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/rwsem.h>
22 #include <linux/pagemap.h>
23 #include <linux/jiffies.h>
24 #include <linux/bootmem.h>
25 #include <linux/memblock.h>
26 #include <linux/compiler.h>
27 #include <linux/kernel.h>
28 #include <linux/kmemcheck.h>
29 #include <linux/kasan.h>
30 #include <linux/module.h>
31 #include <linux/suspend.h>
32 #include <linux/pagevec.h>
33 #include <linux/blkdev.h>
34 #include <linux/slab.h>
35 #include <linux/ratelimit.h>
36 #include <linux/oom.h>
37 #include <linux/notifier.h>
38 #include <linux/topology.h>
39 #include <linux/sysctl.h>
40 #include <linux/cpu.h>
41 #include <linux/cpuset.h>
42 #include <linux/memory_hotplug.h>
43 #include <linux/nodemask.h>
44 #include <linux/vmalloc.h>
45 #include <linux/vmstat.h>
46 #include <linux/mempolicy.h>
47 #include <linux/stop_machine.h>
48 #include <linux/sort.h>
49 #include <linux/pfn.h>
50 #include <linux/backing-dev.h>
51 #include <linux/fault-inject.h>
52 #include <linux/page-isolation.h>
53 #include <linux/page_ext.h>
54 #include <linux/debugobjects.h>
55 #include <linux/kmemleak.h>
56 #include <linux/compaction.h>
57 #include <trace/events/kmem.h>
58 #include <linux/prefetch.h>
59 #include <linux/mm_inline.h>
60 #include <linux/migrate.h>
61 #include <linux/page_ext.h>
62 #include <linux/hugetlb.h>
63 #include <linux/sched/rt.h>
64 #include <linux/page_owner.h>
65 #include <linux/kthread.h>
67 #include <asm/sections.h>
68 #include <asm/tlbflush.h>
69 #include <asm/div64.h>
72 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
73 static DEFINE_MUTEX(pcp_batch_high_lock
);
74 #define MIN_PERCPU_PAGELIST_FRACTION (8)
76 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
77 DEFINE_PER_CPU(int, numa_node
);
78 EXPORT_PER_CPU_SYMBOL(numa_node
);
81 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
83 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
84 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
85 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
86 * defined in <linux/topology.h>.
88 DEFINE_PER_CPU(int, _numa_mem_
); /* Kernel "local memory" node */
89 EXPORT_PER_CPU_SYMBOL(_numa_mem_
);
90 int _node_numa_mem_
[MAX_NUMNODES
];
94 * Array of node states.
96 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
97 [N_POSSIBLE
] = NODE_MASK_ALL
,
98 [N_ONLINE
] = { { [0] = 1UL } },
100 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
101 #ifdef CONFIG_HIGHMEM
102 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
104 #ifdef CONFIG_MOVABLE_NODE
105 [N_MEMORY
] = { { [0] = 1UL } },
107 [N_CPU
] = { { [0] = 1UL } },
110 EXPORT_SYMBOL(node_states
);
112 /* Protect totalram_pages and zone->managed_pages */
113 static DEFINE_SPINLOCK(managed_page_count_lock
);
115 unsigned long totalram_pages __read_mostly
;
116 unsigned long totalreserve_pages __read_mostly
;
117 unsigned long totalcma_pages __read_mostly
;
119 * When calculating the number of globally allowed dirty pages, there
120 * is a certain number of per-zone reserves that should not be
121 * considered dirtyable memory. This is the sum of those reserves
122 * over all existing zones that contribute dirtyable memory.
124 unsigned long dirty_balance_reserve __read_mostly
;
126 int percpu_pagelist_fraction
;
127 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
129 #ifdef CONFIG_PM_SLEEP
131 * The following functions are used by the suspend/hibernate code to temporarily
132 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
133 * while devices are suspended. To avoid races with the suspend/hibernate code,
134 * they should always be called with pm_mutex held (gfp_allowed_mask also should
135 * only be modified with pm_mutex held, unless the suspend/hibernate code is
136 * guaranteed not to run in parallel with that modification).
