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>
69 #include <asm/sections.h>
70 #include <asm/tlbflush.h>
71 #include <asm/div64.h>
74 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
75 static DEFINE_MUTEX(pcp_batch_high_lock
);
76 #define MIN_PERCPU_PAGELIST_FRACTION (8)
78 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
79 DEFINE_PER_CPU(int, numa_node
);
80 EXPORT_PER_CPU_SYMBOL(numa_node
);
83 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
85 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
86 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
87 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
88 * defined in <linux/topology.h>.
90 DEFINE_PER_CPU(int, _numa_mem_
); /* Kernel "local memory" node */
91 EXPORT_PER_CPU_SYMBOL(_numa_mem_
);
92 int _node_numa_mem_
[MAX_NUMNODES
];
95 /* work_structs for global per-cpu drains */
96 DEFINE_MUTEX(pcpu_drain_mutex
);
97 DEFINE_PER_CPU(struct work_struct
, pcpu_drain
);
99 #ifdef CONFIG_GCC_PLUGIN_LATENT_ENTROPY
100 volatile unsigned long latent_entropy __latent_entropy
;
101 EXPORT_SYMBOL(latent_entropy
);
105 * Array of node states.
107 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
108 [N_POSSIBLE
] = NODE_MASK_ALL
,
109 [N_ONLINE
] = { { [0] = 1UL } },
111 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
112 #ifdef CONFIG_HIGHMEM
113 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
115 #ifdef CONFIG_MOVABLE_NODE
116 [N_MEMORY
] = { { [0] = 1UL } },
118 [N_CPU
] = { { [0] = 1UL } },
121 EXPORT_SYMBOL(node_states
);
123 /* Protect totalram_pages and zone->managed_pages */
124 static DEFINE_SPINLOCK(managed_page_count_lock
);
126 unsigned long totalram_pages __read_mostly
;
127 unsigned long totalreserve_pages __read_mostly
;
128 unsigned long totalcma_pages __read_mostly
;
130 int percpu_pagelist_fraction
;
131 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
134 * A cached value of the page's pageblock's migratetype, used when the page is
135 * put on a pcplist. Used to avoid the pageblock migratetype lookup when
136 * freeing from pcplists in most cases, at the cost of possibly becoming stale.
137 * Also the migratetype set in the page does not necessarily match the pcplist
138 * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
139 * other index - this ensures that it will be put on the correct CMA freelist.
141 static inline int get_pcppage_migratetype(struct page
*page
)
146 static inline void set_pcppage_migratetype(struct page
*page
, int migratetype
)
148 page
->index
= migratetype
;
151 #ifdef CONFIG_PM_SLEEP
153 * The following functions are used by the suspend/hibernate code to temporarily
154 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
155 * while devices are suspended. To avoid races with the suspend/hibernate code,
156 * they should always be called with pm_mutex held (gfp_allowed_mask also should
157 * only be modified with pm_mutex held, unless the suspend/hibernate code is
158 * guaranteed not to run in parallel with that modification).
161 static gfp_t saved_gfp_mask
;
163 void pm_restore_gfp_mask(void)
165 WARN_ON(!mutex_is_locked(&pm_mutex
));
166 if (saved_gfp_mask
) {
167 gfp_allowed_mask
= saved_gfp_mask
;
172 void pm_restrict_gfp_mask(void)
174 WARN_ON(!mutex_is_locked(&pm_mutex
));
175 WARN_ON(saved_gfp_mask
);
176 saved_gfp_mask
= gfp_allowed_mask
;
177 gfp_allowed_mask
&= ~(__GFP_IO
| __GFP_FS
);
180 bool pm_suspended_storage(void)
182 if ((gfp_allowed_mask
& (__GFP_IO
| __GFP_FS
)) == (__GFP_IO
| __GFP_FS
))
186 #endif /* CONFIG_PM_SLEEP */
188 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
189 unsigned int pageblock_order __read_mostly
;
192 static void __free_pages_ok(struct page
*page
, unsigned int order
);
195 * results with 256, 32 in the lowmem_reserve sysctl:
196 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
197 * 1G machine -> (16M dma, 784M normal, 224M high)
198 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
199 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
200 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
202 * TBD: should special case ZONE_DMA32 machines here - in those we normally
203 * don't need any ZONE_NORMAL reservation
205 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
206 #ifdef CONFIG_ZONE_DMA
209 #ifdef CONFIG_ZONE_DMA32
212 #ifdef CONFIG_HIGHMEM
218 EXPORT_SYMBOL(totalram_pages
);
220 static char * const zone_names
[MAX_NR_ZONES
] = {
221 #ifdef CONFIG_ZONE_DMA
224 #ifdef CONFIG_ZONE_DMA32
228 #ifdef CONFIG_HIGHMEM
232 #ifdef CONFIG_ZONE_DEVICE
237 char * const migratetype_names
[MIGRATE_TYPES
] = {
245 #ifdef CONFIG_MEMORY_ISOLATION
250 compound_page_dtor
* const compound_page_dtors
[] = {
253 #ifdef CONFIG_HUGETLB_PAGE
256 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
261 int min_free_kbytes
= 1024;
262 int user_min_free_kbytes
= -1;
263 int watermark_scale_factor
= 10;
265 static unsigned long __meminitdata nr_kernel_pages
;
266 static unsigned long __meminitdata nr_all_pages
;
267 static unsigned long __meminitdata dma_reserve
;
269 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
270 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
271 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
272 static unsigned long __initdata required_kernelcore
;
273 static unsigned long __initdata required_movablecore
;
274 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
275 static bool mirrored_kernelcore
;
277 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
279 EXPORT_SYMBOL(movable_zone
);
280 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
283 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
284 int nr_online_nodes __read_mostly
= 1;
285 EXPORT_SYMBOL(nr_node_ids
);
286 EXPORT_SYMBOL(nr_online_nodes
);
289 int page_group_by_mobility_disabled __read_mostly
;
291 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
292 static inline void reset_deferred_meminit(pg_data_t
*pgdat
)
294 pgdat
->first_deferred_pfn
= ULONG_MAX
;
297 /* Returns true if the struct page for the pfn is uninitialised */
298 static inline bool __meminit
early_page_uninitialised(unsigned long pfn
)
300 int nid
= early_pfn_to_nid(pfn
);
302 if (node_online(nid
) && pfn
>= NODE_DATA(nid
)->first_deferred_pfn
)
309 * Returns false when the remaining initialisation should be deferred until
310 * later in the boot cycle when it can be parallelised.
312 static inline bool update_defer_init(pg_data_t
*pgdat
,
313 unsigned long pfn
, unsigned long zone_end
,
314 unsigned long *nr_initialised
)
316 unsigned long max_initialise
;
318 /* Always populate low zones for address-contrained allocations */
319 if (zone_end
< pgdat_end_pfn(pgdat
))
322 * Initialise at least 2G of a node but also take into account that
323 * two large system hashes that can take up 1GB for 0.25TB/node.
325 max_initialise
= max(2UL << (30 - PAGE_SHIFT
),
326 (pgdat
->node_spanned_pages
>> 8));
329 if ((*nr_initialised
> max_initialise
) &&
330 (pfn
& (PAGES_PER_SECTION
- 1)) == 0) {
331 pgdat
->first_deferred_pfn
= pfn
;
338 static inline void reset_deferred_meminit(pg_data_t
*pgdat
)
342 static inline bool early_page_uninitialised(unsigned long pfn
)
347 static inline bool update_defer_init(pg_data_t
*pgdat
,
348 unsigned long pfn
, unsigned long zone_end
,
349 unsigned long *nr_initialised
)
355 /* Return a pointer to the bitmap storing bits affecting a block of pages */
356 static inline unsigned long *get_pageblock_bitmap(struct page
*page
,
359 #ifdef CONFIG_SPARSEMEM
360 return __pfn_to_section(pfn
)->pageblock_flags
;
362 return page_zone(page
)->pageblock_flags
;
363 #endif /* CONFIG_SPARSEMEM */
366 static inline int pfn_to_bitidx(struct page
*page
, unsigned long pfn
)
368 #ifdef CONFIG_SPARSEMEM
369 pfn
&= (PAGES_PER_SECTION
-1);
370 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
372 pfn
= pfn
- round_down(page_zone(page
)->zone_start_pfn
, pageblock_nr_pages
);
373 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
374 #endif /* CONFIG_SPARSEMEM */
378 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
379 * @page: The page within the block of interest
380 * @pfn: The target page frame number
381 * @end_bitidx: The last bit of interest to retrieve
382 * @mask: mask of bits that the caller is interested in
384 * Return: pageblock_bits flags
386 static __always_inline
unsigned long __get_pfnblock_flags_mask(struct page
*page
,
388 unsigned long end_bitidx
,
391 unsigned long *bitmap
;
392 unsigned long bitidx
, word_bitidx
;
395 bitmap
= get_pageblock_bitmap(page
, pfn
);
396 bitidx
= pfn_to_bitidx(page
, pfn
);
397 word_bitidx
= bitidx
/ BITS_PER_LONG
;
398 bitidx
&= (BITS_PER_LONG
-1);
400 word
= bitmap
[word_bitidx
];
401 bitidx
+= end_bitidx
;
402 return (word
>> (BITS_PER_LONG
- bitidx
- 1)) & mask
;
405 unsigned long get_pfnblock_flags_mask(struct page
*page
, unsigned long pfn
,
406 unsigned long end_bitidx
,
409 return __get_pfnblock_flags_mask(page
, pfn
, end_bitidx
, mask
);
412 static __always_inline
int get_pfnblock_migratetype(struct page
*page
, unsigned long pfn
)
414 return __get_pfnblock_flags_mask(page
, pfn
, PB_migrate_end
, MIGRATETYPE_MASK
);
418 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
419 * @page: The page within the block of interest
420 * @flags: The flags to set
421 * @pfn: The target page frame number
422 * @end_bitidx: The last bit of interest
423 * @mask: mask of bits that the caller is interested in
425 void set_pfnblock_flags_mask(struct page
*page
, unsigned long flags
,
427 unsigned long end_bitidx
,
430 unsigned long *bitmap
;
431 unsigned long bitidx
, word_bitidx
;
432 unsigned long old_word
, word
;
434 BUILD_BUG_ON(NR_PAGEBLOCK_BITS
!= 4);
436 bitmap
= get_pageblock_bitmap(page
, pfn
);
437 bitidx
= pfn_to_bitidx(page
, pfn
);
438 word_bitidx
= bitidx
/ BITS_PER_LONG
;
439 bitidx
&= (BITS_PER_LONG
-1);
441 VM_BUG_ON_PAGE(!zone_spans_pfn(page_zone(page
), pfn
), page
);
443 bitidx
+= end_bitidx
;
444 mask
<<= (BITS_PER_LONG
- bitidx
- 1);
445 flags
<<= (BITS_PER_LONG
- bitidx
- 1);
447 word
= READ_ONCE(bitmap
[word_bitidx
]);
449 old_word
= cmpxchg(&bitmap
[word_bitidx
], word
, (word
& ~mask
) | flags
);
450 if (word
== old_word
)
456 void set_pageblock_migratetype(struct page
*page
, int migratetype
)
458 if (unlikely(page_group_by_mobility_disabled
&&
459 migratetype
< MIGRATE_PCPTYPES
))
460 migratetype
= MIGRATE_UNMOVABLE
;
462 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
463 PB_migrate
, PB_migrate_end
);
466 #ifdef CONFIG_DEBUG_VM
467 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
471 unsigned long pfn
= page_to_pfn(page
);
472 unsigned long sp
, start_pfn
;
475 seq
= zone_span_seqbegin(zone
);
476 start_pfn
= zone
->zone_start_pfn
;
477 sp
= zone
->spanned_pages
;
478 if (!zone_spans_pfn(zone
, pfn
))
480 } while (zone_span_seqretry(zone
, seq
));
483 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
484 pfn
, zone_to_nid(zone
), zone
->name
,
485 start_pfn
, start_pfn
+ sp
);
490 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
492 if (!pfn_valid_within(page_to_pfn(page
)))
494 if (zone
!= page_zone(page
))
500 * Temporary debugging check for pages not lying within a given zone.
502 static int bad_range(struct zone
*zone
, struct page
*page
)
504 if (page_outside_zone_boundaries(zone
, page
))
506 if (!page_is_consistent(zone
, page
))
512 static inline int bad_range(struct zone
*zone
, struct page
*page
)
518 static void bad_page(struct page
*page
, const char *reason
,
519 unsigned long bad_flags
)
521 static unsigned long resume
;
522 static unsigned long nr_shown
;
523 static unsigned long nr_unshown
;
526 * Allow a burst of 60 reports, then keep quiet for that minute;
527 * or allow a steady drip of one report per second.
529 if (nr_shown
== 60) {
530 if (time_before(jiffies
, resume
)) {
536 "BUG: Bad page state: %lu messages suppressed\n",
543 resume
= jiffies
+ 60 * HZ
;
545 pr_alert("BUG: Bad page state in process %s pfn:%05lx\n",
546 current
->comm
, page_to_pfn(page
));
547 __dump_page(page
, reason
);
548 bad_flags
&= page
->flags
;
550 pr_alert("bad because of flags: %#lx(%pGp)\n",
551 bad_flags
, &bad_flags
);
552 dump_page_owner(page
);
557 /* Leave bad fields for debug, except PageBuddy could make trouble */
558 page_mapcount_reset(page
); /* remove PageBuddy */
559 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
563 * Higher-order pages are called "compound pages". They are structured thusly:
565 * The first PAGE_SIZE page is called the "head page" and have PG_head set.
567 * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
568 * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
570 * The first tail page's ->compound_dtor holds the offset in array of compound
571 * page destructors. See compound_page_dtors.
573 * The first tail page's ->compound_order holds the order of allocation.
574 * This usage means that zero-order pages may not be compound.
577 void free_compound_page(struct page
*page
)
579 __free_pages_ok(page
, compound_order(page
));
582 void prep_compound_page(struct page
*page
, unsigned int order
)
585 int nr_pages
= 1 << order
;
587 set_compound_page_dtor(page
, COMPOUND_PAGE_DTOR
);
588 set_compound_order(page
, order
);
590 for (i
= 1; i
< nr_pages
; i
++) {
591 struct page
*p
= page
+ i
;
592 set_page_count(p
, 0);
593 p
->mapping
= TAIL_MAPPING
;
594 set_compound_head(p
, page
);
596 atomic_set(compound_mapcount_ptr(page
), -1);
599 #ifdef CONFIG_DEBUG_PAGEALLOC
600 unsigned int _debug_guardpage_minorder
;
601 bool _debug_pagealloc_enabled __read_mostly
602 = IS_ENABLED(CONFIG_DEBUG_PAGEALLOC_ENABLE_DEFAULT
);
603 EXPORT_SYMBOL(_debug_pagealloc_enabled
);
604 bool _debug_guardpage_enabled __read_mostly
;
606 static int __init
early_debug_pagealloc(char *buf
)
610 return kstrtobool(buf
, &_debug_pagealloc_enabled
);
612 early_param("debug_pagealloc", early_debug_pagealloc
);
614 static bool need_debug_guardpage(void)
616 /* If we don't use debug_pagealloc, we don't need guard page */
617 if (!debug_pagealloc_enabled())
620 if (!debug_guardpage_minorder())
626 static void init_debug_guardpage(void)
628 if (!debug_pagealloc_enabled())
631 if (!debug_guardpage_minorder())
634 _debug_guardpage_enabled
= true;
637 struct page_ext_operations debug_guardpage_ops
= {
638 .need
= need_debug_guardpage
,
639 .init
= init_debug_guardpage
,
642 static int __init
debug_guardpage_minorder_setup(char *buf
)
646 if (kstrtoul(buf
, 10, &res
) < 0 || res
> MAX_ORDER
/ 2) {
647 pr_err("Bad debug_guardpage_minorder value\n");
650 _debug_guardpage_minorder
= res
;
651 pr_info("Setting debug_guardpage_minorder to %lu\n", res
);
654 early_param("debug_guardpage_minorder", debug_guardpage_minorder_setup
);
656 static inline bool set_page_guard(struct zone
*zone
, struct page
*page
,
657 unsigned int order
, int migratetype
)
659 struct page_ext
*page_ext
;
661 if (!debug_guardpage_enabled())
664 if (order
>= debug_guardpage_minorder())
667 page_ext
= lookup_page_ext(page
);
668 if (unlikely(!page_ext
))
671 __set_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
673 INIT_LIST_HEAD(&page
->lru
);
674 set_page_private(page
, order
);
675 /* Guard pages are not available for any usage */
676 __mod_zone_freepage_state(zone
, -(1 << order
), migratetype
);
681 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
682 unsigned int order
, int migratetype
)
684 struct page_ext
*page_ext
;
686 if (!debug_guardpage_enabled())
689 page_ext
= lookup_page_ext(page
);
690 if (unlikely(!page_ext
))
693 __clear_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
695 set_page_private(page
, 0);
696 if (!is_migrate_isolate(migratetype
))
697 __mod_zone_freepage_state(zone
, (1 << order
), migratetype
);
700 struct page_ext_operations debug_guardpage_ops
;
701 static inline bool set_page_guard(struct zone
*zone
, struct page
*page
,
702 unsigned int order
, int migratetype
) { return false; }
703 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
704 unsigned int order
, int migratetype
) {}
707 static inline void set_page_order(struct page
*page
, unsigned int order
)
709 set_page_private(page
, order
);
710 __SetPageBuddy(page
);
713 static inline void rmv_page_order(struct page
*page
)
715 __ClearPageBuddy(page
);
716 set_page_private(page
, 0);
720 * This function checks whether a page is free && is the buddy
721 * we can do coalesce a page and its buddy if
722 * (a) the buddy is not in a hole (check before calling!) &&
723 * (b) the buddy is in the buddy system &&
724 * (c) a page and its buddy have the same order &&
725 * (d) a page and its buddy are in the same zone.
727 * For recording whether a page is in the buddy system, we set ->_mapcount
728 * PAGE_BUDDY_MAPCOUNT_VALUE.
729 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
730 * serialized by zone->lock.
732 * For recording page's order, we use page_private(page).
734 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
737 if (page_is_guard(buddy
) && page_order(buddy
) == order
) {
738 if (page_zone_id(page
) != page_zone_id(buddy
))
741 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
746 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
748 * zone check is done late to avoid uselessly
749 * calculating zone/node ids for pages that could
752 if (page_zone_id(page
) != page_zone_id(buddy
))
755 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
763 * Freeing function for a buddy system allocator.
765 * The concept of a buddy system is to maintain direct-mapped table
766 * (containing bit values) for memory blocks of various "orders".
767 * The bottom level table contains the map for the smallest allocatable
768 * units of memory (here, pages), and each level above it describes
769 * pairs of units from the levels below, hence, "buddies".
770 * At a high level, all that happens here is marking the table entry
771 * at the bottom level available, and propagating the changes upward
772 * as necessary, plus some accounting needed to play nicely with other
773 * parts of the VM system.
774 * At each level, we keep a list of pages, which are heads of continuous
775 * free pages of length of (1 << order) and marked with _mapcount
776 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
778 * So when we are allocating or freeing one, we can derive the state of the
779 * other. That is, if we allocate a small block, and both were
780 * free, the remainder of the region must be split into blocks.
781 * If a block is freed, and its buddy is also free, then this
782 * triggers coalescing into a block of larger size.
787 static inline void __free_one_page(struct page
*page
,
789 struct zone
*zone
, unsigned int order
,
792 unsigned long combined_pfn
;
793 unsigned long uninitialized_var(buddy_pfn
);
795 unsigned int max_order
;
797 max_order
= min_t(unsigned int, MAX_ORDER
, pageblock_order
+ 1);
799 VM_BUG_ON(!zone_is_initialized(zone
));
800 VM_BUG_ON_PAGE(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
, page
);
802 VM_BUG_ON(migratetype
== -1);
803 if (likely(!is_migrate_isolate(migratetype
)))
804 __mod_zone_freepage_state(zone
, 1 << order
, migratetype
);
806 VM_BUG_ON_PAGE(pfn
& ((1 << order
) - 1), page
);
807 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
810 while (order
< max_order
- 1) {
811 buddy_pfn
= __find_buddy_pfn(pfn
, order
);
812 buddy
= page
+ (buddy_pfn
- pfn
);
814 if (!pfn_valid_within(buddy_pfn
))
816 if (!page_is_buddy(page
, buddy
, order
))
819 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
820 * merge with it and move up one order.
822 if (page_is_guard(buddy
)) {
823 clear_page_guard(zone
, buddy
, order
, migratetype
);
825 list_del(&buddy
->lru
);
826 zone
->free_area
[order
].nr_free
--;
827 rmv_page_order(buddy
);
829 combined_pfn
= buddy_pfn
& pfn
;
830 page
= page
+ (combined_pfn
- pfn
);
834 if (max_order
< MAX_ORDER
) {
835 /* If we are here, it means order is >= pageblock_order.
836 * We want to prevent merge between freepages on isolate
837 * pageblock and normal pageblock. Without this, pageblock
838 * isolation could cause incorrect freepage or CMA accounting.
840 * We don't want to hit this code for the more frequent
843 if (unlikely(has_isolate_pageblock(zone
))) {
846 buddy_pfn
= __find_buddy_pfn(pfn
, order
);
847 buddy
= page
+ (buddy_pfn
- pfn
);
848 buddy_mt
= get_pageblock_migratetype(buddy
);
850 if (migratetype
!= buddy_mt
851 && (is_migrate_isolate(migratetype
) ||
852 is_migrate_isolate(buddy_mt
)))
856 goto continue_merging
;
860 set_page_order(page
, order
);
863 * If this is not the largest possible page, check if the buddy
864 * of the next-highest order is free. If it is, it's possible
865 * that pages are being freed that will coalesce soon. In case,
866 * that is happening, add the free page to the tail of the list
867 * so it's less likely to be used soon and more likely to be merged
868 * as a higher order page
870 if ((order
< MAX_ORDER
-2) && pfn_valid_within(buddy_pfn
)) {
871 struct page
*higher_page
, *higher_buddy
;
872 combined_pfn
= buddy_pfn
& pfn
;
873 higher_page
= page
+ (combined_pfn
- pfn
);
874 buddy_pfn
= __find_buddy_pfn(combined_pfn
, order
+ 1);
875 higher_buddy
= higher_page
+ (buddy_pfn
- combined_pfn
);
876 if (pfn_valid_within(buddy_pfn
) &&
877 page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
878 list_add_tail(&page
->lru
,
879 &zone
->free_area
[order
].free_list
[migratetype
]);
884 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
886 zone
->free_area
[order
].nr_free
++;
890 * A bad page could be due to a number of fields. Instead of multiple branches,
891 * try and check multiple fields with one check. The caller must do a detailed
892 * check if necessary.
894 static inline bool page_expected_state(struct page
*page
,
895 unsigned long check_flags
)
897 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
900 if (unlikely((unsigned long)page
->mapping
|
901 page_ref_count(page
) |
903 (unsigned long)page
->mem_cgroup
|
905 (page
->flags
& check_flags
)))
911 static void free_pages_check_bad(struct page
*page
)
913 const char *bad_reason
;
914 unsigned long bad_flags
;
919 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
920 bad_reason
= "nonzero mapcount";
921 if (unlikely(page
->mapping
!= NULL
))
922 bad_reason
= "non-NULL mapping";
923 if (unlikely(page_ref_count(page
) != 0))
924 bad_reason
= "nonzero _refcount";
925 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
)) {
926 bad_reason
= "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
927 bad_flags
= PAGE_FLAGS_CHECK_AT_FREE
;
930 if (unlikely(page
->mem_cgroup
))
931 bad_reason
= "page still charged to cgroup";
933 bad_page(page
, bad_reason
, bad_flags
);
936 static inline int free_pages_check(struct page
*page
)
938 if (likely(page_expected_state(page
, PAGE_FLAGS_CHECK_AT_FREE
)))
941 /* Something has gone sideways, find it */
942 free_pages_check_bad(page
);
946 static int free_tail_pages_check(struct page
*head_page
, struct page
*page
)
951 * We rely page->lru.next never has bit 0 set, unless the page
952 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
954 BUILD_BUG_ON((unsigned long)LIST_POISON1
& 1);
956 if (!IS_ENABLED(CONFIG_DEBUG_VM
)) {
960 switch (page
- head_page
) {
962 /* the first tail page: ->mapping is compound_mapcount() */
963 if (unlikely(compound_mapcount(page
))) {
964 bad_page(page
, "nonzero compound_mapcount", 0);
970 * the second tail page: ->mapping is
971 * page_deferred_list().next -- ignore value.
975 if (page
->mapping
!= TAIL_MAPPING
) {
976 bad_page(page
, "corrupted mapping in tail page", 0);
981 if (unlikely(!PageTail(page
))) {
982 bad_page(page
, "PageTail not set", 0);
985 if (unlikely(compound_head(page
) != head_page
)) {
986 bad_page(page
, "compound_head not consistent", 0);
991 page
->mapping
= NULL
;
992 clear_compound_head(page
);
996 static __always_inline
bool free_pages_prepare(struct page
*page
,
997 unsigned int order
, bool check_free
)
1001 VM_BUG_ON_PAGE(PageTail(page
), page
);
1003 trace_mm_page_free(page
, order
);
1004 kmemcheck_free_shadow(page
, order
);
1007 * Check tail pages before head page information is cleared to
1008 * avoid checking PageCompound for order-0 pages.
1010 if (unlikely(order
)) {
1011 bool compound
= PageCompound(page
);
1014 VM_BUG_ON_PAGE(compound
&& compound_order(page
) != order
, page
);
1017 ClearPageDoubleMap(page
);
1018 for (i
= 1; i
< (1 << order
); i
++) {
1020 bad
+= free_tail_pages_check(page
, page
+ i
);
1021 if (unlikely(free_pages_check(page
+ i
))) {
1025 (page
+ i
)->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
1028 if (PageMappingFlags(page
))
1029 page
->mapping
= NULL
;
1030 if (memcg_kmem_enabled() && PageKmemcg(page
))
1031 memcg_kmem_uncharge(page
, order
);
1033 bad
+= free_pages_check(page
);
1037 page_cpupid_reset_last(page
);
1038 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
1039 reset_page_owner(page
, order
);
1041 if (!PageHighMem(page
)) {
1042 debug_check_no_locks_freed(page_address(page
),
1043 PAGE_SIZE
<< order
);
1044 debug_check_no_obj_freed(page_address(page
),
1045 PAGE_SIZE
<< order
);
1047 arch_free_page(page
, order
);
1048 kernel_poison_pages(page
, 1 << order
, 0);
1049 kernel_map_pages(page
, 1 << order
, 0);
1050 kasan_free_pages(page
, order
);
1055 #ifdef CONFIG_DEBUG_VM
1056 static inline bool free_pcp_prepare(struct page
*page
)
1058 return free_pages_prepare(page
, 0, true);
1061 static inline bool bulkfree_pcp_prepare(struct page
*page
)
1066 static bool free_pcp_prepare(struct page
*page
)
1068 return free_pages_prepare(page
, 0, false);
1071 static bool bulkfree_pcp_prepare(struct page
*page
)
1073 return free_pages_check(page
);
1075 #endif /* CONFIG_DEBUG_VM */
1078 * Frees a number of pages from the PCP lists
1079 * Assumes all pages on list are in same zone, and of same order.
