1773 * Go through the free lists for the given migratetype and remove
1774 * the smallest available page from the freelists
1777 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
1780 unsigned int current_order
;
1781 struct free_area
*area
;
1784 /* Find a page of the appropriate size in the preferred list */
1785 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
1786 area
= &(zone
->free_area
[current_order
]);
1787 page
= list_first_entry_or_null(&area
->free_list
[migratetype
],
1791 list_del(&page
->lru
);
1792 rmv_page_order(page
);
1794 expand(zone
, page
, order
, current_order
, area
, migratetype
);
1795 set_pcppage_migratetype(page
, migratetype
);
1804 * This array describes the order lists are fallen back to when
1805 * the free lists for the desirable migrate type are depleted
1807 static int fallbacks
[MIGRATE_TYPES
][4] = {
1808 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1809 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1810 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_TYPES
},
1812 [MIGRATE_CMA
] = { MIGRATE_TYPES
}, /* Never used */
1814 #ifdef CONFIG_MEMORY_ISOLATION
1815 [MIGRATE_ISOLATE
] = { MIGRATE_TYPES
}, /* Never used */
1820 static struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1823 return __rmqueue_smallest(zone
, order
, MIGRATE_CMA
);
1826 static inline struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1827 unsigned int order
) { return NULL
; }
1831 * Move the free pages in a range to the free lists of the requested type.
1832 * Note that start_page and end_pages are not aligned on a pageblock
1833 * boundary. If alignment is required, use move_freepages_block()
1835 int move_freepages(struct zone
*zone
,
1836 struct page
*start_page
, struct page
*end_page
,
1841 int pages_moved
= 0;
1843 #ifndef CONFIG_HOLES_IN_ZONE
1845 * page_zone is not safe to call in this context when
1846 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1847 * anyway as we check zone boundaries in move_freepages_block().
1848 * Remove at a later date when no bug reports exist related to
1849 * grouping pages by mobility
1851 VM_BUG_ON(page_zone(start_page
) != page_zone(end_page
));
1854 for (page
= start_page
; page
<= end_page
;) {
1855 if (!pfn_valid_within(page_to_pfn(page
))) {
1860 /* Make sure we are not inadvertently changing nodes */
1861 VM_BUG_ON_PAGE(page_to_nid(page
) != zone_to_nid(zone
), page
);
1863 if (!PageBuddy(page
)) {
1868 order
= page_order(page
);
1869 list_move(&page
->lru
,
1870 &zone
->free_area
[order
].free_list
[migratetype
]);
1872 pages_moved
+= 1 << order
;
1878 int move_freepages_block(struct zone
*zone
, struct page
*page
,
1881 unsigned long start_pfn
, end_pfn
;
1882 struct page
*start_page
, *end_page
;
1884 start_pfn
= page_to_pfn(page
);
1885 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
1886 start_page
= pfn_to_page(start_pfn
);
1887 end_page
= start_page
+ pageblock_nr_pages
- 1;
1888 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
1890 /* Do not cross zone boundaries */
1891 if (!zone_spans_pfn(zone
, start_pfn
))
1893 if (!zone_spans_pfn(zone
, end_pfn
))
1896 return move_freepages(zone
, start_page
, end_page
, migratetype
);
1899 static void change_pageblock_range(struct page
*pageblock_page
,
1900 int start_order
, int migratetype
)
1902 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
1904 while (nr_pageblocks
--) {
1905 set_pageblock_migratetype(pageblock_page
, migratetype
);
1906 pageblock_page
+= pageblock_nr_pages
;
1911 * When we are falling back to another migratetype during allocation, try to
1912 * steal extra free pages from the same pageblocks to satisfy further
1913 * allocations, instead of polluting multiple pageblocks.
1915 * If we are stealing a relatively large buddy page, it is likely there will
1916 * be more free pages in the pageblock, so try to steal them all. For
1917 * reclaimable and unmovable allocations, we steal regardless of page size,
1918 * as fragmentation caused by those allocations polluting movable pageblocks
1919 * is worse than movable allocations stealing from unmovable and reclaimable
1922 static bool can_steal_fallback(unsigned int order
, int start_mt
)
1925 * Leaving this order check is intended, although there is
1926 * relaxed order check in next check. The reason is that
1927 * we can actually steal whole pageblock if this condition met,
1928 * but, below check doesn't guarantee it and that is just heuristic
1929 * so could be changed anytime.
1931 if (order
>= pageblock_order
)
1934 if (order
>= pageblock_order
/ 2 ||
1935 start_mt
== MIGRATE_RECLAIMABLE
||
1936 start_mt
== MIGRATE_UNMOVABLE
||
1937 page_group_by_mobility_disabled
)
1944 * This function implements actual steal behaviour. If order is large enough,
1945 * we can steal whole pageblock. If not, we first move freepages in this
1946 * pageblock and check whether half of pages are moved or not. If half of
1947 * pages are moved, we can change migratetype of pageblock and permanently
1948 * use it's pages as requested migratetype in the future.
1950 static void steal_suitable_fallback(struct zone
*zone
, struct page
*page
,
1953 unsigned int current_order
= page_order(page
);
1956 /* Take ownership for orders >= pageblock_order */
1957 if (current_order
>= pageblock_order
) {
1958 change_pageblock_range(page
, current_order
, start_type
);
1962 pages
= move_freepages_block(zone
, page
, start_type
);
1964 /* Claim the whole block if over half of it is free */
1965 if (pages
>= (1 << (pageblock_order
-1)) ||
1966 page_group_by_mobility_disabled
)
1967 set_pageblock_migratetype(page
, start_type
);
1971 * Check whether there is a suitable fallback freepage with requested order.
1972 * If only_stealable is true, this function returns fallback_mt only if
1973 * we can steal other freepages all together. This would help to reduce
1974 * fragmentation due to mixed migratetype pages in one pageblock.
1976 int find_suitable_fallback(struct free_area
*area
, unsigned int order
,
1977 int migratetype
, bool only_stealable
, bool *can_steal
)
1982 if (area
->nr_free
== 0)
1987 fallback_mt
= fallbacks
[migratetype
][i
];
1988 if (fallback_mt
== MIGRATE_TYPES
)
1991 if (list_empty(&area
->free_list
[fallback_mt
]))
1994 if (can_steal_fallback(order
, migratetype
))
1997 if (!only_stealable
)
2008 * Reserve a pageblock for exclusive use of high-order atomic allocations if
2009 * there are no empty page blocks that contain a page with a suitable order
2011 static void reserve_highatomic_pageblock(struct page
*page
, struct zone
*zone
,
2012 unsigned int alloc_order
)
2015 unsigned long max_managed
, flags
;
2018 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
2019 * Check is race-prone but harmless.
2021 max_managed
= (zone
->managed_pages
/ 100) + pageblock_nr_pages
;
2022 if (zone
->nr_reserved_highatomic
>= max_managed
)
2025 spin_lock_irqsave(&zone
->lock
, flags
);
2027 /* Recheck the nr_reserved_highatomic limit under the lock */
2028 if (zone
->nr_reserved_highatomic
>= max_managed
)
2032 mt
= get_pageblock_migratetype(page
);
2033 if (!is_migrate_highatomic(mt
) && !is_migrate_isolate(mt
)
2034 && !is_migrate_cma(mt
)) {
2035 zone
->nr_reserved_highatomic
+= pageblock_nr_pages
;
2036 set_pageblock_migratetype(page
, MIGRATE_HIGHATOMIC
);
2037 move_freepages_block(zone
, page
, MIGRATE_HIGHATOMIC
);
2041 spin_unlock_irqrestore(&zone
->lock
, flags
);
2045 * Used when an allocation is about to fail under memory pressure. This
2046 * potentially hurts the reliability of high-order allocations when under
2047 * intense memory pressure but failed atomic allocations should be easier
2048 * to recover from than an OOM.
2050 * If @force is true, try to unreserve a pageblock even though highatomic
2051 * pageblock is exhausted.
2053 static bool unreserve_highatomic_pageblock(const struct alloc_context
*ac
,
2056 struct zonelist
*zonelist
= ac
->zonelist
;
2057 unsigned long flags
;
2064 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
, ac
->high_zoneidx
,
2067 * Preserve at least one pageblock unless memory pressure
2070 if (!force
&& zone
->nr_reserved_highatomic
<=
2074 spin_lock_irqsave(&zone
->lock
, flags
);
2075 for (order
= 0; order
< MAX_ORDER
; order
++) {
2076 struct free_area
*area
= &(zone
->free_area
[order
]);
2078 page
= list_first_entry_or_null(
2079 &area
->free_list
[MIGRATE_HIGHATOMIC
],
2085 * In page freeing path, migratetype change is racy so
2086 * we can counter several free pages in a pageblock
2087 * in this loop althoug we changed the pageblock type
2088 * from highatomic to ac->migratetype. So we should
2089 * adjust the count once.
2091 if (is_migrate_highatomic_page(page
)) {
2093 * It should never happen but changes to
2094 * locking could inadvertently allow a per-cpu
2095 * drain to add pages to MIGRATE_HIGHATOMIC
2096 * while unreserving so be safe and watch for
2099 zone
->nr_reserved_highatomic
-= min(
2101 zone
->nr_reserved_highatomic
);
2105 * Convert to ac->migratetype and avoid the normal
2106 * pageblock stealing heuristics. Minimally, the caller
2107 * is doing the work and needs the pages. More
2108 * importantly, if the block was always converted to
2109 * MIGRATE_UNMOVABLE or another type then the number
2110 * of pageblocks that cannot be completely freed
2113 set_pageblock_migratetype(page
, ac
->migratetype
);
2114 ret
= move_freepages_block(zone
, page
, ac
->migratetype
);
2116 spin_unlock_irqrestore(&zone
->lock
, flags
);
2120 spin_unlock_irqrestore(&zone
->lock
, flags
);
2126 /* Remove an element from the buddy allocator from the fallback list */
2127 static inline struct page
*
2128 __rmqueue_fallback(struct zone
*zone
, unsigned int order
, int start_migratetype
)
2130 struct free_area
*area
;
2131 unsigned int current_order
;
2136 /* Find the largest possible block of pages in the other list */
2137 for (current_order
= MAX_ORDER
-1;
2138 current_order
>= order
&& current_order
<= MAX_ORDER
-1;
2140 area
= &(zone
->free_area
[current_order
]);
2141 fallback_mt
= find_suitable_fallback(area
, current_order
,
2142 start_migratetype
, false, &can_steal
);
2143 if (fallback_mt
== -1)
2146 page
= list_first_entry(&area
->free_list
[fallback_mt
],
2148 if (can_steal
&& !is_migrate_highatomic_page(page
))
2149 steal_suitable_fallback(zone
, page
, start_migratetype
);
2151 /* Remove the page from the freelists */
2153 list_del(&page
->lru
);
2154 rmv_page_order(page
);
2156 expand(zone
, page
, order
, current_order
, area
,
2159 * The pcppage_migratetype may differ from pageblock's
2160 * migratetype depending on the decisions in
2161 * find_suitable_fallback(). This is OK as long as it does not
2162 * differ for MIGRATE_CMA pageblocks. Those can be used as
2163 * fallback only via special __rmqueue_cma_fallback() function
2165 set_pcppage_migratetype(page
, start_migratetype
);
2167 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
2168 start_migratetype
, fallback_mt
);
2177 * Do the hard work of removing an element from the buddy allocator.
2178 * Call me with the zone->lock already held.
2180 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
2185 page
= __rmqueue_smallest(zone
, order
, migratetype
);
2186 if (unlikely(!page
)) {
2187 if (migratetype
== MIGRATE_MOVABLE
)
2188 page
= __rmqueue_cma_fallback(zone
, order
);
2191 page
= __rmqueue_fallback(zone
, order
, migratetype
);
2194 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
2199 * Obtain a specified number of elements from the buddy allocator, all under
2200 * a single hold of the lock, for efficiency. Add them to the supplied list.
2201 * Returns the number of new pages which were placed at *list.
2203 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
2204 unsigned long count
, struct list_head
*list
,
2205 int migratetype
, bool cold
)
2209 spin_lock(&zone
->lock
);
2210 for (i
= 0; i
< count
; ++i
) {
2211 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
2212 if (unlikely(page
== NULL
))
2215 if (unlikely(check_pcp_refill(page
)))
2219 * Split buddy pages returned by expand() are received here
2220 * in physical page order. The page is added to the callers and
2221 * list and the list head then moves forward. From the callers
2222 * perspective, the linked list is ordered by page number in
2223 * some conditions. This is useful for IO devices that can
2224 * merge IO requests if the physical pages are ordered
2228 list_add(&page
->lru
, list
);
2230 list_add_tail(&page
->lru
, list
);
2233 if (is_migrate_cma(get_pcppage_migratetype(page
)))
2234 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
,
2239 * i pages were removed from the buddy list even if some leak due
2240 * to check_pcp_refill failing so adjust NR_FREE_PAGES based
2241 * on i. Do not confuse with 'alloced' which is the number of
2242 * pages added to the pcp list.
2244 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
2245 spin_unlock(&zone
->lock
);
2251 * Called from the vmstat counter updater to drain pagesets of this
2252 * currently executing processor on remote nodes after they have
2255 * Note that this function must be called with the thread pinned to
2256 * a single processor.
2258 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
2260 unsigned long flags
;
2261 int to_drain
, batch
;
2263 local_irq_save(flags
);
2264 batch
= READ_ONCE(pcp
->batch
);
2265 to_drain
= min(pcp
->count
, batch
);
2267 free_pcppages_bulk(zone
, to_drain
, pcp
);
2268 pcp
->count
-= to_drain
;
2270 local_irq_restore(flags
);
2275 * Drain pcplists of the indicated processor and zone.
2277 * The processor must either be the current processor and the
2278 * thread pinned to the current processor or a processor that
2281 static void drain_pages_zone(unsigned int cpu
, struct zone
*zone
)
2283 unsigned long flags
;
2284 struct per_cpu_pageset
*pset
;
2285 struct per_cpu_pages
*pcp
;
2287 local_irq_save(flags
);
2288 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
2292 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
2295 local_irq_restore(flags
);
2299 * Drain pcplists of all zones on the indicated processor.
2301 * The processor must either be the current processor and the
2302 * thread pinned to the current processor or a processor that
2305 static void drain_pages(unsigned int cpu
)
2309 for_each_populated_zone(zone
) {
2310 drain_pages_zone(cpu
, zone
);
2315 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
2317 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2318 * the single zone's pages.
2320 void drain_local_pages(struct zone
*zone
)
2322 int cpu
= smp_processor_id();
2325 drain_pages_zone(cpu
, zone
);
2330 static void drain_local_pages_wq(struct work_struct
*work
)
2333 * drain_all_pages doesn't use proper cpu hotplug protection so
2334 * we can race with cpu offline when the WQ can move this from
2335 * a cpu pinned worker to an unbound one. We can operate on a different
2336 * cpu which is allright but we also have to make sure to not move to
2340 drain_local_pages(NULL
);
2345 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2347 * When zone parameter is non-NULL, spill just the single zone's pages.
2349 * Note that this can be extremely slow as the draining happens in a workqueue.
2351 void drain_all_pages(struct zone
*zone
)
2356 * Allocate in the BSS so we wont require allocation in
2357 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2359 static cpumask_t cpus_with_pcps
;
2362 * Make sure nobody triggers this path before mm_percpu_wq is fully
2365 if (WARN_ON_ONCE(!mm_percpu_wq
))
2368 /* Workqueues cannot recurse */
2369 if (current
->flags
& PF_WQ_WORKER
)
2373 * Do not drain if one is already in progress unless it's specific to
2374 * a zone. Such callers are primarily CMA and memory hotplug and need
2375 * the drain to be complete when the call returns.
2377 if (unlikely(!mutex_trylock(&pcpu_drain_mutex
))) {
2380 mutex_lock(&pcpu_drain_mutex
);
2384 * We don't care about racing with CPU hotplug event
2385 * as offline notification will cause the notified
2386 * cpu to drain that CPU pcps and on_each_cpu_mask
2387 * disables preemption as part of its processing
2389 for_each_online_cpu(cpu
) {
2390 struct per_cpu_pageset
*pcp
;
2392 bool has_pcps
= false;
2395 pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
2399 for_each_populated_zone(z
) {
2400 pcp
= per_cpu_ptr(z
->pageset
, cpu
);
2401 if (pcp
->pcp
.count
) {
2409 cpumask_set_cpu(cpu
, &cpus_with_pcps
);
2411 cpumask_clear_cpu(cpu
, &cpus_with_pcps
);
2414 for_each_cpu(cpu
, &cpus_with_pcps
) {
2415 struct work_struct
*work
= per_cpu_ptr(&pcpu_drain
, cpu
);
2416 INIT_WORK(work
, drain_local_pages_wq
);
2417 queue_work_on(cpu
, mm_percpu_wq
, work
);
2419 for_each_cpu(cpu
, &cpus_with_pcps
)
2420 flush_work(per_cpu_ptr(&pcpu_drain
, cpu
));
2422 mutex_unlock(&pcpu_drain_mutex
);
2425 #ifdef CONFIG_HIBERNATION
2427 void mark_free_pages(struct zone
*zone
)
2429 unsigned long pfn
, max_zone_pfn
;
2430 unsigned long flags
;
2431 unsigned int order
, t
;
2434 if (zone_is_empty(zone
))
2437 spin_lock_irqsave(&zone
->lock
, flags
);
2439 max_zone_pfn
= zone_end_pfn(zone
);
2440 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
2441 if (pfn_valid(pfn
)) {
2442 page
= pfn_to_page(pfn
);
2444 if (page_zone(page
) != zone
)
2447 if (!swsusp_page_is_forbidden(page
))
2448 swsusp_unset_page_free(page
);
2451 for_each_migratetype_order(order
, t
) {
2452 list_for_each_entry(page
,
2453 &zone
->free_area
[order
].free_list
[t
], lru
) {
2456 pfn
= page_to_pfn(page
);
2457 for (i
= 0; i
< (1UL << order
); i
++)
2458 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
2461 spin_unlock_irqrestore(&zone
->lock
, flags
);
2463 #endif /* CONFIG_PM */
2466 * Free a 0-order page
2467 * cold == true ? free a cold page : free a hot page
2469 void free_hot_cold_page(struct page
*page
, bool cold
)
2471 struct zone
*zone
= page_zone(page
);
2472 struct per_cpu_pages
*pcp
;
2473 unsigned long flags
;
2474 unsigned long pfn
= page_to_pfn(page
);
2477 if (!free_pcp_prepare(page
))
2480 migratetype
= get_pfnblock_migratetype(page
, pfn
);
2481 set_pcppage_migratetype(page
, migratetype
);
2482 local_irq_save(flags
);
2483 __count_vm_event(PGFREE
);
2486 * We only track unmovable, reclaimable and movable on pcp lists.
2487 * Free ISOLATE pages back to the allocator because they are being
2488 * offlined but treat HIGHATOMIC as movable pages so we can get those
2489 * areas back if necessary. Otherwise, we may have to free
2490 * excessively into the page allocator
2492 if (migratetype
>= MIGRATE_PCPTYPES
) {
2493 if (unlikely(is_migrate_isolate(migratetype
))) {
2494 free_one_page(zone
, page
, pfn
, 0, migratetype
);
2497 migratetype
= MIGRATE_MOVABLE
;
2500 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2502 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
2504 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
2506 if (pcp
->count
>= pcp
->high
) {
2507 unsigned long batch
= READ_ONCE(pcp
->batch
);
2508 free_pcppages_bulk(zone
, batch
, pcp
);
2509 pcp
->count
-= batch
;
2513 local_irq_restore(flags
);
2517 * Free a list of 0-order pages
2519 void free_hot_cold_page_list(struct list_head
*list
, bool cold
)
2521 struct page
*page
, *next
;
2523 list_for_each_entry_safe(page
, next
, list
, lru
) {
2524 trace_mm_page_free_batched(page
, cold
);
2525 free_hot_cold_page(page
, cold
);
2530 * split_page takes a non-compound higher-order page, and splits it into
2531 * n (1<<order) sub-pages: page[0..n]
2532 * Each sub-page must be freed individually.
2534 * Note: this is probably too low level an operation for use in drivers.
2535 * Please consult with lkml before using this in your driver.
2537 void split_page(struct page
*page
, unsigned int order
)
2541 VM_BUG_ON_PAGE(PageCompound(page
), page
);
2542 VM_BUG_ON_PAGE(!page_count(page
), page
);
2544 #ifdef CONFIG_KMEMCHECK
2546 * Split shadow pages too, because free(page[0]) would
2547 * otherwise free the whole shadow.