139 static gfp_t saved_gfp_mask
;
141 void pm_restore_gfp_mask(void)
143 WARN_ON(!mutex_is_locked(&pm_mutex
));
144 if (saved_gfp_mask
) {
145 gfp_allowed_mask
= saved_gfp_mask
;
150 void pm_restrict_gfp_mask(void)
152 WARN_ON(!mutex_is_locked(&pm_mutex
));
153 WARN_ON(saved_gfp_mask
);
154 saved_gfp_mask
= gfp_allowed_mask
;
155 gfp_allowed_mask
&= ~GFP_IOFS
;
158 bool pm_suspended_storage(void)
160 if ((gfp_allowed_mask
& GFP_IOFS
) == GFP_IOFS
)
164 #endif /* CONFIG_PM_SLEEP */
166 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
167 int pageblock_order __read_mostly
;
170 static void __free_pages_ok(struct page
*page
, unsigned int order
);
173 * results with 256, 32 in the lowmem_reserve sysctl:
174 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
175 * 1G machine -> (16M dma, 784M normal, 224M high)
176 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
177 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
178 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
180 * TBD: should special case ZONE_DMA32 machines here - in those we normally
181 * don't need any ZONE_NORMAL reservation
183 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
184 #ifdef CONFIG_ZONE_DMA
187 #ifdef CONFIG_ZONE_DMA32
190 #ifdef CONFIG_HIGHMEM
196 EXPORT_SYMBOL(totalram_pages
);
198 static char * const zone_names
[MAX_NR_ZONES
] = {
199 #ifdef CONFIG_ZONE_DMA
202 #ifdef CONFIG_ZONE_DMA32
206 #ifdef CONFIG_HIGHMEM
212 int min_free_kbytes
= 1024;
213 int user_min_free_kbytes
= -1;
215 static unsigned long __meminitdata nr_kernel_pages
;
216 static unsigned long __meminitdata nr_all_pages
;
217 static unsigned long __meminitdata dma_reserve
;
219 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
220 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
221 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
222 static unsigned long __initdata required_kernelcore
;
223 static unsigned long __initdata required_movablecore
;
224 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
226 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
228 EXPORT_SYMBOL(movable_zone
);
229 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
232 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
233 int nr_online_nodes __read_mostly
= 1;
234 EXPORT_SYMBOL(nr_node_ids
);
235 EXPORT_SYMBOL(nr_online_nodes
);
238 int page_group_by_mobility_disabled __read_mostly
;
240 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
241 static inline void reset_deferred_meminit(pg_data_t
*pgdat
)
243 pgdat
->first_deferred_pfn
= ULONG_MAX
;
246 /* Returns true if the struct page for the pfn is uninitialised */
247 static inline bool __meminit
early_page_uninitialised(unsigned long pfn
)
249 if (pfn
>= NODE_DATA(early_pfn_to_nid(pfn
))->first_deferred_pfn
)
255 static inline bool early_page_nid_uninitialised(unsigned long pfn
, int nid
)
257 if (pfn
>= NODE_DATA(nid
)->first_deferred_pfn
)
264 * Returns false when the remaining initialisation should be deferred until
265 * later in the boot cycle when it can be parallelised.
267 static inline bool update_defer_init(pg_data_t
*pgdat
,
268 unsigned long pfn
, unsigned long zone_end
,
269 unsigned long *nr_initialised
)
271 /* Always populate low zones for address-contrained allocations */
272 if (zone_end
< pgdat_end_pfn(pgdat
))
275 /* Initialise at least 2G of the highest zone */
277 if (*nr_initialised
> (2UL << (30 - PAGE_SHIFT
)) &&
278 (pfn
& (PAGES_PER_SECTION
- 1)) == 0) {
279 pgdat
->first_deferred_pfn
= pfn
;
286 static inline void reset_deferred_meminit(pg_data_t
*pgdat
)
290 static inline bool early_page_uninitialised(unsigned long pfn
)
295 static inline bool early_page_nid_uninitialised(unsigned long pfn
, int nid
)
300 static inline bool update_defer_init(pg_data_t
*pgdat
,
301 unsigned long pfn
, unsigned long zone_end
,
302 unsigned long *nr_initialised
)
309 void set_pageblock_migratetype(struct page
*page
, int migratetype
)
311 if (unlikely(page_group_by_mobility_disabled
&&
312 migratetype
< MIGRATE_PCPTYPES
))
313 migratetype
= MIGRATE_UNMOVABLE
;
315 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
316 PB_migrate
, PB_migrate_end
);
319 #ifdef CONFIG_DEBUG_VM
320 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
324 unsigned long pfn
= page_to_pfn(page
);
325 unsigned long sp
, start_pfn
;
328 seq
= zone_span_seqbegin(zone
);
329 start_pfn
= zone
->zone_start_pfn
;
330 sp
= zone
->spanned_pages
;
331 if (!zone_spans_pfn(zone
, pfn
))
333 } while (zone_span_seqretry(zone
, seq
));
336 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
337 pfn
, zone_to_nid(zone
), zone
->name
,
338 start_pfn
, start_pfn
+ sp
);
343 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
345 if (!pfn_valid_within(page_to_pfn(page
)))
347 if (zone
!= page_zone(page
))
353 * Temporary debugging check for pages not lying within a given zone.