1080 * count is the number of pages to free.
1082 * If the zone was previously in an "all pages pinned" state then look to
1083 * see if this freeing clears that state.
1085 * And clear the zone's pages_scanned counter, to hold off the "all pages are
1086 * pinned" detection logic.
1088 static void free_pcppages_bulk(struct zone
*zone
, int count
,
1089 struct per_cpu_pages
*pcp
)
1091 int migratetype
= 0;
1093 bool isolated_pageblocks
;
1095 spin_lock(&zone
->lock
);
1096 isolated_pageblocks
= has_isolate_pageblock(zone
);
1100 struct list_head
*list
;
1103 * Remove pages from lists in a round-robin fashion. A
1104 * batch_free count is maintained that is incremented when an
1105 * empty list is encountered. This is so more pages are freed
1106 * off fuller lists instead of spinning excessively around empty
1111 if (++migratetype
== MIGRATE_PCPTYPES
)
1113 list
= &pcp
->lists
[migratetype
];
1114 } while (list_empty(list
));
1116 /* This is the only non-empty list. Free them all. */
1117 if (batch_free
== MIGRATE_PCPTYPES
)
1121 int mt
; /* migratetype of the to-be-freed page */
1123 page
= list_last_entry(list
, struct page
, lru
);
1124 /* must delete as __free_one_page list manipulates */
1125 list_del(&page
->lru
);
1127 mt
= get_pcppage_migratetype(page
);
1128 /* MIGRATE_ISOLATE page should not go to pcplists */
1129 VM_BUG_ON_PAGE(is_migrate_isolate(mt
), page
);
1130 /* Pageblock could have been isolated meanwhile */
1131 if (unlikely(isolated_pageblocks
))
1132 mt
= get_pageblock_migratetype(page
);
1134 if (bulkfree_pcp_prepare(page
))
1137 __free_one_page(page
, page_to_pfn(page
), zone
, 0, mt
);
1138 trace_mm_page_pcpu_drain(page
, 0, mt
);
1139 } while (--count
&& --batch_free
&& !list_empty(list
));
1141 spin_unlock(&zone
->lock
);
1144 static void free_one_page(struct zone
*zone
,
1145 struct page
*page
, unsigned long pfn
,
1149 spin_lock(&zone
->lock
);
1150 if (unlikely(has_isolate_pageblock(zone
) ||
1151 is_migrate_isolate(migratetype
))) {
1152 migratetype
= get_pfnblock_migratetype(page
, pfn
);
1154 __free_one_page(page
, pfn
, zone
, order
, migratetype
);
1155 spin_unlock(&zone
->lock
);
1158 static void __meminit
__init_single_page(struct page
*page
, unsigned long pfn
,
1159 unsigned long zone
, int nid
)
1161 set_page_links(page
, zone
, nid
, pfn
);
1162 init_page_count(page
);
1163 page_mapcount_reset(page
);
1164 page_cpupid_reset_last(page
);
1166 INIT_LIST_HEAD(&page
->lru
);
1167 #ifdef WANT_PAGE_VIRTUAL
1168 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1169 if (!is_highmem_idx(zone
))
1170 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
1174 static void __meminit
__init_single_pfn(unsigned long pfn
, unsigned long zone
,
1177 return __init_single_page(pfn_to_page(pfn
), pfn
, zone
, nid
);
1180 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1181 static void init_reserved_page(unsigned long pfn
)
1186 if (!early_page_uninitialised(pfn
))
1189 nid
= early_pfn_to_nid(pfn
);
1190 pgdat
= NODE_DATA(nid
);
1192 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
1193 struct zone
*zone
= &pgdat
->node_zones
[zid
];
1195 if (pfn
>= zone
->zone_start_pfn
&& pfn
< zone_end_pfn(zone
))
1198 __init_single_pfn(pfn
, zid
, nid
);
1201 static inline void init_reserved_page(unsigned long pfn
)
1204 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1207 * Initialised pages do not have PageReserved set. This function is
1208 * called for each range allocated by the bootmem allocator and
1209 * marks the pages PageReserved. The remaining valid pages are later
1210 * sent to the buddy page allocator.
1212 void __meminit
reserve_bootmem_region(phys_addr_t start
, phys_addr_t end
)
1214 unsigned long start_pfn
= PFN_DOWN(start
);
1215 unsigned long end_pfn
= PFN_UP(end
);
1217 for (; start_pfn
< end_pfn
; start_pfn
++) {
1218 if (pfn_valid(start_pfn
)) {
1219 struct page
*page
= pfn_to_page(start_pfn
);
1221 init_reserved_page(start_pfn
);
1223 /* Avoid false-positive PageTail() */
1224 INIT_LIST_HEAD(&page
->lru
);
1226 SetPageReserved(page
);
1231 static void __free_pages_ok(struct page
*page
, unsigned int order
)
1233 unsigned long flags
;
1235 unsigned long pfn
= page_to_pfn(page
);
1237 if (!free_pages_prepare(page
, order
, true))
1240 migratetype
= get_pfnblock_migratetype(page
, pfn
);
1241 local_irq_save(flags
);
1242 __count_vm_events(PGFREE
, 1 << order
);
1243 free_one_page(page_zone(page
), page
, pfn
, order
, migratetype
);
1244 local_irq_restore(flags
);
1247 static void __init
__free_pages_boot_core(struct page
*page
, unsigned int order
)
1249 unsigned int nr_pages
= 1 << order
;
1250 struct page
*p
= page
;
1254 for (loop
= 0; loop
< (nr_pages
- 1); loop
++, p
++) {
1256 __ClearPageReserved(p
);
1257 set_page_count(p
, 0);
1259 __ClearPageReserved(p
);
1260 set_page_count(p
, 0);
1262 page_zone(page
)->managed_pages
+= nr_pages
;
1263 set_page_refcounted(page
);
1264 __free_pages(page
, order
);
1267 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1268 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1270 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata
;
1272 int __meminit
early_pfn_to_nid(unsigned long pfn
)
1274 static DEFINE_SPINLOCK(early_pfn_lock
);
1277 spin_lock(&early_pfn_lock
);
1278 nid
= __early_pfn_to_nid(pfn
, &early_pfnnid_cache
);
1280 nid
= first_online_node
;
1281 spin_unlock(&early_pfn_lock
);
1287 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1288 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1289 struct mminit_pfnnid_cache
*state
)
1293 nid
= __early_pfn_to_nid(pfn
, state
);
1294 if (nid
>= 0 && nid
!= node
)
1299 /* Only safe to use early in boot when initialisation is single-threaded */
1300 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1302 return meminit_pfn_in_nid(pfn
, node
, &early_pfnnid_cache
);
1307 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1311 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1312 struct mminit_pfnnid_cache
*state
)
1319 void __init
__free_pages_bootmem(struct page
*page
, unsigned long pfn
,
1322 if (early_page_uninitialised(pfn
))
1324 return __free_pages_boot_core(page
, order
);
1328 * Check that the whole (or subset of) a pageblock given by the interval of
1329 * [start_pfn, end_pfn) is valid and within the same zone, before scanning it
1330 * with the migration of free compaction scanner. The scanners then need to
1331 * use only pfn_valid_within() check for arches that allow holes within
1334 * Return struct page pointer of start_pfn, or NULL if checks were not passed.
1336 * It's possible on some configurations to have a setup like node0 node1 node0
1337 * i.e. it's possible that all pages within a zones range of pages do not
1338 * belong to a single zone. We assume that a border between node0 and node1
1339 * can occur within a single pageblock, but not a node0 node1 node0
1340 * interleaving within a single pageblock. It is therefore sufficient to check
1341 * the first and last page of a pageblock and avoid checking each individual
1342 * page in a pageblock.
1344 struct page
*__pageblock_pfn_to_page(unsigned long start_pfn
,
1345 unsigned long end_pfn
, struct zone
*zone
)
1347 struct page
*start_page
;
1348 struct page
*end_page
;
1350 /* end_pfn is one past the range we are checking */
1353 if (!pfn_valid(start_pfn
) || !pfn_valid(end_pfn
))
1356 start_page
= pfn_to_page(start_pfn
);
1358 if (page_zone(start_page
) != zone
)
1361 end_page
= pfn_to_page(end_pfn
);
1363 /* This gives a shorter code than deriving page_zone(end_page) */
1364 if (page_zone_id(start_page
) != page_zone_id(end_page
))
1370 void set_zone_contiguous(struct zone
*zone
)
1372 unsigned long block_start_pfn
= zone
->zone_start_pfn
;
1373 unsigned long block_end_pfn
;
1375 block_end_pfn
= ALIGN(block_start_pfn
+ 1, pageblock_nr_pages
);
1376 for (; block_start_pfn
< zone_end_pfn(zone
);
1377 block_start_pfn
= block_end_pfn
,
1378 block_end_pfn
+= pageblock_nr_pages
) {
1380 block_end_pfn
= min(block_end_pfn
, zone_end_pfn(zone
));
1382 if (!__pageblock_pfn_to_page(block_start_pfn
,
1383 block_end_pfn
, zone
))
1387 /* We confirm that there is no hole */
1388 zone
->contiguous
= true;
1391 void clear_zone_contiguous(struct zone
*zone
)
1393 zone
->contiguous
= false;
1396 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1397 static void __init
deferred_free_range(struct page
*page
,
1398 unsigned long pfn
, int nr_pages
)
1405 /* Free a large naturally-aligned chunk if possible */
1406 if (nr_pages
== pageblock_nr_pages
&&
1407 (pfn
& (pageblock_nr_pages
- 1)) == 0) {
1408 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
1409 __free_pages_boot_core(page
, pageblock_order
);
1413 for (i
= 0; i
< nr_pages
; i
++, page
++, pfn
++) {
1414 if ((pfn
& (pageblock_nr_pages
- 1)) == 0)
1415 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
1416 __free_pages_boot_core(page
, 0);
1420 /* Completion tracking for deferred_init_memmap() threads */
1421 static atomic_t pgdat_init_n_undone __initdata
;
1422 static __initdata
DECLARE_COMPLETION(pgdat_init_all_done_comp
);
1424 static inline void __init
pgdat_init_report_one_done(void)
1426 if (atomic_dec_and_test(&pgdat_init_n_undone
))
1427 complete(&pgdat_init_all_done_comp
);
1430 /* Initialise remaining memory on a node */
1431 static int __init
deferred_init_memmap(void *data
)
1433 pg_data_t
*pgdat
= data
;
1434 int nid
= pgdat
->node_id
;
1435 struct mminit_pfnnid_cache nid_init_state
= { };
1436 unsigned long start
= jiffies
;
1437 unsigned long nr_pages
= 0;
1438 unsigned long walk_start
, walk_end
;
1441 unsigned long first_init_pfn
= pgdat
->first_deferred_pfn
;
1442 const struct cpumask
*cpumask
= cpumask_of_node(pgdat
->node_id
);
1444 if (first_init_pfn
== ULONG_MAX
) {
1445 pgdat_init_report_one_done();
1449 /* Bind memory initialisation thread to a local node if possible */
1450 if (!cpumask_empty(cpumask
))
1451 set_cpus_allowed_ptr(current
, cpumask
);
1453 /* Sanity check boundaries */
1454 BUG_ON(pgdat
->first_deferred_pfn
< pgdat
->node_start_pfn
);
1455 BUG_ON(pgdat
->first_deferred_pfn
> pgdat_end_pfn(pgdat
));
1456 pgdat
->first_deferred_pfn
= ULONG_MAX
;
1458 /* Only the highest zone is deferred so find it */
1459 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
1460 zone
= pgdat
->node_zones
+ zid
;
1461 if (first_init_pfn
< zone_end_pfn(zone
))
1465 for_each_mem_pfn_range(i
, nid
, &walk_start
, &walk_end
, NULL
) {
1466 unsigned long pfn
, end_pfn
;
1467 struct page
*page
= NULL
;
1468 struct page
*free_base_page
= NULL
;
1469 unsigned long free_base_pfn
= 0;
1472 end_pfn
= min(walk_end
, zone_end_pfn(zone
));
1473 pfn
= first_init_pfn
;
1474 if (pfn
< walk_start
)
1476 if (pfn
< zone
->zone_start_pfn
)
1477 pfn
= zone
->zone_start_pfn
;
1479 for (; pfn
< end_pfn
; pfn
++) {
1480 if (!pfn_valid_within(pfn
))
1484 * Ensure pfn_valid is checked every
1485 * pageblock_nr_pages for memory holes
1487 if ((pfn
& (pageblock_nr_pages
- 1)) == 0) {
1488 if (!pfn_valid(pfn
)) {
1494 if (!meminit_pfn_in_nid(pfn
, nid
, &nid_init_state
)) {
1499 /* Minimise pfn page lookups and scheduler checks */
1500 if (page
&& (pfn
& (pageblock_nr_pages
- 1)) != 0) {
1503 nr_pages
+= nr_to_free
;
1504 deferred_free_range(free_base_page
,
1505 free_base_pfn
, nr_to_free
);
1506 free_base_page
= NULL
;
1507 free_base_pfn
= nr_to_free
= 0;
1509 page
= pfn_to_page(pfn
);
1514 VM_BUG_ON(page_zone(page
) != zone
);
1518 __init_single_page(page
, pfn
, zid
, nid
);
1519 if (!free_base_page
) {
1520 free_base_page
= page
;
1521 free_base_pfn
= pfn
;
1526 /* Where possible, batch up pages for a single free */
1529 /* Free the current block of pages to allocator */
1530 nr_pages
+= nr_to_free
;
1531 deferred_free_range(free_base_page
, free_base_pfn
,
1533 free_base_page
= NULL
;
1534 free_base_pfn
= nr_to_free
= 0;
1536 /* Free the last block of pages to allocator */
1537 nr_pages
+= nr_to_free
;
1538 deferred_free_range(free_base_page
, free_base_pfn
, nr_to_free
);
1540 first_init_pfn
= max(end_pfn
, first_init_pfn
);
1543 /* Sanity check that the next zone really is unpopulated */
1544 WARN_ON(++zid
< MAX_NR_ZONES
&& populated_zone(++zone
));
1546 pr_info("node %d initialised, %lu pages in %ums\n", nid
, nr_pages
,
1547 jiffies_to_msecs(jiffies
- start
));
1549 pgdat_init_report_one_done();
1552 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1554 void __init
page_alloc_init_late(void)
1558 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1561 /* There will be num_node_state(N_MEMORY) threads */
1562 atomic_set(&pgdat_init_n_undone
, num_node_state(N_MEMORY
));
1563 for_each_node_state(nid
, N_MEMORY
) {
1564 kthread_run(deferred_init_memmap
, NODE_DATA(nid
), "pgdatinit%d", nid
);
1567 /* Block until all are initialised */
1568 wait_for_completion(&pgdat_init_all_done_comp
);
1570 /* Reinit limits that are based on free pages after the kernel is up */
1571 files_maxfiles_init();
1574 for_each_populated_zone(zone
)
1575 set_zone_contiguous(zone
);
1579 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1580 void __init
init_cma_reserved_pageblock(struct page
*page
)
1582 unsigned i
= pageblock_nr_pages
;
1583 struct page
*p
= page
;
1586 __ClearPageReserved(p
);
1587 set_page_count(p
, 0);
1590 set_pageblock_migratetype(page
, MIGRATE_CMA
);
1592 if (pageblock_order
>= MAX_ORDER
) {
1593 i
= pageblock_nr_pages
;
1596 set_page_refcounted(p
);
1597 __free_pages(p
, MAX_ORDER
- 1);
1598 p
+= MAX_ORDER_NR_PAGES
;
1599 } while (i
-= MAX_ORDER_NR_PAGES
);
1601 set_page_refcounted(page
);
1602 __free_pages(page
, pageblock_order
);
1605 adjust_managed_page_count(page
, pageblock_nr_pages
);
1610 * The order of subdivision here is critical for the IO subsystem.
1611 * Please do not alter this order without good reasons and regression
1612 * testing. Specifically, as large blocks of memory are subdivided,
1613 * the order in which smaller blocks are delivered depends on the order
1614 * they're subdivided in this function. This is the primary factor
1615 * influencing the order in which pages are delivered to the IO
1616 * subsystem according to empirical testing, and this is also justified
1617 * by considering the behavior of a buddy system containing a single
1618 * large block of memory acted on by a series of small allocations.
1619 * This behavior is a critical factor in sglist merging's success.
1623 static inline void expand(struct zone
*zone
, struct page
*page
,
1624 int low
, int high
, struct free_area
*area
,
1627 unsigned long size
= 1 << high
;
1629 while (high
> low
) {
1633 VM_BUG_ON_PAGE(bad_range(zone
, &page
[size
]), &page
[size
]);
1636 * Mark as guard pages (or page), that will allow to
1637 * merge back to allocator when buddy will be freed.
1638 * Corresponding page table entries will not be touched,
1639 * pages will stay not present in virtual address space
1641 if (set_page_guard(zone
, &page
[size
], high
, migratetype
))
1644 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
1646 set_page_order(&page
[size
], high
);
1650 static void check_new_page_bad(struct page
*page
)
1652 const char *bad_reason
= NULL
;
1653 unsigned long bad_flags
= 0;
1655 if (unlikely(atomic_read(&page
->_mapcount
) != -1))
1656 bad_reason
= "nonzero mapcount";
1657 if (unlikely(page
->mapping
!= NULL
))
1658 bad_reason
= "non-NULL mapping";
1659 if (unlikely(page_ref_count(page
) != 0))
1660 bad_reason
= "nonzero _count";
1661 if (unlikely(page
->flags
& __PG_HWPOISON
)) {
1662 bad_reason
= "HWPoisoned (hardware-corrupted)";
1663 bad_flags
= __PG_HWPOISON
;
1664 /* Don't complain about hwpoisoned pages */
1665 page_mapcount_reset(page
); /* remove PageBuddy */
1668 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)) {
1669 bad_reason
= "PAGE_FLAGS_CHECK_AT_PREP flag set";
1670 bad_flags
= PAGE_FLAGS_CHECK_AT_PREP
;
1673 if (unlikely(page
->mem_cgroup
))
1674 bad_reason
= "page still charged to cgroup";
1676 bad_page(page
, bad_reason
, bad_flags
);
1680 * This page is about to be returned from the page allocator
1682 static inline int check_new_page(struct page
*page
)
1684 if (likely(page_expected_state(page
,
1685 PAGE_FLAGS_CHECK_AT_PREP
|__PG_HWPOISON
)))
1688 check_new_page_bad(page
);
1692 static inline bool free_pages_prezeroed(bool poisoned
)
1694 return IS_ENABLED(CONFIG_PAGE_POISONING_ZERO
) &&
1695 page_poisoning_enabled() && poisoned
;
1698 #ifdef CONFIG_DEBUG_VM
1699 static bool check_pcp_refill(struct page
*page
)
1704 static bool check_new_pcp(struct page
*page
)
1706 return check_new_page(page
);
1709 static bool check_pcp_refill(struct page
*page
)
1711 return check_new_page(page
);
1713 static bool check_new_pcp(struct page
*page
)
1717 #endif /* CONFIG_DEBUG_VM */
1719 static bool check_new_pages(struct page
*page
, unsigned int order
)
1722 for (i
= 0; i
< (1 << order
); i
++) {
1723 struct page
*p
= page
+ i
;
1725 if (unlikely(check_new_page(p
)))
1732 inline void post_alloc_hook(struct page
*page
, unsigned int order
,
1735 set_page_private(page
, 0);
1736 set_page_refcounted(page
);
1738 arch_alloc_page(page
, order
);
1739 kernel_map_pages(page
, 1 << order
, 1);
1740 kernel_poison_pages(page
, 1 << order
, 1);
1741 kasan_alloc_pages(page
, order
);
1742 set_page_owner(page
, order
, gfp_flags
);
1745 static void prep_new_page(struct page
*page
, unsigned int order
, gfp_t gfp_flags
,
1746 unsigned int alloc_flags
)
1749 bool poisoned
= true;
1751 for (i
= 0; i
< (1 << order
); i
++) {
1752 struct page
*p
= page
+ i
;
1754 poisoned
&= page_is_poisoned(p
);
1757 post_alloc_hook(page
, order
, gfp_flags
);
1759 if (!free_pages_prezeroed(poisoned
) && (gfp_flags
& __GFP_ZERO
))
1760 for (i
= 0; i
< (1 << order
); i
++)
1761 clear_highpage(page
+ i
);
1763 if (order
&& (gfp_flags
& __GFP_COMP
))
1764 prep_compound_page(page
, order
);
1767 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1768 * allocate the page. The expectation is that the caller is taking
1769 * steps that will free more memory. The caller should avoid the page
1770 * being used for !PFMEMALLOC purposes.
1772 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
1773 set_page_pfmemalloc(page
);
1775 clear_page_pfmemalloc(page
);
1779 * Go through the free lists for the given migratetype and remove
1780 * the smallest available page from the freelists
1783 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
1786 unsigned int current_order
;
1787 struct free_area
*area
;
1790 /* Find a page of the appropriate size in the preferred list */
1791 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
1792 area
= &(zone
->free_area
[current_order
]);
1793 page
= list_first_entry_or_null(&area
->free_list
[migratetype
],
1797 list_del(&page
->lru
);
1798 rmv_page_order(page
);
1800 expand(zone
, page
, order
, current_order
, area
, migratetype
);
1801 set_pcppage_migratetype(page
, migratetype
);
1810 * This array describes the order lists are fallen back to when
1811 * the free lists for the desirable migrate type are depleted
1813 static int fallbacks
[MIGRATE_TYPES
][4] = {
1814 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1815 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1816 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_TYPES
},
1818 [MIGRATE_CMA
] = { MIGRATE_TYPES
}, /* Never used */
1820 #ifdef CONFIG_MEMORY_ISOLATION
1821 [MIGRATE_ISOLATE
] = { MIGRATE_TYPES
}, /* Never used */
1826 static struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1829 return __rmqueue_smallest(zone
, order
, MIGRATE_CMA
);
1832 static inline struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1833 unsigned int order
) { return NULL
; }
1837 * Move the free pages in a range to the free lists of the requested type.
1838 * Note that start_page and end_pages are not aligned on a pageblock
1839 * boundary. If alignment is required, use move_freepages_block()
1841 int move_freepages(struct zone
*zone
,
1842 struct page
*start_page
, struct page
*end_page
,
1847 int pages_moved
= 0;
1849 #ifndef CONFIG_HOLES_IN_ZONE
1851 * page_zone is not safe to call in this context when
1852 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1853 * anyway as we check zone boundaries in move_freepages_block().
1854 * Remove at a later date when no bug reports exist related to
1855 * grouping pages by mobility
1857 VM_BUG_ON(page_zone(start_page
) != page_zone(end_page
));
1860 for (page
= start_page
; page
<= end_page
;) {
1861 if (!pfn_valid_within(page_to_pfn(page
))) {
1866 /* Make sure we are not inadvertently changing nodes */
1867 VM_BUG_ON_PAGE(page_to_nid(page
) != zone_to_nid(zone
), page
);
1869 if (!PageBuddy(page
)) {
1874 order
= page_order(page
);
1875 list_move(&page
->lru
,
1876 &zone
->free_area
[order
].free_list
[migratetype
]);
1878 pages_moved
+= 1 << order
;
1884 int move_freepages_block(struct zone
*zone
, struct page
*page
,
1887 unsigned long start_pfn
, end_pfn
;
1888 struct page
*start_page
, *end_page
;
1890 start_pfn
= page_to_pfn(page
);
1891 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
1892 start_page
= pfn_to_page(start_pfn
);
1893 end_page
= start_page
+ pageblock_nr_pages
- 1;
1894 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
1896 /* Do not cross zone boundaries */
1897 if (!zone_spans_pfn(zone
, start_pfn
))
1899 if (!zone_spans_pfn(zone
, end_pfn
))
1902 return move_freepages(zone
, start_page
, end_page
, migratetype
);
1905 static void change_pageblock_range(struct page
*pageblock_page
,
1906 int start_order
, int migratetype
)
1908 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
1910 while (nr_pageblocks
--) {
1911 set_pageblock_migratetype(pageblock_page
, migratetype
);
1912 pageblock_page
+= pageblock_nr_pages
;
1917 * When we are falling back to another migratetype during allocation, try to
1918 * steal extra free pages from the same pageblocks to satisfy further
1919 * allocations, instead of polluting multiple pageblocks.
1921 * If we are stealing a relatively large buddy page, it is likely there will
1922 * be more free pages in the pageblock, so try to steal them all. For
1923 * reclaimable and unmovable allocations, we steal regardless of page size,
1924 * as fragmentation caused by those allocations polluting movable pageblocks
1925 * is worse than movable allocations stealing from unmovable and reclaimable
1928 static bool can_steal_fallback(unsigned int order
, int start_mt
)
1931 * Leaving this order check is intended, although there is
1932 * relaxed order check in next check. The reason is that
1933 * we can actually steal whole pageblock if this condition met,
1934 * but, below check doesn't guarantee it and that is just heuristic
1935 * so could be changed anytime.
1937 if (order
>= pageblock_order
)
1940 if (order
>= pageblock_order
/ 2 ||
1941 start_mt
== MIGRATE_RECLAIMABLE
||
1942 start_mt
== MIGRATE_UNMOVABLE
||
1943 page_group_by_mobility_disabled
)
1950 * This function implements actual steal behaviour. If order is large enough,
1951 * we can steal whole pageblock. If not, we first move freepages in this
1952 * pageblock and check whether half of pages are moved or not. If half of
1953 * pages are moved, we can change migratetype of pageblock and permanently
1954 * use it's pages as requested migratetype in the future.