2549 if (kmemcheck_page_is_tracked(page
))
2550 split_page(virt_to_page(page
[0].shadow
), order
);
2553 for (i
= 1; i
< (1 << order
); i
++)
2554 set_page_refcounted(page
+ i
);
2555 split_page_owner(page
, order
);
2557 EXPORT_SYMBOL_GPL(split_page
);
2559 int __isolate_free_page(struct page
*page
, unsigned int order
)
2561 unsigned long watermark
;
2565 BUG_ON(!PageBuddy(page
));
2567 zone
= page_zone(page
);
2568 mt
= get_pageblock_migratetype(page
);
2570 if (!is_migrate_isolate(mt
)) {
2572 * Obey watermarks as if the page was being allocated. We can
2573 * emulate a high-order watermark check with a raised order-0
2574 * watermark, because we already know our high-order page
2577 watermark
= min_wmark_pages(zone
) + (1UL << order
);
2578 if (!zone_watermark_ok(zone
, 0, watermark
, 0, ALLOC_CMA
))
2581 __mod_zone_freepage_state(zone
, -(1UL << order
), mt
);
2584 /* Remove page from free list */
2585 list_del(&page
->lru
);
2586 zone
->free_area
[order
].nr_free
--;
2587 rmv_page_order(page
);
2590 * Set the pageblock if the isolated page is at least half of a
2593 if (order
>= pageblock_order
- 1) {
2594 struct page
*endpage
= page
+ (1 << order
) - 1;
2595 for (; page
< endpage
; page
+= pageblock_nr_pages
) {
2596 int mt
= get_pageblock_migratetype(page
);
2597 if (!is_migrate_isolate(mt
) && !is_migrate_cma(mt
)
2598 && !is_migrate_highatomic(mt
))
2599 set_pageblock_migratetype(page
,
2605 return 1UL << order
;
2609 * Update NUMA hit/miss statistics
2611 * Must be called with interrupts disabled.
2613 static inline void zone_statistics(struct zone
*preferred_zone
, struct zone
*z
)
2616 enum zone_stat_item local_stat
= NUMA_LOCAL
;
2618 if (z
->node
!= numa_node_id())
2619 local_stat
= NUMA_OTHER
;
2621 if (z
->node
== preferred_zone
->node
)
2622 __inc_zone_state(z
, NUMA_HIT
);
2624 __inc_zone_state(z
, NUMA_MISS
);
2625 __inc_zone_state(preferred_zone
, NUMA_FOREIGN
);
2627 __inc_zone_state(z
, local_stat
);
2631 /* Remove page from the per-cpu list, caller must protect the list */
2632 static struct page
*__rmqueue_pcplist(struct zone
*zone
, int migratetype
,
2633 bool cold
, struct per_cpu_pages
*pcp
,
2634 struct list_head
*list
)
2639 if (list_empty(list
)) {
2640 pcp
->count
+= rmqueue_bulk(zone
, 0,
2643 if (unlikely(list_empty(list
)))
2648 page
= list_last_entry(list
, struct page
, lru
);
2650 page
= list_first_entry(list
, struct page
, lru
);
2652 list_del(&page
->lru
);
2654 } while (check_new_pcp(page
));
2659 /* Lock and remove page from the per-cpu list */
2660 static struct page
*rmqueue_pcplist(struct zone
*preferred_zone
,
2661 struct zone
*zone
, unsigned int order
,
2662 gfp_t gfp_flags
, int migratetype
)
2664 struct per_cpu_pages
*pcp
;
2665 struct list_head
*list
;
2666 bool cold
= ((gfp_flags
& __GFP_COLD
) != 0);
2668 unsigned long flags
;
2670 local_irq_save(flags
);
2671 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2672 list
= &pcp
->lists
[migratetype
];
2673 page
= __rmqueue_pcplist(zone
, migratetype
, cold
, pcp
, list
);
2675 __count_zid_vm_events(PGALLOC
, page_zonenum(page
), 1 << order
);
2676 zone_statistics(preferred_zone
, zone
);
2678 local_irq_restore(flags
);
2683 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2686 struct page
*rmqueue(struct zone
*preferred_zone
,
2687 struct zone
*zone
, unsigned int order
,
2688 gfp_t gfp_flags
, unsigned int alloc_flags
,
2691 unsigned long flags
;
2694 if (likely(order
== 0)) {
2695 page
= rmqueue_pcplist(preferred_zone
, zone
, order
,
2696 gfp_flags
, migratetype
);
2701 * We most definitely don't want callers attempting to
2702 * allocate greater than order-1 page units with __GFP_NOFAIL.
2704 WARN_ON_ONCE((gfp_flags
& __GFP_NOFAIL
) && (order
> 1));
2705 spin_lock_irqsave(&zone
->lock
, flags
);
2709 if (alloc_flags
& ALLOC_HARDER
) {
2710 page
= __rmqueue_smallest(zone
, order
, MIGRATE_HIGHATOMIC
);
2712 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
2715 page
= __rmqueue(zone
, order
, migratetype
);
2716 } while (page
&& check_new_pages(page
, order
));
2717 spin_unlock(&zone
->lock
);
2720 __mod_zone_freepage_state(zone
, -(1 << order
),
2721 get_pcppage_migratetype(page
));
2723 __count_zid_vm_events(PGALLOC
, page_zonenum(page
), 1 << order
);
2724 zone_statistics(preferred_zone
, zone
);
2725 local_irq_restore(flags
);
2728 VM_BUG_ON_PAGE(page
&& bad_range(zone
, page
), page
);
2732 local_irq_restore(flags
);
2736 #ifdef CONFIG_FAIL_PAGE_ALLOC
2739 struct fault_attr attr
;
2741 bool ignore_gfp_highmem
;
2742 bool ignore_gfp_reclaim
;
2744 } fail_page_alloc
= {
2745 .attr
= FAULT_ATTR_INITIALIZER
,
2746 .ignore_gfp_reclaim
= true,
2747 .ignore_gfp_highmem
= true,
2751 static int __init
setup_fail_page_alloc(char *str
)
2753 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
2755 __setup("fail_page_alloc=", setup_fail_page_alloc
);
2757 static bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2759 if (order
< fail_page_alloc
.min_order
)
2761 if (gfp_mask
& __GFP_NOFAIL
)
2763 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
2765 if (fail_page_alloc
.ignore_gfp_reclaim
&&
2766 (gfp_mask
& __GFP_DIRECT_RECLAIM
))
2769 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
2772 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2774 static int __init
fail_page_alloc_debugfs(void)
2776 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
2779 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
2780 &fail_page_alloc
.attr
);
2782 return PTR_ERR(dir
);
2784 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
2785 &fail_page_alloc
.ignore_gfp_reclaim
))
2787 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
2788 &fail_page_alloc
.ignore_gfp_highmem
))
2790 if (!debugfs_create_u32("min-order", mode
, dir
,
2791 &fail_page_alloc
.min_order
))
2796 debugfs_remove_recursive(dir
);
2801 late_initcall(fail_page_alloc_debugfs
);
2803 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2805 #else /* CONFIG_FAIL_PAGE_ALLOC */
2807 static inline bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2812 #endif /* CONFIG_FAIL_PAGE_ALLOC */
2815 * Return true if free base pages are above 'mark'. For high-order checks it
2816 * will return true of the order-0 watermark is reached and there is at least
2817 * one free page of a suitable size. Checking now avoids taking the zone lock
2818 * to check in the allocation paths if no pages are free.
2820 bool __zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
2821 int classzone_idx
, unsigned int alloc_flags
,
2826 const bool alloc_harder
= (alloc_flags
& ALLOC_HARDER
);
2828 /* free_pages may go negative - that's OK */
2829 free_pages
-= (1 << order
) - 1;
2831 if (alloc_flags
& ALLOC_HIGH
)
2835 * If the caller does not have rights to ALLOC_HARDER then subtract
2836 * the high-atomic reserves. This will over-estimate the size of the
2837 * atomic reserve but it avoids a search.
2839 if (likely(!alloc_harder
))
2840 free_pages
-= z
->nr_reserved_highatomic
;
2845 /* If allocation can't use CMA areas don't use free CMA pages */
2846 if (!(alloc_flags
& ALLOC_CMA
))
2847 free_pages
-= zone_page_state(z
, NR_FREE_CMA_PAGES
);
2851 * Check watermarks for an order-0 allocation request. If these
2852 * are not met, then a high-order request also cannot go ahead
2853 * even if a suitable page happened to be free.
2855 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
2858 /* If this is an order-0 request then the watermark is fine */
2862 /* For a high-order request, check at least one suitable page is free */
2863 for (o
= order
; o
< MAX_ORDER
; o
++) {
2864 struct free_area
*area
= &z
->free_area
[o
];
2873 for (mt
= 0; mt
< MIGRATE_PCPTYPES
; mt
++) {
2874 if (!list_empty(&area
->free_list
[mt
]))
2879 if ((alloc_flags
& ALLOC_CMA
) &&
2880 !list_empty(&area
->free_list
[MIGRATE_CMA
])) {
2888 bool zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
2889 int classzone_idx
, unsigned int alloc_flags
)
2891 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
2892 zone_page_state(z
, NR_FREE_PAGES
));
2895 static inline bool zone_watermark_fast(struct zone
*z
, unsigned int order
,
2896 unsigned long mark
, int classzone_idx
, unsigned int alloc_flags
)
2898 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
2902 /* If allocation can't use CMA areas don't use free CMA pages */
2903 if (!(alloc_flags
& ALLOC_CMA
))
2904 cma_pages
= zone_page_state(z
, NR_FREE_CMA_PAGES
);
2908 * Fast check for order-0 only. If this fails then the reserves
2909 * need to be calculated. There is a corner case where the check
2910 * passes but only the high-order atomic reserve are free. If
2911 * the caller is !atomic then it'll uselessly search the free
2912 * list. That corner case is then slower but it is harmless.
2914 if (!order
&& (free_pages
- cma_pages
) > mark
+ z
->lowmem_reserve
[classzone_idx
])
2917 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
2921 bool zone_watermark_ok_safe(struct zone
*z
, unsigned int order
,
2922 unsigned long mark
, int classzone_idx
)
2924 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
2926 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
2927 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
2929 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, 0,
2934 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2936 return node_distance(zone_to_nid(local_zone
), zone_to_nid(zone
)) <=
2939 #else /* CONFIG_NUMA */
2940 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2944 #endif /* CONFIG_NUMA */
2947 * get_page_from_freelist goes through the zonelist trying to allocate
2950 static struct page
*
2951 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
, int alloc_flags
,
2952 const struct alloc_context
*ac
)
2954 struct zoneref
*z
= ac
->preferred_zoneref
;
2956 struct pglist_data
*last_pgdat_dirty_limit
= NULL
;
2959 * Scan zonelist, looking for a zone with enough free.
2960 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2962 for_next_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
, ac
->high_zoneidx
,
2967 if (cpusets_enabled() &&
2968 (alloc_flags
& ALLOC_CPUSET
) &&
2969 !__cpuset_zone_allowed(zone
, gfp_mask
))
2972 * When allocating a page cache page for writing, we
2973 * want to get it from a node that is within its dirty
2974 * limit, such that no single node holds more than its
2975 * proportional share of globally allowed dirty pages.
2976 * The dirty limits take into account the node's
2977 * lowmem reserves and high watermark so that kswapd
2978 * should be able to balance it without having to
2979 * write pages from its LRU list.
2981 * XXX: For now, allow allocations to potentially
2982 * exceed the per-node dirty limit in the slowpath
2983 * (spread_dirty_pages unset) before going into reclaim,
2984 * which is important when on a NUMA setup the allowed
2985 * nodes are together not big enough to reach the
2986 * global limit. The proper fix for these situations
2987 * will require awareness of nodes in the
2988 * dirty-throttling and the flusher threads.
2990 if (ac
->spread_dirty_pages
) {
2991 if (last_pgdat_dirty_limit
== zone
->zone_pgdat
)
2994 if (!node_dirty_ok(zone
->zone_pgdat
)) {
2995 last_pgdat_dirty_limit
= zone
->zone_pgdat
;
3000 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
3001 if (!zone_watermark_fast(zone
, order
, mark
,
3002 ac_classzone_idx(ac
), alloc_flags
)) {
3005 /* Checked here to keep the fast path fast */
3006 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
3007 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
3010 if (node_reclaim_mode
== 0 ||
3011 !zone_allows_reclaim(ac
->preferred_zoneref
->zone
, zone
))
3014 ret
= node_reclaim(zone
->zone_pgdat
, gfp_mask
, order
);
3016 case NODE_RECLAIM_NOSCAN
:
3019 case NODE_RECLAIM_FULL
:
3020 /* scanned but unreclaimable */
3023 /* did we reclaim enough */
3024 if (zone_watermark_ok(zone
, order
, mark
,
3025 ac_classzone_idx(ac
), alloc_flags
))
3033 page
= rmqueue(ac
->preferred_zoneref
->zone
, zone
, order
,
3034 gfp_mask
, alloc_flags
, ac
->migratetype
);
3036 prep_new_page(page
, order
, gfp_mask
, alloc_flags
);
3039 * If this is a high-order atomic allocation then check
3040 * if the pageblock should be reserved for the future
3042 if (unlikely(order
&& (alloc_flags
& ALLOC_HARDER
)))
3043 reserve_highatomic_pageblock(page
, zone
, order
);
3053 * Large machines with many possible nodes should not always dump per-node
3054 * meminfo in irq context.
3056 static inline bool should_suppress_show_mem(void)
3061 ret
= in_interrupt();
3066 static void warn_alloc_show_mem(gfp_t gfp_mask
, nodemask_t
*nodemask
)
3068 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
3069 static DEFINE_RATELIMIT_STATE(show_mem_rs
, HZ
, 1);
3071 if (should_suppress_show_mem() || !__ratelimit(&show_mem_rs
))
3075 * This documents exceptions given to allocations in certain
3076 * contexts that are allowed to allocate outside current's set
3079 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
3080 if (test_thread_flag(TIF_MEMDIE
) ||
3081 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
3082 filter
&= ~SHOW_MEM_FILTER_NODES
;
3083 if (in_interrupt() || !(gfp_mask
& __GFP_DIRECT_RECLAIM
))
3084 filter
&= ~SHOW_MEM_FILTER_NODES
;
3086 show_mem(filter
, nodemask
);
3089 void warn_alloc(gfp_t gfp_mask
, nodemask_t
*nodemask
, const char *fmt
, ...)
3091 struct va_format vaf
;
3093 static DEFINE_RATELIMIT_STATE(nopage_rs
, DEFAULT_RATELIMIT_INTERVAL
,
3094 DEFAULT_RATELIMIT_BURST
);
3096 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
))
3099 pr_warn("%s: ", current
->comm
);
3101 va_start(args
, fmt
);
3104 pr_cont("%pV", &vaf
);
3107 pr_cont(", mode:%#x(%pGg), nodemask=", gfp_mask
, &gfp_mask
);
3109 pr_cont("%*pbl\n", nodemask_pr_args(nodemask
));
3111 pr_cont("(null)\n");
3113 cpuset_print_current_mems_allowed();
3116 warn_alloc_show_mem(gfp_mask
, nodemask
);
3119 static inline struct page
*
3120 __alloc_pages_cpuset_fallback(gfp_t gfp_mask
, unsigned int order
,
3121 unsigned int alloc_flags
,
3122 const struct alloc_context
*ac
)
3126 page
= get_page_from_freelist(gfp_mask
, order
,
3127 alloc_flags
|ALLOC_CPUSET
, ac
);
3129 * fallback to ignore cpuset restriction if our nodes
3133 page
= get_page_from_freelist(gfp_mask
, order
,
3139 static inline struct page
*
3140 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
3141 const struct alloc_context
*ac
, unsigned long *did_some_progress
)
3143 struct oom_control oc
= {
3144 .zonelist
= ac
->zonelist
,
3145 .nodemask
= ac
->nodemask
,
3147 .gfp_mask
= gfp_mask
,
3152 *did_some_progress
= 0;
3155 * Acquire the oom lock. If that fails, somebody else is
3156 * making progress for us.
3158 if (!mutex_trylock(&oom_lock
)) {
3159 *did_some_progress
= 1;
3160 schedule_timeout_uninterruptible(1);
3165 * Go through the zonelist yet one more time, keep very high watermark
3166 * here, this is only to catch a parallel oom killing, we must fail if
3167 * we're still under heavy pressure.
3169 page
= get_page_from_freelist(gfp_mask
| __GFP_HARDWALL
, order
,
3170 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
, ac
);
3174 /* Coredumps can quickly deplete all memory reserves */
3175 if (current
->flags
& PF_DUMPCORE
)
3177 /* The OOM killer will not help higher order allocs */
3178 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
3180 /* The OOM killer does not needlessly kill tasks for lowmem */
3181 if (ac
->high_zoneidx
< ZONE_NORMAL
)
3183 if (pm_suspended_storage())
3186 * XXX: GFP_NOFS allocations should rather fail than rely on
3187 * other request to make a forward progress.
3188 * We are in an unfortunate situation where out_of_memory cannot
3189 * do much for this context but let's try it to at least get
3190 * access to memory reserved if the current task is killed (see
3191 * out_of_memory). Once filesystems are ready to handle allocation
3192 * failures more gracefully we should just bail out here.
3195 /* The OOM killer may not free memory on a specific node */
3196 if (gfp_mask
& __GFP_THISNODE
)
3199 /* Exhausted what can be done so it's blamo time */
3200 if (out_of_memory(&oc
) || WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
)) {
3201 *did_some_progress
= 1;
3204 * Help non-failing allocations by giving them access to memory
3207 if (gfp_mask
& __GFP_NOFAIL
)
3208 page
= __alloc_pages_cpuset_fallback(gfp_mask
, order
,
3209 ALLOC_NO_WATERMARKS
, ac
);
3212 mutex_unlock(&oom_lock
);
3217 * Maximum number of compaction retries wit a progress before OOM
3218 * killer is consider as the only way to move forward.
3220 #define MAX_COMPACT_RETRIES 16
3222 #ifdef CONFIG_COMPACTION
3223 /* Try memory compaction for high-order allocations before reclaim */
3224 static struct page
*
3225 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
3226 unsigned int alloc_flags
, const struct alloc_context
*ac
,
3227 enum compact_priority prio
, enum compact_result
*compact_result
)
3234 current
->flags
|= PF_MEMALLOC
;
3235 *compact_result
= try_to_compact_pages(gfp_mask
, order
, alloc_flags
, ac
,
3237 current
->flags
&= ~PF_MEMALLOC
;
3239 if (*compact_result
<= COMPACT_INACTIVE
)
3243 * At least in one zone compaction wasn't deferred or skipped, so let's
3244 * count a compaction stall
3246 count_vm_event(COMPACTSTALL
);
3248 page
= get_page_from_freelist(gfp_mask
, order
, alloc_flags
, ac
);
3251 struct zone
*zone
= page_zone(page
);
3253 zone
->compact_blockskip_flush
= false;
3254 compaction_defer_reset(zone
, order
, true);
3255 count_vm_event(COMPACTSUCCESS
);
3260 * It's bad if compaction run occurs and fails. The most likely reason
3261 * is that pages exist, but not enough to satisfy watermarks.
3263 count_vm_event(COMPACTFAIL
);
3271 should_compact_retry(struct alloc_context
*ac
, int order
, int alloc_flags
,
3272 enum compact_result compact_result
,
3273 enum compact_priority
*compact_priority
,
3274 int *compaction_retries
)
3276 int max_retries
= MAX_COMPACT_RETRIES
;
3279 int retries
= *compaction_retries
;
3280 enum compact_priority priority
= *compact_priority
;
3285 if (compaction_made_progress(compact_result
))
3286 (*compaction_retries
)++;
3289 * compaction considers all the zone as desperately out of memory
3290 * so it doesn't really make much sense to retry except when the
3291 * failure could be caused by insufficient priority
3293 if (compaction_failed(compact_result
))
3294 goto check_priority
;
3297 * make sure the compaction wasn't deferred or didn't bail out early
3298 * due to locks contention before we declare that we should give up.
3299 * But do not retry if the given zonelist is not suitable for
3302 if (compaction_withdrawn(compact_result
)) {
3303 ret
= compaction_zonelist_suitable(ac
, order
, alloc_flags
);
3308 * !costly requests are much more important than __GFP_REPEAT
3309 * costly ones because they are de facto nofail and invoke OOM
3310 * killer to move on while costly can fail and users are ready
3311 * to cope with that. 1/4 retries is rather arbitrary but we
3312 * would need much more detailed feedback from compaction to
3313 * make a better decision.