355 static int bad_range(struct zone
*zone
, struct page
*page
)
357 if (page_outside_zone_boundaries(zone
, page
))
359 if (!page_is_consistent(zone
, page
))
365 static inline int bad_range(struct zone
*zone
, struct page
*page
)
371 static void bad_page(struct page
*page
, const char *reason
,
372 unsigned long bad_flags
)
374 static unsigned long resume
;
375 static unsigned long nr_shown
;
376 static unsigned long nr_unshown
;
378 /* Don't complain about poisoned pages */
379 if (PageHWPoison(page
)) {
380 page_mapcount_reset(page
); /* remove PageBuddy */
385 * Allow a burst of 60 reports, then keep quiet for that minute;
386 * or allow a steady drip of one report per second.
388 if (nr_shown
== 60) {
389 if (time_before(jiffies
, resume
)) {
395 "BUG: Bad page state: %lu messages suppressed\n",
402 resume
= jiffies
+ 60 * HZ
;
404 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
405 current
->comm
, page_to_pfn(page
));
406 dump_page_badflags(page
, reason
, bad_flags
);
411 /* Leave bad fields for debug, except PageBuddy could make trouble */
412 page_mapcount_reset(page
); /* remove PageBuddy */
413 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
417 * Higher-order pages are called "compound pages". They are structured thusly:
419 * The first PAGE_SIZE page is called the "head page".
421 * The remaining PAGE_SIZE pages are called "tail pages".
423 * All pages have PG_compound set. All tail pages have their ->first_page
424 * pointing at the head page.
426 * The first tail page's ->lru.next holds the address of the compound page's
427 * put_page() function. Its ->lru.prev holds the order of allocation.
428 * This usage means that zero-order pages may not be compound.
431 static void free_compound_page(struct page
*page
)
433 __free_pages_ok(page
, compound_order(page
));
436 void prep_compound_page(struct page
*page
, unsigned long order
)
439 int nr_pages
= 1 << order
;
441 set_compound_page_dtor(page
, free_compound_page
);
442 set_compound_order(page
, order
);
444 for (i
= 1; i
< nr_pages
; i
++) {
445 struct page
*p
= page
+ i
;
446 set_page_count(p
, 0);
447 p
->first_page
= page
;
448 /* Make sure p->first_page is always valid for PageTail() */
454 #ifdef CONFIG_DEBUG_PAGEALLOC
455 unsigned int _debug_guardpage_minorder
;
456 bool _debug_pagealloc_enabled __read_mostly
;
457 bool _debug_guardpage_enabled __read_mostly
;
459 static int __init
early_debug_pagealloc(char *buf
)
464 if (strcmp(buf
, "on") == 0)
465 _debug_pagealloc_enabled
= true;
469 early_param("debug_pagealloc", early_debug_pagealloc
);
471 static bool need_debug_guardpage(void)
473 /* If we don't use debug_pagealloc, we don't need guard page */
474 if (!debug_pagealloc_enabled())
480 static void init_debug_guardpage(void)
482 if (!debug_pagealloc_enabled())
485 _debug_guardpage_enabled
= true;
488 struct page_ext_operations debug_guardpage_ops
= {
489 .need
= need_debug_guardpage
,
490 .init
= init_debug_guardpage
,
493 static int __init
debug_guardpage_minorder_setup(char *buf
)
497 if (kstrtoul(buf
, 10, &res
) < 0 || res
> MAX_ORDER
/ 2) {
498 printk(KERN_ERR
"Bad debug_guardpage_minorder value\n");
501 _debug_guardpage_minorder
= res
;
502 printk(KERN_INFO
"Setting debug_guardpage_minorder to %lu\n", res
);
505 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup
);
507 static inline void set_page_guard(struct zone
*zone
, struct page
*page
,
508 unsigned int order
, int migratetype
)
510 struct page_ext
*page_ext
;
512 if (!debug_guardpage_enabled())
515 page_ext
= lookup_page_ext(page
);
516 __set_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
518 INIT_LIST_HEAD(&page
->lru
);
519 set_page_private(page
, order
);
520 /* Guard pages are not available for any usage */
521 __mod_zone_freepage_state(zone
, -(1 << order
), migratetype
);
524 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