1956 static void steal_suitable_fallback(struct zone
*zone
, struct page
*page
,
1959 unsigned int current_order
= page_order(page
);
1962 /* Take ownership for orders >= pageblock_order */
1963 if (current_order
>= pageblock_order
) {
1964 change_pageblock_range(page
, current_order
, start_type
);
1968 pages
= move_freepages_block(zone
, page
, start_type
);
1970 /* Claim the whole block if over half of it is free */
1971 if (pages
>= (1 << (pageblock_order
-1)) ||
1972 page_group_by_mobility_disabled
)
1973 set_pageblock_migratetype(page
, start_type
);
1977 * Check whether there is a suitable fallback freepage with requested order.
1978 * If only_stealable is true, this function returns fallback_mt only if
1979 * we can steal other freepages all together. This would help to reduce
1980 * fragmentation due to mixed migratetype pages in one pageblock.
1982 int find_suitable_fallback(struct free_area
*area
, unsigned int order
,
1983 int migratetype
, bool only_stealable
, bool *can_steal
)
1988 if (area
->nr_free
== 0)
1993 fallback_mt
= fallbacks
[migratetype
][i
];
1994 if (fallback_mt
== MIGRATE_TYPES
)
1997 if (list_empty(&area
->free_list
[fallback_mt
]))
2000 if (can_steal_fallback(order
, migratetype
))
2003 if (!only_stealable
)
2014 * Reserve a pageblock for exclusive use of high-order atomic allocations if
2015 * there are no empty page blocks that contain a page with a suitable order
2017 static void reserve_highatomic_pageblock(struct page
*page
, struct zone
*zone
,
2018 unsigned int alloc_order
)
2021 unsigned long max_managed
, flags
;
2024 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
2025 * Check is race-prone but harmless.
2027 max_managed
= (zone
->managed_pages
/ 100) + pageblock_nr_pages
;
2028 if (zone
->nr_reserved_highatomic
>= max_managed
)
2031 spin_lock_irqsave(&zone
->lock
, flags
);
2033 /* Recheck the nr_reserved_highatomic limit under the lock */
2034 if (zone
->nr_reserved_highatomic
>= max_managed
)
2038 mt
= get_pageblock_migratetype(page
);
2039 if (!is_migrate_highatomic(mt
) && !is_migrate_isolate(mt
)
2040 && !is_migrate_cma(mt
)) {
2041 zone
->nr_reserved_highatomic
+= pageblock_nr_pages
;
2042 set_pageblock_migratetype(page
, MIGRATE_HIGHATOMIC
);
2043 move_freepages_block(zone
, page
, MIGRATE_HIGHATOMIC
);
2047 spin_unlock_irqrestore(&zone
->lock
, flags
);
2051 * Used when an allocation is about to fail under memory pressure. This
2052 * potentially hurts the reliability of high-order allocations when under
2053 * intense memory pressure but failed atomic allocations should be easier
2054 * to recover from than an OOM.
2056 * If @force is true, try to unreserve a pageblock even though highatomic
2057 * pageblock is exhausted.
2059 static bool unreserve_highatomic_pageblock(const struct alloc_context
*ac
,
2062 struct zonelist
*zonelist
= ac
->zonelist
;
2063 unsigned long flags
;
2070 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
, ac
->high_zoneidx
,
2073 * Preserve at least one pageblock unless memory pressure
2076 if (!force
&& zone
->nr_reserved_highatomic
<=
2080 spin_lock_irqsave(&zone
->lock
, flags
);
2081 for (order
= 0; order
< MAX_ORDER
; order
++) {
2082 struct free_area
*area
= &(zone
->free_area
[order
]);
2084 page
= list_first_entry_or_null(
2085 &area
->free_list
[MIGRATE_HIGHATOMIC
],
2091 * In page freeing path, migratetype change is racy so
2092 * we can counter several free pages in a pageblock
2093 * in this loop althoug we changed the pageblock type
2094 * from highatomic to ac->migratetype. So we should
2095 * adjust the count once.
2097 if (is_migrate_highatomic_page(page
)) {
2099 * It should never happen but changes to
2100 * locking could inadvertently allow a per-cpu
2101 * drain to add pages to MIGRATE_HIGHATOMIC
2102 * while unreserving so be safe and watch for
2105 zone
->nr_reserved_highatomic
-= min(
2107 zone
->nr_reserved_highatomic
);
2111 * Convert to ac->migratetype and avoid the normal
2112 * pageblock stealing heuristics. Minimally, the caller
2113 * is doing the work and needs the pages. More
2114 * importantly, if the block was always converted to
2115 * MIGRATE_UNMOVABLE or another type then the number
2116 * of pageblocks that cannot be completely freed
2119 set_pageblock_migratetype(page
, ac
->migratetype
);
2120 ret
= move_freepages_block(zone
, page
, ac
->migratetype
);
2122 spin_unlock_irqrestore(&zone
->lock
, flags
);
2126 spin_unlock_irqrestore(&zone
->lock
, flags
);
2132 /* Remove an element from the buddy allocator from the fallback list */
2133 static inline struct page
*
2134 __rmqueue_fallback(struct zone
*zone
, unsigned int order
, int start_migratetype
)
2136 struct free_area
*area
;
2137 unsigned int current_order
;
2142 /* Find the largest possible block of pages in the other list */
2143 for (current_order
= MAX_ORDER
-1;
2144 current_order
>= order
&& current_order
<= MAX_ORDER
-1;
2146 area
= &(zone
->free_area
[current_order
]);
2147 fallback_mt
= find_suitable_fallback(area
, current_order
,
2148 start_migratetype
, false, &can_steal
);
2149 if (fallback_mt
== -1)
2152 page
= list_first_entry(&area
->free_list
[fallback_mt
],
2154 if (can_steal
&& !is_migrate_highatomic_page(page
))
2155 steal_suitable_fallback(zone
, page
, start_migratetype
);
2157 /* Remove the page from the freelists */
2159 list_del(&page
->lru
);
2160 rmv_page_order(page
);
2162 expand(zone
, page
, order
, current_order
, area
,
2165 * The pcppage_migratetype may differ from pageblock's
2166 * migratetype depending on the decisions in
2167 * find_suitable_fallback(). This is OK as long as it does not
2168 * differ for MIGRATE_CMA pageblocks. Those can be used as
2169 * fallback only via special __rmqueue_cma_fallback() function
2171 set_pcppage_migratetype(page
, start_migratetype
);
2173 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
2174 start_migratetype
, fallback_mt
);
2183 * Do the hard work of removing an element from the buddy allocator.
2184 * Call me with the zone->lock already held.
2186 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
2191 page
= __rmqueue_smallest(zone
, order
, migratetype
);
2192 if (unlikely(!page
)) {
2193 if (migratetype
== MIGRATE_MOVABLE
)
2194 page
= __rmqueue_cma_fallback(zone
, order
);
2197 page
= __rmqueue_fallback(zone
, order
, migratetype
);
2200 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
2205 * Obtain a specified number of elements from the buddy allocator, all under
2206 * a single hold of the lock, for efficiency. Add them to the supplied list.
2207 * Returns the number of new pages which were placed at *list.
2209 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
2210 unsigned long count
, struct list_head
*list
,
2211 int migratetype
, bool cold
)
2215 spin_lock(&zone
->lock
);
2216 for (i
= 0; i
< count
; ++i
) {
2217 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
2218 if (unlikely(page
== NULL
))
2221 if (unlikely(check_pcp_refill(page
)))
2225 * Split buddy pages returned by expand() are received here
2226 * in physical page order. The page is added to the callers and
2227 * list and the list head then moves forward. From the callers
2228 * perspective, the linked list is ordered by page number in
2229 * some conditions. This is useful for IO devices that can
2230 * merge IO requests if the physical pages are ordered
2234 list_add(&page
->lru
, list
);
2236 list_add_tail(&page
->lru
, list
);
2239 if (is_migrate_cma(get_pcppage_migratetype(page
)))
2240 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
,
2245 * i pages were removed from the buddy list even if some leak due
2246 * to check_pcp_refill failing so adjust NR_FREE_PAGES based
2247 * on i. Do not confuse with 'alloced' which is the number of
2248 * pages added to the pcp list.
2250 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
2251 spin_unlock(&zone
->lock
);
2257 * Called from the vmstat counter updater to drain pagesets of this
2258 * currently executing processor on remote nodes after they have
2261 * Note that this function must be called with the thread pinned to
2262 * a single processor.
2264 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
2266 unsigned long flags
;
2267 int to_drain
, batch
;
2269 local_irq_save(flags
);
2270 batch
= READ_ONCE(pcp
->batch
);
2271 to_drain
= min(pcp
->count
, batch
);
2273 free_pcppages_bulk(zone
, to_drain
, pcp
);
2274 pcp
->count
-= to_drain
;
2276 local_irq_restore(flags
);
2281 * Drain pcplists of the indicated processor and zone.
2283 * The processor must either be the current processor and the
2284 * thread pinned to the current processor or a processor that
2287 static void drain_pages_zone(unsigned int cpu
, struct zone
*zone
)
2289 unsigned long flags
;
2290 struct per_cpu_pageset
*pset
;
2291 struct per_cpu_pages
*pcp
;
2293 local_irq_save(flags
);
2294 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
2298 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
2301 local_irq_restore(flags
);
2305 * Drain pcplists of all zones on the indicated processor.
2307 * The processor must either be the current processor and the
2308 * thread pinned to the current processor or a processor that
2311 static void drain_pages(unsigned int cpu
)
2315 for_each_populated_zone(zone
) {
2316 drain_pages_zone(cpu
, zone
);
2321 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
2323 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
2324 * the single zone's pages.
2326 void drain_local_pages(struct zone
*zone
)
2328 int cpu
= smp_processor_id();
2331 drain_pages_zone(cpu
, zone
);
2336 static void drain_local_pages_wq(struct work_struct
*work
)
2339 * drain_all_pages doesn't use proper cpu hotplug protection so
2340 * we can race with cpu offline when the WQ can move this from
2341 * a cpu pinned worker to an unbound one. We can operate on a different
2342 * cpu which is allright but we also have to make sure to not move to
2346 drain_local_pages(NULL
);
2351 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
2353 * When zone parameter is non-NULL, spill just the single zone's pages.
2355 * Note that this can be extremely slow as the draining happens in a workqueue.
2357 void drain_all_pages(struct zone
*zone
)
2362 * Allocate in the BSS so we wont require allocation in
2363 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
2365 static cpumask_t cpus_with_pcps
;
2368 * Make sure nobody triggers this path before mm_percpu_wq is fully
2371 if (WARN_ON_ONCE(!mm_percpu_wq
))
2374 /* Workqueues cannot recurse */
2375 if (current
->flags
& PF_WQ_WORKER
)
2379 * Do not drain if one is already in progress unless it's specific to
2380 * a zone. Such callers are primarily CMA and memory hotplug and need
2381 * the drain to be complete when the call returns.
2383 if (unlikely(!mutex_trylock(&pcpu_drain_mutex
))) {
2386 mutex_lock(&pcpu_drain_mutex
);
2390 * We don't care about racing with CPU hotplug event
2391 * as offline notification will cause the notified
2392 * cpu to drain that CPU pcps and on_each_cpu_mask
2393 * disables preemption as part of its processing
2395 for_each_online_cpu(cpu
) {
2396 struct per_cpu_pageset
*pcp
;
2398 bool has_pcps
= false;
2401 pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
2405 for_each_populated_zone(z
) {
2406 pcp
= per_cpu_ptr(z
->pageset
, cpu
);
2407 if (pcp
->pcp
.count
) {
2415 cpumask_set_cpu(cpu
, &cpus_with_pcps
);
2417 cpumask_clear_cpu(cpu
, &cpus_with_pcps
);
2420 for_each_cpu(cpu
, &cpus_with_pcps
) {
2421 struct work_struct
*work
= per_cpu_ptr(&pcpu_drain
, cpu
);
2422 INIT_WORK(work
, drain_local_pages_wq
);
2423 queue_work_on(cpu
, mm_percpu_wq
, work
);
2425 for_each_cpu(cpu
, &cpus_with_pcps
)
2426 flush_work(per_cpu_ptr(&pcpu_drain
, cpu
));
2428 mutex_unlock(&pcpu_drain_mutex
);
2431 #ifdef CONFIG_HIBERNATION
2433 void mark_free_pages(struct zone
*zone
)
2435 unsigned long pfn
, max_zone_pfn
;
2436 unsigned long flags
;
2437 unsigned int order
, t
;
2440 if (zone_is_empty(zone
))
2443 spin_lock_irqsave(&zone
->lock
, flags
);
2445 max_zone_pfn
= zone_end_pfn(zone
);
2446 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
2447 if (pfn_valid(pfn
)) {
2448 page
= pfn_to_page(pfn
);
2450 if (page_zone(page
) != zone
)
2453 if (!swsusp_page_is_forbidden(page
))
2454 swsusp_unset_page_free(page
);
2457 for_each_migratetype_order(order
, t
) {
2458 list_for_each_entry(page
,
2459 &zone
->free_area
[order
].free_list
[t
], lru
) {
2462 pfn
= page_to_pfn(page
);
2463 for (i
= 0; i
< (1UL << order
); i
++)
2464 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
2467 spin_unlock_irqrestore(&zone
->lock
, flags
);
2469 #endif /* CONFIG_PM */
2472 * Free a 0-order page
2473 * cold == true ? free a cold page : free a hot page
2475 void free_hot_cold_page(struct page
*page
, bool cold
)
2477 struct zone
*zone
= page_zone(page
);
2478 struct per_cpu_pages
*pcp
;
2479 unsigned long flags
;
2480 unsigned long pfn
= page_to_pfn(page
);
2483 if (!free_pcp_prepare(page
))
2486 migratetype
= get_pfnblock_migratetype(page
, pfn
);
2487 set_pcppage_migratetype(page
, migratetype
);
2488 local_irq_save(flags
);
2489 __count_vm_event(PGFREE
);
2492 * We only track unmovable, reclaimable and movable on pcp lists.
2493 * Free ISOLATE pages back to the allocator because they are being
2494 * offlined but treat HIGHATOMIC as movable pages so we can get those
2495 * areas back if necessary. Otherwise, we may have to free
2496 * excessively into the page allocator
2498 if (migratetype
>= MIGRATE_PCPTYPES
) {
2499 if (unlikely(is_migrate_isolate(migratetype
))) {
2500 free_one_page(zone
, page
, pfn
, 0, migratetype
);
2503 migratetype
= MIGRATE_MOVABLE
;
2506 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2508 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
2510 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
2512 if (pcp
->count
>= pcp
->high
) {
2513 unsigned long batch
= READ_ONCE(pcp
->batch
);
2514 free_pcppages_bulk(zone
, batch
, pcp
);
2515 pcp
->count
-= batch
;
2519 local_irq_restore(flags
);
2523 * Free a list of 0-order pages
2525 void free_hot_cold_page_list(struct list_head
*list
, bool cold
)
2527 struct page
*page
, *next
;
2529 list_for_each_entry_safe(page
, next
, list
, lru
) {
2530 trace_mm_page_free_batched(page
, cold
);
2531 free_hot_cold_page(page
, cold
);
2536 * split_page takes a non-compound higher-order page, and splits it into
2537 * n (1<<order) sub-pages: page[0..n]
2538 * Each sub-page must be freed individually.
2540 * Note: this is probably too low level an operation for use in drivers.
2541 * Please consult with lkml before using this in your driver.
2543 void split_page(struct page
*page
, unsigned int order
)
2547 VM_BUG_ON_PAGE(PageCompound(page
), page
);
2548 VM_BUG_ON_PAGE(!page_count(page
), page
);
2550 #ifdef CONFIG_KMEMCHECK
2552 * Split shadow pages too, because free(page[0]) would
2553 * otherwise free the whole shadow.
2555 if (kmemcheck_page_is_tracked(page
))
2556 split_page(virt_to_page(page
[0].shadow
), order
);
2559 for (i
= 1; i
< (1 << order
); i
++)
2560 set_page_refcounted(page
+ i
);
2561 split_page_owner(page
, order
);
2563 EXPORT_SYMBOL_GPL(split_page
);
2565 int __isolate_free_page(struct page
*page
, unsigned int order
)
2567 unsigned long watermark
;
2571 BUG_ON(!PageBuddy(page
));
2573 zone
= page_zone(page
);
2574 mt
= get_pageblock_migratetype(page
);
2576 if (!is_migrate_isolate(mt
)) {
2578 * Obey watermarks as if the page was being allocated. We can
2579 * emulate a high-order watermark check with a raised order-0
2580 * watermark, because we already know our high-order page
2583 watermark
= min_wmark_pages(zone
) + (1UL << order
);
2584 if (!zone_watermark_ok(zone
, 0, watermark
, 0, ALLOC_CMA
))
2587 __mod_zone_freepage_state(zone
, -(1UL << order
), mt
);
2590 /* Remove page from free list */
2591 list_del(&page
->lru
);
2592 zone
->free_area
[order
].nr_free
--;
2593 rmv_page_order(page
);
2596 * Set the pageblock if the isolated page is at least half of a
2599 if (order
>= pageblock_order
- 1) {
2600 struct page
*endpage
= page
+ (1 << order
) - 1;
2601 for (; page
< endpage
; page
+= pageblock_nr_pages
) {
2602 int mt
= get_pageblock_migratetype(page
);
2603 if (!is_migrate_isolate(mt
) && !is_migrate_cma(mt
)
2604 && !is_migrate_highatomic(mt
))
2605 set_pageblock_migratetype(page
,
2611 return 1UL << order
;
2615 * Update NUMA hit/miss statistics
2617 * Must be called with interrupts disabled.
2619 static inline void zone_statistics(struct zone
*preferred_zone
, struct zone
*z
)
2622 enum zone_stat_item local_stat
= NUMA_LOCAL
;
2624 if (z
->node
!= numa_node_id())
2625 local_stat
= NUMA_OTHER
;
2627 if (z
->node
== preferred_zone
->node
)
2628 __inc_zone_state(z
, NUMA_HIT
);
2630 __inc_zone_state(z
, NUMA_MISS
);
2631 __inc_zone_state(preferred_zone
, NUMA_FOREIGN
);
2633 __inc_zone_state(z
, local_stat
);
2637 /* Remove page from the per-cpu list, caller must protect the list */
2638 static struct page
*__rmqueue_pcplist(struct zone
*zone
, int migratetype
,
2639 bool cold
, struct per_cpu_pages
*pcp
,
2640 struct list_head
*list
)
2645 if (list_empty(list
)) {
2646 pcp
->count
+= rmqueue_bulk(zone
, 0,
2649 if (unlikely(list_empty(list
)))
2654 page
= list_last_entry(list
, struct page
, lru
);
2656 page
= list_first_entry(list
, struct page
, lru
);
2658 list_del(&page
->lru
);
2660 } while (check_new_pcp(page
));
2665 /* Lock and remove page from the per-cpu list */
2666 static struct page
*rmqueue_pcplist(struct zone
*preferred_zone
,
2667 struct zone
*zone
, unsigned int order
,
2668 gfp_t gfp_flags
, int migratetype
)
2670 struct per_cpu_pages
*pcp
;
2671 struct list_head
*list
;
2672 bool cold
= ((gfp_flags
& __GFP_COLD
) != 0);
2674 unsigned long flags
;
2676 local_irq_save(flags
);
2677 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2678 list
= &pcp
->lists
[migratetype
];
2679 page
= __rmqueue_pcplist(zone
, migratetype
, cold
, pcp
, list
);
2681 __count_zid_vm_events(PGALLOC
, page_zonenum(page
), 1 << order
);
2682 zone_statistics(preferred_zone
, zone
);
2684 local_irq_restore(flags
);
2689 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2692 struct page
*rmqueue(struct zone
*preferred_zone
,
2693 struct zone
*zone
, unsigned int order
,
2694 gfp_t gfp_flags
, unsigned int alloc_flags
,
2697 unsigned long flags
;
2700 if (likely(order
== 0)) {
2701 page
= rmqueue_pcplist(preferred_zone
, zone
, order
,
2702 gfp_flags
, migratetype
);
2707 * We most definitely don't want callers attempting to
2708 * allocate greater than order-1 page units with __GFP_NOFAIL.
2710 WARN_ON_ONCE((gfp_flags
& __GFP_NOFAIL
) && (order
> 1));
2711 spin_lock_irqsave(&zone
->lock
, flags
);
2715 if (alloc_flags
& ALLOC_HARDER
) {
2716 page
= __rmqueue_smallest(zone
, order
, MIGRATE_HIGHATOMIC
);
2718 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
2721 page
= __rmqueue(zone
, order
, migratetype
);
2722 } while (page
&& check_new_pages(page
, order
));
2723 spin_unlock(&zone
->lock
);
2726 __mod_zone_freepage_state(zone
, -(1 << order
),
2727 get_pcppage_migratetype(page
));
2729 __count_zid_vm_events(PGALLOC
, page_zonenum(page
), 1 << order
);
2730 zone_statistics(preferred_zone
, zone
);
2731 local_irq_restore(flags
);
2734 VM_BUG_ON_PAGE(page
&& bad_range(zone
, page
), page
);
2738 local_irq_restore(flags
);
2742 #ifdef CONFIG_FAIL_PAGE_ALLOC
2745 struct fault_attr attr
;
2747 bool ignore_gfp_highmem
;
2748 bool ignore_gfp_reclaim
;
2750 } fail_page_alloc
= {
2751 .attr
= FAULT_ATTR_INITIALIZER
,
2752 .ignore_gfp_reclaim
= true,
2753 .ignore_gfp_highmem
= true,
2757 static int __init
setup_fail_page_alloc(char *str
)
2759 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
2761 __setup("fail_page_alloc=", setup_fail_page_alloc
);
2763 static bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2765 if (order
< fail_page_alloc
.min_order
)
2767 if (gfp_mask
& __GFP_NOFAIL
)
2769 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
2771 if (fail_page_alloc
.ignore_gfp_reclaim
&&
2772 (gfp_mask
& __GFP_DIRECT_RECLAIM
))
2775 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
2778 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2780 static int __init
fail_page_alloc_debugfs(void)
2782 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
2785 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
2786 &fail_page_alloc
.attr
);
2788 return PTR_ERR(dir
);
2790 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
2791 &fail_page_alloc
.ignore_gfp_reclaim
))
2793 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
2794 &fail_page_alloc
.ignore_gfp_highmem
))
2796 if (!debugfs_create_u32("min-order", mode
, dir
,
2797 &fail_page_alloc
.min_order
))
2802 debugfs_remove_recursive(dir
);
2807 late_initcall(fail_page_alloc_debugfs
);
2809 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2811 #else /* CONFIG_FAIL_PAGE_ALLOC */
2813 static inline bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2818 #endif /* CONFIG_FAIL_PAGE_ALLOC */
2821 * Return true if free base pages are above 'mark'. For high-order checks it
2822 * will return true of the order-0 watermark is reached and there is at least
2823 * one free page of a suitable size. Checking now avoids taking the zone lock
2824 * to check in the allocation paths if no pages are free.
2826 bool __zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
2827 int classzone_idx
, unsigned int alloc_flags
,
2832 const bool alloc_harder
= (alloc_flags
& ALLOC_HARDER
);
2834 /* free_pages may go negative - that's OK */
2835 free_pages
-= (1 << order
) - 1;
2837 if (alloc_flags
& ALLOC_HIGH
)
2841 * If the caller does not have rights to ALLOC_HARDER then subtract
2842 * the high-atomic reserves. This will over-estimate the size of the
2843 * atomic reserve but it avoids a search.
2845 if (likely(!alloc_harder
))
2846 free_pages
-= z
->nr_reserved_highatomic
;
2851 /* If allocation can't use CMA areas don't use free CMA pages */
2852 if (!(alloc_flags
& ALLOC_CMA
))
2853 free_pages
-= zone_page_state(z
, NR_FREE_CMA_PAGES
);
2857 * Check watermarks for an order-0 allocation request. If these
2858 * are not met, then a high-order request also cannot go ahead
2859 * even if a suitable page happened to be free.
2861 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
2864 /* If this is an order-0 request then the watermark is fine */
2868 /* For a high-order request, check at least one suitable page is free */
2869 for (o
= order
; o
< MAX_ORDER
; o
++) {
2870 struct free_area
*area
= &z
->free_area
[o
];
2879 for (mt
= 0; mt
< MIGRATE_PCPTYPES
; mt
++) {
2880 if (!list_empty(&area
->free_list
[mt
]))
2885 if ((alloc_flags
& ALLOC_CMA
) &&
2886 !list_empty(&area
->free_list
[MIGRATE_CMA
])) {
2894 bool zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
2895 int classzone_idx
, unsigned int alloc_flags
)
2897 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
2898 zone_page_state(z
, NR_FREE_PAGES
));
2901 static inline bool zone_watermark_fast(struct zone
*z
, unsigned int order
,
2902 unsigned long mark
, int classzone_idx
, unsigned int alloc_flags
)
2904 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
2908 /* If allocation can't use CMA areas don't use free CMA pages */
2909 if (!(alloc_flags
& ALLOC_CMA
))
2910 cma_pages
= zone_page_state(z
, NR_FREE_CMA_PAGES
);
2914 * Fast check for order-0 only. If this fails then the reserves
2915 * need to be calculated. There is a corner case where the check
2916 * passes but only the high-order atomic reserve are free. If
2917 * the caller is !atomic then it'll uselessly search the free
2918 * list. That corner case is then slower but it is harmless.
2920 if (!order
&& (free_pages
- cma_pages
) > mark
+ z
->lowmem_reserve
[classzone_idx
])
2923 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
2927 bool zone_watermark_ok_safe(struct zone
*z
, unsigned int order
,
2928 unsigned long mark
, int classzone_idx
)
2930 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
2932 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
2933 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
2935 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, 0,
2940 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2942 return node_distance(zone_to_nid(local_zone
), zone_to_nid(zone
)) <=
2945 #else /* CONFIG_NUMA */
2946 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2950 #endif /* CONFIG_NUMA */
2953 * get_page_from_freelist goes through the zonelist trying to allocate
2956 static struct page
*
2957 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
, int alloc_flags
,
2958 const struct alloc_context
*ac
)
2960 struct zoneref
*z
= ac
->preferred_zoneref
;
2962 struct pglist_data
*last_pgdat_dirty_limit
= NULL
;
2965 * Scan zonelist, looking for a zone with enough free.
2966 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2968 for_next_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
, ac
->high_zoneidx
,
2973 if (cpusets_enabled() &&
2974 (alloc_flags
& ALLOC_CPUSET
) &&
2975 !__cpuset_zone_allowed(zone
, gfp_mask
))
2978 * When allocating a page cache page for writing, we
2979 * want to get it from a node that is within its dirty
2980 * limit, such that no single node holds more than its
2981 * proportional share of globally allowed dirty pages.