3315 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
3317 if (*compaction_retries
<= max_retries
) {
3323 * Make sure there are attempts at the highest priority if we exhausted
3324 * all retries or failed at the lower priorities.
3327 min_priority
= (order
> PAGE_ALLOC_COSTLY_ORDER
) ?
3328 MIN_COMPACT_COSTLY_PRIORITY
: MIN_COMPACT_PRIORITY
;
3330 if (*compact_priority
> min_priority
) {
3331 (*compact_priority
)--;
3332 *compaction_retries
= 0;
3336 trace_compact_retry(order
, priority
, compact_result
, retries
, max_retries
, ret
);
3340 static inline struct page
*
3341 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
3342 unsigned int alloc_flags
, const struct alloc_context
*ac
,
3343 enum compact_priority prio
, enum compact_result
*compact_result
)
3345 *compact_result
= COMPACT_SKIPPED
;
3350 should_compact_retry(struct alloc_context
*ac
, unsigned int order
, int alloc_flags
,
3351 enum compact_result compact_result
,
3352 enum compact_priority
*compact_priority
,
3353 int *compaction_retries
)
3358 if (!order
|| order
> PAGE_ALLOC_COSTLY_ORDER
)
3362 * There are setups with compaction disabled which would prefer to loop
3363 * inside the allocator rather than hit the oom killer prematurely.
3364 * Let's give them a good hope and keep retrying while the order-0
3365 * watermarks are OK.
3367 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
, ac
->high_zoneidx
,
3369 if (zone_watermark_ok(zone
, 0, min_wmark_pages(zone
),
3370 ac_classzone_idx(ac
), alloc_flags
))
3375 #endif /* CONFIG_COMPACTION */
3377 /* Perform direct synchronous page reclaim */
3379 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
,
3380 const struct alloc_context
*ac
)
3382 struct reclaim_state reclaim_state
;
3387 /* We now go into synchronous reclaim */
3388 cpuset_memory_pressure_bump();
3389 current
->flags
|= PF_MEMALLOC
;
3390 lockdep_set_current_reclaim_state(gfp_mask
);
3391 reclaim_state
.reclaimed_slab
= 0;
3392 current
->reclaim_state
= &reclaim_state
;
3394 progress
= try_to_free_pages(ac
->zonelist
, order
, gfp_mask
,
3397 current
->reclaim_state
= NULL
;
3398 lockdep_clear_current_reclaim_state();
3399 current
->flags
&= ~PF_MEMALLOC
;
3406 /* The really slow allocator path where we enter direct reclaim */
3407 static inline struct page
*
3408 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
3409 unsigned int alloc_flags
, const struct alloc_context
*ac
,
3410 unsigned long *did_some_progress
)
3412 struct page
*page
= NULL
;
3413 bool drained
= false;
3415 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, ac
);
3416 if (unlikely(!(*did_some_progress
)))
3420 page
= get_page_from_freelist(gfp_mask
, order
, alloc_flags
, ac
);
3423 * If an allocation failed after direct reclaim, it could be because
3424 * pages are pinned on the per-cpu lists or in high alloc reserves.
3425 * Shrink them them and try again
3427 if (!page
&& !drained
) {
3428 unreserve_highatomic_pageblock(ac
, false);
3429 drain_all_pages(NULL
);
3437 static void wake_all_kswapds(unsigned int order
, const struct alloc_context
*ac
)
3441 pg_data_t
*last_pgdat
= NULL
;
3443 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
,
3444 ac
->high_zoneidx
, ac
->nodemask
) {
3445 if (last_pgdat
!= zone
->zone_pgdat
)
3446 wakeup_kswapd(zone
, order
, ac
->high_zoneidx
);
3447 last_pgdat
= zone
->zone_pgdat
;
3451 static inline unsigned int
3452 gfp_to_alloc_flags(gfp_t gfp_mask
)
3454 unsigned int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
3456 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
3457 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
3460 * The caller may dip into page reserves a bit more if the caller
3461 * cannot run direct reclaim, or if the caller has realtime scheduling
3462 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
3463 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
3465 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
3467 if (gfp_mask
& __GFP_ATOMIC
) {
3469 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
3470 * if it can't schedule.
3472 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
3473 alloc_flags
|= ALLOC_HARDER
;
3475 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
3476 * comment for __cpuset_node_allowed().
3478 alloc_flags
&= ~ALLOC_CPUSET
;
3479 } else if (unlikely(rt_task(current
)) && !in_interrupt())
3480 alloc_flags
|= ALLOC_HARDER
;
3483 if (gfpflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
3484 alloc_flags
|= ALLOC_CMA
;
3489 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask
)
3491 if (unlikely(gfp_mask
& __GFP_NOMEMALLOC
))
3494 if (gfp_mask
& __GFP_MEMALLOC
)
3496 if (in_serving_softirq() && (current
->flags
& PF_MEMALLOC
))
3498 if (!in_interrupt() &&
3499 ((current
->flags
& PF_MEMALLOC
) ||
3500 unlikely(test_thread_flag(TIF_MEMDIE
))))
3507 * Checks whether it makes sense to retry the reclaim to make a forward progress
3508 * for the given allocation request.
3510 * We give up when we either have tried MAX_RECLAIM_RETRIES in a row
3511 * without success, or when we couldn't even meet the watermark if we
3512 * reclaimed all remaining pages on the LRU lists.
3514 * Returns true if a retry is viable or false to enter the oom path.
3517 should_reclaim_retry(gfp_t gfp_mask
, unsigned order
,
3518 struct alloc_context
*ac
, int alloc_flags
,
3519 bool did_some_progress
, int *no_progress_loops
)
3525 * Costly allocations might have made a progress but this doesn't mean
3526 * their order will become available due to high fragmentation so
3527 * always increment the no progress counter for them
3529 if (did_some_progress
&& order
<= PAGE_ALLOC_COSTLY_ORDER
)
3530 *no_progress_loops
= 0;
3532 (*no_progress_loops
)++;
3535 * Make sure we converge to OOM if we cannot make any progress
3536 * several times in the row.
3538 if (*no_progress_loops
> MAX_RECLAIM_RETRIES
) {
3539 /* Before OOM, exhaust highatomic_reserve */
3540 return unreserve_highatomic_pageblock(ac
, true);
3544 * Keep reclaiming pages while there is a chance this will lead
3545 * somewhere. If none of the target zones can satisfy our allocation
3546 * request even if all reclaimable pages are considered then we are
3547 * screwed and have to go OOM.
3549 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
, ac
->high_zoneidx
,
3551 unsigned long available
;
3552 unsigned long reclaimable
;
3553 unsigned long min_wmark
= min_wmark_pages(zone
);
3556 available
= reclaimable
= zone_reclaimable_pages(zone
);
3557 available
+= zone_page_state_snapshot(zone
, NR_FREE_PAGES
);
3560 * Would the allocation succeed if we reclaimed all
3561 * reclaimable pages?
3563 wmark
= __zone_watermark_ok(zone
, order
, min_wmark
,
3564 ac_classzone_idx(ac
), alloc_flags
, available
);
3565 trace_reclaim_retry_zone(z
, order
, reclaimable
,
3566 available
, min_wmark
, *no_progress_loops
, wmark
);
3569 * If we didn't make any progress and have a lot of
3570 * dirty + writeback pages then we should wait for
3571 * an IO to complete to slow down the reclaim and
3572 * prevent from pre mature OOM
3574 if (!did_some_progress
) {
3575 unsigned long write_pending
;
3577 write_pending
= zone_page_state_snapshot(zone
,
3578 NR_ZONE_WRITE_PENDING
);
3580 if (2 * write_pending
> reclaimable
) {
3581 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3587 * Memory allocation/reclaim might be called from a WQ
3588 * context and the current implementation of the WQ
3589 * concurrency control doesn't recognize that
3590 * a particular WQ is congested if the worker thread is
3591 * looping without ever sleeping. Therefore we have to
3592 * do a short sleep here rather than calling
3595 if (current
->flags
& PF_WQ_WORKER
)
3596 schedule_timeout_uninterruptible(1);
3607 static inline struct page
*
3608 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
3609 struct alloc_context
*ac
)
3611 bool can_direct_reclaim
= gfp_mask
& __GFP_DIRECT_RECLAIM
;
3612 struct page
*page
= NULL
;
3613 unsigned int alloc_flags
;
3614 unsigned long did_some_progress
;
3615 enum compact_priority compact_priority
;
3616 enum compact_result compact_result
;
3617 int compaction_retries
;
3618 int no_progress_loops
;
3619 unsigned long alloc_start
= jiffies
;
3620 unsigned int stall_timeout
= 10 * HZ
;
3621 unsigned int cpuset_mems_cookie
;
3624 * In the slowpath, we sanity check order to avoid ever trying to
3625 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3626 * be using allocators in order of preference for an area that is
3629 if (order
>= MAX_ORDER
) {
3630 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
3635 * We also sanity check to catch abuse of atomic reserves being used by
3636 * callers that are not in atomic context.
3638 if (WARN_ON_ONCE((gfp_mask
& (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)) ==
3639 (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)))
3640 gfp_mask
&= ~__GFP_ATOMIC
;
3643 compaction_retries
= 0;
3644 no_progress_loops
= 0;
3645 compact_priority
= DEF_COMPACT_PRIORITY
;
3646 cpuset_mems_cookie
= read_mems_allowed_begin();
3649 * The fast path uses conservative alloc_flags to succeed only until
3650 * kswapd needs to be woken up, and to avoid the cost of setting up
3651 * alloc_flags precisely. So we do that now.
3653 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
3656 * We need to recalculate the starting point for the zonelist iterator
3657 * because we might have used different nodemask in the fast path, or
3658 * there was a cpuset modification and we are retrying - otherwise we
3659 * could end up iterating over non-eligible zones endlessly.
3661 ac
->preferred_zoneref
= first_zones_zonelist(ac
->zonelist
,
3662 ac
->high_zoneidx
, ac
->nodemask
);
3663 if (!ac
->preferred_zoneref
->zone
)
3666 if (gfp_mask
& __GFP_KSWAPD_RECLAIM
)
3667 wake_all_kswapds(order
, ac
);
3670 * The adjusted alloc_flags might result in immediate success, so try
3673 page
= get_page_from_freelist(gfp_mask
, order
, alloc_flags
, ac
);
3678 * For costly allocations, try direct compaction first, as it's likely
3679 * that we have enough base pages and don't need to reclaim. Don't try
3680 * that for allocations that are allowed to ignore watermarks, as the
3681 * ALLOC_NO_WATERMARKS attempt didn't yet happen.
3683 if (can_direct_reclaim
&& order
> PAGE_ALLOC_COSTLY_ORDER
&&
3684 !gfp_pfmemalloc_allowed(gfp_mask
)) {
3685 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
3687 INIT_COMPACT_PRIORITY
,
3693 * Checks for costly allocations with __GFP_NORETRY, which
3694 * includes THP page fault allocations
3696 if (gfp_mask
& __GFP_NORETRY
) {
3698 * If compaction is deferred for high-order allocations,
3699 * it is because sync compaction recently failed. If
3700 * this is the case and the caller requested a THP
3701 * allocation, we do not want to heavily disrupt the
3702 * system, so we fail the allocation instead of entering
3705 if (compact_result
== COMPACT_DEFERRED
)
3709 * Looks like reclaim/compaction is worth trying, but
3710 * sync compaction could be very expensive, so keep
3711 * using async compaction.
3713 compact_priority
= INIT_COMPACT_PRIORITY
;
3718 /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */
3719 if (gfp_mask
& __GFP_KSWAPD_RECLAIM
)
3720 wake_all_kswapds(order
, ac
);
3722 if (gfp_pfmemalloc_allowed(gfp_mask
))
3723 alloc_flags
= ALLOC_NO_WATERMARKS
;
3726 * Reset the zonelist iterators if memory policies can be ignored.
3727 * These allocations are high priority and system rather than user
3730 if (!(alloc_flags
& ALLOC_CPUSET
) || (alloc_flags
& ALLOC_NO_WATERMARKS
)) {
3731 ac
->zonelist
= node_zonelist(numa_node_id(), gfp_mask
);
3732 ac
->preferred_zoneref
= first_zones_zonelist(ac
->zonelist
,
3733 ac
->high_zoneidx
, ac
->nodemask
);
3736 /* Attempt with potentially adjusted zonelist and alloc_flags */
3737 page
= get_page_from_freelist(gfp_mask
, order
, alloc_flags
, ac
);
3741 /* Caller is not willing to reclaim, we can't balance anything */
3742 if (!can_direct_reclaim
)
3745 /* Make sure we know about allocations which stall for too long */
3746 if (time_after(jiffies
, alloc_start
+ stall_timeout
)) {
3747 warn_alloc(gfp_mask
& ~__GFP_NOWARN
, ac
->nodemask
,
3748 "page allocation stalls for %ums, order:%u",
3749 jiffies_to_msecs(jiffies
-alloc_start
), order
);
3750 stall_timeout
+= 10 * HZ
;
3753 /* Avoid recursion of direct reclaim */
3754 if (current
->flags
& PF_MEMALLOC
)
3757 /* Try direct reclaim and then allocating */
3758 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
, alloc_flags
, ac
,
3759 &did_some_progress
);
3763 /* Try direct compaction and then allocating */
3764 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
, ac
,
3765 compact_priority
, &compact_result
);
3769 /* Do not loop if specifically requested */
3770 if (gfp_mask
& __GFP_NORETRY
)
3774 * Do not retry costly high order allocations unless they are
3777 if (order
> PAGE_ALLOC_COSTLY_ORDER
&& !(gfp_mask
& __GFP_REPEAT
))
3780 if (should_reclaim_retry(gfp_mask
, order
, ac
, alloc_flags
,
3781 did_some_progress
> 0, &no_progress_loops
))
3785 * It doesn't make any sense to retry for the compaction if the order-0
3786 * reclaim is not able to make any progress because the current
3787 * implementation of the compaction depends on the sufficient amount
3788 * of free memory (see __compaction_suitable)
3790 if (did_some_progress
> 0 &&
3791 should_compact_retry(ac
, order
, alloc_flags
,
3792 compact_result
, &compact_priority
,
3793 &compaction_retries
))
3797 * It's possible we raced with cpuset update so the OOM would be
3798 * premature (see below the nopage: label for full explanation).
3800 if (read_mems_allowed_retry(cpuset_mems_cookie
))
3803 /* Reclaim has failed us, start killing things */
3804 page
= __alloc_pages_may_oom(gfp_mask
, order
, ac
, &did_some_progress
);
3808 /* Avoid allocations with no watermarks from looping endlessly */
3809 if (test_thread_flag(TIF_MEMDIE
))
3812 /* Retry as long as the OOM killer is making progress */
3813 if (did_some_progress
) {
3814 no_progress_loops
= 0;
3820 * When updating a task's mems_allowed or mempolicy nodemask, it is
3821 * possible to race with parallel threads in such a way that our
3822 * allocation can fail while the mask is being updated. If we are about
3823 * to fail, check if the cpuset changed during allocation and if so,
3826 if (read_mems_allowed_retry(cpuset_mems_cookie
))
3830 * Make sure that __GFP_NOFAIL request doesn't leak out and make sure
3833 if (gfp_mask
& __GFP_NOFAIL
) {
3835 * All existing users of the __GFP_NOFAIL are blockable, so warn
3836 * of any new users that actually require GFP_NOWAIT
3838 if (WARN_ON_ONCE(!can_direct_reclaim
))
3842 * PF_MEMALLOC request from this context is rather bizarre
3843 * because we cannot reclaim anything and only can loop waiting
3844 * for somebody to do a work for us
3846 WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
);
3849 * non failing costly orders are a hard requirement which we
3850 * are not prepared for much so let's warn about these users
3851 * so that we can identify them and convert them to something
3854 WARN_ON_ONCE(order
> PAGE_ALLOC_COSTLY_ORDER
);
3857 * Help non-failing allocations by giving them access to memory
3858 * reserves but do not use ALLOC_NO_WATERMARKS because this
3859 * could deplete whole memory reserves which would just make
3860 * the situation worse
3862 page
= __alloc_pages_cpuset_fallback(gfp_mask
, order
, ALLOC_HARDER
, ac
);
3870 warn_alloc(gfp_mask
, ac
->nodemask
,
3871 "page allocation failure: order:%u", order
);
3876 static inline bool prepare_alloc_pages(gfp_t gfp_mask
, unsigned int order
,
3877 struct zonelist
*zonelist
, nodemask_t
*nodemask
,
3878 struct alloc_context
*ac
, gfp_t
*alloc_mask
,
3879 unsigned int *alloc_flags
)
3881 ac
->high_zoneidx
= gfp_zone(gfp_mask
);
3882 ac
->zonelist
= zonelist
;
3883 ac
->nodemask
= nodemask
;
3884 ac
->migratetype
= gfpflags_to_migratetype(gfp_mask
);
3886 if (cpusets_enabled()) {
3887 *alloc_mask
|= __GFP_HARDWALL
;
3889 ac
->nodemask
= &cpuset_current_mems_allowed
;
3891 *alloc_flags
|= ALLOC_CPUSET
;
3894 lockdep_trace_alloc(gfp_mask
);
3896 might_sleep_if(gfp_mask
& __GFP_DIRECT_RECLAIM
);
3898 if (should_fail_alloc_page(gfp_mask
, order
))
3901 if (IS_ENABLED(CONFIG_CMA
) && ac
->migratetype
== MIGRATE_MOVABLE
)
3902 *alloc_flags
|= ALLOC_CMA
;
3907 /* Determine whether to spread dirty pages and what the first usable zone */
3908 static inline void finalise_ac(gfp_t gfp_mask
,
3909 unsigned int order
, struct alloc_context
*ac
)
3911 /* Dirty zone balancing only done in the fast path */
3912 ac
->spread_dirty_pages
= (gfp_mask
& __GFP_WRITE
);
3915 * The preferred zone is used for statistics but crucially it is
3916 * also used as the starting point for the zonelist iterator. It
3917 * may get reset for allocations that ignore memory policies.
3919 ac
->preferred_zoneref
= first_zones_zonelist(ac
->zonelist
,
3920 ac
->high_zoneidx
, ac
->nodemask
);
3924 * This is the 'heart' of the zoned buddy allocator.
3927 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
3928 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
3931 unsigned int alloc_flags
= ALLOC_WMARK_LOW
;
3932 gfp_t alloc_mask
= gfp_mask
; /* The gfp_t that was actually used for allocation */
3933 struct alloc_context ac
= { };
3935 gfp_mask
&= gfp_allowed_mask
;
3936 if (!prepare_alloc_pages(gfp_mask
, order
, zonelist
, nodemask
, &ac
, &alloc_mask
, &alloc_flags
))
3939 finalise_ac(gfp_mask
, order
, &ac
);
3941 /* First allocation attempt */
3942 page
= get_page_from_freelist(alloc_mask
, order
, alloc_flags
, &ac
);
3947 * Apply scoped allocation constraints. This is mainly about GFP_NOFS
3948 * resp. GFP_NOIO which has to be inherited for all allocation requests
3949 * from a particular context which has been marked by
3950 * memalloc_no{fs,io}_{save,restore}.
3952 alloc_mask
= current_gfp_context(gfp_mask
);
3953 ac
.spread_dirty_pages
= false;
3956 * Restore the original nodemask if it was potentially replaced with
3957 * &cpuset_current_mems_allowed to optimize the fast-path attempt.
3959 if (unlikely(ac
.nodemask
!= nodemask
))
3960 ac
.nodemask
= nodemask
;
3962 page
= __alloc_pages_slowpath(alloc_mask
, order
, &ac
);
3965 if (memcg_kmem_enabled() && (gfp_mask
& __GFP_ACCOUNT
) && page
&&
3966 unlikely(memcg_kmem_charge(page
, gfp_mask
, order
) != 0)) {
3967 __free_pages(page
, order
);
3971 if (kmemcheck_enabled
&& page
)
3972 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
3974 trace_mm_page_alloc(page
, order
, alloc_mask
, ac
.migratetype
);
3978 EXPORT_SYMBOL(__alloc_pages_nodemask
);
3981 * Common helper functions.