525 unsigned int order
, int migratetype
)
527 struct page_ext
*page_ext
;
529 if (!debug_guardpage_enabled())
532 page_ext
= lookup_page_ext(page
);
533 __clear_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
535 set_page_private(page
, 0);
536 if (!is_migrate_isolate(migratetype
))
537 __mod_zone_freepage_state(zone
, (1 << order
), migratetype
);
540 struct page_ext_operations debug_guardpage_ops
= { NULL
, };
541 static inline void set_page_guard(struct zone
*zone
, struct page
*page
,
542 unsigned int order
, int migratetype
) {}
543 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
544 unsigned int order
, int migratetype
) {}
547 static inline void set_page_order(struct page
*page
, unsigned int order
)
549 set_page_private(page
, order
);
550 __SetPageBuddy(page
);
553 static inline void rmv_page_order(struct page
*page
)
555 __ClearPageBuddy(page
);
556 set_page_private(page
, 0);
560 * This function checks whether a page is free && is the buddy
561 * we can do coalesce a page and its buddy if
562 * (a) the buddy is not in a hole &&
563 * (b) the buddy is in the buddy system &&
564 * (c) a page and its buddy have the same order &&
565 * (d) a page and its buddy are in the same zone.
567 * For recording whether a page is in the buddy system, we set ->_mapcount
568 * PAGE_BUDDY_MAPCOUNT_VALUE.
569 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
570 * serialized by zone->lock.
572 * For recording page's order, we use page_private(page).
574 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
577 if (!pfn_valid_within(page_to_pfn(buddy
)))
580 if (page_is_guard(buddy
) && page_order(buddy
) == order
) {
581 if (page_zone_id(page
) != page_zone_id(buddy
))
584 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
589 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
591 * zone check is done late to avoid uselessly
592 * calculating zone/node ids for pages that could
595 if (page_zone_id(page
) != page_zone_id(buddy
))
598 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
606 * Freeing function for a buddy system allocator.
608 * The concept of a buddy system is to maintain direct-mapped table
609 * (containing bit values) for memory blocks of various "orders".
610 * The bottom level table contains the map for the smallest allocatable
611 * units of memory (here, pages), and each level above it describes
612 * pairs of units from the levels below, hence, "buddies".
613 * At a high level, all that happens here is marking the table entry
614 * at the bottom level available, and propagating the changes upward
615 * as necessary, plus some accounting needed to play nicely with other
616 * parts of the VM system.
617 * At each level, we keep a list of pages, which are heads of continuous
618 * free pages of length of (1 << order) and marked with _mapcount
619 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
621 * So when we are allocating or freeing one, we can derive the state of the
622 * other. That is, if we allocate a small block, and both were
623 * free, the remainder of the region must be split into blocks.
624 * If a block is freed, and its buddy is also free, then this
625 * triggers coalescing into a block of larger size.
630 static inline void __free_one_page(struct page
*page
,
632 struct zone
*zone
, unsigned int order
,
635 unsigned long page_idx
;
636 unsigned long combined_idx
;
637 unsigned long uninitialized_var(buddy_idx
);
639 int max_order
= MAX_ORDER
;
641 VM_BUG_ON(!zone_is_initialized(zone
));
642 VM_BUG_ON_PAGE(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
, page
);
644 VM_BUG_ON(migratetype
== -1);
645 if (is_migrate_isolate(migratetype
)) {
647 * We restrict max order of merging to prevent merge
648 * between freepages on isolate pageblock and normal
649 * pageblock. Without this, pageblock isolation
650 * could cause incorrect freepage accounting.