2982 * The dirty limits take into account the node's
2983 * lowmem reserves and high watermark so that kswapd
2984 * should be able to balance it without having to
2985 * write pages from its LRU list.
2987 * XXX: For now, allow allocations to potentially
2988 * exceed the per-node dirty limit in the slowpath
2989 * (spread_dirty_pages unset) before going into reclaim,
2990 * which is important when on a NUMA setup the allowed
2991 * nodes are together not big enough to reach the
2992 * global limit. The proper fix for these situations
2993 * will require awareness of nodes in the
2994 * dirty-throttling and the flusher threads.
2996 if (ac
->spread_dirty_pages
) {
2997 if (last_pgdat_dirty_limit
== zone
->zone_pgdat
)
3000 if (!node_dirty_ok(zone
->zone_pgdat
)) {
3001 last_pgdat_dirty_limit
= zone
->zone_pgdat
;
3006 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
3007 if (!zone_watermark_fast(zone
, order
, mark
,
3008 ac_classzone_idx(ac
), alloc_flags
)) {
3011 /* Checked here to keep the fast path fast */
3012 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
3013 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
3016 if (node_reclaim_mode
== 0 ||
3017 !zone_allows_reclaim(ac
->preferred_zoneref
->zone
, zone
))
3020 ret
= node_reclaim(zone
->zone_pgdat
, gfp_mask
, order
);
3022 case NODE_RECLAIM_NOSCAN
:
3025 case NODE_RECLAIM_FULL
:
3026 /* scanned but unreclaimable */
3029 /* did we reclaim enough */
3030 if (zone_watermark_ok(zone
, order
, mark
,
3031 ac_classzone_idx(ac
), alloc_flags
))
3039 page
= rmqueue(ac
->preferred_zoneref
->zone
, zone
, order
,
3040 gfp_mask
, alloc_flags
, ac
->migratetype
);
3042 prep_new_page(page
, order
, gfp_mask
, alloc_flags
);
3045 * If this is a high-order atomic allocation then check
3046 * if the pageblock should be reserved for the future
3048 if (unlikely(order
&& (alloc_flags
& ALLOC_HARDER
)))
3049 reserve_highatomic_pageblock(page
, zone
, order
);
3059 * Large machines with many possible nodes should not always dump per-node
3060 * meminfo in irq context.
3062 static inline bool should_suppress_show_mem(void)
3067 ret
= in_interrupt();
3072 static void warn_alloc_show_mem(gfp_t gfp_mask
, nodemask_t
*nodemask
)
3074 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
3075 static DEFINE_RATELIMIT_STATE(show_mem_rs
, HZ
, 1);
3077 if (should_suppress_show_mem() || !__ratelimit(&show_mem_rs
))
3081 * This documents exceptions given to allocations in certain
3082 * contexts that are allowed to allocate outside current's set
3085 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
3086 if (test_thread_flag(TIF_MEMDIE
) ||
3087 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
3088 filter
&= ~SHOW_MEM_FILTER_NODES
;
3089 if (in_interrupt() || !(gfp_mask
& __GFP_DIRECT_RECLAIM
))
3090 filter
&= ~SHOW_MEM_FILTER_NODES
;
3092 show_mem(filter
, nodemask
);
3095 void warn_alloc(gfp_t gfp_mask
, nodemask_t
*nodemask
, const char *fmt
, ...)
3097 struct va_format vaf
;
3099 static DEFINE_RATELIMIT_STATE(nopage_rs
, DEFAULT_RATELIMIT_INTERVAL
,
3100 DEFAULT_RATELIMIT_BURST
);
3102 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
) ||
3103 debug_guardpage_minorder() > 0)
3106 pr_warn("%s: ", current
->comm
);
3108 va_start(args
, fmt
);
3111 pr_cont("%pV", &vaf
);
3114 pr_cont(", mode:%#x(%pGg), nodemask=", gfp_mask
, &gfp_mask
);
3116 pr_cont("%*pbl\n", nodemask_pr_args(nodemask
));
3118 pr_cont("(null)\n");
3120 cpuset_print_current_mems_allowed();
3123 warn_alloc_show_mem(gfp_mask
, nodemask
);
3126 static inline struct page
*
3127 __alloc_pages_cpuset_fallback(gfp_t gfp_mask
, unsigned int order
,
3128 unsigned int alloc_flags
,
3129 const struct alloc_context
*ac
)
3133 page
= get_page_from_freelist(gfp_mask
, order
,
3134 alloc_flags
|ALLOC_CPUSET
, ac
);
3136 * fallback to ignore cpuset restriction if our nodes
3140 page
= get_page_from_freelist(gfp_mask
, order
,
3146 static inline struct page
*
3147 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
3148 const struct alloc_context
*ac
, unsigned long *did_some_progress
)
3150 struct oom_control oc
= {
3151 .zonelist
= ac
->zonelist
,
3152 .nodemask
= ac
->nodemask
,
3154 .gfp_mask
= gfp_mask
,
3159 *did_some_progress
= 0;
3162 * Acquire the oom lock. If that fails, somebody else is
3163 * making progress for us.
3165 if (!mutex_trylock(&oom_lock
)) {
3166 *did_some_progress
= 1;
3167 schedule_timeout_uninterruptible(1);
3172 * Go through the zonelist yet one more time, keep very high watermark
3173 * here, this is only to catch a parallel oom killing, we must fail if
3174 * we're still under heavy pressure.
3176 page
= get_page_from_freelist(gfp_mask
| __GFP_HARDWALL
, order
,
3177 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
, ac
);
3181 /* Coredumps can quickly deplete all memory reserves */
3182 if (current
->flags
& PF_DUMPCORE
)
3184 /* The OOM killer will not help higher order allocs */
3185 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
3187 /* The OOM killer does not needlessly kill tasks for lowmem */
3188 if (ac
->high_zoneidx
< ZONE_NORMAL
)
3190 if (pm_suspended_storage())
3193 * XXX: GFP_NOFS allocations should rather fail than rely on
3194 * other request to make a forward progress.
3195 * We are in an unfortunate situation where out_of_memory cannot
3196 * do much for this context but let's try it to at least get
3197 * access to memory reserved if the current task is killed (see
3198 * out_of_memory). Once filesystems are ready to handle allocation
3199 * failures more gracefully we should just bail out here.
3202 /* The OOM killer may not free memory on a specific node */
3203 if (gfp_mask
& __GFP_THISNODE
)
3206 /* Exhausted what can be done so it's blamo time */
3207 if (out_of_memory(&oc
) || WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
)) {
3208 *did_some_progress
= 1;
3211 * Help non-failing allocations by giving them access to memory
3214 if (gfp_mask
& __GFP_NOFAIL
)
3215 page
= __alloc_pages_cpuset_fallback(gfp_mask
, order
,
3216 ALLOC_NO_WATERMARKS
, ac
);
3219 mutex_unlock(&oom_lock
);
3224 * Maximum number of compaction retries wit a progress before OOM
3225 * killer is consider as the only way to move forward.
3227 #define MAX_COMPACT_RETRIES 16
3229 #ifdef CONFIG_COMPACTION
3230 /* Try memory compaction for high-order allocations before reclaim */
3231 static struct page
*
3232 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
3233 unsigned int alloc_flags
, const struct alloc_context
*ac
,
3234 enum compact_priority prio
, enum compact_result
*compact_result
)
3241 current
->flags
|= PF_MEMALLOC
;
3242 *compact_result
= try_to_compact_pages(gfp_mask
, order
, alloc_flags
, ac
,
3244 current
->flags
&= ~PF_MEMALLOC
;
3246 if (*compact_result
<= COMPACT_INACTIVE
)
3250 * At least in one zone compaction wasn't deferred or skipped, so let's
3251 * count a compaction stall
3253 count_vm_event(COMPACTSTALL
);
3255 page
= get_page_from_freelist(gfp_mask
, order
, alloc_flags
, ac
);
3258 struct zone
*zone
= page_zone(page
);
3260 zone
->compact_blockskip_flush
= false;
3261 compaction_defer_reset(zone
, order
, true);
3262 count_vm_event(COMPACTSUCCESS
);
3267 * It's bad if compaction run occurs and fails. The most likely reason
3268 * is that pages exist, but not enough to satisfy watermarks.
3270 count_vm_event(COMPACTFAIL
);
3278 should_compact_retry(struct alloc_context
*ac
, int order
, int alloc_flags
,
3279 enum compact_result compact_result
,
3280 enum compact_priority
*compact_priority
,
3281 int *compaction_retries
)
3283 int max_retries
= MAX_COMPACT_RETRIES
;
3286 int retries
= *compaction_retries
;
3287 enum compact_priority priority
= *compact_priority
;
3292 if (compaction_made_progress(compact_result
))
3293 (*compaction_retries
)++;
3296 * compaction considers all the zone as desperately out of memory
3297 * so it doesn't really make much sense to retry except when the
3298 * failure could be caused by insufficient priority
3300 if (compaction_failed(compact_result
))
3301 goto check_priority
;
3304 * make sure the compaction wasn't deferred or didn't bail out early
3305 * due to locks contention before we declare that we should give up.
3306 * But do not retry if the given zonelist is not suitable for
3309 if (compaction_withdrawn(compact_result
)) {
3310 ret
= compaction_zonelist_suitable(ac
, order
, alloc_flags
);
3315 * !costly requests are much more important than __GFP_REPEAT
3316 * costly ones because they are de facto nofail and invoke OOM
3317 * killer to move on while costly can fail and users are ready
3318 * to cope with that. 1/4 retries is rather arbitrary but we
3319 * would need much more detailed feedback from compaction to
3320 * make a better decision.
3322 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
3324 if (*compaction_retries
<= max_retries
) {
3330 * Make sure there are attempts at the highest priority if we exhausted
3331 * all retries or failed at the lower priorities.
3334 min_priority
= (order
> PAGE_ALLOC_COSTLY_ORDER
) ?
3335 MIN_COMPACT_COSTLY_PRIORITY
: MIN_COMPACT_PRIORITY
;
3337 if (*compact_priority
> min_priority
) {
3338 (*compact_priority
)--;
3339 *compaction_retries
= 0;
3343 trace_compact_retry(order
, priority
, compact_result
, retries
, max_retries
, ret
);
3347 static inline struct page
*
3348 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
3349 unsigned int alloc_flags
, const struct alloc_context
*ac
,
3350 enum compact_priority prio
, enum compact_result
*compact_result
)
3352 *compact_result
= COMPACT_SKIPPED
;
3357 should_compact_retry(struct alloc_context
*ac
, unsigned int order
, int alloc_flags
,
3358 enum compact_result compact_result
,
3359 enum compact_priority
*compact_priority
,
3360 int *compaction_retries
)
3365 if (!order
|| order
> PAGE_ALLOC_COSTLY_ORDER
)
3369 * There are setups with compaction disabled which would prefer to loop
3370 * inside the allocator rather than hit the oom killer prematurely.
3371 * Let's give them a good hope and keep retrying while the order-0
3372 * watermarks are OK.
3374 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
, ac
->high_zoneidx
,
3376 if (zone_watermark_ok(zone
, 0, min_wmark_pages(zone
),
3377 ac_classzone_idx(ac
), alloc_flags
))
3382 #endif /* CONFIG_COMPACTION */
3384 /* Perform direct synchronous page reclaim */
3386 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
,
3387 const struct alloc_context
*ac
)
3389 struct reclaim_state reclaim_state
;
3394 /* We now go into synchronous reclaim */
3395 cpuset_memory_pressure_bump();
3396 current
->flags
|= PF_MEMALLOC
;
3397 lockdep_set_current_reclaim_state(gfp_mask
);
3398 reclaim_state
.reclaimed_slab
= 0;
3399 current
->reclaim_state
= &reclaim_state
;
3401 progress
= try_to_free_pages(ac
->zonelist
, order
, gfp_mask
,
3404 current
->reclaim_state
= NULL
;
3405 lockdep_clear_current_reclaim_state();
3406 current
->flags
&= ~PF_MEMALLOC
;
3413 /* The really slow allocator path where we enter direct reclaim */
3414 static inline struct page
*
3415 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
3416 unsigned int alloc_flags
, const struct alloc_context
*ac
,
3417 unsigned long *did_some_progress
)
3419 struct page
*page
= NULL
;
3420 bool drained
= false;
3422 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, ac
);
3423 if (unlikely(!(*did_some_progress
)))
3427 page
= get_page_from_freelist(gfp_mask
, order
, alloc_flags
, ac
);
3430 * If an allocation failed after direct reclaim, it could be because
3431 * pages are pinned on the per-cpu lists or in high alloc reserves.
3432 * Shrink them them and try again
3434 if (!page
&& !drained
) {
3435 unreserve_highatomic_pageblock(ac
, false);
3436 drain_all_pages(NULL
);
3444 static void wake_all_kswapds(unsigned int order
, const struct alloc_context
*ac
)
3448 pg_data_t
*last_pgdat
= NULL
;
3450 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
,
3451 ac
->high_zoneidx
, ac
->nodemask
) {
3452 if (last_pgdat
!= zone
->zone_pgdat
)
3453 wakeup_kswapd(zone
, order
, ac
->high_zoneidx
);
3454 last_pgdat
= zone
->zone_pgdat
;
3458 static inline unsigned int
3459 gfp_to_alloc_flags(gfp_t gfp_mask
)
3461 unsigned int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
3463 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
3464 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
3467 * The caller may dip into page reserves a bit more if the caller
3468 * cannot run direct reclaim, or if the caller has realtime scheduling
3469 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
3470 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
3472 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
3474 if (gfp_mask
& __GFP_ATOMIC
) {
3476 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
3477 * if it can't schedule.
3479 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
3480 alloc_flags
|= ALLOC_HARDER
;
3482 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
3483 * comment for __cpuset_node_allowed().
3485 alloc_flags
&= ~ALLOC_CPUSET
;
3486 } else if (unlikely(rt_task(current
)) && !in_interrupt())
3487 alloc_flags
|= ALLOC_HARDER
;
3490 if (gfpflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
3491 alloc_flags
|= ALLOC_CMA
;
3496 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask
)
3498 if (unlikely(gfp_mask
& __GFP_NOMEMALLOC
))
3501 if (gfp_mask
& __GFP_MEMALLOC
)
3503 if (in_serving_softirq() && (current
->flags
& PF_MEMALLOC
))
3505 if (!in_interrupt() &&
3506 ((current
->flags
& PF_MEMALLOC
) ||
3507 unlikely(test_thread_flag(TIF_MEMDIE
))))
3514 * Checks whether it makes sense to retry the reclaim to make a forward progress
3515 * for the given allocation request.
3517 * We give up when we either have tried MAX_RECLAIM_RETRIES in a row
3518 * without success, or when we couldn't even meet the watermark if we
3519 * reclaimed all remaining pages on the LRU lists.
3521 * Returns true if a retry is viable or false to enter the oom path.
3524 should_reclaim_retry(gfp_t gfp_mask
, unsigned order
,
3525 struct alloc_context
*ac
, int alloc_flags
,
3526 bool did_some_progress
, int *no_progress_loops
)
3532 * Costly allocations might have made a progress but this doesn't mean
3533 * their order will become available due to high fragmentation so
3534 * always increment the no progress counter for them
3536 if (did_some_progress
&& order
<= PAGE_ALLOC_COSTLY_ORDER
)
3537 *no_progress_loops
= 0;
3539 (*no_progress_loops
)++;
3542 * Make sure we converge to OOM if we cannot make any progress
3543 * several times in the row.
3545 if (*no_progress_loops
> MAX_RECLAIM_RETRIES
) {
3546 /* Before OOM, exhaust highatomic_reserve */
3547 return unreserve_highatomic_pageblock(ac
, true);
3551 * Keep reclaiming pages while there is a chance this will lead
3552 * somewhere. If none of the target zones can satisfy our allocation
3553 * request even if all reclaimable pages are considered then we are
3554 * screwed and have to go OOM.
3556 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
, ac
->high_zoneidx
,
3558 unsigned long available
;
3559 unsigned long reclaimable
;
3560 unsigned long min_wmark
= min_wmark_pages(zone
);
3563 available
= reclaimable
= zone_reclaimable_pages(zone
);
3564 available
+= zone_page_state_snapshot(zone
, NR_FREE_PAGES
);
3567 * Would the allocation succeed if we reclaimed all
3568 * reclaimable pages?
3570 wmark
= __zone_watermark_ok(zone
, order
, min_wmark
,
3571 ac_classzone_idx(ac
), alloc_flags
, available
);
3572 trace_reclaim_retry_zone(z
, order
, reclaimable
,
3573 available
, min_wmark
, *no_progress_loops
, wmark
);
3576 * If we didn't make any progress and have a lot of
3577 * dirty + writeback pages then we should wait for
3578 * an IO to complete to slow down the reclaim and
3579 * prevent from pre mature OOM
3581 if (!did_some_progress
) {
3582 unsigned long write_pending
;
3584 write_pending
= zone_page_state_snapshot(zone
,
3585 NR_ZONE_WRITE_PENDING
);
3587 if (2 * write_pending
> reclaimable
) {
3588 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3594 * Memory allocation/reclaim might be called from a WQ
3595 * context and the current implementation of the WQ
3596 * concurrency control doesn't recognize that
3597 * a particular WQ is congested if the worker thread is
3598 * looping without ever sleeping. Therefore we have to
3599 * do a short sleep here rather than calling
3602 if (current
->flags
& PF_WQ_WORKER
)
3603 schedule_timeout_uninterruptible(1);
3614 static inline struct page
*
3615 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
3616 struct alloc_context
*ac
)
3618 bool can_direct_reclaim
= gfp_mask
& __GFP_DIRECT_RECLAIM
;
3619 struct page
*page
= NULL
;
3620 unsigned int alloc_flags
;
3621 unsigned long did_some_progress
;
3622 enum compact_priority compact_priority
;
3623 enum compact_result compact_result
;
3624 int compaction_retries
;
3625 int no_progress_loops
;
3626 unsigned long alloc_start
= jiffies
;
3627 unsigned int stall_timeout
= 10 * HZ
;
3628 unsigned int cpuset_mems_cookie
;
3631 * In the slowpath, we sanity check order to avoid ever trying to
3632 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
3633 * be using allocators in order of preference for an area that is
3636 if (order
>= MAX_ORDER
) {
3637 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
3642 * We also sanity check to catch abuse of atomic reserves being used by
3643 * callers that are not in atomic context.
3645 if (WARN_ON_ONCE((gfp_mask
& (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)) ==
3646 (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)))
3647 gfp_mask
&= ~__GFP_ATOMIC
;
3650 compaction_retries
= 0;
3651 no_progress_loops
= 0;
3652 compact_priority
= DEF_COMPACT_PRIORITY
;
3653 cpuset_mems_cookie
= read_mems_allowed_begin();
3656 * The fast path uses conservative alloc_flags to succeed only until
3657 * kswapd needs to be woken up, and to avoid the cost of setting up
3658 * alloc_flags precisely. So we do that now.
3660 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
3663 * We need to recalculate the starting point for the zonelist iterator
3664 * because we might have used different nodemask in the fast path, or
3665 * there was a cpuset modification and we are retrying - otherwise we
3666 * could end up iterating over non-eligible zones endlessly.
3668 ac
->preferred_zoneref
= first_zones_zonelist(ac
->zonelist
,
3669 ac
->high_zoneidx
, ac
->nodemask
);
3670 if (!ac
->preferred_zoneref
->zone
)
3673 if (gfp_mask
& __GFP_KSWAPD_RECLAIM
)
3674 wake_all_kswapds(order
, ac
);
3677 * The adjusted alloc_flags might result in immediate success, so try
3680 page
= get_page_from_freelist(gfp_mask
, order
, alloc_flags
, ac
);
3685 * For costly allocations, try direct compaction first, as it's likely
3686 * that we have enough base pages and don't need to reclaim. Don't try
3687 * that for allocations that are allowed to ignore watermarks, as the
3688 * ALLOC_NO_WATERMARKS attempt didn't yet happen.
3690 if (can_direct_reclaim
&& order
> PAGE_ALLOC_COSTLY_ORDER
&&
3691 !gfp_pfmemalloc_allowed(gfp_mask
)) {
3692 page
= __alloc_pages_direct_compact(gfp_mask
, order
,
3694 INIT_COMPACT_PRIORITY
,
3700 * Checks for costly allocations with __GFP_NORETRY, which
3701 * includes THP page fault allocations
3703 if (gfp_mask
& __GFP_NORETRY
) {
3705 * If compaction is deferred for high-order allocations,
3706 * it is because sync compaction recently failed. If
3707 * this is the case and the caller requested a THP
3708 * allocation, we do not want to heavily disrupt the
3709 * system, so we fail the allocation instead of entering
3712 if (compact_result
== COMPACT_DEFERRED
)
3716 * Looks like reclaim/compaction is worth trying, but
3717 * sync compaction could be very expensive, so keep
3718 * using async compaction.
3720 compact_priority
= INIT_COMPACT_PRIORITY
;
3725 /* Ensure kswapd doesn't accidentally go to sleep as long as we loop */
3726 if (gfp_mask
& __GFP_KSWAPD_RECLAIM
)
3727 wake_all_kswapds(order
, ac
);
3729 if (gfp_pfmemalloc_allowed(gfp_mask
))
3730 alloc_flags
= ALLOC_NO_WATERMARKS
;
3733 * Reset the zonelist iterators if memory policies can be ignored.
3734 * These allocations are high priority and system rather than user
3737 if (!(alloc_flags
& ALLOC_CPUSET
) || (alloc_flags
& ALLOC_NO_WATERMARKS
)) {
3738 ac
->zonelist
= node_zonelist(numa_node_id(), gfp_mask
);
3739 ac
->preferred_zoneref
= first_zones_zonelist(ac
->zonelist
,
3740 ac
->high_zoneidx
, ac
->nodemask
);
3743 /* Attempt with potentially adjusted zonelist and alloc_flags */
3744 page
= get_page_from_freelist(gfp_mask
, order
, alloc_flags
, ac
);
3748 /* Caller is not willing to reclaim, we can't balance anything */
3749 if (!can_direct_reclaim
)
3752 /* Make sure we know about allocations which stall for too long */
3753 if (time_after(jiffies
, alloc_start
+ stall_timeout
)) {
3754 warn_alloc(gfp_mask
, ac
->nodemask
,
3755 "page allocation stalls for %ums, order:%u",
3756 jiffies_to_msecs(jiffies
-alloc_start
), order
);
3757 stall_timeout
+= 10 * HZ
;
3760 /* Avoid recursion of direct reclaim */
3761 if (current
->flags
& PF_MEMALLOC
)
3764 /* Try direct reclaim and then allocating */
3765 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
, alloc_flags
, ac
,
3766 &did_some_progress
);
3770 /* Try direct compaction and then allocating */
3771 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
, ac
,
3772 compact_priority
, &compact_result
);
3776 /* Do not loop if specifically requested */
3777 if (gfp_mask
& __GFP_NORETRY
)
3781 * Do not retry costly high order allocations unless they are
3784 if (order
> PAGE_ALLOC_COSTLY_ORDER
&& !(gfp_mask
& __GFP_REPEAT
))
3787 if (should_reclaim_retry(gfp_mask
, order
, ac
, alloc_flags
,
3788 did_some_progress
> 0, &no_progress_loops
))
3792 * It doesn't make any sense to retry for the compaction if the order-0
3793 * reclaim is not able to make any progress because the current
3794 * implementation of the compaction depends on the sufficient amount
3795 * of free memory (see __compaction_suitable)
3797 if (did_some_progress
> 0 &&
3798 should_compact_retry(ac
, order
, alloc_flags
,
3799 compact_result
, &compact_priority
,
3800 &compaction_retries
))
3804 * It's possible we raced with cpuset update so the OOM would be
3805 * premature (see below the nopage: label for full explanation).
3807 if (read_mems_allowed_retry(cpuset_mems_cookie
))
3810 /* Reclaim has failed us, start killing things */
3811 page
= __alloc_pages_may_oom(gfp_mask
, order
, ac
, &did_some_progress
);
3815 /* Avoid allocations with no watermarks from looping endlessly */
3816 if (test_thread_flag(TIF_MEMDIE
))
3819 /* Retry as long as the OOM killer is making progress */
3820 if (did_some_progress
) {
3821 no_progress_loops
= 0;
3827 * When updating a task's mems_allowed or mempolicy nodemask, it is
3828 * possible to race with parallel threads in such a way that our
3829 * allocation can fail while the mask is being updated. If we are about
3830 * to fail, check if the cpuset changed during allocation and if so,
3833 if (read_mems_allowed_retry(cpuset_mems_cookie
))
3837 * Make sure that __GFP_NOFAIL request doesn't leak out and make sure
3840 if (gfp_mask
& __GFP_NOFAIL
) {
3842 * All existing users of the __GFP_NOFAIL are blockable, so warn
3843 * of any new users that actually require GFP_NOWAIT
3845 if (WARN_ON_ONCE(!can_direct_reclaim
))
3849 * PF_MEMALLOC request from this context is rather bizarre
3850 * because we cannot reclaim anything and only can loop waiting
3851 * for somebody to do a work for us
3853 WARN_ON_ONCE(current
->flags
& PF_MEMALLOC
);
3856 * non failing costly orders are a hard requirement which we
3857 * are not prepared for much so let's warn about these users
3858 * so that we can identify them and convert them to something
3861 WARN_ON_ONCE(order
> PAGE_ALLOC_COSTLY_ORDER
);
3864 * Help non-failing allocations by giving them access to memory
3865 * reserves but do not use ALLOC_NO_WATERMARKS because this
3866 * could deplete whole memory reserves which would just make
3867 * the situation worse
3869 page
= __alloc_pages_cpuset_fallback(gfp_mask
, order
, ALLOC_HARDER
, ac
);
3877 warn_alloc(gfp_mask
, ac
->nodemask
,
3878 "page allocation failure: order:%u", order
);
3883 static inline bool prepare_alloc_pages(gfp_t gfp_mask
, unsigned int order
,
3884 struct zonelist
*zonelist
, nodemask_t
*nodemask
,
3885 struct alloc_context
*ac
, gfp_t
*alloc_mask
,
3886 unsigned int *alloc_flags
)
3888 ac
->high_zoneidx
= gfp_zone(gfp_mask
);
3889 ac
->zonelist
= zonelist
;
3890 ac
->nodemask
= nodemask
;
3891 ac
->migratetype
= gfpflags_to_migratetype(gfp_mask
);
3893 if (cpusets_enabled()) {
3894 *alloc_mask
|= __GFP_HARDWALL
;
3896 ac
->nodemask
= &cpuset_current_mems_allowed
;
3898 *alloc_flags
|= ALLOC_CPUSET
;
3901 lockdep_trace_alloc(gfp_mask
);
3903 might_sleep_if(gfp_mask
& __GFP_DIRECT_RECLAIM
);
3905 if (should_fail_alloc_page(gfp_mask
, order
))
3908 if (IS_ENABLED(CONFIG_CMA
) && ac
->migratetype
== MIGRATE_MOVABLE
)
3909 *alloc_flags
|= ALLOC_CMA
;
3914 /* Determine whether to spread dirty pages and what the first usable zone */
3915 static inline void finalise_ac(gfp_t gfp_mask
,
3916 unsigned int order
, struct alloc_context
*ac
)
3918 /* Dirty zone balancing only done in the fast path */
3919 ac
->spread_dirty_pages
= (gfp_mask
& __GFP_WRITE
);
3922 * The preferred zone is used for statistics but crucially it is
3923 * also used as the starting point for the zonelist iterator. It
3924 * may get reset for allocations that ignore memory policies.