3983 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
3988 * __get_free_pages() returns a 32-bit address, which cannot represent
3991 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
3993 page
= alloc_pages(gfp_mask
, order
);
3996 return (unsigned long) page_address(page
);
3998 EXPORT_SYMBOL(__get_free_pages
);
4000 unsigned long get_zeroed_page(gfp_t gfp_mask
)
4002 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
4004 EXPORT_SYMBOL(get_zeroed_page
);
4006 void __free_pages(struct page
*page
, unsigned int order
)
4008 if (put_page_testzero(page
)) {
4010 free_hot_cold_page(page
, false);
4012 __free_pages_ok(page
, order
);
4016 EXPORT_SYMBOL(__free_pages
);
4018 void free_pages(unsigned long addr
, unsigned int order
)
4021 VM_BUG_ON(!virt_addr_valid((void *)addr
));
4022 __free_pages(virt_to_page((void *)addr
), order
);
4026 EXPORT_SYMBOL(free_pages
);
4030 * An arbitrary-length arbitrary-offset area of memory which resides
4031 * within a 0 or higher order page. Multiple fragments within that page
4032 * are individually refcounted, in the page's reference counter.
4034 * The page_frag functions below provide a simple allocation framework for
4035 * page fragments. This is used by the network stack and network device
4036 * drivers to provide a backing region of memory for use as either an
4037 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
4039 static struct page
*__page_frag_cache_refill(struct page_frag_cache
*nc
,
4042 struct page
*page
= NULL
;
4043 gfp_t gfp
= gfp_mask
;
4045 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4046 gfp_mask
|= __GFP_COMP
| __GFP_NOWARN
| __GFP_NORETRY
|
4048 page
= alloc_pages_node(NUMA_NO_NODE
, gfp_mask
,
4049 PAGE_FRAG_CACHE_MAX_ORDER
);
4050 nc
->size
= page
? PAGE_FRAG_CACHE_MAX_SIZE
: PAGE_SIZE
;
4052 if (unlikely(!page
))
4053 page
= alloc_pages_node(NUMA_NO_NODE
, gfp
, 0);
4055 nc
->va
= page
? page_address(page
) : NULL
;
4060 void __page_frag_cache_drain(struct page
*page
, unsigned int count
)
4062 VM_BUG_ON_PAGE(page_ref_count(page
) == 0, page
);
4064 if (page_ref_sub_and_test(page
, count
)) {
4065 unsigned int order
= compound_order(page
);
4068 free_hot_cold_page(page
, false);
4070 __free_pages_ok(page
, order
);
4073 EXPORT_SYMBOL(__page_frag_cache_drain
);
4075 void *page_frag_alloc(struct page_frag_cache
*nc
,
4076 unsigned int fragsz
, gfp_t gfp_mask
)
4078 unsigned int size
= PAGE_SIZE
;
4082 if (unlikely(!nc
->va
)) {
4084 page
= __page_frag_cache_refill(nc
, gfp_mask
);
4088 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4089 /* if size can vary use size else just use PAGE_SIZE */
4092 /* Even if we own the page, we do not use atomic_set().
4093 * This would break get_page_unless_zero() users.
4095 page_ref_add(page
, size
- 1);
4097 /* reset page count bias and offset to start of new frag */
4098 nc
->pfmemalloc
= page_is_pfmemalloc(page
);
4099 nc
->pagecnt_bias
= size
;
4103 offset
= nc
->offset
- fragsz
;
4104 if (unlikely(offset
< 0)) {
4105 page
= virt_to_page(nc
->va
);
4107 if (!page_ref_sub_and_test(page
, nc
->pagecnt_bias
))
4110 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4111 /* if size can vary use size else just use PAGE_SIZE */
4114 /* OK, page count is 0, we can safely set it */
4115 set_page_count(page
, size
);
4117 /* reset page count bias and offset to start of new frag */
4118 nc
->pagecnt_bias
= size
;
4119 offset
= size
- fragsz
;
4123 nc
->offset
= offset
;
4125 return nc
->va
+ offset
;
4127 EXPORT_SYMBOL(page_frag_alloc
);
4130 * Frees a page fragment allocated out of either a compound or order 0 page.
4132 void page_frag_free(void *addr
)
4134 struct page
*page
= virt_to_head_page(addr
);
4136 if (unlikely(put_page_testzero(page
)))
4137 __free_pages_ok(page
, compound_order(page
));
4139 EXPORT_SYMBOL(page_frag_free
);
4141 static void *make_alloc_exact(unsigned long addr
, unsigned int order
,
4145 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
4146 unsigned long used
= addr
+ PAGE_ALIGN(size
);
4148 split_page(virt_to_page((void *)addr
), order
);
4149 while (used
< alloc_end
) {
4154 return (void *)addr
;
4158 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
4159 * @size: the number of bytes to allocate
4160 * @gfp_mask: GFP flags for the allocation
4162 * This function is similar to alloc_pages(), except that it allocates the
4163 * minimum number of pages to satisfy the request. alloc_pages() can only
4164 * allocate memory in power-of-two pages.
4166 * This function is also limited by MAX_ORDER.
4168 * Memory allocated by this function must be released by free_pages_exact().
4170 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
4172 unsigned int order
= get_order(size
);
4175 addr
= __get_free_pages(gfp_mask
, order
);
4176 return make_alloc_exact(addr
, order
, size
);
4178 EXPORT_SYMBOL(alloc_pages_exact
);
4181 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
4183 * @nid: the preferred node ID where memory should be allocated
4184 * @size: the number of bytes to allocate
4185 * @gfp_mask: GFP flags for the allocation
4187 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
4190 void * __meminit
alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
4192 unsigned int order
= get_order(size
);
4193 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
4196 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
4200 * free_pages_exact - release memory allocated via alloc_pages_exact()
4201 * @virt: the value returned by alloc_pages_exact.
4202 * @size: size of allocation, same value as passed to alloc_pages_exact().
4204 * Release the memory allocated by a previous call to alloc_pages_exact.
4206 void free_pages_exact(void *virt
, size_t size
)
4208 unsigned long addr
= (unsigned long)virt
;
4209 unsigned long end
= addr
+ PAGE_ALIGN(size
);
4211 while (addr
< end
) {
4216 EXPORT_SYMBOL(free_pages_exact
);
4219 * nr_free_zone_pages - count number of pages beyond high watermark
4220 * @offset: The zone index of the highest zone
4222 * nr_free_zone_pages() counts the number of counts pages which are beyond the
4223 * high watermark within all zones at or below a given zone index. For each
4224 * zone, the number of pages is calculated as:
4226 * nr_free_zone_pages = managed_pages - high_pages
4228 static unsigned long nr_free_zone_pages(int offset
)
4233 /* Just pick one node, since fallback list is circular */
4234 unsigned long sum
= 0;
4236 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
4238 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
4239 unsigned long size
= zone
->managed_pages
;
4240 unsigned long high
= high_wmark_pages(zone
);
4249 * nr_free_buffer_pages - count number of pages beyond high watermark
4251 * nr_free_buffer_pages() counts the number of pages which are beyond the high
4252 * watermark within ZONE_DMA and ZONE_NORMAL.
4254 unsigned long nr_free_buffer_pages(void)
4256 return nr_free_zone_pages(gfp_zone(GFP_USER
));
4258 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
4261 * nr_free_pagecache_pages - count number of pages beyond high watermark
4263 * nr_free_pagecache_pages() counts the number of pages which are beyond the
4264 * high watermark within all zones.
4266 unsigned long nr_free_pagecache_pages(void)
4268 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
4271 static inline void show_node(struct zone
*zone
)
4273 if (IS_ENABLED(CONFIG_NUMA
))
4274 printk("Node %d ", zone_to_nid(zone
));
4277 long si_mem_available(void)
4280 unsigned long pagecache
;
4281 unsigned long wmark_low
= 0;
4282 unsigned long pages
[NR_LRU_LISTS
];
4286 for (lru
= LRU_BASE
; lru
< NR_LRU_LISTS
; lru
++)
4287 pages
[lru
] = global_node_page_state(NR_LRU_BASE
+ lru
);
4290 wmark_low
+= zone
->watermark
[WMARK_LOW
];
4293 * Estimate the amount of memory available for userspace allocations,
4294 * without causing swapping.
4296 available
= global_page_state(NR_FREE_PAGES
) - totalreserve_pages
;
4299 * Not all the page cache can be freed, otherwise the system will
4300 * start swapping. Assume at least half of the page cache, or the
4301 * low watermark worth of cache, needs to stay.
4303 pagecache
= pages
[LRU_ACTIVE_FILE
] + pages
[LRU_INACTIVE_FILE
];
4304 pagecache
-= min(pagecache
/ 2, wmark_low
);
4305 available
+= pagecache
;
4308 * Part of the reclaimable slab consists of items that are in use,
4309 * and cannot be freed. Cap this estimate at the low watermark.
4311 available
+= global_page_state(NR_SLAB_RECLAIMABLE
) -
4312 min(global_page_state(NR_SLAB_RECLAIMABLE
) / 2, wmark_low
);
4318 EXPORT_SYMBOL_GPL(si_mem_available
);
4320 void si_meminfo(struct sysinfo
*val
)
4322 val
->totalram
= totalram_pages
;
4323 val
->sharedram
= global_node_page_state(NR_SHMEM
);
4324 val
->freeram
= global_page_state(NR_FREE_PAGES
);
4325 val
->bufferram
= nr_blockdev_pages();
4326 val
->totalhigh
= totalhigh_pages
;
4327 val
->freehigh
= nr_free_highpages();
4328 val
->mem_unit
= PAGE_SIZE
;
4331 EXPORT_SYMBOL(si_meminfo
);
4334 void si_meminfo_node(struct sysinfo
*val
, int nid
)
4336 int zone_type
; /* needs to be signed */
4337 unsigned long managed_pages
= 0;
4338 unsigned long managed_highpages
= 0;
4339 unsigned long free_highpages
= 0;
4340 pg_data_t
*pgdat
= NODE_DATA(nid
);
4342 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++)
4343 managed_pages
+= pgdat
->node_zones
[zone_type
].managed_pages
;
4344 val
->totalram
= managed_pages
;
4345 val
->sharedram
= node_page_state(pgdat
, NR_SHMEM
);
4346 val
->freeram
= sum_zone_node_page_state(nid
, NR_FREE_PAGES
);
4347 #ifdef CONFIG_HIGHMEM
4348 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
4349 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4351 if (is_highmem(zone
)) {
4352 managed_highpages
+= zone
->managed_pages
;
4353 free_highpages
+= zone_page_state(zone
, NR_FREE_PAGES
);
4356 val
->totalhigh
= managed_highpages
;
4357 val
->freehigh
= free_highpages
;
4359 val
->totalhigh
= managed_highpages
;
4360 val
->freehigh
= free_highpages
;
4362 val
->mem_unit
= PAGE_SIZE
;
4367 * Determine whether the node should be displayed or not, depending on whether
4368 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
4370 static bool show_mem_node_skip(unsigned int flags
, int nid
, nodemask_t
*nodemask
)
4372 if (!(flags
& SHOW_MEM_FILTER_NODES
))
4376 * no node mask - aka implicit memory numa policy. Do not bother with
4377 * the synchronization - read_mems_allowed_begin - because we do not
4378 * have to be precise here.
4381 nodemask
= &cpuset_current_mems_allowed
;
4383 return !node_isset(nid
, *nodemask
);
4386 #define K(x) ((x) << (PAGE_SHIFT-10))
4388 static void show_migration_types(unsigned char type
)
4390 static const char types
[MIGRATE_TYPES
] = {
4391 [MIGRATE_UNMOVABLE
] = 'U',
4392 [MIGRATE_MOVABLE
] = 'M',
4393 [MIGRATE_RECLAIMABLE
] = 'E',
4394 [MIGRATE_HIGHATOMIC
] = 'H',
4396 [MIGRATE_CMA
] = 'C',
4398 #ifdef CONFIG_MEMORY_ISOLATION
4399 [MIGRATE_ISOLATE
] = 'I',
4402 char tmp
[MIGRATE_TYPES
+ 1];
4406 for (i
= 0; i
< MIGRATE_TYPES
; i
++) {
4407 if (type
& (1 << i
))
4412 printk(KERN_CONT
"(%s) ", tmp
);
4416 * Show free area list (used inside shift_scroll-lock stuff)
4417 * We also calculate the percentage fragmentation. We do this by counting the
4418 * memory on each free list with the exception of the first item on the list.
4421 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
4424 void show_free_areas(unsigned int filter
, nodemask_t
*nodemask
)
4426 unsigned long free_pcp
= 0;
4431 for_each_populated_zone(zone
) {
4432 if (show_mem_node_skip(filter
, zone_to_nid(zone
), nodemask
))
4435 for_each_online_cpu(cpu
)
4436 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
4439 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
4440 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
4441 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
4442 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
4443 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
4444 " free:%lu free_pcp:%lu free_cma:%lu\n",
4445 global_node_page_state(NR_ACTIVE_ANON
),
4446 global_node_page_state(NR_INACTIVE_ANON
),
4447 global_node_page_state(NR_ISOLATED_ANON
),
4448 global_node_page_state(NR_ACTIVE_FILE
),
4449 global_node_page_state(NR_INACTIVE_FILE
),
4450 global_node_page_state(NR_ISOLATED_FILE
),
4451 global_node_page_state(NR_UNEVICTABLE
),
4452 global_node_page_state(NR_FILE_DIRTY
),
4453 global_node_page_state(NR_WRITEBACK
),
4454 global_node_page_state(NR_UNSTABLE_NFS
),
4455 global_page_state(NR_SLAB_RECLAIMABLE
),
4456 global_page_state(NR_SLAB_UNRECLAIMABLE
),
4457 global_node_page_state(NR_FILE_MAPPED
),
4458 global_node_page_state(NR_SHMEM
),
4459 global_page_state(NR_PAGETABLE
),
4460 global_page_state(NR_BOUNCE
),
4461 global_page_state(NR_FREE_PAGES
),
4463 global_page_state(NR_FREE_CMA_PAGES
));
4465 for_each_online_pgdat(pgdat
) {
4466 if (show_mem_node_skip(filter
, pgdat
->node_id
, nodemask
))
4470 " active_anon:%lukB"
4471 " inactive_anon:%lukB"
4472 " active_file:%lukB"
4473 " inactive_file:%lukB"
4474 " unevictable:%lukB"
4475 " isolated(anon):%lukB"
4476 " isolated(file):%lukB"
4481 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4483 " shmem_pmdmapped: %lukB"
4486 " writeback_tmp:%lukB"
4488 " all_unreclaimable? %s"
4491 K(node_page_state(pgdat
, NR_ACTIVE_ANON
)),
4492 K(node_page_state(pgdat
, NR_INACTIVE_ANON
)),
4493 K(node_page_state(pgdat
, NR_ACTIVE_FILE
)),
4494 K(node_page_state(pgdat
, NR_INACTIVE_FILE
)),
4495 K(node_page_state(pgdat
, NR_UNEVICTABLE
)),
4496 K(node_page_state(pgdat
, NR_ISOLATED_ANON
)),
4497 K(node_page_state(pgdat
, NR_ISOLATED_FILE
)),
4498 K(node_page_state(pgdat
, NR_FILE_MAPPED
)),
4499 K(node_page_state(pgdat
, NR_FILE_DIRTY
)),
4500 K(node_page_state(pgdat
, NR_WRITEBACK
)),
4501 K(node_page_state(pgdat
, NR_SHMEM
)),
4502 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4503 K(node_page_state(pgdat
, NR_SHMEM_THPS
) * HPAGE_PMD_NR
),
4504 K(node_page_state(pgdat
, NR_SHMEM_PMDMAPPED
)
4506 K(node_page_state(pgdat
, NR_ANON_THPS
) * HPAGE_PMD_NR
),
4508 K(node_page_state(pgdat
, NR_WRITEBACK_TEMP
)),
4509 K(node_page_state(pgdat
, NR_UNSTABLE_NFS
)),
4510 pgdat
->kswapd_failures
>= MAX_RECLAIM_RETRIES
?
4514 for_each_populated_zone(zone
) {
4517 if (show_mem_node_skip(filter
, zone_to_nid(zone
), nodemask
))
4521 for_each_online_cpu(cpu
)
4522 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
4531 " active_anon:%lukB"
4532 " inactive_anon:%lukB"
4533 " active_file:%lukB"
4534 " inactive_file:%lukB"
4535 " unevictable:%lukB"
4536 " writepending:%lukB"
4540 " slab_reclaimable:%lukB"
4541 " slab_unreclaimable:%lukB"
4542 " kernel_stack:%lukB"
4550 K(zone_page_state(zone
, NR_FREE_PAGES
)),
4551 K(min_wmark_pages(zone
)),
4552 K(low_wmark_pages(zone
)),
4553 K(high_wmark_pages(zone
)),
4554 K(zone_page_state(zone
, NR_ZONE_ACTIVE_ANON
)),
4555 K(zone_page_state(zone
, NR_ZONE_INACTIVE_ANON
)),
4556 K(zone_page_state(zone
, NR_ZONE_ACTIVE_FILE
)),
4557 K(zone_page_state(zone
, NR_ZONE_INACTIVE_FILE
)),
4558 K(zone_page_state(zone
, NR_ZONE_UNEVICTABLE
)),
4559 K(zone_page_state(zone
, NR_ZONE_WRITE_PENDING
)),
4560 K(zone
->present_pages
),
4561 K(zone
->managed_pages
),
4562 K(zone_page_state(zone
, NR_MLOCK
)),
4563 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
4564 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
4565 zone_page_state(zone
, NR_KERNEL_STACK_KB
),
4566 K(zone_page_state(zone
, NR_PAGETABLE
)),
4567 K(zone_page_state(zone
, NR_BOUNCE
)),
4569 K(this_cpu_read(zone
->pageset
->pcp
.count
)),
4570 K(zone_page_state(zone
, NR_FREE_CMA_PAGES
)));
4571 printk("lowmem_reserve[]:");
4572 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4573 printk(KERN_CONT
" %ld", zone
->lowmem_reserve
[i
]);
4574 printk(KERN_CONT
"\n");
4577 for_each_populated_zone(zone
) {
4579 unsigned long nr
[MAX_ORDER
], flags
, total
= 0;
4580 unsigned char types
[MAX_ORDER
];
4582 if (show_mem_node_skip(filter
, zone_to_nid(zone
), nodemask
))
4585 printk(KERN_CONT
"%s: ", zone
->name
);
4587 spin_lock_irqsave(&zone
->lock
, flags
);
4588 for (order
= 0; order
< MAX_ORDER
; order
++) {
4589 struct free_area
*area
= &zone
->free_area
[order
];
4592 nr
[order
] = area
->nr_free
;
4593 total
+= nr
[order
] << order
;
4596 for (type
= 0; type
< MIGRATE_TYPES
; type
++) {
4597 if (!list_empty(&area
->free_list
[type
]))
4598 types
[order
] |= 1 << type
;
4601 spin_unlock_irqrestore(&zone
->lock
, flags
);
4602 for (order
= 0; order
< MAX_ORDER
; order
++) {
4603 printk(KERN_CONT
"%lu*%lukB ",
4604 nr
[order
], K(1UL) << order
);
4606 show_migration_types(types
[order
]);
4608 printk(KERN_CONT
"= %lukB\n", K(total
));
4611 hugetlb_show_meminfo();
4613 printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES
));
4615 show_swap_cache_info();
4618 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
4620 zoneref
->zone
= zone
;
4621 zoneref
->zone_idx
= zone_idx(zone
);
4625 * Builds allocation fallback zone lists.
4627 * Add all populated zones of a node to the zonelist.
4629 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
4633 enum zone_type zone_type
= MAX_NR_ZONES
;
4637 zone
= pgdat
->node_zones
+ zone_type
;
4638 if (managed_zone(zone
)) {
4639 zoneref_set_zone(zone
,
4640 &zonelist
->_zonerefs
[nr_zones
++]);
4641 check_highest_zone(zone_type
);
4643 } while (zone_type
);
4651 * 0 = automatic detection of better ordering.
4652 * 1 = order by ([node] distance, -zonetype)
4653 * 2 = order by (-zonetype, [node] distance)
4655 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
4656 * the same zonelist. So only NUMA can configure this param.
4658 #define ZONELIST_ORDER_DEFAULT 0
4659 #define ZONELIST_ORDER_NODE 1
4660 #define ZONELIST_ORDER_ZONE 2
4662 /* zonelist order in the kernel.
4663 * set_zonelist_order() will set this to NODE or ZONE.