652 max_order
= min(MAX_ORDER
, pageblock_order
+ 1);
654 __mod_zone_freepage_state(zone
, 1 << order
, migratetype
);
657 page_idx
= pfn
& ((1 << max_order
) - 1);
659 VM_BUG_ON_PAGE(page_idx
& ((1 << order
) - 1), page
);
660 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
662 while (order
< max_order
- 1) {
663 buddy_idx
= __find_buddy_index(page_idx
, order
);
664 buddy
= page
+ (buddy_idx
- page_idx
);
665 if (!page_is_buddy(page
, buddy
, order
))
668 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
669 * merge with it and move up one order.
671 if (page_is_guard(buddy
)) {
672 clear_page_guard(zone
, buddy
, order
, migratetype
);
674 list_del(&buddy
->lru
);
675 zone
->free_area
[order
].nr_free
--;
676 rmv_page_order(buddy
);
678 combined_idx
= buddy_idx
& page_idx
;
679 page
= page
+ (combined_idx
- page_idx
);
680 page_idx
= combined_idx
;
683 set_page_order(page
, order
);
686 * If this is not the largest possible page, check if the buddy
687 * of the next-highest order is free. If it is, it's possible
688 * that pages are being freed that will coalesce soon. In case,
689 * that is happening, add the free page to the tail of the list
690 * so it's less likely to be used soon and more likely to be merged
691 * as a higher order page
693 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
694 struct page
*higher_page
, *higher_buddy
;
695 combined_idx
= buddy_idx
& page_idx
;
696 higher_page
= page
+ (combined_idx
- page_idx
);
697 buddy_idx
= __find_buddy_index(combined_idx
, order
+ 1);
698 higher_buddy
= higher_page
+ (buddy_idx
- combined_idx
);
699 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
700 list_add_tail(&page
->lru
,
701 &zone
->free_area
[order
].free_list
[migratetype
]);
706 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
708 zone
->free_area
[order
].nr_free
++;
711 static inline int free_pages_check(struct page
*page
)
713 const char *bad_reason
= NULL
;
714 unsigned long bad_flags
= 0;
716 if (unlikely(page_mapcount(page
)))
717 bad_reason
= "nonzero mapcount";
718 if (unlikely(page
->mapping
!= NULL
))
719 bad_reason
= "non-NULL mapping";
720 if (unlikely(atomic_read(&page
->_count
) != 0))
721 bad_reason
= "nonzero _count";
722 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
)) {
723 bad_reason
= "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
724 bad_flags
= PAGE_FLAGS_CHECK_AT_FREE
;
727 if (unlikely(page
->mem_cgroup
))
728 bad_reason
= "page still charged to cgroup";
730 if (unlikely(bad_reason
)) {
731 bad_page(page
, bad_reason
, bad_flags
);
734 page_cpupid_reset_last(page
);
735 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
736 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
741 * Frees a number of pages from the PCP lists
742 * Assumes all pages on list are in same zone, and of same order.
743 * count is the number of pages to free.
745 * If the zone was previously in an "all pages pinned" state then look to
746 * see if this freeing clears that state.
748 * And clear the zone's pages_scanned counter, to hold off the "all pages are
749 * pinned" detection logic.
751 static void free_pcppages_bulk(struct zone
*zone
, int count
,
752 struct per_cpu_pages
*pcp
)
757 unsigned long nr_scanned
;
759 spin_lock(&zone
->lock
);
760 nr_scanned
= zone_page_state(zone
, NR_PAGES_SCANNED
);
762 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, -nr_scanned
);
766 struct list_head
*list
;
769 * Remove pages from lists in a round-robin fashion. A
770 * batch_free count is maintained that is incremented when an
771 * empty list is encountered. This is so more pages are freed
772 * off fuller lists instead of spinning excessively around empty
777 if (++migratetype
== MIGRATE_PCPTYPES
)
779 list
= &pcp
->lists
[migratetype
];
780 } while (list_empty(list
));
782 /* This is the only non-empty list. Free them all. */
783 if (batch_free
== MIGRATE_PCPTYPES
)
784 batch_free
= to_free
;
787 int mt
; /* migratetype of the to-be-freed page */