3926 ac
->preferred_zoneref
= first_zones_zonelist(ac
->zonelist
,
3927 ac
->high_zoneidx
, ac
->nodemask
);
3931 * This is the 'heart' of the zoned buddy allocator.
3934 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
3935 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
3938 unsigned int alloc_flags
= ALLOC_WMARK_LOW
;
3939 gfp_t alloc_mask
= gfp_mask
; /* The gfp_t that was actually used for allocation */
3940 struct alloc_context ac
= { };
3942 gfp_mask
&= gfp_allowed_mask
;
3943 if (!prepare_alloc_pages(gfp_mask
, order
, zonelist
, nodemask
, &ac
, &alloc_mask
, &alloc_flags
))
3946 finalise_ac(gfp_mask
, order
, &ac
);
3948 /* First allocation attempt */
3949 page
= get_page_from_freelist(alloc_mask
, order
, alloc_flags
, &ac
);
3954 * Runtime PM, block IO and its error handling path can deadlock
3955 * because I/O on the device might not complete.
3957 alloc_mask
= memalloc_noio_flags(gfp_mask
);
3958 ac
.spread_dirty_pages
= false;
3961 * Restore the original nodemask if it was potentially replaced with
3962 * &cpuset_current_mems_allowed to optimize the fast-path attempt.
3964 if (unlikely(ac
.nodemask
!= nodemask
))
3965 ac
.nodemask
= nodemask
;
3967 page
= __alloc_pages_slowpath(alloc_mask
, order
, &ac
);
3970 if (memcg_kmem_enabled() && (gfp_mask
& __GFP_ACCOUNT
) && page
&&
3971 unlikely(memcg_kmem_charge(page
, gfp_mask
, order
) != 0)) {
3972 __free_pages(page
, order
);
3976 if (kmemcheck_enabled
&& page
)
3977 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
3979 trace_mm_page_alloc(page
, order
, alloc_mask
, ac
.migratetype
);
3983 EXPORT_SYMBOL(__alloc_pages_nodemask
);
3986 * Common helper functions.
3988 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
3993 * __get_free_pages() returns a 32-bit address, which cannot represent
3996 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
3998 page
= alloc_pages(gfp_mask
, order
);
4001 return (unsigned long) page_address(page
);
4003 EXPORT_SYMBOL(__get_free_pages
);
4005 unsigned long get_zeroed_page(gfp_t gfp_mask
)
4007 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
4009 EXPORT_SYMBOL(get_zeroed_page
);
4011 void __free_pages(struct page
*page
, unsigned int order
)
4013 if (put_page_testzero(page
)) {
4015 free_hot_cold_page(page
, false);
4017 __free_pages_ok(page
, order
);
4021 EXPORT_SYMBOL(__free_pages
);
4023 void free_pages(unsigned long addr
, unsigned int order
)
4026 VM_BUG_ON(!virt_addr_valid((void *)addr
));
4027 __free_pages(virt_to_page((void *)addr
), order
);
4031 EXPORT_SYMBOL(free_pages
);
4035 * An arbitrary-length arbitrary-offset area of memory which resides
4036 * within a 0 or higher order page. Multiple fragments within that page
4037 * are individually refcounted, in the page's reference counter.
4039 * The page_frag functions below provide a simple allocation framework for
4040 * page fragments. This is used by the network stack and network device
4041 * drivers to provide a backing region of memory for use as either an
4042 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
4044 static struct page
*__page_frag_cache_refill(struct page_frag_cache
*nc
,
4047 struct page
*page
= NULL
;
4048 gfp_t gfp
= gfp_mask
;
4050 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4051 gfp_mask
|= __GFP_COMP
| __GFP_NOWARN
| __GFP_NORETRY
|
4053 page
= alloc_pages_node(NUMA_NO_NODE
, gfp_mask
,
4054 PAGE_FRAG_CACHE_MAX_ORDER
);
4055 nc
->size
= page
? PAGE_FRAG_CACHE_MAX_SIZE
: PAGE_SIZE
;
4057 if (unlikely(!page
))
4058 page
= alloc_pages_node(NUMA_NO_NODE
, gfp
, 0);
4060 nc
->va
= page
? page_address(page
) : NULL
;
4065 void __page_frag_cache_drain(struct page
*page
, unsigned int count
)
4067 VM_BUG_ON_PAGE(page_ref_count(page
) == 0, page
);
4069 if (page_ref_sub_and_test(page
, count
)) {
4070 unsigned int order
= compound_order(page
);
4073 free_hot_cold_page(page
, false);
4075 __free_pages_ok(page
, order
);
4078 EXPORT_SYMBOL(__page_frag_cache_drain
);
4080 void *page_frag_alloc(struct page_frag_cache
*nc
,
4081 unsigned int fragsz
, gfp_t gfp_mask
)
4083 unsigned int size
= PAGE_SIZE
;
4087 if (unlikely(!nc
->va
)) {
4089 page
= __page_frag_cache_refill(nc
, gfp_mask
);
4093 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4094 /* if size can vary use size else just use PAGE_SIZE */
4097 /* Even if we own the page, we do not use atomic_set().
4098 * This would break get_page_unless_zero() users.
4100 page_ref_add(page
, size
- 1);
4102 /* reset page count bias and offset to start of new frag */
4103 nc
->pfmemalloc
= page_is_pfmemalloc(page
);
4104 nc
->pagecnt_bias
= size
;
4108 offset
= nc
->offset
- fragsz
;
4109 if (unlikely(offset
< 0)) {
4110 page
= virt_to_page(nc
->va
);
4112 if (!page_ref_sub_and_test(page
, nc
->pagecnt_bias
))
4115 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
4116 /* if size can vary use size else just use PAGE_SIZE */
4119 /* OK, page count is 0, we can safely set it */
4120 set_page_count(page
, size
);
4122 /* reset page count bias and offset to start of new frag */
4123 nc
->pagecnt_bias
= size
;
4124 offset
= size
- fragsz
;
4128 nc
->offset
= offset
;
4130 return nc
->va
+ offset
;
4132 EXPORT_SYMBOL(page_frag_alloc
);
4135 * Frees a page fragment allocated out of either a compound or order 0 page.
4137 void page_frag_free(void *addr
)
4139 struct page
*page
= virt_to_head_page(addr
);
4141 if (unlikely(put_page_testzero(page
)))
4142 __free_pages_ok(page
, compound_order(page
));
4144 EXPORT_SYMBOL(page_frag_free
);
4146 static void *make_alloc_exact(unsigned long addr
, unsigned int order
,
4150 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
4151 unsigned long used
= addr
+ PAGE_ALIGN(size
);
4153 split_page(virt_to_page((void *)addr
), order
);
4154 while (used
< alloc_end
) {
4159 return (void *)addr
;
4163 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
4164 * @size: the number of bytes to allocate
4165 * @gfp_mask: GFP flags for the allocation
4167 * This function is similar to alloc_pages(), except that it allocates the
4168 * minimum number of pages to satisfy the request. alloc_pages() can only
4169 * allocate memory in power-of-two pages.
4171 * This function is also limited by MAX_ORDER.
4173 * Memory allocated by this function must be released by free_pages_exact().
4175 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
4177 unsigned int order
= get_order(size
);
4180 addr
= __get_free_pages(gfp_mask
, order
);
4181 return make_alloc_exact(addr
, order
, size
);
4183 EXPORT_SYMBOL(alloc_pages_exact
);
4186 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
4188 * @nid: the preferred node ID where memory should be allocated
4189 * @size: the number of bytes to allocate
4190 * @gfp_mask: GFP flags for the allocation
4192 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
4195 void * __meminit
alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
4197 unsigned int order
= get_order(size
);
4198 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
4201 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
4205 * free_pages_exact - release memory allocated via alloc_pages_exact()
4206 * @virt: the value returned by alloc_pages_exact.
4207 * @size: size of allocation, same value as passed to alloc_pages_exact().
4209 * Release the memory allocated by a previous call to alloc_pages_exact.
4211 void free_pages_exact(void *virt
, size_t size
)
4213 unsigned long addr
= (unsigned long)virt
;
4214 unsigned long end
= addr
+ PAGE_ALIGN(size
);
4216 while (addr
< end
) {
4221 EXPORT_SYMBOL(free_pages_exact
);
4224 * nr_free_zone_pages - count number of pages beyond high watermark
4225 * @offset: The zone index of the highest zone
4227 * nr_free_zone_pages() counts the number of counts pages which are beyond the
4228 * high watermark within all zones at or below a given zone index. For each
4229 * zone, the number of pages is calculated as:
4231 * nr_free_zone_pages = managed_pages - high_pages
4233 static unsigned long nr_free_zone_pages(int offset
)
4238 /* Just pick one node, since fallback list is circular */
4239 unsigned long sum
= 0;
4241 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
4243 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
4244 unsigned long size
= zone
->managed_pages
;
4245 unsigned long high
= high_wmark_pages(zone
);
4254 * nr_free_buffer_pages - count number of pages beyond high watermark
4256 * nr_free_buffer_pages() counts the number of pages which are beyond the high
4257 * watermark within ZONE_DMA and ZONE_NORMAL.
4259 unsigned long nr_free_buffer_pages(void)
4261 return nr_free_zone_pages(gfp_zone(GFP_USER
));
4263 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
4266 * nr_free_pagecache_pages - count number of pages beyond high watermark
4268 * nr_free_pagecache_pages() counts the number of pages which are beyond the
4269 * high watermark within all zones.
4271 unsigned long nr_free_pagecache_pages(void)
4273 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
4276 static inline void show_node(struct zone
*zone
)
4278 if (IS_ENABLED(CONFIG_NUMA
))
4279 printk("Node %d ", zone_to_nid(zone
));
4282 long si_mem_available(void)
4285 unsigned long pagecache
;
4286 unsigned long wmark_low
= 0;
4287 unsigned long pages
[NR_LRU_LISTS
];
4291 for (lru
= LRU_BASE
; lru
< NR_LRU_LISTS
; lru
++)
4292 pages
[lru
] = global_node_page_state(NR_LRU_BASE
+ lru
);
4295 wmark_low
+= zone
->watermark
[WMARK_LOW
];
4298 * Estimate the amount of memory available for userspace allocations,
4299 * without causing swapping.
4301 available
= global_page_state(NR_FREE_PAGES
) - totalreserve_pages
;
4304 * Not all the page cache can be freed, otherwise the system will
4305 * start swapping. Assume at least half of the page cache, or the
4306 * low watermark worth of cache, needs to stay.
4308 pagecache
= pages
[LRU_ACTIVE_FILE
] + pages
[LRU_INACTIVE_FILE
];
4309 pagecache
-= min(pagecache
/ 2, wmark_low
);
4310 available
+= pagecache
;
4313 * Part of the reclaimable slab consists of items that are in use,
4314 * and cannot be freed. Cap this estimate at the low watermark.
4316 available
+= global_page_state(NR_SLAB_RECLAIMABLE
) -
4317 min(global_page_state(NR_SLAB_RECLAIMABLE
) / 2, wmark_low
);
4323 EXPORT_SYMBOL_GPL(si_mem_available
);
4325 void si_meminfo(struct sysinfo
*val
)
4327 val
->totalram
= totalram_pages
;
4328 val
->sharedram
= global_node_page_state(NR_SHMEM
);
4329 val
->freeram
= global_page_state(NR_FREE_PAGES
);
4330 val
->bufferram
= nr_blockdev_pages();
4331 val
->totalhigh
= totalhigh_pages
;
4332 val
->freehigh
= nr_free_highpages();
4333 val
->mem_unit
= PAGE_SIZE
;
4336 EXPORT_SYMBOL(si_meminfo
);
4339 void si_meminfo_node(struct sysinfo
*val
, int nid
)
4341 int zone_type
; /* needs to be signed */
4342 unsigned long managed_pages
= 0;
4343 unsigned long managed_highpages
= 0;
4344 unsigned long free_highpages
= 0;
4345 pg_data_t
*pgdat
= NODE_DATA(nid
);
4347 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++)
4348 managed_pages
+= pgdat
->node_zones
[zone_type
].managed_pages
;
4349 val
->totalram
= managed_pages
;
4350 val
->sharedram
= node_page_state(pgdat
, NR_SHMEM
);
4351 val
->freeram
= sum_zone_node_page_state(nid
, NR_FREE_PAGES
);
4352 #ifdef CONFIG_HIGHMEM
4353 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
4354 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4356 if (is_highmem(zone
)) {
4357 managed_highpages
+= zone
->managed_pages
;
4358 free_highpages
+= zone_page_state(zone
, NR_FREE_PAGES
);
4361 val
->totalhigh
= managed_highpages
;
4362 val
->freehigh
= free_highpages
;
4364 val
->totalhigh
= managed_highpages
;
4365 val
->freehigh
= free_highpages
;
4367 val
->mem_unit
= PAGE_SIZE
;
4372 * Determine whether the node should be displayed or not, depending on whether
4373 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
4375 static bool show_mem_node_skip(unsigned int flags
, int nid
, nodemask_t
*nodemask
)
4377 if (!(flags
& SHOW_MEM_FILTER_NODES
))
4381 * no node mask - aka implicit memory numa policy. Do not bother with
4382 * the synchronization - read_mems_allowed_begin - because we do not
4383 * have to be precise here.
4386 nodemask
= &cpuset_current_mems_allowed
;
4388 return !node_isset(nid
, *nodemask
);
4391 #define K(x) ((x) << (PAGE_SHIFT-10))
4393 static void show_migration_types(unsigned char type
)
4395 static const char types
[MIGRATE_TYPES
] = {
4396 [MIGRATE_UNMOVABLE
] = 'U',
4397 [MIGRATE_MOVABLE
] = 'M',
4398 [MIGRATE_RECLAIMABLE
] = 'E',
4399 [MIGRATE_HIGHATOMIC
] = 'H',
4401 [MIGRATE_CMA
] = 'C',
4403 #ifdef CONFIG_MEMORY_ISOLATION
4404 [MIGRATE_ISOLATE
] = 'I',
4407 char tmp
[MIGRATE_TYPES
+ 1];
4411 for (i
= 0; i
< MIGRATE_TYPES
; i
++) {
4412 if (type
& (1 << i
))
4417 printk(KERN_CONT
"(%s) ", tmp
);
4421 * Show free area list (used inside shift_scroll-lock stuff)
4422 * We also calculate the percentage fragmentation. We do this by counting the
4423 * memory on each free list with the exception of the first item on the list.
4426 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
4429 void show_free_areas(unsigned int filter
, nodemask_t
*nodemask
)
4431 unsigned long free_pcp
= 0;
4436 for_each_populated_zone(zone
) {
4437 if (show_mem_node_skip(filter
, zone_to_nid(zone
), nodemask
))
4440 for_each_online_cpu(cpu
)
4441 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
4444 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
4445 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
4446 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
4447 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
4448 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
4449 " free:%lu free_pcp:%lu free_cma:%lu\n",
4450 global_node_page_state(NR_ACTIVE_ANON
),
4451 global_node_page_state(NR_INACTIVE_ANON
),
4452 global_node_page_state(NR_ISOLATED_ANON
),
4453 global_node_page_state(NR_ACTIVE_FILE
),
4454 global_node_page_state(NR_INACTIVE_FILE
),
4455 global_node_page_state(NR_ISOLATED_FILE
),
4456 global_node_page_state(NR_UNEVICTABLE
),
4457 global_node_page_state(NR_FILE_DIRTY
),
4458 global_node_page_state(NR_WRITEBACK
),
4459 global_node_page_state(NR_UNSTABLE_NFS
),
4460 global_page_state(NR_SLAB_RECLAIMABLE
),
4461 global_page_state(NR_SLAB_UNRECLAIMABLE
),
4462 global_node_page_state(NR_FILE_MAPPED
),
4463 global_node_page_state(NR_SHMEM
),
4464 global_page_state(NR_PAGETABLE
),
4465 global_page_state(NR_BOUNCE
),
4466 global_page_state(NR_FREE_PAGES
),
4468 global_page_state(NR_FREE_CMA_PAGES
));
4470 for_each_online_pgdat(pgdat
) {
4471 if (show_mem_node_skip(filter
, pgdat
->node_id
, nodemask
))
4475 " active_anon:%lukB"
4476 " inactive_anon:%lukB"
4477 " active_file:%lukB"
4478 " inactive_file:%lukB"
4479 " unevictable:%lukB"
4480 " isolated(anon):%lukB"
4481 " isolated(file):%lukB"
4486 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4488 " shmem_pmdmapped: %lukB"
4491 " writeback_tmp:%lukB"
4493 " all_unreclaimable? %s"
4496 K(node_page_state(pgdat
, NR_ACTIVE_ANON
)),
4497 K(node_page_state(pgdat
, NR_INACTIVE_ANON
)),
4498 K(node_page_state(pgdat
, NR_ACTIVE_FILE
)),
4499 K(node_page_state(pgdat
, NR_INACTIVE_FILE
)),
4500 K(node_page_state(pgdat
, NR_UNEVICTABLE
)),
4501 K(node_page_state(pgdat
, NR_ISOLATED_ANON
)),
4502 K(node_page_state(pgdat
, NR_ISOLATED_FILE
)),
4503 K(node_page_state(pgdat
, NR_FILE_MAPPED
)),
4504 K(node_page_state(pgdat
, NR_FILE_DIRTY
)),
4505 K(node_page_state(pgdat
, NR_WRITEBACK
)),
4506 K(node_page_state(pgdat
, NR_SHMEM
)),
4507 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4508 K(node_page_state(pgdat
, NR_SHMEM_THPS
) * HPAGE_PMD_NR
),
4509 K(node_page_state(pgdat
, NR_SHMEM_PMDMAPPED
)
4511 K(node_page_state(pgdat
, NR_ANON_THPS
) * HPAGE_PMD_NR
),
4513 K(node_page_state(pgdat
, NR_WRITEBACK_TEMP
)),
4514 K(node_page_state(pgdat
, NR_UNSTABLE_NFS
)),
4515 pgdat
->kswapd_failures
>= MAX_RECLAIM_RETRIES
?
4519 for_each_populated_zone(zone
) {
4522 if (show_mem_node_skip(filter
, zone_to_nid(zone
), nodemask
))
4526 for_each_online_cpu(cpu
)
4527 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
4536 " active_anon:%lukB"
4537 " inactive_anon:%lukB"
4538 " active_file:%lukB"
4539 " inactive_file:%lukB"
4540 " unevictable:%lukB"
4541 " writepending:%lukB"
4545 " slab_reclaimable:%lukB"
4546 " slab_unreclaimable:%lukB"
4547 " kernel_stack:%lukB"
4555 K(zone_page_state(zone
, NR_FREE_PAGES
)),
4556 K(min_wmark_pages(zone
)),
4557 K(low_wmark_pages(zone
)),
4558 K(high_wmark_pages(zone
)),
4559 K(zone_page_state(zone
, NR_ZONE_ACTIVE_ANON
)),
4560 K(zone_page_state(zone
, NR_ZONE_INACTIVE_ANON
)),
4561 K(zone_page_state(zone
, NR_ZONE_ACTIVE_FILE
)),
4562 K(zone_page_state(zone
, NR_ZONE_INACTIVE_FILE
)),
4563 K(zone_page_state(zone
, NR_ZONE_UNEVICTABLE
)),
4564 K(zone_page_state(zone
, NR_ZONE_WRITE_PENDING
)),
4565 K(zone
->present_pages
),
4566 K(zone
->managed_pages
),
4567 K(zone_page_state(zone
, NR_MLOCK
)),
4568 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
4569 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
4570 zone_page_state(zone
, NR_KERNEL_STACK_KB
),
4571 K(zone_page_state(zone
, NR_PAGETABLE
)),
4572 K(zone_page_state(zone
, NR_BOUNCE
)),
4574 K(this_cpu_read(zone
->pageset
->pcp
.count
)),
4575 K(zone_page_state(zone
, NR_FREE_CMA_PAGES
)));
4576 printk("lowmem_reserve[]:");
4577 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
4578 printk(KERN_CONT
" %ld", zone
->lowmem_reserve
[i
]);
4579 printk(KERN_CONT
"\n");
4582 for_each_populated_zone(zone
) {
4584 unsigned long nr
[MAX_ORDER
], flags
, total
= 0;
4585 unsigned char types
[MAX_ORDER
];
4587 if (show_mem_node_skip(filter
, zone_to_nid(zone
), nodemask
))
4590 printk(KERN_CONT
"%s: ", zone
->name
);
4592 spin_lock_irqsave(&zone
->lock
, flags
);
4593 for (order
= 0; order
< MAX_ORDER
; order
++) {
4594 struct free_area
*area
= &zone
->free_area
[order
];
4597 nr
[order
] = area
->nr_free
;
4598 total
+= nr
[order
] << order
;
4601 for (type
= 0; type
< MIGRATE_TYPES
; type
++) {
4602 if (!list_empty(&area
->free_list
[type
]))
4603 types
[order
] |= 1 << type
;
4606 spin_unlock_irqrestore(&zone
->lock
, flags
);
4607 for (order
= 0; order
< MAX_ORDER
; order
++) {
4608 printk(KERN_CONT
"%lu*%lukB ",
4609 nr
[order
], K(1UL) << order
);
4611 show_migration_types(types
[order
]);
4613 printk(KERN_CONT
"= %lukB\n", K(total
));
4616 hugetlb_show_meminfo();
4618 printk("%ld total pagecache pages\n", global_node_page_state(NR_FILE_PAGES
));
4620 show_swap_cache_info();
4623 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
4625 zoneref
->zone
= zone
;
4626 zoneref
->zone_idx
= zone_idx(zone
);
4630 * Builds allocation fallback zone lists.
4632 * Add all populated zones of a node to the zonelist.
4634 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
4638 enum zone_type zone_type
= MAX_NR_ZONES
;
4642 zone
= pgdat
->node_zones
+ zone_type
;
4643 if (managed_zone(zone
)) {
4644 zoneref_set_zone(zone
,
4645 &zonelist
->_zonerefs
[nr_zones
++]);
4646 check_highest_zone(zone_type
);
4648 } while (zone_type
);
4656 * 0 = automatic detection of better ordering.
4657 * 1 = order by ([node] distance, -zonetype)
4658 * 2 = order by (-zonetype, [node] distance)
4660 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
4661 * the same zonelist. So only NUMA can configure this param.
4663 #define ZONELIST_ORDER_DEFAULT 0
4664 #define ZONELIST_ORDER_NODE 1
4665 #define ZONELIST_ORDER_ZONE 2
4667 /* zonelist order in the kernel.
4668 * set_zonelist_order() will set this to NODE or ZONE.
4670 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4671 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
4675 /* The value user specified ....changed by config */
4676 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4677 /* string for sysctl */
4678 #define NUMA_ZONELIST_ORDER_LEN 16
4679 char numa_zonelist_order
[16] = "default";
4682 * interface for configure zonelist ordering.
4683 * command line option "numa_zonelist_order"
4684 * = "[dD]efault - default, automatic configuration.
4685 * = "[nN]ode - order by node locality, then by zone within node
4686 * = "[zZ]one - order by zone, then by locality within zone
4689 static int __parse_numa_zonelist_order(char *s
)
4691 if (*s
== 'd' || *s
== 'D') {
4692 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
4693 } else if (*s
== 'n' || *s
== 'N') {
4694 user_zonelist_order
= ZONELIST_ORDER_NODE
;
4695 } else if (*s
== 'z' || *s
== 'Z') {
4696 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
4698 pr_warn("Ignoring invalid numa_zonelist_order value: %s\n", s
);
4704 static __init
int setup_numa_zonelist_order(char *s
)
4711 ret
= __parse_numa_zonelist_order(s
);
4713 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
4717 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
4720 * sysctl handler for numa_zonelist_order
4722 int numa_zonelist_order_handler(struct ctl_table
*table
, int write
,
4723 void __user
*buffer
, size_t *length
,
4726 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
4728 static DEFINE_MUTEX(zl_order_mutex
);
4730 mutex_lock(&zl_order_mutex
);
4732 if (strlen((char *)table
->data
) >= NUMA_ZONELIST_ORDER_LEN
) {
4736 strcpy(saved_string
, (char *)table
->data
);
4738 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
4742 int oldval
= user_zonelist_order
;
4744 ret
= __parse_numa_zonelist_order((char *)table
->data
);
4747 * bogus value. restore saved string
4749 strncpy((char *)table
->data
, saved_string
,
4750 NUMA_ZONELIST_ORDER_LEN
);
4751 user_zonelist_order
= oldval
;
4752 } else if (oldval
!= user_zonelist_order
) {
4753 mutex_lock(&zonelists_mutex
);
4754 build_all_zonelists(NULL
, NULL
);
4755 mutex_unlock(&zonelists_mutex
);
4759 mutex_unlock(&zl_order_mutex
);
4764 #define MAX_NODE_LOAD (nr_online_nodes)
4765 static int node_load
[MAX_NUMNODES
];
4768 * find_next_best_node - find the next node that should appear in a given node's fallback list
4769 * @node: node whose fallback list we're appending
4770 * @used_node_mask: nodemask_t of already used nodes
4772 * We use a number of factors to determine which is the next node that should
4773 * appear on a given node's fallback list. The node should not have appeared
4774 * already in @node's fallback list, and it should be the next closest node
4775 * according to the distance array (which contains arbitrary distance values
4776 * from each node to each node in the system), and should also prefer nodes
4777 * with no CPUs, since presumably they'll have very little allocation pressure
4778 * on them otherwise.
4779 * It returns -1 if no node is found.