4665 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4666 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
4670 /* The value user specified ....changed by config */
4671 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4672 /* string for sysctl */
4673 #define NUMA_ZONELIST_ORDER_LEN 16
4674 char numa_zonelist_order
[16] = "default";
4677 * interface for configure zonelist ordering.
4678 * command line option "numa_zonelist_order"
4679 * = "[dD]efault - default, automatic configuration.
4680 * = "[nN]ode - order by node locality, then by zone within node
4681 * = "[zZ]one - order by zone, then by locality within zone
4684 static int __parse_numa_zonelist_order(char *s
)
4686 if (*s
== 'd' || *s
== 'D') {
4687 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4688 } else if (*s
== 'n' || *s
== 'N') {
4689 user_zonelist_order
= ZONELIST_ORDER_NODE
;
4690 } else if (*s
== 'z' || *s
== 'Z') {
4691 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
4693 pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s
);
4699 static __init
int setup_numa_zonelist_order(char *s
)
4706 ret
= __parse_numa_zonelist_order(s
);
4708 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
4712 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
4715 * sysctl handler for numa_zonelist_order
4717 int numa_zonelist_order_handler(struct ctl_table
*table
, int write
,
4718 void __user
*buffer
, size_t *length
,
4721 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
4723 static DEFINE_MUTEX(zl_order_mutex
);
4725 mutex_lock(&zl_order_mutex
);
4727 if (strlen((char *)table
->data
) >= NUMA_ZONELIST_ORDER_LEN
) {
4731 strcpy(saved_string
, (char *)table
->data
);
4733 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
4737 int oldval
= user_zonelist_order
;
4739 ret
= __parse_numa_zonelist_order((char *)table
->data
);
4742 * bogus value. restore saved string
4744 strncpy((char *)table
->data
, saved_string
,
4745 NUMA_ZONELIST_ORDER_LEN
);
4746 user_zonelist_order
= oldval
;
4747 } else if (oldval
!= user_zonelist_order
) {
4748 mutex_lock(&zonelists_mutex
);
4749 build_all_zonelists(NULL
, NULL
);
4750 mutex_unlock(&zonelists_mutex
);
4754 mutex_unlock(&zl_order_mutex
);
4759 #define MAX_NODE_LOAD (nr_online_nodes)
4760 static int node_load
[MAX_NUMNODES
];
4763 * find_next_best_node - find the next node that should appear in a given node's fallback list
4764 * @node: node whose fallback list we're appending
4765 * @used_node_mask: nodemask_t of already used nodes
4767 * We use a number of factors to determine which is the next node that should
4768 * appear on a given node's fallback list. The node should not have appeared
4769 * already in @node's fallback list, and it should be the next closest node
4770 * according to the distance array (which contains arbitrary distance values
4771 * from each node to each node in the system), and should also prefer nodes
4772 * with no CPUs, since presumably they'll have very little allocation pressure
4773 * on them otherwise.
4774 * It returns -1 if no node is found.
4776 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
4779 int min_val
= INT_MAX
;
4780 int best_node
= NUMA_NO_NODE
;
4781 const struct cpumask
*tmp
= cpumask_of_node(0);
4783 /* Use the local node if we haven't already */
4784 if (!node_isset(node
, *used_node_mask
)) {
4785 node_set(node
, *used_node_mask
);
4789 for_each_node_state(n
, N_MEMORY
) {
4791 /* Don't want a node to appear more than once */
4792 if (node_isset(n
, *used_node_mask
))
4795 /* Use the distance array to find the distance */
4796 val
= node_distance(node
, n
);
4798 /* Penalize nodes under us ("prefer the next node") */
4801 /* Give preference to headless and unused nodes */
4802 tmp
= cpumask_of_node(n
);
4803 if (!cpumask_empty(tmp
))
4804 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
4806 /* Slight preference for less loaded node */
4807 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
4808 val
+= node_load
[n
];
4810 if (val
< min_val
) {
4817 node_set(best_node
, *used_node_mask
);
4824 * Build zonelists ordered by node and zones within node.
4825 * This results in maximum locality--normal zone overflows into local
4826 * DMA zone, if any--but risks exhausting DMA zone.
4828 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
4831 struct zonelist
*zonelist
;
4833 zonelist
= &pgdat
->node_zonelists
[ZONELIST_FALLBACK
];
4834 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
4836 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4837 zonelist
->_zonerefs
[j
].zone
= NULL
;
4838 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4842 * Build gfp_thisnode zonelists
4844 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
4847 struct zonelist
*zonelist
;
4849 zonelist
= &pgdat
->node_zonelists
[ZONELIST_NOFALLBACK
];
4850 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4851 zonelist
->_zonerefs
[j
].zone
= NULL
;
4852 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4856 * Build zonelists ordered by zone and nodes within zones.
4857 * This results in conserving DMA zone[s] until all Normal memory is
4858 * exhausted, but results in overflowing to remote node while memory
4859 * may still exist in local DMA zone.
4861 static int node_order
[MAX_NUMNODES
];
4863 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
4866 int zone_type
; /* needs to be signed */
4868 struct zonelist
*zonelist
;
4870 zonelist
= &pgdat
->node_zonelists
[ZONELIST_FALLBACK
];
4872 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
4873 for (j
= 0; j
< nr_nodes
; j
++) {
4874 node
= node_order
[j
];
4875 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
4876 if (managed_zone(z
)) {
4878 &zonelist
->_zonerefs
[pos
++]);
4879 check_highest_zone(zone_type
);
4883 zonelist
->_zonerefs
[pos
].zone
= NULL
;
4884 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
4887 #if defined(CONFIG_64BIT)
4889 * Devices that require DMA32/DMA are relatively rare and do not justify a
4890 * penalty to every machine in case the specialised case applies. Default
4891 * to Node-ordering on 64-bit NUMA machines
4893 static int default_zonelist_order(void)
4895 return ZONELIST_ORDER_NODE
;
4899 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4900 * by the kernel. If processes running on node 0 deplete the low memory zone
4901 * then reclaim will occur more frequency increasing stalls and potentially
4902 * be easier to OOM if a large percentage of the zone is under writeback or
4903 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4904 * Hence, default to zone ordering on 32-bit.
4906 static int default_zonelist_order(void)
4908 return ZONELIST_ORDER_ZONE
;
4910 #endif /* CONFIG_64BIT */
4912 static void set_zonelist_order(void)
4914 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
4915 current_zonelist_order
= default_zonelist_order();
4917 current_zonelist_order
= user_zonelist_order
;
4920 static void build_zonelists(pg_data_t
*pgdat
)
4923 nodemask_t used_mask
;
4924 int local_node
, prev_node
;
4925 struct zonelist
*zonelist
;
4926 unsigned int order
= current_zonelist_order
;
4928 /* initialize zonelists */
4929 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
4930 zonelist
= pgdat
->node_zonelists
+ i
;
4931 zonelist
->_zonerefs
[0].zone
= NULL
;
4932 zonelist
->_zonerefs
[0].zone_idx
= 0;
4935 /* NUMA-aware ordering of nodes */
4936 local_node
= pgdat
->node_id
;
4937 load
= nr_online_nodes
;
4938 prev_node
= local_node
;
4939 nodes_clear(used_mask
);
4941 memset(node_order
, 0, sizeof(node_order
));
4944 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
4946 * We don't want to pressure a particular node.
4947 * So adding penalty to the first node in same
4948 * distance group to make it round-robin.
4950 if (node_distance(local_node
, node
) !=
4951 node_distance(local_node
, prev_node
))
4952 node_load
[node
] = load
;
4956 if (order
== ZONELIST_ORDER_NODE
)
4957 build_zonelists_in_node_order(pgdat
, node
);
4959 node_order
[i
++] = node
; /* remember order */
4962 if (order
== ZONELIST_ORDER_ZONE
) {
4963 /* calculate node order -- i.e., DMA last! */
4964 build_zonelists_in_zone_order(pgdat
, i
);
4967 build_thisnode_zonelists(pgdat
);
4970 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4972 * Return node id of node used for "local" allocations.
4973 * I.e., first node id of first zone in arg node's generic zonelist.
4974 * Used for initializing percpu 'numa_mem', which is used primarily
4975 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4977 int local_memory_node(int node
)
4981 z
= first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
4982 gfp_zone(GFP_KERNEL
),
4984 return z
->zone
->node
;
4988 static void setup_min_unmapped_ratio(void);
4989 static void setup_min_slab_ratio(void);
4990 #else /* CONFIG_NUMA */
4992 static void set_zonelist_order(void)
4994 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
4997 static void build_zonelists(pg_data_t
*pgdat
)
4999 int node
, local_node
;
5001 struct zonelist
*zonelist
;
5003 local_node
= pgdat
->node_id
;
5005 zonelist
= &pgdat
->node_zonelists
[ZONELIST_FALLBACK
];
5006 j
= build_zonelists_node(pgdat
, zonelist
, 0);
5009 * Now we build the zonelist so that it contains the zones
5010 * of all the other nodes.
5011 * We don't want to pressure a particular node, so when
5012 * building the zones for node N, we make sure that the
5013 * zones coming right after the local ones are those from
5014 * node N+1 (modulo N)
5016 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
5017 if (!node_online(node
))
5019 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
5021 for (node
= 0; node
< local_node
; node
++) {
5022 if (!node_online(node
))
5024 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
5027 zonelist
->_zonerefs
[j
].zone
= NULL
;
5028 zonelist
->_zonerefs
[j
].zone_idx
= 0;
5031 #endif /* CONFIG_NUMA */
5034 * Boot pageset table. One per cpu which is going to be used for all
5035 * zones and all nodes. The parameters will be set in such a way
5036 * that an item put on a list will immediately be handed over to
5037 * the buddy list. This is safe since pageset manipulation is done
5038 * with interrupts disabled.
5040 * The boot_pagesets must be kept even after bootup is complete for
5041 * unused processors and/or zones. They do play a role for bootstrapping
5042 * hotplugged processors.
5044 * zoneinfo_show() and maybe other functions do
5045 * not check if the processor is online before following the pageset pointer.
5046 * Other parts of the kernel may not check if the zone is available.
5048 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
5049 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
5050 static void setup_zone_pageset(struct zone
*zone
);
5053 * Global mutex to protect against size modification of zonelists
5054 * as well as to serialize pageset setup for the new populated zone.
5056 DEFINE_MUTEX(zonelists_mutex
);
5058 /* return values int ....just for stop_machine() */
5059 static int __build_all_zonelists(void *data
)
5063 pg_data_t
*self
= data
;
5066 memset(node_load
, 0, sizeof(node_load
));
5069 if (self
&& !node_online(self
->node_id
)) {
5070 build_zonelists(self
);
5073 for_each_online_node(nid
) {
5074 pg_data_t
*pgdat
= NODE_DATA(nid
);
5076 build_zonelists(pgdat
);
5080 * Initialize the boot_pagesets that are going to be used
5081 * for bootstrapping processors. The real pagesets for
5082 * each zone will be allocated later when the per cpu
5083 * allocator is available.
5085 * boot_pagesets are used also for bootstrapping offline
5086 * cpus if the system is already booted because the pagesets
5087 * are needed to initialize allocators on a specific cpu too.
5088 * F.e. the percpu allocator needs the page allocator which
5089 * needs the percpu allocator in order to allocate its pagesets
5090 * (a chicken-egg dilemma).
5092 for_each_possible_cpu(cpu
) {
5093 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
5095 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
5097 * We now know the "local memory node" for each node--
5098 * i.e., the node of the first zone in the generic zonelist.
5099 * Set up numa_mem percpu variable for on-line cpus. During
5100 * boot, only the boot cpu should be on-line; we'll init the
5101 * secondary cpus' numa_mem as they come on-line. During
5102 * node/memory hotplug, we'll fixup all on-line cpus.
5104 if (cpu_online(cpu
))
5105 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
5112 static noinline
void __init
5113 build_all_zonelists_init(void)
5115 __build_all_zonelists(NULL
);
5116 mminit_verify_zonelist();
5117 cpuset_init_current_mems_allowed();
5121 * Called with zonelists_mutex held always
5122 * unless system_state == SYSTEM_BOOTING.
5124 * __ref due to (1) call of __meminit annotated setup_zone_pageset
5125 * [we're only called with non-NULL zone through __meminit paths] and
5126 * (2) call of __init annotated helper build_all_zonelists_init
5127 * [protected by SYSTEM_BOOTING].
5129 void __ref
build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
)
5131 set_zonelist_order();
5133 if (system_state
== SYSTEM_BOOTING
) {
5134 build_all_zonelists_init();
5136 #ifdef CONFIG_MEMORY_HOTPLUG
5138 setup_zone_pageset(zone
);
5140 /* we have to stop all cpus to guarantee there is no user
5142 stop_machine(__build_all_zonelists
, pgdat
, NULL
);
5143 /* cpuset refresh routine should be here */
5145 vm_total_pages
= nr_free_pagecache_pages();
5147 * Disable grouping by mobility if the number of pages in the
5148 * system is too low to allow the mechanism to work. It would be
5149 * more accurate, but expensive to check per-zone. This check is
5150 * made on memory-hotadd so a system can start with mobility
5151 * disabled and enable it later
5153 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
5154 page_group_by_mobility_disabled
= 1;
5156 page_group_by_mobility_disabled
= 0;
5158 pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n",
5160 zonelist_order_name
[current_zonelist_order
],
5161 page_group_by_mobility_disabled
? "off" : "on",
5164 pr_info("Policy zone: %s\n", zone_names
[policy_zone
]);
5169 * Initially all pages are reserved - free ones are freed
5170 * up by free_all_bootmem() once the early boot process is
5171 * done. Non-atomic initialization, single-pass.
5173 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
5174 unsigned long start_pfn
, enum memmap_context context
)
5176 struct vmem_altmap
*altmap
= to_vmem_altmap(__pfn_to_phys(start_pfn
));
5177 unsigned long end_pfn
= start_pfn
+ size
;
5178 pg_data_t
*pgdat
= NODE_DATA(nid
);
5180 unsigned long nr_initialised
= 0;
5181 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5182 struct memblock_region
*r
= NULL
, *tmp
;
5185 if (highest_memmap_pfn
< end_pfn
- 1)
5186 highest_memmap_pfn
= end_pfn
- 1;
5189 * Honor reservation requested by the driver for this ZONE_DEVICE
5192 if (altmap
&& start_pfn
== altmap
->base_pfn
)
5193 start_pfn
+= altmap
->reserve
;
5195 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
5197 * There can be holes in boot-time mem_map[]s handed to this
5198 * function. They do not exist on hotplugged memory.
5200 if (context
!= MEMMAP_EARLY
)
5203 if (!early_pfn_valid(pfn
)) {
5204 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5206 * Skip to the pfn preceding the next valid one (or
5207 * end_pfn), such that we hit a valid pfn (or end_pfn)
5208 * on our next iteration of the loop.
5210 pfn
= memblock_next_valid_pfn(pfn
, end_pfn
) - 1;
5214 if (!early_pfn_in_nid(pfn
, nid
))
5216 if (!update_defer_init(pgdat
, pfn
, end_pfn
, &nr_initialised
))
5219 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5221 * Check given memblock attribute by firmware which can affect
5222 * kernel memory layout. If zone==ZONE_MOVABLE but memory is
5223 * mirrored, it's an overlapped memmap init. skip it.
5225 if (mirrored_kernelcore
&& zone
== ZONE_MOVABLE
) {
5226 if (!r
|| pfn
>= memblock_region_memory_end_pfn(r
)) {
5227 for_each_memblock(memory
, tmp
)
5228 if (pfn
< memblock_region_memory_end_pfn(tmp
))
5232 if (pfn
>= memblock_region_memory_base_pfn(r
) &&
5233 memblock_is_mirror(r
)) {
5234 /* already initialized as NORMAL */
5235 pfn
= memblock_region_memory_end_pfn(r
);
5243 * Mark the block movable so that blocks are reserved for
5244 * movable at startup. This will force kernel allocations
5245 * to reserve their blocks rather than leaking throughout
5246 * the address space during boot when many long-lived
5247 * kernel allocations are made.
5249 * bitmap is created for zone's valid pfn range. but memmap
5250 * can be created for invalid pages (for alignment)
5251 * check here not to call set_pageblock_migratetype() against
5254 if (!(pfn
& (pageblock_nr_pages
- 1))) {
5255 struct page
*page
= pfn_to_page(pfn
);
5257 __init_single_page(page
, pfn
, zone
, nid
);
5258 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
5260 __init_single_pfn(pfn
, zone
, nid
);
5265 static void __meminit
zone_init_free_lists(struct zone
*zone
)
5267 unsigned int order
, t
;
5268 for_each_migratetype_order(order
, t
) {
5269 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
5270 zone
->free_area
[order
].nr_free
= 0;
5274 #ifndef __HAVE_ARCH_MEMMAP_INIT
5275 #define memmap_init(size, nid, zone, start_pfn) \
5276 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
5279 static int zone_batchsize(struct zone
*zone
)
5285 * The per-cpu-pages pools are set to around 1000th of the
5286 * size of the zone. But no more than 1/2 of a meg.
5288 * OK, so we don't know how big the cache is. So guess.
5290 batch
= zone
->managed_pages
/ 1024;
5291 if (batch
* PAGE_SIZE
> 512 * 1024)
5292 batch
= (512 * 1024) / PAGE_SIZE
;
5293 batch
/= 4; /* We effectively *= 4 below */
5298 * Clamp the batch to a 2^n - 1 value. Having a power
5299 * of 2 value was found to be more likely to have
5300 * suboptimal cache aliasing properties in some cases.
5302 * For example if 2 tasks are alternately allocating
5303 * batches of pages, one task can end up with a lot
5304 * of pages of one half of the possible page colors
5305 * and the other with pages of the other colors.
5307 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
5312 /* The deferral and batching of frees should be suppressed under NOMMU
5315 * The problem is that NOMMU needs to be able to allocate large chunks
5316 * of contiguous memory as there's no hardware page translation to
5317 * assemble apparent contiguous memory from discontiguous pages.
5319 * Queueing large contiguous runs of pages for batching, however,
5320 * causes the pages to actually be freed in smaller chunks. As there
5321 * can be a significant delay between the individual batches being
5322 * recycled, this leads to the once large chunks of space being
5323 * fragmented and becoming unavailable for high-order allocations.
5330 * pcp->high and pcp->batch values are related and dependent on one another:
5331 * ->batch must never be higher then ->high.
5332 * The following function updates them in a safe manner without read side
5335 * Any new users of pcp->batch and pcp->high should ensure they can cope with
5336 * those fields changing asynchronously (acording the the above rule).
5338 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
5339 * outside of boot time (or some other assurance that no concurrent updaters
5342 static void pageset_update(struct per_cpu_pages
*pcp
, unsigned long high
,
5343 unsigned long batch
)
5345 /* start with a fail safe value for batch */
5349 /* Update high, then batch, in order */
5356 /* a companion to pageset_set_high() */
5357 static void pageset_set_batch(struct per_cpu_pageset
*p
, unsigned long batch
)
5359 pageset_update(&p
->pcp
, 6 * batch
, max(1UL, 1 * batch
));
5362 static void pageset_init(struct per_cpu_pageset
*p
)
5364 struct per_cpu_pages
*pcp
;
5367 memset(p
, 0, sizeof(*p
));
5371 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
5372 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
5375 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
5378 pageset_set_batch(p
, batch
);
5382 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
5383 * to the value high for the pageset p.
5385 static void pageset_set_high(struct per_cpu_pageset
*p
,
5388 unsigned long batch
= max(1UL, high
/ 4);
5389 if ((high
/ 4) > (PAGE_SHIFT
* 8))
5390 batch
= PAGE_SHIFT
* 8;
5392 pageset_update(&p
->pcp
, high
, batch
);
5395 static void pageset_set_high_and_batch(struct zone
*zone
,
5396 struct per_cpu_pageset
*pcp
)
5398 if (percpu_pagelist_fraction
)
5399 pageset_set_high(pcp
,
5400 (zone
->managed_pages
/
5401 percpu_pagelist_fraction
));
5403 pageset_set_batch(pcp
, zone_batchsize(zone
));
5406 static void __meminit
zone_pageset_init(struct zone
*zone
, int cpu
)
5408 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
5411 pageset_set_high_and_batch(zone
, pcp
);
5414 static void __meminit
setup_zone_pageset(struct zone
*zone
)
5417 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
5418 for_each_possible_cpu(cpu
)
5419 zone_pageset_init(zone
, cpu
);
5423 * Allocate per cpu pagesets and initialize them.
5424 * Before this call only boot pagesets were available.