4781 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
4784 int min_val
= INT_MAX
;
4785 int best_node
= NUMA_NO_NODE
;
4786 const struct cpumask
*tmp
= cpumask_of_node(0);
4788 /* Use the local node if we haven't already */
4789 if (!node_isset(node
, *used_node_mask
)) {
4790 node_set(node
, *used_node_mask
);
4794 for_each_node_state(n
, N_MEMORY
) {
4796 /* Don't want a node to appear more than once */
4797 if (node_isset(n
, *used_node_mask
))
4800 /* Use the distance array to find the distance */
4801 val
= node_distance(node
, n
);
4803 /* Penalize nodes under us ("prefer the next node") */
4806 /* Give preference to headless and unused nodes */
4807 tmp
= cpumask_of_node(n
);
4808 if (!cpumask_empty(tmp
))
4809 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
4811 /* Slight preference for less loaded node */
4812 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
4813 val
+= node_load
[n
];
4815 if (val
< min_val
) {
4822 node_set(best_node
, *used_node_mask
);
4829 * Build zonelists ordered by node and zones within node.
4830 * This results in maximum locality--normal zone overflows into local
4831 * DMA zone, if any--but risks exhausting DMA zone.
4833 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
4836 struct zonelist
*zonelist
;
4838 zonelist
= &pgdat
->node_zonelists
[ZONELIST_FALLBACK
];
4839 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
4841 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4842 zonelist
->_zonerefs
[j
].zone
= NULL
;
4843 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4847 * Build gfp_thisnode zonelists
4849 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
4852 struct zonelist
*zonelist
;
4854 zonelist
= &pgdat
->node_zonelists
[ZONELIST_NOFALLBACK
];
4855 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4856 zonelist
->_zonerefs
[j
].zone
= NULL
;
4857 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4861 * Build zonelists ordered by zone and nodes within zones.
4862 * This results in conserving DMA zone[s] until all Normal memory is
4863 * exhausted, but results in overflowing to remote node while memory
4864 * may still exist in local DMA zone.
4866 static int node_order
[MAX_NUMNODES
];
4868 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
4871 int zone_type
; /* needs to be signed */
4873 struct zonelist
*zonelist
;
4875 zonelist
= &pgdat
->node_zonelists
[ZONELIST_FALLBACK
];
4877 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
4878 for (j
= 0; j
< nr_nodes
; j
++) {
4879 node
= node_order
[j
];
4880 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
4881 if (managed_zone(z
)) {
4883 &zonelist
->_zonerefs
[pos
++]);
4884 check_highest_zone(zone_type
);
4888 zonelist
->_zonerefs
[pos
].zone
= NULL
;
4889 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
4892 #if defined(CONFIG_64BIT)
4894 * Devices that require DMA32/DMA are relatively rare and do not justify a
4895 * penalty to every machine in case the specialised case applies. Default
4896 * to Node-ordering on 64-bit NUMA machines
4898 static int default_zonelist_order(void)
4900 return ZONELIST_ORDER_NODE
;
4904 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4905 * by the kernel. If processes running on node 0 deplete the low memory zone
4906 * then reclaim will occur more frequency increasing stalls and potentially
4907 * be easier to OOM if a large percentage of the zone is under writeback or
4908 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4909 * Hence, default to zone ordering on 32-bit.
4911 static int default_zonelist_order(void)
4913 return ZONELIST_ORDER_ZONE
;
4915 #endif /* CONFIG_64BIT */
4917 static void set_zonelist_order(void)
4919 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
4920 current_zonelist_order
= default_zonelist_order();
4922 current_zonelist_order
= user_zonelist_order
;
4925 static void build_zonelists(pg_data_t
*pgdat
)
4928 nodemask_t used_mask
;
4929 int local_node
, prev_node
;
4930 struct zonelist
*zonelist
;
4931 unsigned int order
= current_zonelist_order
;
4933 /* initialize zonelists */
4934 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
4935 zonelist
= pgdat
->node_zonelists
+ i
;
4936 zonelist
->_zonerefs
[0].zone
= NULL
;
4937 zonelist
->_zonerefs
[0].zone_idx
= 0;
4940 /* NUMA-aware ordering of nodes */
4941 local_node
= pgdat
->node_id
;
4942 load
= nr_online_nodes
;
4943 prev_node
= local_node
;
4944 nodes_clear(used_mask
);
4946 memset(node_order
, 0, sizeof(node_order
));
4949 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
4951 * We don't want to pressure a particular node.
4952 * So adding penalty to the first node in same
4953 * distance group to make it round-robin.
4955 if (node_distance(local_node
, node
) !=
4956 node_distance(local_node
, prev_node
))
4957 node_load
[node
] = load
;
4961 if (order
== ZONELIST_ORDER_NODE
)
4962 build_zonelists_in_node_order(pgdat
, node
);
4964 node_order
[i
++] = node
; /* remember order */
4967 if (order
== ZONELIST_ORDER_ZONE
) {
4968 /* calculate node order -- i.e., DMA last! */
4969 build_zonelists_in_zone_order(pgdat
, i
);
4972 build_thisnode_zonelists(pgdat
);
4975 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4977 * Return node id of node used for "local" allocations.
4978 * I.e., first node id of first zone in arg node's generic zonelist.
4979 * Used for initializing percpu 'numa_mem', which is used primarily
4980 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4982 int local_memory_node(int node
)
4986 z
= first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
4987 gfp_zone(GFP_KERNEL
),
4989 return z
->zone
->node
;
4993 static void setup_min_unmapped_ratio(void);
4994 static void setup_min_slab_ratio(void);
4995 #else /* CONFIG_NUMA */
4997 static void set_zonelist_order(void)
4999 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
5002 static void build_zonelists(pg_data_t
*pgdat
)
5004 int node
, local_node
;
5006 struct zonelist
*zonelist
;
5008 local_node
= pgdat
->node_id
;
5010 zonelist
= &pgdat
->node_zonelists
[ZONELIST_FALLBACK
];
5011 j
= build_zonelists_node(pgdat
, zonelist
, 0);
5014 * Now we build the zonelist so that it contains the zones
5015 * of all the other nodes.
5016 * We don't want to pressure a particular node, so when
5017 * building the zones for node N, we make sure that the
5018 * zones coming right after the local ones are those from
5019 * node N+1 (modulo N)
5021 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
5022 if (!node_online(node
))
5024 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
5026 for (node
= 0; node
< local_node
; node
++) {
5027 if (!node_online(node
))
5029 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
5032 zonelist
->_zonerefs
[j
].zone
= NULL
;
5033 zonelist
->_zonerefs
[j
].zone_idx
= 0;
5036 #endif /* CONFIG_NUMA */
5039 * Boot pageset table. One per cpu which is going to be used for all
5040 * zones and all nodes. The parameters will be set in such a way
5041 * that an item put on a list will immediately be handed over to
5042 * the buddy list. This is safe since pageset manipulation is done
5043 * with interrupts disabled.
5045 * The boot_pagesets must be kept even after bootup is complete for
5046 * unused processors and/or zones. They do play a role for bootstrapping
5047 * hotplugged processors.
5049 * zoneinfo_show() and maybe other functions do
5050 * not check if the processor is online before following the pageset pointer.
5051 * Other parts of the kernel may not check if the zone is available.
5053 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
5054 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
5055 static void setup_zone_pageset(struct zone
*zone
);
5058 * Global mutex to protect against size modification of zonelists
5059 * as well as to serialize pageset setup for the new populated zone.
5061 DEFINE_MUTEX(zonelists_mutex
);
5063 /* return values int ....just for stop_machine() */
5064 static int __build_all_zonelists(void *data
)
5068 pg_data_t
*self
= data
;
5071 memset(node_load
, 0, sizeof(node_load
));
5074 if (self
&& !node_online(self
->node_id
)) {
5075 build_zonelists(self
);
5078 for_each_online_node(nid
) {
5079 pg_data_t
*pgdat
= NODE_DATA(nid
);
5081 build_zonelists(pgdat
);
5085 * Initialize the boot_pagesets that are going to be used
5086 * for bootstrapping processors. The real pagesets for
5087 * each zone will be allocated later when the per cpu
5088 * allocator is available.
5090 * boot_pagesets are used also for bootstrapping offline
5091 * cpus if the system is already booted because the pagesets
5092 * are needed to initialize allocators on a specific cpu too.
5093 * F.e. the percpu allocator needs the page allocator which
5094 * needs the percpu allocator in order to allocate its pagesets
5095 * (a chicken-egg dilemma).
5097 for_each_possible_cpu(cpu
) {
5098 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
5100 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
5102 * We now know the "local memory node" for each node--
5103 * i.e., the node of the first zone in the generic zonelist.
5104 * Set up numa_mem percpu variable for on-line cpus. During
5105 * boot, only the boot cpu should be on-line; we'll init the
5106 * secondary cpus' numa_mem as they come on-line. During
5107 * node/memory hotplug, we'll fixup all on-line cpus.
5109 if (cpu_online(cpu
))
5110 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
5117 static noinline
void __init
5118 build_all_zonelists_init(void)
5120 __build_all_zonelists(NULL
);
5121 mminit_verify_zonelist();
5122 cpuset_init_current_mems_allowed();
5126 * Called with zonelists_mutex held always
5127 * unless system_state == SYSTEM_BOOTING.
5129 * __ref due to (1) call of __meminit annotated setup_zone_pageset
5130 * [we're only called with non-NULL zone through __meminit paths] and
5131 * (2) call of __init annotated helper build_all_zonelists_init
5132 * [protected by SYSTEM_BOOTING].
5134 void __ref
build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
)
5136 set_zonelist_order();
5138 if (system_state
== SYSTEM_BOOTING
) {
5139 build_all_zonelists_init();
5141 #ifdef CONFIG_MEMORY_HOTPLUG
5143 setup_zone_pageset(zone
);
5145 /* we have to stop all cpus to guarantee there is no user
5147 stop_machine(__build_all_zonelists
, pgdat
, NULL
);
5148 /* cpuset refresh routine should be here */
5150 vm_total_pages
= nr_free_pagecache_pages();
5152 * Disable grouping by mobility if the number of pages in the
5153 * system is too low to allow the mechanism to work. It would be
5154 * more accurate, but expensive to check per-zone. This check is
5155 * made on memory-hotadd so a system can start with mobility
5156 * disabled and enable it later
5158 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
5159 page_group_by_mobility_disabled
= 1;
5161 page_group_by_mobility_disabled
= 0;
5163 pr_info("Built %i zonelists in %s order, mobility grouping %s. Total pages: %ld\n",
5165 zonelist_order_name
[current_zonelist_order
],
5166 page_group_by_mobility_disabled
? "off" : "on",
5169 pr_info("Policy zone: %s\n", zone_names
[policy_zone
]);
5174 * Initially all pages are reserved - free ones are freed
5175 * up by free_all_bootmem() once the early boot process is
5176 * done. Non-atomic initialization, single-pass.
5178 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
5179 unsigned long start_pfn
, enum memmap_context context
)
5181 struct vmem_altmap
*altmap
= to_vmem_altmap(__pfn_to_phys(start_pfn
));
5182 unsigned long end_pfn
= start_pfn
+ size
;
5183 pg_data_t
*pgdat
= NODE_DATA(nid
);
5185 unsigned long nr_initialised
= 0;
5186 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5187 struct memblock_region
*r
= NULL
, *tmp
;
5190 if (highest_memmap_pfn
< end_pfn
- 1)
5191 highest_memmap_pfn
= end_pfn
- 1;
5194 * Honor reservation requested by the driver for this ZONE_DEVICE
5197 if (altmap
&& start_pfn
== altmap
->base_pfn
)
5198 start_pfn
+= altmap
->reserve
;
5200 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
5202 * There can be holes in boot-time mem_map[]s handed to this
5203 * function. They do not exist on hotplugged memory.
5205 if (context
!= MEMMAP_EARLY
)
5208 if (!early_pfn_valid(pfn
)) {
5209 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5211 * Skip to the pfn preceding the next valid one (or
5212 * end_pfn), such that we hit a valid pfn (or end_pfn)
5213 * on our next iteration of the loop.
5215 pfn
= memblock_next_valid_pfn(pfn
, end_pfn
) - 1;
5219 if (!early_pfn_in_nid(pfn
, nid
))
5221 if (!update_defer_init(pgdat
, pfn
, end_pfn
, &nr_initialised
))
5224 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5226 * Check given memblock attribute by firmware which can affect
5227 * kernel memory layout. If zone==ZONE_MOVABLE but memory is
5228 * mirrored, it's an overlapped memmap init. skip it.
5230 if (mirrored_kernelcore
&& zone
== ZONE_MOVABLE
) {
5231 if (!r
|| pfn
>= memblock_region_memory_end_pfn(r
)) {
5232 for_each_memblock(memory
, tmp
)
5233 if (pfn
< memblock_region_memory_end_pfn(tmp
))
5237 if (pfn
>= memblock_region_memory_base_pfn(r
) &&
5238 memblock_is_mirror(r
)) {
5239 /* already initialized as NORMAL */
5240 pfn
= memblock_region_memory_end_pfn(r
);
5248 * Mark the block movable so that blocks are reserved for
5249 * movable at startup. This will force kernel allocations
5250 * to reserve their blocks rather than leaking throughout
5251 * the address space during boot when many long-lived
5252 * kernel allocations are made.
5254 * bitmap is created for zone's valid pfn range. but memmap
5255 * can be created for invalid pages (for alignment)
5256 * check here not to call set_pageblock_migratetype() against
5259 if (!(pfn
& (pageblock_nr_pages
- 1))) {
5260 struct page
*page
= pfn_to_page(pfn
);
5262 __init_single_page(page
, pfn
, zone
, nid
);
5263 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
5265 __init_single_pfn(pfn
, zone
, nid
);
5270 static void __meminit
zone_init_free_lists(struct zone
*zone
)
5272 unsigned int order
, t
;
5273 for_each_migratetype_order(order
, t
) {
5274 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
5275 zone
->free_area
[order
].nr_free
= 0;
5279 #ifndef __HAVE_ARCH_MEMMAP_INIT
5280 #define memmap_init(size, nid, zone, start_pfn) \
5281 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
5284 static int zone_batchsize(struct zone
*zone
)
5290 * The per-cpu-pages pools are set to around 1000th of the
5291 * size of the zone. But no more than 1/2 of a meg.
5293 * OK, so we don't know how big the cache is. So guess.
5295 batch
= zone
->managed_pages
/ 1024;
5296 if (batch
* PAGE_SIZE
> 512 * 1024)
5297 batch
= (512 * 1024) / PAGE_SIZE
;
5298 batch
/= 4; /* We effectively *= 4 below */
5303 * Clamp the batch to a 2^n - 1 value. Having a power
5304 * of 2 value was found to be more likely to have
5305 * suboptimal cache aliasing properties in some cases.
5307 * For example if 2 tasks are alternately allocating
5308 * batches of pages, one task can end up with a lot
5309 * of pages of one half of the possible page colors
5310 * and the other with pages of the other colors.
5312 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
5317 /* The deferral and batching of frees should be suppressed under NOMMU
5320 * The problem is that NOMMU needs to be able to allocate large chunks
5321 * of contiguous memory as there's no hardware page translation to
5322 * assemble apparent contiguous memory from discontiguous pages.
5324 * Queueing large contiguous runs of pages for batching, however,
5325 * causes the pages to actually be freed in smaller chunks. As there
5326 * can be a significant delay between the individual batches being
5327 * recycled, this leads to the once large chunks of space being
5328 * fragmented and becoming unavailable for high-order allocations.
5335 * pcp->high and pcp->batch values are related and dependent on one another:
5336 * ->batch must never be higher then ->high.
5337 * The following function updates them in a safe manner without read side
5340 * Any new users of pcp->batch and pcp->high should ensure they can cope with
5341 * those fields changing asynchronously (acording the the above rule).
5343 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
5344 * outside of boot time (or some other assurance that no concurrent updaters
5347 static void pageset_update(struct per_cpu_pages
*pcp
, unsigned long high
,
5348 unsigned long batch
)
5350 /* start with a fail safe value for batch */
5354 /* Update high, then batch, in order */
5361 /* a companion to pageset_set_high() */
5362 static void pageset_set_batch(struct per_cpu_pageset
*p
, unsigned long batch
)
5364 pageset_update(&p
->pcp
, 6 * batch
, max(1UL, 1 * batch
));
5367 static void pageset_init(struct per_cpu_pageset
*p
)
5369 struct per_cpu_pages
*pcp
;
5372 memset(p
, 0, sizeof(*p
));
5376 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
5377 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
5380 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
5383 pageset_set_batch(p
, batch
);
5387 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
5388 * to the value high for the pageset p.
5390 static void pageset_set_high(struct per_cpu_pageset
*p
,
5393 unsigned long batch
= max(1UL, high
/ 4);
5394 if ((high
/ 4) > (PAGE_SHIFT
* 8))
5395 batch
= PAGE_SHIFT
* 8;
5397 pageset_update(&p
->pcp
, high
, batch
);
5400 static void pageset_set_high_and_batch(struct zone
*zone
,
5401 struct per_cpu_pageset
*pcp
)
5403 if (percpu_pagelist_fraction
)
5404 pageset_set_high(pcp
,
5405 (zone
->managed_pages
/
5406 percpu_pagelist_fraction
));
5408 pageset_set_batch(pcp
, zone_batchsize(zone
));
5411 static void __meminit
zone_pageset_init(struct zone
*zone
, int cpu
)
5413 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
5416 pageset_set_high_and_batch(zone
, pcp
);
5419 static void __meminit
setup_zone_pageset(struct zone
*zone
)
5422 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
5423 for_each_possible_cpu(cpu
)
5424 zone_pageset_init(zone
, cpu
);
5428 * Allocate per cpu pagesets and initialize them.
5429 * Before this call only boot pagesets were available.
5431 void __init
setup_per_cpu_pageset(void)
5433 struct pglist_data
*pgdat
;
5436 for_each_populated_zone(zone
)
5437 setup_zone_pageset(zone
);
5439 for_each_online_pgdat(pgdat
)
5440 pgdat
->per_cpu_nodestats
=
5441 alloc_percpu(struct per_cpu_nodestat
);
5444 static __meminit
void zone_pcp_init(struct zone
*zone
)
5447 * per cpu subsystem is not up at this point. The following code
5448 * relies on the ability of the linker to provide the
5449 * offset of a (static) per cpu variable into the per cpu area.
5451 zone
->pageset
= &boot_pageset
;
5453 if (populated_zone(zone
))
5454 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
5455 zone
->name
, zone
->present_pages
,
5456 zone_batchsize(zone
));
5459 int __meminit
init_currently_empty_zone(struct zone
*zone
,
5460 unsigned long zone_start_pfn
,
5463 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
5465 pgdat
->nr_zones
= zone_idx(zone
) + 1;
5467 zone
->zone_start_pfn
= zone_start_pfn
;
5469 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
5470 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
5472 (unsigned long)zone_idx(zone
),
5473 zone_start_pfn
, (zone_start_pfn
+ size
));
5475 zone_init_free_lists(zone
);
5476 zone
->initialized
= 1;
5481 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5482 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
5485 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
5487 int __meminit
__early_pfn_to_nid(unsigned long pfn
,
5488 struct mminit_pfnnid_cache
*state
)
5490 unsigned long start_pfn
, end_pfn
;
5493 if (state
->last_start
<= pfn
&& pfn
< state
->last_end
)
5494 return state
->last_nid
;
5496 nid
= memblock_search_pfn_nid(pfn
, &start_pfn
, &end_pfn
);
5498 state
->last_start
= start_pfn
;
5499 state
->last_end
= end_pfn
;
5500 state
->last_nid
= nid
;
5505 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
5508 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
5509 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
5510 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
5512 * If an architecture guarantees that all ranges registered contain no holes
5513 * and may be freed, this this function may be used instead of calling
5514 * memblock_free_early_nid() manually.
5516 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
5518 unsigned long start_pfn
, end_pfn
;
5521 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
5522 start_pfn
= min(start_pfn
, max_low_pfn
);
5523 end_pfn
= min(end_pfn
, max_low_pfn
);
5525 if (start_pfn
< end_pfn
)
5526 memblock_free_early_nid(PFN_PHYS(start_pfn
),
5527 (end_pfn
- start_pfn
) << PAGE_SHIFT
,
5533 * sparse_memory_present_with_active_regions - Call memory_present for each active range
5534 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
5536 * If an architecture guarantees that all ranges registered contain no holes and may
5537 * be freed, this function may be used instead of calling memory_present() manually.
5539 void __init
sparse_memory_present_with_active_regions(int nid
)
5541 unsigned long start_pfn
, end_pfn
;
5544 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
5545 memory_present(this_nid
, start_pfn
, end_pfn
);
5549 * get_pfn_range_for_nid - Return the start and end page frames for a node
5550 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
5551 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
5552 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
5554 * It returns the start and end page frame of a node based on information
5555 * provided by memblock_set_node(). If called for a node
5556 * with no available memory, a warning is printed and the start and end
5559 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
5560 unsigned long *start_pfn
, unsigned long *end_pfn
)
5562 unsigned long this_start_pfn
, this_end_pfn
;
5568 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
5569 *start_pfn
= min(*start_pfn
, this_start_pfn
);
5570 *end_pfn
= max(*end_pfn
, this_end_pfn
);
5573 if (*start_pfn
== -1UL)
5578 * This finds a zone that can be used for ZONE_MOVABLE pages. The
5579 * assumption is made that zones within a node are ordered in monotonic
5580 * increasing memory addresses so that the "highest" populated zone is used
5582 static void __init
find_usable_zone_for_movable(void)
5585 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
5586 if (zone_index
== ZONE_MOVABLE
)
5589 if (arch_zone_highest_possible_pfn
[zone_index
] >
5590 arch_zone_lowest_possible_pfn
[zone_index
])
5594 VM_BUG_ON(zone_index
== -1);
5595 movable_zone
= zone_index
;
5599 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
5600 * because it is sized independent of architecture. Unlike the other zones,
5601 * the starting point for ZONE_MOVABLE is not fixed. It may be different
5602 * in each node depending on the size of each node and how evenly kernelcore
5603 * is distributed. This helper function adjusts the zone ranges
5604 * provided by the architecture for a given node by using the end of the
5605 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5606 * zones within a node are in order of monotonic increases memory addresses
5608 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
5609 unsigned long zone_type
,
5610 unsigned long node_start_pfn
,
5611 unsigned long node_end_pfn
,
5612 unsigned long *zone_start_pfn
,
5613 unsigned long *zone_end_pfn
)
5615 /* Only adjust if ZONE_MOVABLE is on this node */
5616 if (zone_movable_pfn
[nid
]) {
5617 /* Size ZONE_MOVABLE */
5618 if (zone_type
== ZONE_MOVABLE
) {
5619 *zone_start_pfn
= zone_movable_pfn
[nid
];
5620 *zone_end_pfn
= min(node_end_pfn
,
5621 arch_zone_highest_possible_pfn
[movable_zone
]);
5623 /* Adjust for ZONE_MOVABLE starting within this range */
5624 } else if (!mirrored_kernelcore
&&
5625 *zone_start_pfn
< zone_movable_pfn
[nid
] &&
5626 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
5627 *zone_end_pfn
= zone_movable_pfn
[nid
];
5629 /* Check if this whole range is within ZONE_MOVABLE */
5630 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
5631 *zone_start_pfn
= *zone_end_pfn
;
5636 * Return the number of pages a zone spans in a node, including holes
5637 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5639 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5640 unsigned long zone_type
,
5641 unsigned long node_start_pfn
,
5642 unsigned long node_end_pfn
,
5643 unsigned long *zone_start_pfn
,
5644 unsigned long *zone_end_pfn
,
5645 unsigned long *ignored
)
5647 /* When hotadd a new node from cpu_up(), the node should be empty */
5648 if (!node_start_pfn
&& !node_end_pfn
)
5651 /* Get the start and end of the zone */
5652 *zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
5653 *zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
5654 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5655 node_start_pfn
, node_end_pfn
,
5656 zone_start_pfn
, zone_end_pfn
);
5658 /* Check that this node has pages within the zone's required range */
5659 if (*zone_end_pfn
< node_start_pfn
|| *zone_start_pfn
> node_end_pfn
)
5662 /* Move the zone boundaries inside the node if necessary */
5663 *zone_end_pfn
= min(*zone_end_pfn
, node_end_pfn
);
5664 *zone_start_pfn
= max(*zone_start_pfn
, node_start_pfn
);
5666 /* Return the spanned pages */
5667 return *zone_end_pfn
- *zone_start_pfn
;
5671 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
5672 * then all holes in the requested range will be accounted for.
5674 unsigned long __meminit
__absent_pages_in_range(int nid
,
5675 unsigned long range_start_pfn
,
5676 unsigned long range_end_pfn
)
5678 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
5679 unsigned long start_pfn
, end_pfn
;
5682 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5683 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
5684 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
5685 nr_absent
-= end_pfn
- start_pfn
;
5691 * absent_pages_in_range - Return number of page frames in holes within a range
5692 * @start_pfn: The start PFN to start searching for holes
5693 * @end_pfn: The end PFN to stop searching for holes
5695 * It returns the number of pages frames in memory holes within a range.
5697 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
5698 unsigned long end_pfn
)
5700 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
5703 /* Return the number of page frames in holes in a zone on a node */
5704 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5705 unsigned long zone_type
,
5706 unsigned long node_start_pfn
,
5707 unsigned long node_end_pfn
,
5708 unsigned long *ignored
)
5710 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
5711 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
5712 unsigned long zone_start_pfn
, zone_end_pfn
;
5713 unsigned long nr_absent
;
5715 /* When hotadd a new node from cpu_up(), the node should be empty */
5716 if (!node_start_pfn
&& !node_end_pfn
)
5719 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
5720 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
5722 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5723 node_start_pfn
, node_end_pfn
,
5724 &zone_start_pfn
, &zone_end_pfn
);
5725 nr_absent
= __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
5728 * ZONE_MOVABLE handling.
5729 * Treat pages to be ZONE_MOVABLE in ZONE_NORMAL as absent pages
5732 if (mirrored_kernelcore
&& zone_movable_pfn
[nid
]) {
5733 unsigned long start_pfn
, end_pfn
;
5734 struct memblock_region
*r
;
5736 for_each_memblock(memory
, r
) {
5737 start_pfn
= clamp(memblock_region_memory_base_pfn(r
),
5738 zone_start_pfn
, zone_end_pfn
);
5739 end_pfn
= clamp(memblock_region_memory_end_pfn(r
),
5740 zone_start_pfn
, zone_end_pfn
);
5742 if (zone_type
== ZONE_MOVABLE
&&
5743 memblock_is_mirror(r
))
5744 nr_absent
+= end_pfn
- start_pfn
;
5746 if (zone_type
== ZONE_NORMAL
&&
5747 !memblock_is_mirror(r
))
5748 nr_absent
+= end_pfn
- start_pfn
;
5755 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5756 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5757 unsigned long zone_type
,
5758 unsigned long node_start_pfn
,
5759 unsigned long node_end_pfn
,
5760 unsigned long *zone_start_pfn
,
5761 unsigned long *zone_end_pfn
,
5762 unsigned long *zones_size
)
5766 *zone_start_pfn
= node_start_pfn
;
5767 for (zone
= 0; zone
< zone_type
; zone
++)
5768 *zone_start_pfn
+= zones_size
[zone
];
5770 *zone_end_pfn
= *zone_start_pfn
+ zones_size
[zone_type
];
5772 return zones_size
[zone_type
];
5775 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5776 unsigned long zone_type
,
5777 unsigned long node_start_pfn
,
5778 unsigned long node_end_pfn
,
5779 unsigned long *zholes_size
)
5784 return zholes_size
[zone_type
];
5787 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5789 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
5790 unsigned long node_start_pfn
,
5791 unsigned long node_end_pfn
,
5792 unsigned long *zones_size
,
5793 unsigned long *zholes_size
)
5795 unsigned long realtotalpages
= 0, totalpages
= 0;
5798 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5799 struct zone
*zone
= pgdat
->node_zones
+ i
;
5800 unsigned long zone_start_pfn
, zone_end_pfn
;
5801 unsigned long size
, real_size
;
5803 size
= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
5809 real_size
= size
- zone_absent_pages_in_node(pgdat
->node_id
, i
,
5810 node_start_pfn
, node_end_pfn
,
5813 zone
->zone_start_pfn
= zone_start_pfn
;
5815 zone
->zone_start_pfn
= 0;
5816 zone
->spanned_pages
= size
;
5817 zone
->present_pages
= real_size
;
5820 realtotalpages
+= real_size
;
5823 pgdat
->node_spanned_pages
= totalpages
;
5824 pgdat
->node_present_pages
= realtotalpages
;
5825 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
5829 #ifndef CONFIG_SPARSEMEM
5831 * Calculate the size of the zone->blockflags rounded to an unsigned long
5832 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5833 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
5834 * round what is now in bits to nearest long in bits, then return it in
5837 static unsigned long __init
usemap_size(unsigned long zone_start_pfn
, unsigned long zonesize
)
5839 unsigned long usemapsize
;
5841 zonesize
+= zone_start_pfn
& (pageblock_nr_pages
-1);
5842 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
5843 usemapsize
= usemapsize
>> pageblock_order
;
5844 usemapsize
*= NR_PAGEBLOCK_BITS
;
5845 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
5847 return usemapsize
/ 8;
5850 static void __init
setup_usemap(struct pglist_data
*pgdat
,
5852 unsigned long zone_start_pfn
,
5853 unsigned long zonesize
)
5855 unsigned long usemapsize
= usemap_size(zone_start_pfn
, zonesize
);
5856 zone
->pageblock_flags
= NULL
;
5858 zone
->pageblock_flags
=
5859 memblock_virt_alloc_node_nopanic(usemapsize
,
5863 static inline void setup_usemap(struct pglist_data
*pgdat
, struct zone
*zone
,
5864 unsigned long zone_start_pfn
, unsigned long zonesize
) {}
5865 #endif /* CONFIG_SPARSEMEM */
5867 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
5869 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
5870 void __paginginit
set_pageblock_order(void)
5874 /* Check that pageblock_nr_pages has not already been setup */
5875 if (pageblock_order
)
5878 if (HPAGE_SHIFT
> PAGE_SHIFT
)
5879 order
= HUGETLB_PAGE_ORDER
;
5881 order
= MAX_ORDER
- 1;
5884 * Assume the largest contiguous order of interest is a huge page.
5885 * This value may be variable depending on boot parameters on IA64 and
5888 pageblock_order
= order
;
5890 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5893 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
5894 * is unused as pageblock_order is set at compile-time. See
5895 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5898 void __paginginit
set_pageblock_order(void)
5902 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5904 static unsigned long __paginginit
calc_memmap_size(unsigned long spanned_pages
,
5905 unsigned long present_pages
)
5907 unsigned long pages
= spanned_pages
;
5910 * Provide a more accurate estimation if there are holes within
5911 * the zone and SPARSEMEM is in use. If there are holes within the
5912 * zone, each populated memory region may cost us one or two extra
5913 * memmap pages due to alignment because memmap pages for each
5914 * populated regions may not be naturally aligned on page boundary.
5915 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5917 if (spanned_pages
> present_pages
+ (present_pages
>> 4) &&
5918 IS_ENABLED(CONFIG_SPARSEMEM
))
5919 pages
= present_pages
;
5921 return PAGE_ALIGN(pages
* sizeof(struct page
)) >> PAGE_SHIFT
;
5925 * Set up the zone data structures:
5926 * - mark all pages reserved
5927 * - mark all memory queues empty
5928 * - clear the memory bitmaps
5930 * NOTE: pgdat should get zeroed by caller.
5932 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
)
5935 int nid
= pgdat
->node_id
;
5938 pgdat_resize_init(pgdat
);
5939 #ifdef CONFIG_NUMA_BALANCING
5940 spin_lock_init(&pgdat
->numabalancing_migrate_lock
);
5941 pgdat
->numabalancing_migrate_nr_pages
= 0;
5942 pgdat
->numabalancing_migrate_next_window
= jiffies
;
5944 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5945 spin_lock_init(&pgdat
->split_queue_lock
);
5946 INIT_LIST_HEAD(&pgdat
->split_queue
);
5947 pgdat
->split_queue_len
= 0;
5949 init_waitqueue_head(&pgdat
->kswapd_wait
);
5950 init_waitqueue_head(&pgdat
->pfmemalloc_wait
);
5951 #ifdef CONFIG_COMPACTION
5952 init_waitqueue_head(&pgdat
->kcompactd_wait
);
5954 pgdat_page_ext_init(pgdat
);
5955 spin_lock_init(&pgdat
->lru_lock
);
5956 lruvec_init(node_lruvec(pgdat
));
5958 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5959 struct zone
*zone
= pgdat
->node_zones
+ j
;
5960 unsigned long size
, realsize
, freesize
, memmap_pages
;
5961 unsigned long zone_start_pfn
= zone
->zone_start_pfn
;
5963 size
= zone
->spanned_pages
;
5964 realsize
= freesize
= zone
->present_pages
;
5967 * Adjust freesize so that it accounts for how much memory
5968 * is used by this zone for memmap. This affects the watermark
5969 * and per-cpu initialisations
5971 memmap_pages
= calc_memmap_size(size
, realsize
);
5972 if (!is_highmem_idx(j
)) {
5973 if (freesize
>= memmap_pages
) {
5974 freesize
-= memmap_pages
;
5977 " %s zone: %lu pages used for memmap\n",
5978 zone_names
[j
], memmap_pages
);
5980 pr_warn(" %s zone: %lu pages exceeds freesize %lu\n",
5981 zone_names
[j
], memmap_pages
, freesize
);
5984 /* Account for reserved pages */
5985 if (j
== 0 && freesize
> dma_reserve
) {
5986 freesize
-= dma_reserve
;
5987 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
5988 zone_names
[0], dma_reserve
);
5991 if (!is_highmem_idx(j
))
5992 nr_kernel_pages
+= freesize
;
5993 /* Charge for highmem memmap if there are enough kernel pages */
5994 else if (nr_kernel_pages
> memmap_pages
* 2)
5995 nr_kernel_pages
-= memmap_pages
;
5996 nr_all_pages
+= freesize
;
5999 * Set an approximate value for lowmem here, it will be adjusted
6000 * when the bootmem allocator frees pages into the buddy system.
6001 * And all highmem pages will be managed by the buddy system.
6003 zone
->managed_pages
= is_highmem_idx(j
) ? realsize
: freesize
;
6007 zone
->name
= zone_names
[j
];
6008 zone
->zone_pgdat
= pgdat
;
6009 spin_lock_init(&zone
->lock
);
6010 zone_seqlock_init(zone
);
6011 zone_pcp_init(zone
);
6016 set_pageblock_order();
6017 setup_usemap(pgdat
, zone
, zone_start_pfn
, size
);
6018 ret
= init_currently_empty_zone(zone
, zone_start_pfn
, size
);
6020 memmap_init(size
, nid
, j
, zone_start_pfn
);
6024 static void __ref
alloc_node_mem_map(struct pglist_data
*pgdat
)
6026 unsigned long __maybe_unused start
= 0;
6027 unsigned long __maybe_unused offset
= 0;
6029 /* Skip empty nodes */
6030 if (!pgdat
->node_spanned_pages
)
6033 #ifdef CONFIG_FLAT_NODE_MEM_MAP
6034 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
6035 offset
= pgdat
->node_start_pfn
- start
;
6036 /* ia64 gets its own node_mem_map, before this, without bootmem */
6037 if (!pgdat
->node_mem_map
) {
6038 unsigned long size
, end
;
6042 * The zone's endpoints aren't required to be MAX_ORDER
6043 * aligned but the node_mem_map endpoints must be in order
6044 * for the buddy allocator to function correctly.
6046 end
= pgdat_end_pfn(pgdat
);
6047 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
6048 size
= (end
- start
) * sizeof(struct page
);
6049 map
= alloc_remap(pgdat
->node_id
, size
);
6051 map
= memblock_virt_alloc_node_nopanic(size
,
6053 pgdat
->node_mem_map
= map
+ offset
;
6055 #ifndef CONFIG_NEED_MULTIPLE_NODES
6057 * With no DISCONTIG, the global mem_map is just set as node 0's
6059 if (pgdat
== NODE_DATA(0)) {
6060 mem_map
= NODE_DATA(0)->node_mem_map
;
6061 #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
6062 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
6064 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6067 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
6070 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
6071 unsigned long node_start_pfn
, unsigned long *zholes_size
)
6073 pg_data_t
*pgdat
= NODE_DATA(nid
);
6074 unsigned long start_pfn
= 0;
6075 unsigned long end_pfn
= 0;
6077 /* pg_data_t should be reset to zero when it's allocated */
6078 WARN_ON(pgdat
->nr_zones
|| pgdat
->kswapd_classzone_idx
);
6080 reset_deferred_meminit(pgdat
);
6081 pgdat
->node_id
= nid
;
6082 pgdat
->node_start_pfn
= node_start_pfn
;
6083 pgdat
->per_cpu_nodestats
= NULL
;
6084 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6085 get_pfn_range_for_nid(nid
, &start_pfn
, &end_pfn
);
6086 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid
,
6087 (u64
)start_pfn
<< PAGE_SHIFT
,
6088 end_pfn
? ((u64
)end_pfn
<< PAGE_SHIFT
) - 1 : 0);
6090 start_pfn
= node_start_pfn
;
6092 calculate_node_totalpages(pgdat
, start_pfn
, end_pfn
,
6093 zones_size
, zholes_size
);
6095 alloc_node_mem_map(pgdat
);
6096 #ifdef CONFIG_FLAT_NODE_MEM_MAP
6097 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
6098 nid
, (unsigned long)pgdat
,
6099 (unsigned long)pgdat
->node_mem_map
);
6102 free_area_init_core(pgdat
);
6105 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
6107 #if MAX_NUMNODES > 1
6109 * Figure out the number of possible node ids.
6111 void __init
setup_nr_node_ids(void)
6113 unsigned int highest
;
6115 highest
= find_last_bit(node_possible_map
.bits
, MAX_NUMNODES
);
6116 nr_node_ids
= highest
+ 1;
6121 * node_map_pfn_alignment - determine the maximum internode alignment
6123 * This function should be called after node map is populated and sorted.
6124 * It calculates the maximum power of two alignment which can distinguish
6127 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
6128 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
6129 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
6130 * shifted, 1GiB is enough and this function will indicate so.
6132 * This is used to test whether pfn -> nid mapping of the chosen memory
6133 * model has fine enough granularity to avoid incorrect mapping for the
6134 * populated node map.
6136 * Returns the determined alignment in pfn's. 0 if there is no alignment
6137 * requirement (single node).
6139 unsigned long __init
node_map_pfn_alignment(void)
6141 unsigned long accl_mask
= 0, last_end
= 0;
6142 unsigned long start
, end
, mask
;
6146 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
6147 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
6154 * Start with a mask granular enough to pin-point to the
6155 * start pfn and tick off bits one-by-one until it becomes
6156 * too coarse to separate the current node from the last.
6158 mask
= ~((1 << __ffs(start
)) - 1);
6159 while (mask
&& last_end
<= (start
& (mask
<< 1)))
6162 /* accumulate all internode masks */
6166 /* convert mask to number of pages */
6167 return ~accl_mask
+ 1;
6170 /* Find the lowest pfn for a node */
6171 static unsigned long __init
find_min_pfn_for_node(int nid
)
6173 unsigned long min_pfn
= ULONG_MAX
;
6174 unsigned long start_pfn
;
6177 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
6178 min_pfn
= min(min_pfn
, start_pfn
);
6180 if (min_pfn
== ULONG_MAX
) {
6181 pr_warn("Could not find start_pfn for node %d\n", nid
);
6189 * find_min_pfn_with_active_regions - Find the minimum PFN registered
6191 * It returns the minimum PFN based on information provided via
6192 * memblock_set_node().
6194 unsigned long __init
find_min_pfn_with_active_regions(void)
6196 return find_min_pfn_for_node(MAX_NUMNODES
);
6200 * early_calculate_totalpages()
6201 * Sum pages in active regions for movable zone.
6202 * Populate N_MEMORY for calculating usable_nodes.
6204 static unsigned long __init
early_calculate_totalpages(void)
6206 unsigned long totalpages
= 0;
6207 unsigned long start_pfn
, end_pfn
;
6210 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
6211 unsigned long pages
= end_pfn
- start_pfn
;
6213 totalpages
+= pages
;
6215 node_set_state(nid
, N_MEMORY
);
6221 * Find the PFN the Movable zone begins in each node. Kernel memory
6222 * is spread evenly between nodes as long as the nodes have enough
6223 * memory. When they don't, some nodes will have more kernelcore than
6226 static void __init
find_zone_movable_pfns_for_nodes(void)
6229 unsigned long usable_startpfn
;
6230 unsigned long kernelcore_node
, kernelcore_remaining
;
6231 /* save the state before borrow the nodemask */
6232 nodemask_t saved_node_state
= node_states
[N_MEMORY
];
6233 unsigned long totalpages
= early_calculate_totalpages();
6234 int usable_nodes
= nodes_weight(node_states
[N_MEMORY
]);
6235 struct memblock_region
*r
;
6237 /* Need to find movable_zone earlier when movable_node is specified. */
6238 find_usable_zone_for_movable();
6241 * If movable_node is specified, ignore kernelcore and movablecore
6244 if (movable_node_is_enabled()) {
6245 for_each_memblock(memory
, r
) {
6246 if (!memblock_is_hotpluggable(r
))
6251 usable_startpfn
= PFN_DOWN(r
->base
);
6252 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
6253 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
6261 * If kernelcore=mirror is specified, ignore movablecore option
6263 if (mirrored_kernelcore
) {
6264 bool mem_below_4gb_not_mirrored
= false;
6266 for_each_memblock(memory
, r
) {
6267 if (memblock_is_mirror(r
))
6272 usable_startpfn
= memblock_region_memory_base_pfn(r
);
6274 if (usable_startpfn
< 0x100000) {
6275 mem_below_4gb_not_mirrored
= true;
6279 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
6280 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
6284 if (mem_below_4gb_not_mirrored
)
6285 pr_warn("This configuration results in unmirrored kernel memory.");
6291 * If movablecore=nn[KMG] was specified, calculate what size of
6292 * kernelcore that corresponds so that memory usable for
6293 * any allocation type is evenly spread. If both kernelcore
6294 * and movablecore are specified, then the value of kernelcore
6295 * will be used for required_kernelcore if it's greater than
6296 * what movablecore would have allowed.
6298 if (required_movablecore
) {
6299 unsigned long corepages
;
6302 * Round-up so that ZONE_MOVABLE is at least as large as what
6303 * was requested by the user
6305 required_movablecore
=
6306 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
6307 required_movablecore
= min(totalpages
, required_movablecore
);
6308 corepages
= totalpages
- required_movablecore
;
6310 required_kernelcore
= max(required_kernelcore
, corepages
);
6314 * If kernelcore was not specified or kernelcore size is larger
6315 * than totalpages, there is no ZONE_MOVABLE.
6317 if (!required_kernelcore
|| required_kernelcore
>= totalpages
)
6320 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
6321 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
6324 /* Spread kernelcore memory as evenly as possible throughout nodes */
6325 kernelcore_node
= required_kernelcore
/ usable_nodes
;
6326 for_each_node_state(nid
, N_MEMORY
) {
6327 unsigned long start_pfn
, end_pfn
;
6330 * Recalculate kernelcore_node if the division per node
6331 * now exceeds what is necessary to satisfy the requested
6332 * amount of memory for the kernel
6334 if (required_kernelcore
< kernelcore_node
)
6335 kernelcore_node
= required_kernelcore
/ usable_nodes
;
6338 * As the map is walked, we track how much memory is usable
6339 * by the kernel using kernelcore_remaining. When it is
6340 * 0, the rest of the node is usable by ZONE_MOVABLE
6342 kernelcore_remaining
= kernelcore_node
;
6344 /* Go through each range of PFNs within this node */
6345 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
6346 unsigned long size_pages
;
6348 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
6349 if (start_pfn
>= end_pfn
)
6352 /* Account for what is only usable for kernelcore */
6353 if (start_pfn
< usable_startpfn
) {
6354 unsigned long kernel_pages
;
6355 kernel_pages
= min(end_pfn
, usable_startpfn
)
6358 kernelcore_remaining
-= min(kernel_pages
,
6359 kernelcore_remaining
);
6360 required_kernelcore
-= min(kernel_pages
,
6361 required_kernelcore
);
6363 /* Continue if range is now fully accounted */
6364 if (end_pfn
<= usable_startpfn
) {
6367 * Push zone_movable_pfn to the end so
6368 * that if we have to rebalance
6369 * kernelcore across nodes, we will
6370 * not double account here
6372 zone_movable_pfn
[nid
] = end_pfn
;
6375 start_pfn
= usable_startpfn
;
6379 * The usable PFN range for ZONE_MOVABLE is from
6380 * start_pfn->end_pfn. Calculate size_pages as the
6381 * number of pages used as kernelcore
6383 size_pages
= end_pfn
- start_pfn
;
6384 if (size_pages
> kernelcore_remaining
)
6385 size_pages
= kernelcore_remaining
;
6386 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
6389 * Some kernelcore has been met, update counts and
6390 * break if the kernelcore for this node has been
6393 required_kernelcore
-= min(required_kernelcore
,
6395 kernelcore_remaining
-= size_pages
;
6396 if (!kernelcore_remaining
)
6402 * If there is still required_kernelcore, we do another pass with one
6403 * less node in the count. This will push zone_movable_pfn[nid] further
6404 * along on the nodes that still have memory until kernelcore is
6408 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
6412 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
6413 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
6414 zone_movable_pfn
[nid
] =
6415 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
6418 /* restore the node_state */
6419 node_states
[N_MEMORY
] = saved_node_state
;
6422 /* Any regular or high memory on that node ? */
6423 static void check_for_memory(pg_data_t
*pgdat
, int nid
)
6425 enum zone_type zone_type
;
6427 if (N_MEMORY
== N_NORMAL_MEMORY
)
6430 for (zone_type
= 0; zone_type
<= ZONE_MOVABLE
- 1; zone_type
++) {
6431 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
6432 if (populated_zone(zone
)) {
6433 node_set_state(nid
, N_HIGH_MEMORY
);
6434 if (N_NORMAL_MEMORY
!= N_HIGH_MEMORY
&&
6435 zone_type
<= ZONE_NORMAL
)
6436 node_set_state(nid
, N_NORMAL_MEMORY
);
6443 * free_area_init_nodes - Initialise all pg_data_t and zone data
6444 * @max_zone_pfn: an array of max PFNs for each zone
6446 * This will call free_area_init_node() for each active node in the system.
6447 * Using the page ranges provided by memblock_set_node(), the size of each
6448 * zone in each node and their holes is calculated. If the maximum PFN
6449 * between two adjacent zones match, it is assumed that the zone is empty.
6450 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
6451 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
6452 * starts where the previous one ended. For example, ZONE_DMA32 starts
6453 * at arch_max_dma_pfn.
6455 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
6457 unsigned long start_pfn
, end_pfn
;
6460 /* Record where the zone boundaries are */
6461 memset(arch_zone_lowest_possible_pfn
, 0,
6462 sizeof(arch_zone_lowest_possible_pfn
));
6463 memset(arch_zone_highest_possible_pfn
, 0,
6464 sizeof(arch_zone_highest_possible_pfn
));
6466 start_pfn
= find_min_pfn_with_active_regions();
6468 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
6469 if (i
== ZONE_MOVABLE
)
6472 end_pfn
= max(max_zone_pfn
[i
], start_pfn
);
6473 arch_zone_lowest_possible_pfn
[i
] = start_pfn
;
6474 arch_zone_highest_possible_pfn
[i
] = end_pfn
;
6476 start_pfn
= end_pfn
;
6479 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
6480 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
6481 find_zone_movable_pfns_for_nodes();
6483 /* Print out the zone ranges */
6484 pr_info("Zone ranges:\n");
6485 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
6486 if (i
== ZONE_MOVABLE
)
6488 pr_info(" %-8s ", zone_names
[i
]);
6489 if (arch_zone_lowest_possible_pfn
[i
] ==
6490 arch_zone_highest_possible_pfn
[i
])
6493 pr_cont("[mem %#018Lx-%#018Lx]\n",
6494 (u64
)arch_zone_lowest_possible_pfn
[i
]
6496 ((u64
)arch_zone_highest_possible_pfn
[i
]
6497 << PAGE_SHIFT
) - 1);
6500 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
6501 pr_info("Movable zone start for each node\n");
6502 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
6503 if (zone_movable_pfn
[i
])
6504 pr_info(" Node %d: %#018Lx\n", i
,
6505 (u64
)zone_movable_pfn
[i
] << PAGE_SHIFT
);
6508 /* Print out the early node map */
6509 pr_info("Early memory node ranges\n");
6510 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
6511 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid
,
6512 (u64
)start_pfn
<< PAGE_SHIFT
,
6513 ((u64
)end_pfn
<< PAGE_SHIFT
) - 1);
6515 /* Initialise every node */
6516 mminit_verify_pageflags_layout();
6517 setup_nr_node_ids();
6518 for_each_online_node(nid
) {
6519 pg_data_t
*pgdat
= NODE_DATA(nid
);
6520 free_area_init_node(nid
, NULL
,
6521 find_min_pfn_for_node(nid
), NULL
);
6523 /* Any memory on that node */
6524 if (pgdat
->node_present_pages
)
6525 node_set_state(nid
, N_MEMORY
);
6526 check_for_memory(pgdat
, nid
);
6530 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
6532 unsigned long long coremem
;
6536 coremem
= memparse(p
, &p
);
6537 *core
= coremem
>> PAGE_SHIFT
;
6539 /* Paranoid check that UL is enough for the coremem value */
6540 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
6546 * kernelcore=size sets the amount of memory for use for allocations that
6547 * cannot be reclaimed or migrated.
6549 static int __init
cmdline_parse_kernelcore(char *p
)
6551 /* parse kernelcore=mirror */
6552 if (parse_option_str(p
, "mirror")) {
6553 mirrored_kernelcore
= true;
6557 return cmdline_parse_core(p
, &required_kernelcore
);
6561 * movablecore=size sets the amount of memory for use for allocations that
6562 * can be reclaimed or migrated.
6564 static int __init
cmdline_parse_movablecore(char *p
)
6566 return cmdline_parse_core(p
, &required_movablecore
);
6569 early_param("kernelcore", cmdline_parse_kernelcore
);
6570 early_param("movablecore", cmdline_parse_movablecore
);
6572 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
6574 void adjust_managed_page_count(struct page
*page
, long count
)
6576 spin_lock(&managed_page_count_lock
);
6577 page_zone(page
)->managed_pages
+= count
;
6578 totalram_pages
+= count
;
6579 #ifdef CONFIG_HIGHMEM
6580 if (PageHighMem(page
))
6581 totalhigh_pages
+= count
;
6583 spin_unlock(&managed_page_count_lock
);
6585 EXPORT_SYMBOL(adjust_managed_page_count
);
6587 unsigned long free_reserved_area(void *start
, void *end
, int poison
, char *s
)
6590 unsigned long pages
= 0;
6592 start
= (void *)PAGE_ALIGN((unsigned long)start
);
6593 end
= (void *)((unsigned long)end
& PAGE_MASK
);
6594 for (pos
= start
; pos
< end
; pos
+= PAGE_SIZE
, pages
++) {
6595 if ((unsigned int)poison
<= 0xFF)
6596 memset(pos
, poison
, PAGE_SIZE
);
6597 free_reserved_page(virt_to_page(pos
));
6601 pr_info("Freeing %s memory: %ldK\n",
6602 s
, pages
<< (PAGE_SHIFT
- 10));
6606 EXPORT_SYMBOL(free_reserved_area
);
6608 #ifdef CONFIG_HIGHMEM
6609 void free_highmem_page(struct page
*page
)
6611 __free_reserved_page(page
);
6613 page_zone(page
)->managed_pages
++;
6619 void __init
mem_init_print_info(const char *str
)
6621 unsigned long physpages
, codesize
, datasize
, rosize
, bss_size
;
6622 unsigned long init_code_size
, init_data_size
;
6624 physpages
= get_num_physpages();
6625 codesize
= _etext
- _stext
;
6626 datasize
= _edata
- _sdata
;
6627 rosize
= __end_rodata
- __start_rodata
;
6628 bss_size
= __bss_stop
- __bss_start
;
6629 init_data_size
= __init_end
- __init_begin
;
6630 init_code_size
= _einittext
- _sinittext
;
6633 * Detect special cases and adjust section sizes accordingly:
6634 * 1) .init.* may be embedded into .data sections
6635 * 2) .init.text.* may be out of [__init_begin, __init_end],
6636 * please refer to arch/tile/kernel/vmlinux.lds.S.
6637 * 3) .rodata.* may be embedded into .text or .data sections.