5426 void __init
setup_per_cpu_pageset(void)
5428 struct pglist_data
*pgdat
;
5431 for_each_populated_zone(zone
)
5432 setup_zone_pageset(zone
);
5434 for_each_online_pgdat(pgdat
)
5435 pgdat
->per_cpu_nodestats
=
5436 alloc_percpu(struct per_cpu_nodestat
);
5439 static __meminit
void zone_pcp_init(struct zone
*zone
)
5442 * per cpu subsystem is not up at this point. The following code
5443 * relies on the ability of the linker to provide the
5444 * offset of a (static) per cpu variable into the per cpu area.
5446 zone
->pageset
= &boot_pageset
;
5448 if (populated_zone(zone
))
5449 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
5450 zone
->name
, zone
->present_pages
,
5451 zone_batchsize(zone
));
5454 int __meminit
init_currently_empty_zone(struct zone
*zone
,
5455 unsigned long zone_start_pfn
,
5458 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
5460 pgdat
->nr_zones
= zone_idx(zone
) + 1;
5462 zone
->zone_start_pfn
= zone_start_pfn
;
5464 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
5465 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
5467 (unsigned long)zone_idx(zone
),
5468 zone_start_pfn
, (zone_start_pfn
+ size
));
5470 zone_init_free_lists(zone
);
5471 zone
->initialized
= 1;
5476 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5477 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
5480 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
5482 int __meminit
__early_pfn_to_nid(unsigned long pfn
,
5483 struct mminit_pfnnid_cache
*state
)
5485 unsigned long start_pfn
, end_pfn
;
5488 if (state
->last_start
<= pfn
&& pfn
< state
->last_end
)
5489 return state
->last_nid
;
5491 nid
= memblock_search_pfn_nid(pfn
, &start_pfn
, &end_pfn
);
5493 state
->last_start
= start_pfn
;
5494 state
->last_end
= end_pfn
;
5495 state
->last_nid
= nid
;
5500 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
5503 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
5504 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
5505 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
5507 * If an architecture guarantees that all ranges registered contain no holes
5508 * and may be freed, this this function may be used instead of calling
5509 * memblock_free_early_nid() manually.
5511 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
5513 unsigned long start_pfn
, end_pfn
;
5516 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
5517 start_pfn
= min(start_pfn
, max_low_pfn
);
5518 end_pfn
= min(end_pfn
, max_low_pfn
);
5520 if (start_pfn
< end_pfn
)
5521 memblock_free_early_nid(PFN_PHYS(start_pfn
),
5522 (end_pfn
- start_pfn
) << PAGE_SHIFT
,
5528 * sparse_memory_present_with_active_regions - Call memory_present for each active range
5529 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
5531 * If an architecture guarantees that all ranges registered contain no holes and may
5532 * be freed, this function may be used instead of calling memory_present() manually.
5534 void __init
sparse_memory_present_with_active_regions(int nid
)
5536 unsigned long start_pfn
, end_pfn
;
5539 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
5540 memory_present(this_nid
, start_pfn
, end_pfn
);
5544 * get_pfn_range_for_nid - Return the start and end page frames for a node
5545 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
5546 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
5547 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
5549 * It returns the start and end page frame of a node based on information
5550 * provided by memblock_set_node(). If called for a node
5551 * with no available memory, a warning is printed and the start and end
5554 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
5555 unsigned long *start_pfn
, unsigned long *end_pfn
)
5557 unsigned long this_start_pfn
, this_end_pfn
;
5563 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
5564 *start_pfn
= min(*start_pfn
, this_start_pfn
);
5565 *end_pfn
= max(*end_pfn
, this_end_pfn
);
5568 if (*start_pfn
== -1UL)
5573 * This finds a zone that can be used for ZONE_MOVABLE pages. The
5574 * assumption is made that zones within a node are ordered in monotonic
5575 * increasing memory addresses so that the "highest" populated zone is used
5577 static void __init
find_usable_zone_for_movable(void)
5580 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
5581 if (zone_index
== ZONE_MOVABLE
)
5584 if (arch_zone_highest_possible_pfn
[zone_index
] >
5585 arch_zone_lowest_possible_pfn
[zone_index
])
5589 VM_BUG_ON(zone_index
== -1);
5590 movable_zone
= zone_index
;
5594 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
5595 * because it is sized independent of architecture. Unlike the other zones,
5596 * the starting point for ZONE_MOVABLE is not fixed. It may be different
5597 * in each node depending on the size of each node and how evenly kernelcore
5598 * is distributed. This helper function adjusts the zone ranges
5599 * provided by the architecture for a given node by using the end of the
5600 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5601 * zones within a node are in order of monotonic increases memory addresses
5603 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
5604 unsigned long zone_type
,
5605 unsigned long node_start_pfn
,
5606 unsigned long node_end_pfn
,
5607 unsigned long *zone_start_pfn
,
5608 unsigned long *zone_end_pfn
)
5610 /* Only adjust if ZONE_MOVABLE is on this node */
5611 if (zone_movable_pfn
[nid
]) {
5612 /* Size ZONE_MOVABLE */
5613 if (zone_type
== ZONE_MOVABLE
) {
5614 *zone_start_pfn
= zone_movable_pfn
[nid
];
5615 *zone_end_pfn
= min(node_end_pfn
,
5616 arch_zone_highest_possible_pfn
[movable_zone
]);
5618 /* Adjust for ZONE_MOVABLE starting within this range */
5619 } else if (!mirrored_kernelcore
&&
5620 *zone_start_pfn
< zone_movable_pfn
[nid
] &&
5621 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
5622 *zone_end_pfn
= zone_movable_pfn
[nid
];
5624 /* Check if this whole range is within ZONE_MOVABLE */
5625 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
5626 *zone_start_pfn
= *zone_end_pfn
;
5631 * Return the number of pages a zone spans in a node, including holes
5632 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5634 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5635 unsigned long zone_type
,
5636 unsigned long node_start_pfn
,
5637 unsigned long node_end_pfn
,
5638 unsigned long *zone_start_pfn
,
5639 unsigned long *zone_end_pfn
,
5640 unsigned long *ignored
)
5642 /* When hotadd a new node from cpu_up(), the node should be empty */
5643 if (!node_start_pfn
&& !node_end_pfn
)
5646 /* Get the start and end of the zone */
5647 *zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
5648 *zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
5649 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5650 node_start_pfn
, node_end_pfn
,
5651 zone_start_pfn
, zone_end_pfn
);
5653 /* Check that this node has pages within the zone's required range */
5654 if (*zone_end_pfn
< node_start_pfn
|| *zone_start_pfn
> node_end_pfn
)
5657 /* Move the zone boundaries inside the node if necessary */
5658 *zone_end_pfn
= min(*zone_end_pfn
, node_end_pfn
);
5659 *zone_start_pfn
= max(*zone_start_pfn
, node_start_pfn
);
5661 /* Return the spanned pages */
5662 return *zone_end_pfn
- *zone_start_pfn
;
5666 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
5667 * then all holes in the requested range will be accounted for.
5669 unsigned long __meminit
__absent_pages_in_range(int nid
,
5670 unsigned long range_start_pfn
,
5671 unsigned long range_end_pfn
)
5673 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
5674 unsigned long start_pfn
, end_pfn
;
5677 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5678 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
5679 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
5680 nr_absent
-= end_pfn
- start_pfn
;
5686 * absent_pages_in_range - Return number of page frames in holes within a range
5687 * @start_pfn: The start PFN to start searching for holes
5688 * @end_pfn: The end PFN to stop searching for holes
5690 * It returns the number of pages frames in memory holes within a range.
5692 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
5693 unsigned long end_pfn
)
5695 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
5698 /* Return the number of page frames in holes in a zone on a node */
5699 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5700 unsigned long zone_type
,
5701 unsigned long node_start_pfn
,
5702 unsigned long node_end_pfn
,
5703 unsigned long *ignored
)
5705 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
5706 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
5707 unsigned long zone_start_pfn
, zone_end_pfn
;
5708 unsigned long nr_absent
;
5710 /* When hotadd a new node from cpu_up(), the node should be empty */
5711 if (!node_start_pfn
&& !node_end_pfn
)
5714 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
5715 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
5717 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5718 node_start_pfn
, node_end_pfn
,
5719 &zone_start_pfn
, &zone_end_pfn
);
5720 nr_absent
= __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
5723 * ZONE_MOVABLE handling.
5724 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
5727 if (mirrored_kernelcore
&& zone_movable_pfn
[nid
]) {
5728 unsigned long start_pfn
, end_pfn
;
5729 struct memblock_region
*r
;
5731 for_each_memblock(memory
, r
) {
5732 start_pfn
= clamp(memblock_region_memory_base_pfn(r
),
5733 zone_start_pfn
, zone_end_pfn
);
5734 end_pfn
= clamp(memblock_region_memory_end_pfn(r
),
5735 zone_start_pfn
, zone_end_pfn
);
5737 if (zone_type
== ZONE_MOVABLE
&&
5738 memblock_is_mirror(r
))
5739 nr_absent
+= end_pfn
- start_pfn
;
5741 if (zone_type
== ZONE_NORMAL
&&
5742 !memblock_is_mirror(r
))
5743 nr_absent
+= end_pfn
- start_pfn
;
5750 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5751 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5752 unsigned long zone_type
,
5753 unsigned long node_start_pfn
,
5754 unsigned long node_end_pfn
,
5755 unsigned long *zone_start_pfn
,
5756 unsigned long *zone_end_pfn
,
5757 unsigned long *zones_size
)
5761 *zone_start_pfn
= node_start_pfn
;
5762 for (zone
= 0; zone
< zone_type
; zone
++)
5763 *zone_start_pfn
+= zones_size
[zone
];
5765 *zone_end_pfn
= *zone_start_pfn
+ zones_size
[zone_type
];
5767 return zones_size
[zone_type
];
5770 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5771 unsigned long zone_type
,
5772 unsigned long node_start_pfn
,
5773 unsigned long node_end_pfn
,
5774 unsigned long *zholes_size
)
5779 return zholes_size
[zone_type
];
5782 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5784 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
5785 unsigned long node_start_pfn
,
5786 unsigned long node_end_pfn
,
5787 unsigned long *zones_size
,
5788 unsigned long *zholes_size
)
5790 unsigned long realtotalpages
= 0, totalpages
= 0;
5793 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5794 struct zone
*zone
= pgdat
->node_zones
+ i
;
5795 unsigned long zone_start_pfn
, zone_end_pfn
;
5796 unsigned long size
, real_size
;
5798 size
= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
5804 real_size
= size
- zone_absent_pages_in_node(pgdat
->node_id
, i
,
5805 node_start_pfn
, node_end_pfn
,
5808 zone
->zone_start_pfn
= zone_start_pfn
;
5810 zone
->zone_start_pfn
= 0;
5811 zone
->spanned_pages
= size
;
5812 zone
->present_pages
= real_size
;
5815 realtotalpages
+= real_size
;
5818 pgdat
->node_spanned_pages
= totalpages
;
5819 pgdat
->node_present_pages
= realtotalpages
;
5820 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
5824 #ifndef CONFIG_SPARSEMEM
5826 * Calculate the size of the zone->blockflags rounded to an unsigned long
5827 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5828 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
5829 * round what is now in bits to nearest long in bits, then return it in
5832 static unsigned long __init
usemap_size(unsigned long zone_start_pfn
, unsigned long zonesize
)
5834 unsigned long usemapsize
;
5836 zonesize
+= zone_start_pfn
& (pageblock_nr_pages
-1);
5837 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
5838 usemapsize
= usemapsize
>> pageblock_order
;
5839 usemapsize
*= NR_PAGEBLOCK_BITS
;
5840 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
5842 return usemapsize
/ 8;
5845 static void __init
setup_usemap(struct pglist_data
*pgdat
,
5847 unsigned long zone_start_pfn
,
5848 unsigned long zonesize
)
5850 unsigned long usemapsize
= usemap_size(zone_start_pfn
, zonesize
);
5851 zone
->pageblock_flags
= NULL
;
5853 zone
->pageblock_flags
=
5854 memblock_virt_alloc_node_nopanic(usemapsize
,
5858 static inline void setup_usemap(struct pglist_data
*pgdat
, struct zone
*zone
,
5859 unsigned long zone_start_pfn
, unsigned long zonesize
) {}
5860 #endif /* CONFIG_SPARSEMEM */
5862 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
5864 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
5865 void __paginginit
set_pageblock_order(void)
5869 /* Check that pageblock_nr_pages has not already been setup */
5870 if (pageblock_order
)
5873 if (HPAGE_SHIFT
> PAGE_SHIFT
)
5874 order
= HUGETLB_PAGE_ORDER
;
5876 order
= MAX_ORDER
- 1;
5879 * Assume the largest contiguous order of interest is a huge page.
5880 * This value may be variable depending on boot parameters on IA64 and
5883 pageblock_order
= order
;
5885 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5888 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
5889 * is unused as pageblock_order is set at compile-time. See
5890 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5893 void __paginginit
set_pageblock_order(void)
5897 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5899 static unsigned long __paginginit
calc_memmap_size(unsigned long spanned_pages
,
5900 unsigned long present_pages
)
5902 unsigned long pages
= spanned_pages
;
5905 * Provide a more accurate estimation if there are holes within
5906 * the zone and SPARSEMEM is in use. If there are holes within the
5907 * zone, each populated memory region may cost us one or two extra
5908 * memmap pages due to alignment because memmap pages for each
5909 * populated regions may not be naturally aligned on page boundary.
5910 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5912 if (spanned_pages
> present_pages
+ (present_pages
>> 4) &&
5913 IS_ENABLED(CONFIG_SPARSEMEM
))
5914 pages
= present_pages
;
5916 return PAGE_ALIGN(pages
* sizeof(struct page
)) >> PAGE_SHIFT
;
5920 * Set up the zone data structures:
5921 * - mark all pages reserved
5922 * - mark all memory queues empty
5923 * - clear the memory bitmaps
5925 * NOTE: pgdat should get zeroed by caller.
5927 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
)
5930 int nid
= pgdat
->node_id
;
5933 pgdat_resize_init(pgdat
);
5934 #ifdef CONFIG_NUMA_BALANCING
5935 spin_lock_init(&pgdat
->numabalancing_migrate_lock
);
5936 pgdat
->numabalancing_migrate_nr_pages
= 0;
5937 pgdat
->numabalancing_migrate_next_window
= jiffies
;
5939 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5940 spin_lock_init(&pgdat
->split_queue_lock
);
5941 INIT_LIST_HEAD(&pgdat
->split_queue
);
5942 pgdat
->split_queue_len
= 0;
5944 init_waitqueue_head(&pgdat
->kswapd_wait
);
5945 init_waitqueue_head(&pgdat
->pfmemalloc_wait
);
5946 #ifdef CONFIG_COMPACTION
5947 init_waitqueue_head(&pgdat
->kcompactd_wait
);
5949 pgdat_page_ext_init(pgdat
);
5950 spin_lock_init(&pgdat
->lru_lock
);
5951 lruvec_init(node_lruvec(pgdat
));
5953 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5954 struct zone
*zone
= pgdat
->node_zones
+ j
;
5955 unsigned long size
, realsize
, freesize
, memmap_pages
;
5956 unsigned long zone_start_pfn
= zone
->zone_start_pfn
;
5958 size
= zone
->spanned_pages
;
5959 realsize
= freesize
= zone
->present_pages
;
5962 * Adjust freesize so that it accounts for how much memory
5963 * is used by this zone for memmap. This affects the watermark
5964 * and per-cpu initialisations
5966 memmap_pages
= calc_memmap_size(size
, realsize
);
5967 if (!is_highmem_idx(j
)) {
5968 if (freesize
>= memmap_pages
) {
5969 freesize
-= memmap_pages
;
5972 " %s zone: %lu pages used for memmap\n",
5973 zone_names
[j
], memmap_pages
);
5975 pr_warn(" %s zone: %lu pages exceeds freesize %lu\n",
5976 zone_names
[j
], memmap_pages
, freesize
);
5979 /* Account for reserved pages */
5980 if (j
== 0 && freesize
> dma_reserve
) {
5981 freesize
-= dma_reserve
;
5982 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
5983 zone_names
[0], dma_reserve
);
5986 if (!is_highmem_idx(j
))
5987 nr_kernel_pages
+= freesize
;
5988 /* Charge for highmem memmap if there are enough kernel pages */
5989 else if (nr_kernel_pages
> memmap_pages
* 2)
5990 nr_kernel_pages
-= memmap_pages
;
5991 nr_all_pages
+= freesize
;
5994 * Set an approximate value for lowmem here, it will be adjusted
5995 * when the bootmem allocator frees pages into the buddy system.
5996 * And all highmem pages will be managed by the buddy system.
5998 zone
->managed_pages
= is_highmem_idx(j
) ? realsize
: freesize
;
6002 zone
->name
= zone_names
[j
];
6003 zone
->zone_pgdat
= pgdat
;
6004 spin_lock_init(&zone
->lock
);
6005 zone_seqlock_init(zone
);
6006 zone_pcp_init(zone
);
6011 set_pageblock_order();
6012 setup_usemap(pgdat
, zone
, zone_start_pfn
, size
);
6013 ret
= init_currently_empty_zone(zone
, zone_start_pfn
, size
);
6015 memmap_init(size
, nid
, j
, zone_start_pfn
);
6019 static void __ref
alloc_node_mem_map(struct pglist_data
*pgdat
)
6021 unsigned long __maybe_unused start
= 0;
6022 unsigned long __maybe_unused offset
= 0;
6024 /* Skip empty nodes */
6025 if (!pgdat
->node_spanned_pages
)
6028 #ifdef CONFIG_FLAT_NODE_MEM_MAP
6029 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
6030 offset
= pgdat
->node_start_pfn
- start
;
6031 /* ia64 gets its own node_mem_map, before this, without bootmem */
6032 if (!pgdat
->node_mem_map
) {
6033 unsigned long size
, end
;
6037 * The zone's endpoints aren't required to be MAX_ORDER
6038 * aligned but the node_mem_map endpoints must be in order
6039 * for the buddy allocator to function correctly.
6041 end
= pgdat_end_pfn(pgdat
);
6042 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
6043 size
= (end
- start
) * sizeof(struct page
);
6044 map
= alloc_remap(pgdat
->node_id
, size
);
6046 map
= memblock_virt_alloc_node_nopanic(size
,
6048 pgdat
->node_mem_map
= map
+ offset
;
6050 #ifndef CONFIG_NEED_MULTIPLE_NODES
6052 * With no DISCONTIG, the global mem_map is just set as node 0's
6054 if (pgdat
== NODE_DATA(0)) {
6055 mem_map
= NODE_DATA(0)->node_mem_map
;
6056 #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
6057 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
6059 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6062 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
6065 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
6066 unsigned long node_start_pfn
, unsigned long *zholes_size
)
6068 pg_data_t
*pgdat
= NODE_DATA(nid
);
6069 unsigned long start_pfn
= 0;
6070 unsigned long end_pfn
= 0;
6072 /* pg_data_t should be reset to zero when it's allocated */
6073 WARN_ON(pgdat
->nr_zones
|| pgdat
->kswapd_classzone_idx
);
6075 reset_deferred_meminit(pgdat
);
6076 pgdat
->node_id
= nid
;
6077 pgdat
->node_start_pfn
= node_start_pfn
;
6078 pgdat
->per_cpu_nodestats
= NULL
;
6079 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6080 get_pfn_range_for_nid(nid
, &start_pfn
, &end_pfn
);
6081 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid
,
6082 (u64
)start_pfn
<< PAGE_SHIFT
,
6083 end_pfn
? ((u64
)end_pfn
<< PAGE_SHIFT
) - 1 : 0);
6085 start_pfn
= node_start_pfn
;
6087 calculate_node_totalpages(pgdat
, start_pfn
, end_pfn
,
6088 zones_size
, zholes_size
);
6090 alloc_node_mem_map(pgdat
);
6091 #ifdef CONFIG_FLAT_NODE_MEM_MAP
6092 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
6093 nid
, (unsigned long)pgdat
,
6094 (unsigned long)pgdat
->node_mem_map
);
6097 free_area_init_core(pgdat
);
6100 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6102 #if MAX_NUMNODES > 1
6104 * Figure out the number of possible node ids.
6106 void __init
setup_nr_node_ids(void)
6108 unsigned int highest
;
6110 highest
= find_last_bit(node_possible_map
.bits
, MAX_NUMNODES
);
6111 nr_node_ids
= highest
+ 1;
6116 * node_map_pfn_alignment - determine the maximum internode alignment
6118 * This function should be called after node map is populated and sorted.
6119 * It calculates the maximum power of two alignment which can distinguish
6122 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
6123 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
6124 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
6125 * shifted, 1GiB is enough and this function will indicate so.
6127 * This is used to test whether pfn -> nid mapping of the chosen memory
6128 * model has fine enough granularity to avoid incorrect mapping for the
6129 * populated node map.
6131 * Returns the determined alignment in pfn's. 0 if there is no alignment
6132 * requirement (single node).
6134 unsigned long __init
node_map_pfn_alignment(void)
6136 unsigned long accl_mask
= 0, last_end
= 0;
6137 unsigned long start
, end
, mask
;
6141 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
6142 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
6149 * Start with a mask granular enough to pin-point to the
6150 * start pfn and tick off bits one-by-one until it becomes
6151 * too coarse to separate the current node from the last.
6153 mask
= ~((1 << __ffs(start
)) - 1);
6154 while (mask
&& last_end
<= (start
& (mask
<< 1)))
6157 /* accumulate all internode masks */
6161 /* convert mask to number of pages */
6162 return ~accl_mask
+ 1;
6165 /* Find the lowest pfn for a node */
6166 static unsigned long __init
find_min_pfn_for_node(int nid
)
6168 unsigned long min_pfn
= ULONG_MAX
;
6169 unsigned long start_pfn
;
6172 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
6173 min_pfn
= min(min_pfn
, start_pfn
);
6175 if (min_pfn
== ULONG_MAX
) {
6176 pr_warn("Could not find start_pfn for node %d\n", nid
);
6184 * find_min_pfn_with_active_regions - Find the minimum PFN registered
6186 * It returns the minimum PFN based on information provided via
6187 * memblock_set_node().
6189 unsigned long __init
find_min_pfn_with_active_regions(void)
6191 return find_min_pfn_for_node(MAX_NUMNODES
);
6195 * early_calculate_totalpages()
6196 * Sum pages in active regions for movable zone.
6197 * Populate N_MEMORY for calculating usable_nodes.
6199 static unsigned long __init
early_calculate_totalpages(void)
6201 unsigned long totalpages
= 0;
6202 unsigned long start_pfn
, end_pfn
;
6205 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
6206 unsigned long pages
= end_pfn
- start_pfn
;
6208 totalpages
+= pages
;
6210 node_set_state(nid
, N_MEMORY
);
6216 * Find the PFN the Movable zone begins in each node. Kernel memory
6217 * is spread evenly between nodes as long as the nodes have enough
6218 * memory. When they don't, some nodes will have more kernelcore than
6221 static void __init
find_zone_movable_pfns_for_nodes(void)
6224 unsigned long usable_startpfn
;
6225 unsigned long kernelcore_node
, kernelcore_remaining
;
6226 /* save the state before borrow the nodemask */
6227 nodemask_t saved_node_state
= node_states
[N_MEMORY
];
6228 unsigned long totalpages
= early_calculate_totalpages();
6229 int usable_nodes
= nodes_weight(node_states
[N_MEMORY
]);
6230 struct memblock_region
*r
;
6232 /* Need to find movable_zone earlier when movable_node is specified. */
6233 find_usable_zone_for_movable();
6236 * If movable_node is specified, ignore kernelcore and movablecore
6239 if (movable_node_is_enabled()) {
6240 for_each_memblock(memory
, r
) {
6241 if (!memblock_is_hotpluggable(r
))
6246 usable_startpfn
= PFN_DOWN(r
->base
);
6247 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
6248 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
6256 * If kernelcore=mirror is specified, ignore movablecore option
6258 if (mirrored_kernelcore
) {
6259 bool mem_below_4gb_not_mirrored
= false;
6261 for_each_memblock(memory
, r
) {
6262 if (memblock_is_mirror(r
))
6267 usable_startpfn
= memblock_region_memory_base_pfn(r
);
6269 if (usable_startpfn
< 0x100000) {
6270 mem_below_4gb_not_mirrored
= true;
6274 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
6275 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
6279 if (mem_below_4gb_not_mirrored
)
6280 pr_warn("This configuration results in unmirrored kernel memory.");
6286 * If movablecore=nn[KMG] was specified, calculate what size of
6287 * kernelcore that corresponds so that memory usable for
6288 * any allocation type is evenly spread. If both kernelcore
6289 * and movablecore are specified, then the value of kernelcore
6290 * will be used for required_kernelcore if it's greater than
6291 * what movablecore would have allowed.
6293 if (required_movablecore
) {
6294 unsigned long corepages
;
6297 * Round-up so that ZONE_MOVABLE is at least as large as what
6298 * was requested by the user
6300 required_movablecore
=
6301 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
6302 required_movablecore
= min(totalpages
, required_movablecore
);
6303 corepages
= totalpages
- required_movablecore
;
6305 required_kernelcore
= max(required_kernelcore
, corepages
);
6309 * If kernelcore was not specified or kernelcore size is larger
6310 * than totalpages, there is no ZONE_MOVABLE.
6312 if (!required_kernelcore
|| required_kernelcore
>= totalpages
)
6315 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
6316 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
6319 /* Spread kernelcore memory as evenly as possible throughout nodes */
6320 kernelcore_node
= required_kernelcore
/ usable_nodes
;
6321 for_each_node_state(nid
, N_MEMORY
) {
6322 unsigned long start_pfn
, end_pfn
;
6325 * Recalculate kernelcore_node if the division per node
6326 * now exceeds what is necessary to satisfy the requested
6327 * amount of memory for the kernel
6329 if (required_kernelcore
< kernelcore_node
)
6330 kernelcore_node
= required_kernelcore
/ usable_nodes
;
6333 * As the map is walked, we track how much memory is usable
6334 * by the kernel using kernelcore_remaining. When it is
6335 * 0, the rest of the node is usable by ZONE_MOVABLE
6337 kernelcore_remaining
= kernelcore_node
;
6339 /* Go through each range of PFNs within this node */
6340 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
6341 unsigned long size_pages
;
6343 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
6344 if (start_pfn
>= end_pfn
)
6347 /* Account for what is only usable for kernelcore */
6348 if (start_pfn
< usable_startpfn
) {
6349 unsigned long kernel_pages
;
6350 kernel_pages
= min(end_pfn
, usable_startpfn
)
6353 kernelcore_remaining
-= min(kernel_pages
,
6354 kernelcore_remaining
);
6355 required_kernelcore
-= min(kernel_pages
,
6356 required_kernelcore
);
6358 /* Continue if range is now fully accounted */
6359 if (end_pfn
<= usable_startpfn
) {
6362 * Push zone_movable_pfn to the end so
6363 * that if we have to rebalance
6364 * kernelcore across nodes, we will
6365 * not double account here
6367 zone_movable_pfn
[nid
] = end_pfn
;
6370 start_pfn
= usable_startpfn
;
6374 * The usable PFN range for ZONE_MOVABLE is from
6375 * start_pfn->end_pfn. Calculate size_pages as the
6376 * number of pages used as kernelcore
6378 size_pages
= end_pfn
- start_pfn
;
6379 if (size_pages
> kernelcore_remaining
)
6380 size_pages
= kernelcore_remaining
;
6381 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
6384 * Some kernelcore has been met, update counts and
6385 * break if the kernelcore for this node has been
6388 required_kernelcore
-= min(required_kernelcore
,
6390 kernelcore_remaining
-= size_pages
;
6391 if (!kernelcore_remaining
)
6397 * If there is still required_kernelcore, we do another pass with one
6398 * less node in the count. This will push zone_movable_pfn[nid] further
6399 * along on the nodes that still have memory until kernelcore is
6403 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
6407 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
6408 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
6409 zone_movable_pfn
[nid
] =
6410 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
6413 /* restore the node_state */
6414 node_states
[N_MEMORY
] = saved_node_state
;
6417 /* Any regular or high memory on that node ? */
6418 static void check_for_memory(pg_data_t
*pgdat
, int nid
)
6420 enum zone_type zone_type
;
6422 if (N_MEMORY
== N_NORMAL_MEMORY
)
6425 for (zone_type
= 0; zone_type
<= ZONE_MOVABLE
- 1; zone_type
++) {
6426 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
6427 if (populated_zone(zone
)) {
6428 node_set_state(nid
, N_HIGH_MEMORY
);
6429 if (N_NORMAL_MEMORY
!= N_HIGH_MEMORY
&&
6430 zone_type
<= ZONE_NORMAL
)
6431 node_set_state(nid
, N_NORMAL_MEMORY
);
6438 * free_area_init_nodes - Initialise all pg_data_t and zone data
6439 * @max_zone_pfn: an array of max PFNs for each zone
6441 * This will call free_area_init_node() for each active node in the system.
6442 * Using the page ranges provided by memblock_set_node(), the size of each
6443 * zone in each node and their holes is calculated. If the maximum PFN
6444 * between two adjacent zones match, it is assumed that the zone is empty.
6445 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
6446 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
6447 * starts where the previous one ended. For example, ZONE_DMA32 starts
6448 * at arch_max_dma_pfn.
6450 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
6452 unsigned long start_pfn
, end_pfn
;
6455 /* Record where the zone boundaries are */
6456 memset(arch_zone_lowest_possible_pfn
, 0,
6457 sizeof(arch_zone_lowest_possible_pfn
));
6458 memset(arch_zone_highest_possible_pfn
, 0,
6459 sizeof(arch_zone_highest_possible_pfn
));
6461 start_pfn
= find_min_pfn_with_active_regions();
6463 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
6464 if (i
== ZONE_MOVABLE
)
6467 end_pfn
= max(max_zone_pfn
[i
], start_pfn
);
6468 arch_zone_lowest_possible_pfn
[i
] = start_pfn
;
6469 arch_zone_highest_possible_pfn
[i
] = end_pfn
;
6471 start_pfn
= end_pfn
;
6474 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
6475 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
6476 find_zone_movable_pfns_for_nodes();
6478 /* Print out the zone ranges */
6479 pr_info("Zone ranges:\n");
6480 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
6481 if (i
== ZONE_MOVABLE
)
6483 pr_info(" %-8s ", zone_names
[i
]);
6484 if (arch_zone_lowest_possible_pfn
[i
] ==
6485 arch_zone_highest_possible_pfn
[i
])
6488 pr_cont("[mem %#018Lx-%#018Lx]\n",
6489 (u64
)arch_zone_lowest_possible_pfn
[i
]
6491 ((u64
)arch_zone_highest_possible_pfn
[i
]
6492 << PAGE_SHIFT
) - 1);
6495 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
6496 pr_info("Movable zone start for each node\n");
6497 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
6498 if (zone_movable_pfn
[i
])
6499 pr_info(" Node %d: %#018Lx\n", i
,
6500 (u64
)zone_movable_pfn
[i
] << PAGE_SHIFT
);
6503 /* Print out the early node map */
6504 pr_info("Early memory node ranges\n");
6505 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
6506 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid
,
6507 (u64
)start_pfn
<< PAGE_SHIFT
,
6508 ((u64
)end_pfn
<< PAGE_SHIFT
) - 1);
6510 /* Initialise every node */
6511 mminit_verify_pageflags_layout();
6512 setup_nr_node_ids();
6513 for_each_online_node(nid
) {
6514 pg_data_t
*pgdat
= NODE_DATA(nid
);
6515 free_area_init_node(nid
, NULL
,
6516 find_min_pfn_for_node(nid
), NULL
);
6518 /* Any memory on that node */
6519 if (pgdat
->node_present_pages
)
6520 node_set_state(nid
, N_MEMORY
);
6521 check_for_memory(pgdat
, nid
);
6525 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
6527 unsigned long long coremem
;
6531 coremem
= memparse(p
, &p
);
6532 *core
= coremem
>> PAGE_SHIFT
;
6534 /* Paranoid check that UL is enough for the coremem value */
6535 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
6541 * kernelcore=size sets the amount of memory for use for allocations that
6542 * cannot be reclaimed or migrated.
6544 static int __init
cmdline_parse_kernelcore(char *p
)
6546 /* parse kernelcore=mirror */
6547 if (parse_option_str(p
, "mirror")) {
6548 mirrored_kernelcore
= true;
6552 return cmdline_parse_core(p
, &required_kernelcore
);
6556 * movablecore=size sets the amount of memory for use for allocations that
6557 * can be reclaimed or migrated.
6559 static int __init
cmdline_parse_movablecore(char *p
)
6561 return cmdline_parse_core(p
, &required_movablecore
);
6564 early_param("kernelcore", cmdline_parse_kernelcore
);
6565 early_param("movablecore", cmdline_parse_movablecore
);
6567 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6569 void adjust_managed_page_count(struct page
*page
, long count
)
6571 spin_lock(&managed_page_count_lock
);
6572 page_zone(page
)->managed_pages
+= count
;
6573 totalram_pages
+= count
;
6574 #ifdef CONFIG_HIGHMEM
6575 if (PageHighMem(page
))
6576 totalhigh_pages
+= count
;
6578 spin_unlock(&managed_page_count_lock
);
6580 EXPORT_SYMBOL(adjust_managed_page_count
);
6582 unsigned long free_reserved_area(void *start
, void *end
, int poison
, char *s
)
6585 unsigned long pages
= 0;
6587 start
= (void *)PAGE_ALIGN((unsigned long)start
);
6588 end
= (void *)((unsigned long)end
& PAGE_MASK
);
6589 for (pos
= start
; pos
< end
; pos
+= PAGE_SIZE
, pages
++) {
6590 if ((unsigned int)poison
<= 0xFF)
6591 memset(pos
, poison
, PAGE_SIZE
);
6592 free_reserved_page(virt_to_page(pos
));
6596 pr_info("Freeing %s memory: %ldK\n",
6597 s
, pages
<< (PAGE_SHIFT
- 10));
6601 EXPORT_SYMBOL(free_reserved_area
);
6603 #ifdef CONFIG_HIGHMEM
6604 void free_highmem_page(struct page
*page
)
6606 __free_reserved_page(page
);
6608 page_zone(page
)->managed_pages
++;
6614 void __init
mem_init_print_info(const char *str
)
6616 unsigned long physpages
, codesize
, datasize
, rosize
, bss_size
;
6617 unsigned long init_code_size
, init_data_size
;
6619 physpages
= get_num_physpages();
6620 codesize
= _etext
- _stext
;
6621 datasize
= _edata
- _sdata
;
6622 rosize
= __end_rodata
- __start_rodata
;
6623 bss_size
= __bss_stop
- __bss_start
;
6624 init_data_size
= __init_end
- __init_begin
;
6625 init_code_size
= _einittext
- _sinittext
;
6628 * Detect special cases and adjust section sizes accordingly:
6629 * 1) .init.* may be embedded into .data sections
6630 * 2) .init.text.* may be out of [__init_begin, __init_end],
6631 * please refer to arch/tile/kernel/vmlinux.lds.S.
6632 * 3) .rodata.* may be embedded into .text or .data sections.
6634 #define adj_init_size(start, end, size, pos, adj) \
6636 if (start <= pos && pos < end && size > adj) \
6640 adj_init_size(__init_begin
, __init_end
, init_data_size
,
6641 _sinittext
, init_code_size
);
6642 adj_init_size(_stext
, _etext
, codesize
, _sinittext
, init_code_size
);
6643 adj_init_size(_sdata
, _edata
, datasize
, __init_begin
, init_data_size
);
6644 adj_init_size(_stext
, _etext
, codesize
, __start_rodata
, rosize
);
6645 adj_init_size(_sdata
, _edata
, datasize
, __start_rodata
, rosize
);
6647 #undef adj_init_size
6649 pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
6650 #ifdef CONFIG_HIGHMEM
6654 nr_free_pages() << (PAGE_SHIFT
- 10),
6655 physpages
<< (PAGE_SHIFT
- 10),
6656 codesize
>> 10, datasize
>> 10, rosize
>> 10,
6657 (init_data_size
+ init_code_size
) >> 10, bss_size
>> 10,
6658 (physpages
- totalram_pages
- totalcma_pages
) << (PAGE_SHIFT
- 10),
6659 totalcma_pages
<< (PAGE_SHIFT
- 10),
6660 #ifdef CONFIG_HIGHMEM
6661 totalhigh_pages
<< (PAGE_SHIFT
- 10),
6663 str
? ", " : "", str
? str
: "");
6667 * set_dma_reserve - set the specified number of pages reserved in the first zone
6668 * @new_dma_reserve: The number of pages to mark reserved
6670 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
6671 * In the DMA zone, a significant percentage may be consumed by kernel image
6672 * and other unfreeable allocations which can skew the watermarks badly. This
6673 * function may optionally be used to account for unfreeable pages in the
6674 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6675 * smaller per-cpu batchsize.
6677 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
6679 dma_reserve
= new_dma_reserve
;
6682 void __init
free_area_init(unsigned long *zones_size
)
6684 free_area_init_node(0, zones_size
,
6685 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
6688 static int page_alloc_cpu_dead(unsigned int cpu
)
6691 lru_add_drain_cpu(cpu
);
6695 * Spill the event counters of the dead processor
6696 * into the current processors event counters.
6697 * This artificially elevates the count of the current
6700 vm_events_fold_cpu(cpu
);
6703 * Zero the differential counters of the dead processor
6704 * so that the vm statistics are consistent.
6706 * This is only okay since the processor is dead and cannot
6707 * race with what we are doing.
6709 cpu_vm_stats_fold(cpu
);
6713 void __init
page_alloc_init(void)
6717 ret
= cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD
,
6718 "mm/page_alloc:dead", NULL
,
6719 page_alloc_cpu_dead
);
6724 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
6725 * or min_free_kbytes changes.
6727 static void calculate_totalreserve_pages(void)
6729 struct pglist_data
*pgdat
;
6730 unsigned long reserve_pages
= 0;
6731 enum zone_type i
, j
;
6733 for_each_online_pgdat(pgdat
) {
6735 pgdat
->totalreserve_pages
= 0;
6737 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
6738 struct zone
*zone
= pgdat
->node_zones
+ i
;
6741 /* Find valid and maximum lowmem_reserve in the zone */
6742 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
6743 if (zone
->lowmem_reserve
[j
] > max
)
6744 max
= zone
->lowmem_reserve
[j
];
6747 /* we treat the high watermark as reserved pages. */
6748 max
+= high_wmark_pages(zone
);
6750 if (max
> zone
->managed_pages
)
6751 max
= zone
->managed_pages
;
6753 pgdat
->totalreserve_pages
+= max
;
6755 reserve_pages
+= max
;
6758 totalreserve_pages
= reserve_pages
;
6762 * setup_per_zone_lowmem_reserve - called whenever
6763 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
6764 * has a correct pages reserved value, so an adequate number of
6765 * pages are left in the zone after a successful __alloc_pages().
6767 static void setup_per_zone_lowmem_reserve(void)
6769 struct pglist_data
*pgdat
;
6770 enum zone_type j
, idx
;
6772 for_each_online_pgdat(pgdat
) {
6773 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
6774 struct zone
*zone
= pgdat
->node_zones
+ j
;
6775 unsigned long managed_pages
= zone
->managed_pages
;
6777 zone
->lowmem_reserve
[j
] = 0;
6781 struct zone
*lower_zone
;
6785 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
6786 sysctl_lowmem_reserve_ratio
[idx
] = 1;
6788 lower_zone
= pgdat
->node_zones
+ idx
;
6789 lower_zone
->lowmem_reserve
[j
] = managed_pages
/
6790 sysctl_lowmem_reserve_ratio
[idx
];
6791 managed_pages
+= lower_zone
->managed_pages
;
6796 /* update totalreserve_pages */
6797 calculate_totalreserve_pages();
6800 static void __setup_per_zone_wmarks(void)
6802 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
6803 unsigned long lowmem_pages
= 0;
6805 unsigned long flags
;
6807 /* Calculate total number of !ZONE_HIGHMEM pages */
6808 for_each_zone(zone
) {
6809 if (!is_highmem(zone
))
6810 lowmem_pages
+= zone
->managed_pages
;
6813 for_each_zone(zone
) {
6816 spin_lock_irqsave(&zone
->lock
, flags
);
6817 tmp
= (u64
)pages_min
* zone
->managed_pages
;
6818 do_div(tmp
, lowmem_pages
);
6819 if (is_highmem(zone
)) {
6821 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6822 * need highmem pages, so cap pages_min to a small
6825 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
6826 * deltas control asynch page reclaim, and so should
6827 * not be capped for highmem.