6639 #define adj_init_size(start, end, size, pos, adj) \
6641 if (start <= pos && pos < end && size > adj) \
6645 adj_init_size(__init_begin
, __init_end
, init_data_size
,
6646 _sinittext
, init_code_size
);
6647 adj_init_size(_stext
, _etext
, codesize
, _sinittext
, init_code_size
);
6648 adj_init_size(_sdata
, _edata
, datasize
, __init_begin
, init_data_size
);
6649 adj_init_size(_stext
, _etext
, codesize
, __start_rodata
, rosize
);
6650 adj_init_size(_sdata
, _edata
, datasize
, __start_rodata
, rosize
);
6652 #undef adj_init_size
6654 pr_info("Memory: %luK/%luK available (%luK kernel code, %luK rwdata, %luK rodata, %luK init, %luK bss, %luK reserved, %luK cma-reserved"
6655 #ifdef CONFIG_HIGHMEM
6659 nr_free_pages() << (PAGE_SHIFT
- 10),
6660 physpages
<< (PAGE_SHIFT
- 10),
6661 codesize
>> 10, datasize
>> 10, rosize
>> 10,
6662 (init_data_size
+ init_code_size
) >> 10, bss_size
>> 10,
6663 (physpages
- totalram_pages
- totalcma_pages
) << (PAGE_SHIFT
- 10),
6664 totalcma_pages
<< (PAGE_SHIFT
- 10),
6665 #ifdef CONFIG_HIGHMEM
6666 totalhigh_pages
<< (PAGE_SHIFT
- 10),
6668 str
? ", " : "", str
? str
: "");
6672 * set_dma_reserve - set the specified number of pages reserved in the first zone
6673 * @new_dma_reserve: The number of pages to mark reserved
6675 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
6676 * In the DMA zone, a significant percentage may be consumed by kernel image
6677 * and other unfreeable allocations which can skew the watermarks badly. This
6678 * function may optionally be used to account for unfreeable pages in the
6679 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
6680 * smaller per-cpu batchsize.
6682 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
6684 dma_reserve
= new_dma_reserve
;
6687 void __init
free_area_init(unsigned long *zones_size
)
6689 free_area_init_node(0, zones_size
,
6690 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
6693 static int page_alloc_cpu_dead(unsigned int cpu
)
6696 lru_add_drain_cpu(cpu
);
6700 * Spill the event counters of the dead processor
6701 * into the current processors event counters.
6702 * This artificially elevates the count of the current
6705 vm_events_fold_cpu(cpu
);
6708 * Zero the differential counters of the dead processor
6709 * so that the vm statistics are consistent.
6711 * This is only okay since the processor is dead and cannot
6712 * race with what we are doing.
6714 cpu_vm_stats_fold(cpu
);
6718 void __init
page_alloc_init(void)
6722 ret
= cpuhp_setup_state_nocalls(CPUHP_PAGE_ALLOC_DEAD
,
6723 "mm/page_alloc:dead", NULL
,
6724 page_alloc_cpu_dead
);
6729 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
6730 * or min_free_kbytes changes.
6732 static void calculate_totalreserve_pages(void)
6734 struct pglist_data
*pgdat
;
6735 unsigned long reserve_pages
= 0;
6736 enum zone_type i
, j
;
6738 for_each_online_pgdat(pgdat
) {
6740 pgdat
->totalreserve_pages
= 0;
6742 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
6743 struct zone
*zone
= pgdat
->node_zones
+ i
;
6746 /* Find valid and maximum lowmem_reserve in the zone */
6747 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
6748 if (zone
->lowmem_reserve
[j
] > max
)
6749 max
= zone
->lowmem_reserve
[j
];
6752 /* we treat the high watermark as reserved pages. */
6753 max
+= high_wmark_pages(zone
);
6755 if (max
> zone
->managed_pages
)
6756 max
= zone
->managed_pages
;
6758 pgdat
->totalreserve_pages
+= max
;
6760 reserve_pages
+= max
;
6763 totalreserve_pages
= reserve_pages
;
6767 * setup_per_zone_lowmem_reserve - called whenever
6768 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
6769 * has a correct pages reserved value, so an adequate number of
6770 * pages are left in the zone after a successful __alloc_pages().
6772 static void setup_per_zone_lowmem_reserve(void)
6774 struct pglist_data
*pgdat
;
6775 enum zone_type j
, idx
;
6777 for_each_online_pgdat(pgdat
) {
6778 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
6779 struct zone
*zone
= pgdat
->node_zones
+ j
;
6780 unsigned long managed_pages
= zone
->managed_pages
;
6782 zone
->lowmem_reserve
[j
] = 0;
6786 struct zone
*lower_zone
;
6790 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
6791 sysctl_lowmem_reserve_ratio
[idx
] = 1;
6793 lower_zone
= pgdat
->node_zones
+ idx
;
6794 lower_zone
->lowmem_reserve
[j
] = managed_pages
/
6795 sysctl_lowmem_reserve_ratio
[idx
];
6796 managed_pages
+= lower_zone
->managed_pages
;
6801 /* update totalreserve_pages */
6802 calculate_totalreserve_pages();
6805 static void __setup_per_zone_wmarks(void)
6807 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
6808 unsigned long lowmem_pages
= 0;
6810 unsigned long flags
;
6812 /* Calculate total number of !ZONE_HIGHMEM pages */
6813 for_each_zone(zone
) {
6814 if (!is_highmem(zone
))
6815 lowmem_pages
+= zone
->managed_pages
;
6818 for_each_zone(zone
) {
6821 spin_lock_irqsave(&zone
->lock
, flags
);
6822 tmp
= (u64
)pages_min
* zone
->managed_pages
;
6823 do_div(tmp
, lowmem_pages
);
6824 if (is_highmem(zone
)) {
6826 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6827 * need highmem pages, so cap pages_min to a small
6830 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
6831 * deltas control asynch page reclaim, and so should
6832 * not be capped for highmem.
6834 unsigned long min_pages
;
6836 min_pages
= zone
->managed_pages
/ 1024;
6837 min_pages
= clamp(min_pages
, SWAP_CLUSTER_MAX
, 128UL);
6838 zone
->watermark
[WMARK_MIN
] = min_pages
;
6841 * If it's a lowmem zone, reserve a number of pages
6842 * proportionate to the zone's size.
6844 zone
->watermark
[WMARK_MIN
] = tmp
;
6848 * Set the kswapd watermarks distance according to the
6849 * scale factor in proportion to available memory, but
6850 * ensure a minimum size on small systems.
6852 tmp
= max_t(u64
, tmp
>> 2,
6853 mult_frac(zone
->managed_pages
,
6854 watermark_scale_factor
, 10000));
6856 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + tmp
;
6857 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + tmp
* 2;
6859 spin_unlock_irqrestore(&zone
->lock
, flags
);
6862 /* update totalreserve_pages */
6863 calculate_totalreserve_pages();
6867 * setup_per_zone_wmarks - called when min_free_kbytes changes
6868 * or when memory is hot-{added|removed}
6870 * Ensures that the watermark[min,low,high] values for each zone are set
6871 * correctly with respect to min_free_kbytes.
6873 void setup_per_zone_wmarks(void)
6875 mutex_lock(&zonelists_mutex
);
6876 __setup_per_zone_wmarks();
6877 mutex_unlock(&zonelists_mutex
);
6881 * Initialise min_free_kbytes.
6883 * For small machines we want it small (128k min). For large machines
6884 * we want it large (64MB max). But it is not linear, because network
6885 * bandwidth does not increase linearly with machine size. We use
6887 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
6888 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6904 int __meminit
init_per_zone_wmark_min(void)
6906 unsigned long lowmem_kbytes
;
6907 int new_min_free_kbytes
;
6909 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
6910 new_min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
6912 if (new_min_free_kbytes
> user_min_free_kbytes
) {
6913 min_free_kbytes
= new_min_free_kbytes
;
6914 if (min_free_kbytes
< 128)
6915 min_free_kbytes
= 128;
6916 if (min_free_kbytes
> 65536)
6917 min_free_kbytes
= 65536;
6919 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6920 new_min_free_kbytes
, user_min_free_kbytes
);
6922 setup_per_zone_wmarks();
6923 refresh_zone_stat_thresholds();
6924 setup_per_zone_lowmem_reserve();
6927 setup_min_unmapped_ratio();
6928 setup_min_slab_ratio();
6933 core_initcall(init_per_zone_wmark_min
)
6936 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
6937 * that we can call two helper functions whenever min_free_kbytes
6940 int min_free_kbytes_sysctl_handler(struct ctl_table
*table
, int write
,
6941 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6945 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6950 user_min_free_kbytes
= min_free_kbytes
;
6951 setup_per_zone_wmarks();
6956 int watermark_scale_factor_sysctl_handler(struct ctl_table
*table
, int write
,
6957 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6961 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6966 setup_per_zone_wmarks();
6972 static void setup_min_unmapped_ratio(void)
6977 for_each_online_pgdat(pgdat
)
6978 pgdat
->min_unmapped_pages
= 0;
6981 zone
->zone_pgdat
->min_unmapped_pages
+= (zone
->managed_pages
*
6982 sysctl_min_unmapped_ratio
) / 100;
6986 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6987 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6991 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6995 setup_min_unmapped_ratio();
7000 static void setup_min_slab_ratio(void)
7005 for_each_online_pgdat(pgdat
)
7006 pgdat
->min_slab_pages
= 0;
7009 zone
->zone_pgdat
->min_slab_pages
+= (zone
->managed_pages
*
7010 sysctl_min_slab_ratio
) / 100;
7013 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
7014 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
7018 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
7022 setup_min_slab_ratio();
7029 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
7030 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
7031 * whenever sysctl_lowmem_reserve_ratio changes.
7033 * The reserve ratio obviously has absolutely no relation with the
7034 * minimum watermarks. The lowmem reserve ratio can only make sense
7035 * if in function of the boot time zone sizes.
7037 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
7038 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
7040 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
7041 setup_per_zone_lowmem_reserve();
7046 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
7047 * cpu. It is the fraction of total pages in each zone that a hot per cpu
7048 * pagelist can have before it gets flushed back to buddy allocator.
7050 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table
*table
, int write
,
7051 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
7054 int old_percpu_pagelist_fraction
;
7057 mutex_lock(&pcp_batch_high_lock
);
7058 old_percpu_pagelist_fraction
= percpu_pagelist_fraction
;
7060 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
7061 if (!write
|| ret
< 0)
7064 /* Sanity checking to avoid pcp imbalance */
7065 if (percpu_pagelist_fraction
&&
7066 percpu_pagelist_fraction
< MIN_PERCPU_PAGELIST_FRACTION
) {
7067 percpu_pagelist_fraction
= old_percpu_pagelist_fraction
;
7073 if (percpu_pagelist_fraction
== old_percpu_pagelist_fraction
)
7076 for_each_populated_zone(zone
) {
7079 for_each_possible_cpu(cpu
)
7080 pageset_set_high_and_batch(zone
,
7081 per_cpu_ptr(zone
->pageset
, cpu
));
7084 mutex_unlock(&pcp_batch_high_lock
);
7089 int hashdist
= HASHDIST_DEFAULT
;
7091 static int __init
set_hashdist(char *str
)
7095 hashdist
= simple_strtoul(str
, &str
, 0);
7098 __setup("hashdist=", set_hashdist
);
7101 #ifndef __HAVE_ARCH_RESERVED_KERNEL_PAGES
7103 * Returns the number of pages that arch has reserved but
7104 * is not known to alloc_large_system_hash().
7106 static unsigned long __init
arch_reserved_kernel_pages(void)
7113 * allocate a large system hash table from bootmem
7114 * - it is assumed that the hash table must contain an exact power-of-2
7115 * quantity of entries
7116 * - limit is the number of hash buckets, not the total allocation size
7118 void *__init
alloc_large_system_hash(const char *tablename
,
7119 unsigned long bucketsize
,
7120 unsigned long numentries
,
7123 unsigned int *_hash_shift
,
7124 unsigned int *_hash_mask
,
7125 unsigned long low_limit
,
7126 unsigned long high_limit
)
7128 unsigned long long max
= high_limit
;
7129 unsigned long log2qty
, size
;
7132 /* allow the kernel cmdline to have a say */
7134 /* round applicable memory size up to nearest megabyte */
7135 numentries
= nr_kernel_pages
;
7136 numentries
-= arch_reserved_kernel_pages();
7138 /* It isn't necessary when PAGE_SIZE >= 1MB */
7139 if (PAGE_SHIFT
< 20)
7140 numentries
= round_up(numentries
, (1<<20)/PAGE_SIZE
);
7142 /* limit to 1 bucket per 2^scale bytes of low memory */
7143 if (scale
> PAGE_SHIFT
)
7144 numentries
>>= (scale
- PAGE_SHIFT
);
7146 numentries
<<= (PAGE_SHIFT
- scale
);
7148 /* Make sure we've got at least a 0-order allocation.. */
7149 if (unlikely(flags
& HASH_SMALL
)) {
7150 /* Makes no sense without HASH_EARLY */
7151 WARN_ON(!(flags
& HASH_EARLY
));
7152 if (!(numentries
>> *_hash_shift
)) {
7153 numentries
= 1UL << *_hash_shift
;
7154 BUG_ON(!numentries
);
7156 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
7157 numentries
= PAGE_SIZE
/ bucketsize
;
7159 numentries
= roundup_pow_of_two(numentries
);
7161 /* limit allocation size to 1/16 total memory by default */
7163 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
7164 do_div(max
, bucketsize
);
7166 max
= min(max
, 0x80000000ULL
);
7168 if (numentries
< low_limit
)
7169 numentries
= low_limit
;
7170 if (numentries
> max
)
7173 log2qty
= ilog2(numentries
);
7176 size
= bucketsize
<< log2qty
;
7177 if (flags
& HASH_EARLY
)
7178 table
= memblock_virt_alloc_nopanic(size
, 0);
7180 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
7183 * If bucketsize is not a power-of-two, we may free
7184 * some pages at the end of hash table which
7185 * alloc_pages_exact() automatically does
7187 if (get_order(size
) < MAX_ORDER
) {
7188 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
7189 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
7192 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
7195 panic("Failed to allocate %s hash table\n", tablename
);
7197 pr_info("%s hash table entries: %ld (order: %d, %lu bytes)\n",
7198 tablename
, 1UL << log2qty
, ilog2(size
) - PAGE_SHIFT
, size
);
7201 *_hash_shift
= log2qty
;
7203 *_hash_mask
= (1 << log2qty
) - 1;
7209 * This function checks whether pageblock includes unmovable pages or not.
7210 * If @count is not zero, it is okay to include less @count unmovable pages
7212 * PageLRU check without isolation or lru_lock could race so that
7213 * MIGRATE_MOVABLE block might include unmovable pages. And __PageMovable
7214 * check without lock_page also may miss some movable non-lru pages at
7215 * race condition. So you can't expect this function should be exact.
7217 bool has_unmovable_pages(struct zone
*zone
, struct page
*page
, int count
,
7218 bool skip_hwpoisoned_pages
)
7220 unsigned long pfn
, iter
, found
;
7224 * For avoiding noise data, lru_add_drain_all() should be called
7225 * If ZONE_MOVABLE, the zone never contains unmovable pages
7227 if (zone_idx(zone
) == ZONE_MOVABLE
)
7229 mt
= get_pageblock_migratetype(page
);
7230 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
7233 pfn
= page_to_pfn(page
);
7234 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
7235 unsigned long check
= pfn
+ iter
;
7237 if (!pfn_valid_within(check
))
7240 page
= pfn_to_page(check
);
7243 * Hugepages are not in LRU lists, but they're movable.
7244 * We need not scan over tail pages bacause we don't
7245 * handle each tail page individually in migration.
7247 if (PageHuge(page
)) {
7248 iter
= round_up(iter
+ 1, 1<<compound_order(page
)) - 1;
7253 * We can't use page_count without pin a page
7254 * because another CPU can free compound page.
7255 * This check already skips compound tails of THP
7256 * because their page->_refcount is zero at all time.
7258 if (!page_ref_count(page
)) {
7259 if (PageBuddy(page
))
7260 iter
+= (1 << page_order(page
)) - 1;
7265 * The HWPoisoned page may be not in buddy system, and
7266 * page_count() is not 0.
7268 if (skip_hwpoisoned_pages
&& PageHWPoison(page
))
7271 if (__PageMovable(page
))
7277 * If there are RECLAIMABLE pages, we need to check
7278 * it. But now, memory offline itself doesn't call
7279 * shrink_node_slabs() and it still to be fixed.
7282 * If the page is not RAM, page_count()should be 0.
7283 * we don't need more check. This is an _used_ not-movable page.
7285 * The problematic thing here is PG_reserved pages. PG_reserved
7286 * is set to both of a memory hole page and a _used_ kernel
7295 bool is_pageblock_removable_nolock(struct page
*page
)
7301 * We have to be careful here because we are iterating over memory
7302 * sections which are not zone aware so we might end up outside of
7303 * the zone but still within the section.
7304 * We have to take care about the node as well. If the node is offline
7305 * its NODE_DATA will be NULL - see page_zone.
7307 if (!node_online(page_to_nid(page
)))
7310 zone
= page_zone(page
);
7311 pfn
= page_to_pfn(page
);
7312 if (!zone_spans_pfn(zone
, pfn
))
7315 return !has_unmovable_pages(zone
, page
, 0, true);
7318 #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
7320 static unsigned long pfn_max_align_down(unsigned long pfn
)
7322 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
7323 pageblock_nr_pages
) - 1);
7326 static unsigned long pfn_max_align_up(unsigned long pfn
)
7328 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
7329 pageblock_nr_pages
));
7332 /* [start, end) must belong to a single zone. */
7333 static int __alloc_contig_migrate_range(struct compact_control
*cc
,
7334 unsigned long start
, unsigned long end
)
7336 /* This function is based on compact_zone() from compaction.c. */
7337 unsigned long nr_reclaimed
;
7338 unsigned long pfn
= start
;
7339 unsigned int tries
= 0;
7344 while (pfn
< end
|| !list_empty(&cc
->migratepages
)) {
7345 if (fatal_signal_pending(current
)) {
7350 if (list_empty(&cc
->migratepages
)) {
7351 cc
->nr_migratepages
= 0;
7352 pfn
= isolate_migratepages_range(cc
, pfn
, end
);
7358 } else if (++tries
== 5) {
7359 ret
= ret
< 0 ? ret
: -EBUSY
;
7363 nr_reclaimed
= reclaim_clean_pages_from_list(cc
->zone
,
7365 cc
->nr_migratepages
-= nr_reclaimed
;
7367 ret
= migrate_pages(&cc
->migratepages
, alloc_migrate_target
,
7368 NULL
, 0, cc
->mode
, MR_CMA
);
7371 putback_movable_pages(&cc
->migratepages
);
7378 * alloc_contig_range() -- tries to allocate given range of pages
7379 * @start: start PFN to allocate
7380 * @end: one-past-the-last PFN to allocate
7381 * @migratetype: migratetype of the underlaying pageblocks (either
7382 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
7383 * in range must have the same migratetype and it must
7384 * be either of the two.
7385 * @gfp_mask: GFP mask to use during compaction
7387 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
7388 * aligned, however it's the caller's responsibility to guarantee that
7389 * we are the only thread that changes migrate type of pageblocks the
7392 * The PFN range must belong to a single zone.
7394 * Returns zero on success or negative error code. On success all
7395 * pages which PFN is in [start, end) are allocated for the caller and
7396 * need to be freed with free_contig_range().
7398 int alloc_contig_range(unsigned long start
, unsigned long end
,
7399 unsigned migratetype
, gfp_t gfp_mask
)
7401 unsigned long outer_start
, outer_end
;
7405 struct compact_control cc
= {
7406 .nr_migratepages
= 0,
7408 .zone
= page_zone(pfn_to_page(start
)),
7409 .mode
= MIGRATE_SYNC
,
7410 .ignore_skip_hint
= true,
7411 .gfp_mask
= memalloc_noio_flags(gfp_mask
),
7413 INIT_LIST_HEAD(&cc
.migratepages
);
7416 * What we do here is we mark all pageblocks in range as
7417 * MIGRATE_ISOLATE. Because pageblock and max order pages may
7418 * have different sizes, and due to the way page allocator
7419 * work, we align the range to biggest of the two pages so
7420 * that page allocator won't try to merge buddies from
7421 * different pageblocks and change MIGRATE_ISOLATE to some
7422 * other migration type.
7424 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
7425 * migrate the pages from an unaligned range (ie. pages that
7426 * we are interested in). This will put all the pages in
7427 * range back to page allocator as MIGRATE_ISOLATE.
7429 * When this is done, we take the pages in range from page
7430 * allocator removing them from the buddy system. This way
7431 * page allocator will never consider using them.
7433 * This lets us mark the pageblocks back as
7434 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
7435 * aligned range but not in the unaligned, original range are
7436 * put back to page allocator so that buddy can use them.
7439 ret
= start_isolate_page_range(pfn_max_align_down(start
),
7440 pfn_max_align_up(end
), migratetype
,
7446 * In case of -EBUSY, we'd like to know which page causes problem.
7447 * So, just fall through. We will check it in test_pages_isolated().
7449 ret
= __alloc_contig_migrate_range(&cc
, start
, end
);
7450 if (ret
&& ret
!= -EBUSY
)
7454 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
7455 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
7456 * more, all pages in [start, end) are free in page allocator.
7457 * What we are going to do is to allocate all pages from
7458 * [start, end) (that is remove them from page allocator).
7460 * The only problem is that pages at the beginning and at the
7461 * end of interesting range may be not aligned with pages that
7462 * page allocator holds, ie. they can be part of higher order
7463 * pages. Because of this, we reserve the bigger range and
7464 * once this is done free the pages we are not interested in.
7466 * We don't have to hold zone->lock here because the pages are
7467 * isolated thus they won't get removed from buddy.
7470 lru_add_drain_all();
7471 drain_all_pages(cc
.zone
);
7474 outer_start
= start
;
7475 while (!PageBuddy(pfn_to_page(outer_start
))) {
7476 if (++order
>= MAX_ORDER
) {
7477 outer_start
= start
;
7480 outer_start
&= ~0UL << order
;
7483 if (outer_start
!= start
) {
7484 order
= page_order(pfn_to_page(outer_start
));
7487 * outer_start page could be small order buddy page and
7488 * it doesn't include start page. Adjust outer_start
7489 * in this case to report failed page properly
7490 * on tracepoint in test_pages_isolated()
7492 if (outer_start
+ (1UL << order
) <= start
)
7493 outer_start
= start
;
7496 /* Make sure the range is really isolated. */
7497 if (test_pages_isolated(outer_start
, end
, false)) {
7498 pr_info("%s: [%lx, %lx) PFNs busy\n",
7499 __func__
, outer_start
, end
);
7504 /* Grab isolated pages from freelists. */
7505 outer_end
= isolate_freepages_range(&cc
, outer_start
, end
);
7511 /* Free head and tail (if any) */
7512 if (start
!= outer_start
)
7513 free_contig_range(outer_start
, start
- outer_start
);
7514 if (end
!= outer_end
)
7515 free_contig_range(end
, outer_end
- end
);
7518 undo_isolate_page_range(pfn_max_align_down(start
),
7519 pfn_max_align_up(end
), migratetype
);
7523 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
7525 unsigned int count
= 0;
7527 for (; nr_pages
--; pfn
++) {
7528 struct page
*page
= pfn_to_page(pfn
);
7530 count
+= page_count(page
) != 1;
7533 WARN(count
!= 0, "%d pages are still in use!\n", count
);
7537 #ifdef CONFIG_MEMORY_HOTPLUG
7539 * The zone indicated has a new number of managed_pages; batch sizes and percpu
7540 * page high values need to be recalulated.
7542 void __meminit
zone_pcp_update(struct zone
*zone
)
7545 mutex_lock(&pcp_batch_high_lock
);
7546 for_each_possible_cpu(cpu
)
7547 pageset_set_high_and_batch(zone
,
7548 per_cpu_ptr(zone
->pageset
, cpu
));
7549 mutex_unlock(&pcp_batch_high_lock
);
7553 void zone_pcp_reset(struct zone
*zone
)
7555 unsigned long flags
;
7557 struct per_cpu_pageset
*pset
;
7559 /* avoid races with drain_pages() */
7560 local_irq_save(flags
);
7561 if (zone
->pageset
!= &boot_pageset
) {
7562 for_each_online_cpu(cpu
) {
7563 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
7564 drain_zonestat(zone
, pset
);
7566 free_percpu(zone
->pageset
);
7567 zone
->pageset
= &boot_pageset
;
7569 local_irq_restore(flags
);
7572 #ifdef CONFIG_MEMORY_HOTREMOVE
7574 * All pages in the range must be in a single zone and isolated
7575 * before calling this.
7578 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
7582 unsigned int order
, i
;
7584 unsigned long flags
;
7585 /* find the first valid pfn */
7586 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
7591 zone
= page_zone(pfn_to_page(pfn
));
7592 spin_lock_irqsave(&zone
->lock
, flags
);
7594 while (pfn
< end_pfn
) {
7595 if (!pfn_valid(pfn
)) {
7599 page
= pfn_to_page(pfn
);
7601 * The HWPoisoned page may be not in buddy system, and
7602 * page_count() is not 0.
7604 if (unlikely(!PageBuddy(page
) && PageHWPoison(page
))) {
7606 SetPageReserved(page
);
7610 BUG_ON(page_count(page
));
7611 BUG_ON(!PageBuddy(page
));
7612 order
= page_order(page
);
7613 #ifdef CONFIG_DEBUG_VM
7614 pr_info("remove from free list %lx %d %lx\n",
7615 pfn
, 1 << order
, end_pfn
);
7617 list_del(&page
->lru
);
7618 rmv_page_order(page
);
7619 zone
->free_area
[order
].nr_free
--;
7620 for (i
= 0; i
< (1 << order
); i
++)
7621 SetPageReserved((page
+i
));
7622 pfn
+= (1 << order
);
7624 spin_unlock_irqrestore(&zone
->lock
, flags
);
7628 bool is_free_buddy_page(struct page
*page
)
7630 struct zone
*zone
= page_zone(page
);
7631 unsigned long pfn
= page_to_pfn(page
);
7632 unsigned long flags
;
7635 spin_lock_irqsave(&zone
->lock
, flags
);
7636 for (order
= 0; order
< MAX_ORDER
; order
++) {
7637 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
7639 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
7642 spin_unlock_irqrestore(&zone
->lock
, flags
);
7644 return order
< MAX_ORDER
;