6829 unsigned long min_pages
;
6831 min_pages
= zone
->managed_pages
/ 1024;
6832 min_pages
= clamp(min_pages
, SWAP_CLUSTER_MAX
, 128UL);
6833 zone
->watermark
[WMARK_MIN
] = min_pages
;
6836 * If it's a lowmem zone, reserve a number of pages
6837 * proportionate to the zone's size.
6839 zone
->watermark
[WMARK_MIN
] = tmp
;
6843 * Set the kswapd watermarks distance according to the
6844 * scale factor in proportion to available memory, but
6845 * ensure a minimum size on small systems.
6847 tmp
= max_t(u64
, tmp
>> 2,
6848 mult_frac(zone
->managed_pages
,
6849 watermark_scale_factor
, 10000));
6851 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + tmp
;
6852 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + tmp
* 2;
6854 spin_unlock_irqrestore(&zone
->lock
, flags
);
6857 /* update totalreserve_pages */
6858 calculate_totalreserve_pages();
6862 * setup_per_zone_wmarks - called when min_free_kbytes changes
6863 * or when memory is hot-{added|removed}
6865 * Ensures that the watermark[min,low,high] values for each zone are set
6866 * correctly with respect to min_free_kbytes.
6868 void setup_per_zone_wmarks(void)
6870 mutex_lock(&zonelists_mutex
);
6871 __setup_per_zone_wmarks();
6872 mutex_unlock(&zonelists_mutex
);
6876 * Initialise min_free_kbytes.
6878 * For small machines we want it small (128k min). For large machines
6879 * we want it large (64MB max). But it is not linear, because network
6880 * bandwidth does not increase linearly with machine size. We use
6882 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
6883 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6899 int __meminit
init_per_zone_wmark_min(void)
6901 unsigned long lowmem_kbytes
;
6902 int new_min_free_kbytes
;
6904 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
6905 new_min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
6907 if (new_min_free_kbytes
> user_min_free_kbytes
) {
6908 min_free_kbytes
= new_min_free_kbytes
;
6909 if (min_free_kbytes
< 128)
6910 min_free_kbytes
= 128;
6911 if (min_free_kbytes
> 65536)
6912 min_free_kbytes
= 65536;
6914 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6915 new_min_free_kbytes
, user_min_free_kbytes
);
6917 setup_per_zone_wmarks();
6918 refresh_zone_stat_thresholds();
6919 setup_per_zone_lowmem_reserve();
6922 setup_min_unmapped_ratio();
6923 setup_min_slab_ratio();
6928 core_initcall(init_per_zone_wmark_min
)
6931 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
6932 * that we can call two helper functions whenever min_free_kbytes
6935 int min_free_kbytes_sysctl_handler(struct ctl_table
*table
, int write
,
6936 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6940 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6945 user_min_free_kbytes
= min_free_kbytes
;
6946 setup_per_zone_wmarks();
6951 int watermark_scale_factor_sysctl_handler(struct ctl_table
*table
, int write
,
6952 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6956 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6961 setup_per_zone_wmarks();
6967 static void setup_min_unmapped_ratio(void)
6972 for_each_online_pgdat(pgdat
)
6973 pgdat
->min_unmapped_pages
= 0;
6976 zone
->zone_pgdat
->min_unmapped_pages
+= (zone
->managed_pages
*
6977 sysctl_min_unmapped_ratio
) / 100;
6981 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6982 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6986 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6990 setup_min_unmapped_ratio();
6995 static void setup_min_slab_ratio(void)
7000 for_each_online_pgdat(pgdat
)
7001 pgdat
->min_slab_pages
= 0;
7004 zone
->zone_pgdat
->min_slab_pages
+= (zone
->managed_pages
*
7005 sysctl_min_slab_ratio
) / 100;
7008 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
7009 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
7013 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
7017 setup_min_slab_ratio();
7024 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
7025 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
7026 * whenever sysctl_lowmem_reserve_ratio changes.
7028 * The reserve ratio obviously has absolutely no relation with the
7029 * minimum watermarks. The lowmem reserve ratio can only make sense
7030 * if in function of the boot time zone sizes.
7032 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
7033 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
7035 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
7036 setup_per_zone_lowmem_reserve();
7041 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
7042 * cpu. It is the fraction of total pages in each zone that a hot per cpu
7043 * pagelist can have before it gets flushed back to buddy allocator.
7045 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table
*table
, int write
,
7046 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
7049 int old_percpu_pagelist_fraction
;
7052 mutex_lock(&pcp_batch_high_lock
);
7053 old_percpu_pagelist_fraction
= percpu_pagelist_fraction
;
7055 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
7056 if (!write
|| ret
< 0)
7059 /* Sanity checking to avoid pcp imbalance */
7060 if (percpu_pagelist_fraction
&&
7061 percpu_pagelist_fraction
< MIN_PERCPU_PAGELIST_FRACTION
) {
7062 percpu_pagelist_fraction
= old_percpu_pagelist_fraction
;
7068 if (percpu_pagelist_fraction
== old_percpu_pagelist_fraction
)
7071 for_each_populated_zone(zone
) {
7074 for_each_possible_cpu(cpu
)
7075 pageset_set_high_and_batch(zone
,
7076 per_cpu_ptr(zone
->pageset
, cpu
));
7079 mutex_unlock(&pcp_batch_high_lock
);
7084 int hashdist
= HASHDIST_DEFAULT
;
7086 static int __init
set_hashdist(char *str
)
7090 hashdist
= simple_strtoul(str
, &str
, 0);
7093 __setup("hashdist=", set_hashdist
);
7096 #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES
7098 * Returns the number of pages that arch has reserved but
7099 * is not known to alloc_large_system_hash().
7101 static unsigned long __init
arch_reserved_kernel_pages(void)
7108 * allocate a large system hash table from bootmem
7109 * - it is assumed that the hash table must contain an exact power-of-2
7110 * quantity of entries
7111 * - limit is the number of hash buckets, not the total allocation size
7113 void *__init
alloc_large_system_hash(const char *tablename
,
7114 unsigned long bucketsize
,
7115 unsigned long numentries
,
7118 unsigned int *_hash_shift
,
7119 unsigned int *_hash_mask
,
7120 unsigned long low_limit
,
7121 unsigned long high_limit
)
7123 unsigned long long max
= high_limit
;
7124 unsigned long log2qty
, size
;
7127 /* allow the kernel cmdline to have a say */
7129 /* round applicable memory size up to nearest megabyte */
7130 numentries
= nr_kernel_pages
;
7131 numentries
-= arch_reserved_kernel_pages();
7133 /* It isn't necessary when PAGE_SIZE >= 1MB */
7134 if (PAGE_SHIFT
< 20)
7135 numentries
= round_up(numentries
, (1<<20)/PAGE_SIZE
);
7137 /* limit to 1 bucket per 2^scale bytes of low memory */
7138 if (scale
> PAGE_SHIFT
)
7139 numentries
>>= (scale
- PAGE_SHIFT
);
7141 numentries
<<= (PAGE_SHIFT
- scale
);
7143 /* Make sure we've got at least a 0-order allocation.. */
7144 if (unlikely(flags
& HASH_SMALL
)) {
7145 /* Makes no sense without HASH_EARLY */
7146 WARN_ON(!(flags
& HASH_EARLY
));
7147 if (!(numentries
>> *_hash_shift
)) {
7148 numentries
= 1UL << *_hash_shift
;
7149 BUG_ON(!numentries
);
7151 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
7152 numentries
= PAGE_SIZE
/ bucketsize
;
7154 numentries
= roundup_pow_of_two(numentries
);
7156 /* limit allocation size to 1/16 total memory by default */
7158 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
7159 do_div(max
, bucketsize
);
7161 max
= min(max
, 0x80000000ULL
);
7163 if (numentries
< low_limit
)
7164 numentries
= low_limit
;
7165 if (numentries
> max
)
7168 log2qty
= ilog2(numentries
);
7171 size
= bucketsize
<< log2qty
;
7172 if (flags
& HASH_EARLY
)
7173 table
= memblock_virt_alloc_nopanic(size
, 0);
7175 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
7178 * If bucketsize is not a power-of-two, we may free
7179 * some pages at the end of hash table which
7180 * alloc_pages_exact() automatically does
7182 if (get_order(size
) < MAX_ORDER
) {
7183 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
7184 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
7187 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
7190 panic("Failed to allocate %s hash table\n", tablename
);
7192 pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n",
7193 tablename
, 1UL << log2qty
, ilog2(size
) - PAGE_SHIFT
, size
);
7196 *_hash_shift
= log2qty
;
7198 *_hash_mask
= (1 << log2qty
) - 1;
7204 * This function checks whether pageblock includes unmovable pages or not.
7205 * If @count is not zero, it is okay to include less @count unmovable pages
7207 * PageLRU check without isolation or lru_lock could race so that
7208 * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable
7209 * check without lock_page also may miss some movable non-lru pages at
7210 * race condition. So you can't expect this function should be exact.
7212 bool has_unmovable_pages(struct zone
*zone
, struct page
*page
, int count
,
7213 bool skip_hwpoisoned_pages
)
7215 unsigned long pfn
, iter
, found
;
7219 * For avoiding noise data, lru_add_drain_all() should be called
7220 * If ZONE_MOVABLE, the zone never contains unmovable pages
7222 if (zone_idx(zone
) == ZONE_MOVABLE
)
7224 mt
= get_pageblock_migratetype(page
);
7225 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
7228 pfn
= page_to_pfn(page
);
7229 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
7230 unsigned long check
= pfn
+ iter
;
7232 if (!pfn_valid_within(check
))
7235 page
= pfn_to_page(check
);
7238 * Hugepages are not in LRU lists, but they're movable.
7239 * We need not scan over tail pages bacause we don't
7240 * handle each tail page individually in migration.
7242 if (PageHuge(page
)) {
7243 iter
= round_up(iter
+ 1, 1<<compound_order(page
)) - 1;
7248 * We can't use page_count without pin a page
7249 * because another CPU can free compound page.
7250 * This check already skips compound tails of THP
7251 * because their page->_refcount is zero at all time.
7253 if (!page_ref_count(page
)) {
7254 if (PageBuddy(page
))
7255 iter
+= (1 << page_order(page
)) - 1;
7260 * The HWPoisoned page may be not in buddy system, and
7261 * page_count() is not 0.
7263 if (skip_hwpoisoned_pages
&& PageHWPoison(page
))
7266 if (__PageMovable(page
))
7272 * If there are RECLAIMABLE pages, we need to check
7273 * it. But now, memory offline itself doesn't call
7274 * shrink_node_slabs() and it still to be fixed.
7277 * If the page is not RAM, page_count()should be 0.
7278 * we don't need more check. This is an _used_ not-movable page.
7280 * The problematic thing here is PG_reserved pages. PG_reserved
7281 * is set to both of a memory hole page and a _used_ kernel
7290 bool is_pageblock_removable_nolock(struct page
*page
)
7296 * We have to be careful here because we are iterating over memory
7297 * sections which are not zone aware so we might end up outside of
7298 * the zone but still within the section.
7299 * We have to take care about the node as well. If the node is offline
7300 * its NODE_DATA will be NULL - see page_zone.
7302 if (!node_online(page_to_nid(page
)))
7305 zone
= page_zone(page
);
7306 pfn
= page_to_pfn(page
);
7307 if (!zone_spans_pfn(zone
, pfn
))
7310 return !has_unmovable_pages(zone
, page
, 0, true);
7313 #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
7315 static unsigned long pfn_max_align_down(unsigned long pfn
)
7317 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
7318 pageblock_nr_pages
) - 1);
7321 static unsigned long pfn_max_align_up(unsigned long pfn
)
7323 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
7324 pageblock_nr_pages
));
7327 /* [start, end) must belong to a single zone. */
7328 static int __alloc_contig_migrate_range(struct compact_control
*cc
,
7329 unsigned long start
, unsigned long end
)
7331 /* This function is based on compact_zone() from compaction.c. */
7332 unsigned long nr_reclaimed
;
7333 unsigned long pfn
= start
;
7334 unsigned int tries
= 0;
7339 while (pfn
< end
|| !list_empty(&cc
->migratepages
)) {
7340 if (fatal_signal_pending(current
)) {
7345 if (list_empty(&cc
->migratepages
)) {
7346 cc
->nr_migratepages
= 0;
7347 pfn
= isolate_migratepages_range(cc
, pfn
, end
);
7353 } else if (++tries
== 5) {
7354 ret
= ret
< 0 ? ret
: -EBUSY
;
7358 nr_reclaimed
= reclaim_clean_pages_from_list(cc
->zone
,
7360 cc
->nr_migratepages
-= nr_reclaimed
;
7362 ret
= migrate_pages(&cc
->migratepages
, alloc_migrate_target
,
7363 NULL
, 0, cc
->mode
, MR_CMA
);
7366 putback_movable_pages(&cc
->migratepages
);
7373 * alloc_contig_range() -- tries to allocate given range of pages
7374 * @start: start PFN to allocate
7375 * @end: one-past-the-last PFN to allocate
7376 * @migratetype: migratetype of the underlaying pageblocks (either
7377 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
7378 * in range must have the same migratetype and it must
7379 * be either of the two.
7380 * @gfp_mask: GFP mask to use during compaction
7382 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
7383 * aligned, however it's the caller's responsibility to guarantee that
7384 * we are the only thread that changes migrate type of pageblocks the
7387 * The PFN range must belong to a single zone.
7389 * Returns zero on success or negative error code. On success all
7390 * pages which PFN is in [start, end) are allocated for the caller and
7391 * need to be freed with free_contig_range().
7393 int alloc_contig_range(unsigned long start
, unsigned long end
,
7394 unsigned migratetype
, gfp_t gfp_mask
)
7396 unsigned long outer_start
, outer_end
;
7400 struct compact_control cc
= {
7401 .nr_migratepages
= 0,
7403 .zone
= page_zone(pfn_to_page(start
)),
7404 .mode
= MIGRATE_SYNC
,
7405 .ignore_skip_hint
= true,
7406 .gfp_mask
= current_gfp_context(gfp_mask
),
7408 INIT_LIST_HEAD(&cc
.migratepages
);
7411 * What we do here is we mark all pageblocks in range as
7412 * MIGRATE_ISOLATE. Because pageblock and max order pages may
7413 * have different sizes, and due to the way page allocator
7414 * work, we align the range to biggest of the two pages so
7415 * that page allocator won't try to merge buddies from
7416 * different pageblocks and change MIGRATE_ISOLATE to some
7417 * other migration type.
7419 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
7420 * migrate the pages from an unaligned range (ie. pages that
7421 * we are interested in). This will put all the pages in
7422 * range back to page allocator as MIGRATE_ISOLATE.
7424 * When this is done, we take the pages in range from page
7425 * allocator removing them from the buddy system. This way
7426 * page allocator will never consider using them.
7428 * This lets us mark the pageblocks back as
7429 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
7430 * aligned range but not in the unaligned, original range are
7431 * put back to page allocator so that buddy can use them.
7434 ret
= start_isolate_page_range(pfn_max_align_down(start
),
7435 pfn_max_align_up(end
), migratetype
,
7441 * In case of -EBUSY, we'd like to know which page causes problem.
7442 * So, just fall through. We will check it in test_pages_isolated().
7444 ret
= __alloc_contig_migrate_range(&cc
, start
, end
);
7445 if (ret
&& ret
!= -EBUSY
)
7449 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
7450 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
7451 * more, all pages in [start, end) are free in page allocator.
7452 * What we are going to do is to allocate all pages from
7453 * [start, end) (that is remove them from page allocator).
7455 * The only problem is that pages at the beginning and at the
7456 * end of interesting range may be not aligned with pages that
7457 * page allocator holds, ie. they can be part of higher order
7458 * pages. Because of this, we reserve the bigger range and
7459 * once this is done free the pages we are not interested in.
7461 * We don't have to hold zone->lock here because the pages are
7462 * isolated thus they won't get removed from buddy.
7465 lru_add_drain_all();
7466 drain_all_pages(cc
.zone
);
7469 outer_start
= start
;
7470 while (!PageBuddy(pfn_to_page(outer_start
))) {
7471 if (++order
>= MAX_ORDER
) {
7472 outer_start
= start
;
7475 outer_start
&= ~0UL << order
;
7478 if (outer_start
!= start
) {
7479 order
= page_order(pfn_to_page(outer_start
));
7482 * outer_start page could be small order buddy page and
7483 * it doesn't include start page. Adjust outer_start
7484 * in this case to report failed page properly
7485 * on tracepoint in test_pages_isolated()
7487 if (outer_start
+ (1UL << order
) <= start
)
7488 outer_start
= start
;
7491 /* Make sure the range is really isolated. */
7492 if (test_pages_isolated(outer_start
, end
, false)) {
7493 pr_info("%s: [%lx, %lx) PFNs busy\n",
7494 __func__
, outer_start
, end
);
7499 /* Grab isolated pages from freelists. */
7500 outer_end
= isolate_freepages_range(&cc
, outer_start
, end
);
7506 /* Free head and tail (if any) */
7507 if (start
!= outer_start
)
7508 free_contig_range(outer_start
, start
- outer_start
);
7509 if (end
!= outer_end
)
7510 free_contig_range(end
, outer_end
- end
);
7513 undo_isolate_page_range(pfn_max_align_down(start
),
7514 pfn_max_align_up(end
), migratetype
);
7518 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
7520 unsigned int count
= 0;
7522 for (; nr_pages
--; pfn
++) {
7523 struct page
*page
= pfn_to_page(pfn
);
7525 count
+= page_count(page
) != 1;
7528 WARN(count
!= 0, "%d pages are still in use!\n", count
);
7532 #ifdef CONFIG_MEMORY_HOTPLUG
7534 * The zone indicated has a new number of managed_pages; batch sizes and percpu
7535 * page high values need to be recalulated.
7537 void __meminit
zone_pcp_update(struct zone
*zone
)
7540 mutex_lock(&pcp_batch_high_lock
);
7541 for_each_possible_cpu(cpu
)
7542 pageset_set_high_and_batch(zone
,
7543 per_cpu_ptr(zone
->pageset
, cpu
));
7544 mutex_unlock(&pcp_batch_high_lock
);
7548 void zone_pcp_reset(struct zone
*zone
)
7550 unsigned long flags
;
7552 struct per_cpu_pageset
*pset
;
7554 /* avoid races with drain_pages() */
7555 local_irq_save(flags
);
7556 if (zone
->pageset
!= &boot_pageset
) {
7557 for_each_online_cpu(cpu
) {
7558 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
7559 drain_zonestat(zone
, pset
);
7561 free_percpu(zone
->pageset
);
7562 zone
->pageset
= &boot_pageset
;
7564 local_irq_restore(flags
);
7567 #ifdef CONFIG_MEMORY_HOTREMOVE
7569 * All pages in the range must be in a single zone and isolated
7570 * before calling this.
7573 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
7577 unsigned int order
, i
;
7579 unsigned long flags
;
7580 /* find the first valid pfn */
7581 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
7586 zone
= page_zone(pfn_to_page(pfn
));
7587 spin_lock_irqsave(&zone
->lock
, flags
);
7589 while (pfn
< end_pfn
) {
7590 if (!pfn_valid(pfn
)) {
7594 page
= pfn_to_page(pfn
);
7596 * The HWPoisoned page may be not in buddy system, and
7597 * page_count() is not 0.
7599 if (unlikely(!PageBuddy(page
) && PageHWPoison(page
))) {
7601 SetPageReserved(page
);
7605 BUG_ON(page_count(page
));
7606 BUG_ON(!PageBuddy(page
));
7607 order
= page_order(page
);
7608 #ifdef CONFIG_DEBUG_VM
7609 pr_info("remove from free list %lx %d %lx\n",
7610 pfn
, 1 << order
, end_pfn
);
7612 list_del(&page
->lru
);
7613 rmv_page_order(page
);
7614 zone
->free_area
[order
].nr_free
--;
7615 for (i
= 0; i
< (1 << order
); i
++)
7616 SetPageReserved((page
+i
));
7617 pfn
+= (1 << order
);
7619 spin_unlock_irqrestore(&zone
->lock
, flags
);
7623 bool is_free_buddy_page(struct page
*page
)
7625 struct zone
*zone
= page_zone(page
);
7626 unsigned long pfn
= page_to_pfn(page
);
7627 unsigned long flags
;
7630 spin_lock_irqsave(&zone
->lock
, flags
);
7631 for (order
= 0; order
< MAX_ORDER
; order
++) {
7632 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
7634 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
7637 spin_unlock_irqrestore(&zone
->lock
, flags
);
7639 return order
< MAX_ORDER
;