789 page
= list_entry(list
->prev
, struct page
, lru
);
790 /* must delete as __free_one_page list manipulates */
791 list_del(&page
->lru
);
792 mt
= get_freepage_migratetype(page
);
793 if (unlikely(has_isolate_pageblock(zone
)))
794 mt
= get_pageblock_migratetype(page
);
796 /* MIGRATE_MOVABLE list may include MIGRATE_RESERVEs */
797 __free_one_page(page
, page_to_pfn(page
), zone
, 0, mt
);
798 trace_mm_page_pcpu_drain(page
, 0, mt
);
799 } while (--to_free
&& --batch_free
&& !list_empty(list
));
801 spin_unlock(&zone
->lock
);
804 static void free_one_page(struct zone
*zone
,
805 struct page
*page
, unsigned long pfn
,
809 unsigned long nr_scanned
;
810 spin_lock(&zone
->lock
);
811 nr_scanned
= zone_page_state(zone
, NR_PAGES_SCANNED
);
813 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, -nr_scanned
);
815 if (unlikely(has_isolate_pageblock(zone
) ||
816 is_migrate_isolate(migratetype
))) {
817 migratetype
= get_pfnblock_migratetype(page
, pfn
);
819 __free_one_page(page
, pfn
, zone
, order
, migratetype
);
820 spin_unlock(&zone
->lock
);
823 static int free_tail_pages_check(struct page
*head_page
, struct page
*page
)
825 if (!IS_ENABLED(CONFIG_DEBUG_VM
))
827 if (unlikely(!PageTail(page
))) {
828 bad_page(page
, "PageTail not set", 0);
831 if (unlikely(page
->first_page
!= head_page
)) {
832 bad_page(page
, "first_page not consistent", 0);
838 static void __meminit
__init_single_page(struct page
*page
, unsigned long pfn
,
839 unsigned long zone
, int nid
)
841 set_page_links(page
, zone
, nid
, pfn
);
842 init_page_count(page
);
843 page_mapcount_reset(page
);
844 page_cpupid_reset_last(page
);
846 INIT_LIST_HEAD(&page
->lru
);
847 #ifdef WANT_PAGE_VIRTUAL
848 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
849 if (!is_highmem_idx(zone
))
850 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
854 static void __meminit
__init_single_pfn(unsigned long pfn
, unsigned long zone
,
857 return __init_single_page(pfn_to_page(pfn
), pfn
, zone
, nid
);
860 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
861 static void init_reserved_page(unsigned long pfn
)
866 if (!early_page_uninitialised(pfn
))
869 nid
= early_pfn_to_nid(pfn
);
870 pgdat
= NODE_DATA(nid
);
872 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
873 struct zone
*zone
= &pgdat
->node_zones
[zid
];
875 if (pfn
>= zone
->zone_start_pfn
&& pfn
< zone_end_pfn(zone
))
878 __init_single_pfn(pfn
, zid
, nid
);
881 static inline void init_reserved_page(unsigned long pfn
)
884 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
887 * Initialised pages do not have PageReserved set. This function is
888 * called for each range allocated by the bootmem allocator and
889 * marks the pages PageReserved. The remaining valid pages are later
890 * sent to the buddy page allocator.
892 void __meminit
reserve_bootmem_region(unsigned long start
, unsigned long end
)
894 unsigned long start_pfn
= PFN_DOWN(start
);
895 unsigned long end_pfn
= PFN_UP(end
);
897 for (; start_pfn
< end_pfn
; start_pfn
++) {
898 if (pfn_valid(start_pfn
)) {
899 struct page
*page
= pfn_to_page(start_pfn
);
901 init_reserved_page(start_pfn
);
902 SetPageReserved(page
);
907 static bool free_pages_prepare(struct page
*page
, unsigned int order
)
909 bool compound
= PageCompound(page
);
912 VM_BUG_ON_PAGE(PageTail(page
), page
);
913 VM_BUG_ON_PAGE(compound
&& compound_order(page
) != order
, page
);
915 trace_mm_page_free(page
, order
);
916 kmemcheck_free_shadow(page
, order
);
917 kasan_free_pages(page
, order
);
920 page
->mapping
= NULL
;
921 bad
+= free_pages_check(page
);
922 for (i
= 1; i
< (1 << order
); i
++) {
924 bad
+= free_tail_pages_check(page
, page
+ i
);
925 bad
+= free_pages_check(page
+ i
);
930 reset_page_owner(page
, order
);
932 if (!PageHighMem(page
)) {
933 debug_check_no_locks_freed(page_address(page
),
935 debug_check_no_obj_freed(page_address(page
),
938 arch_free_page(page
, order
);
939 kernel_map_pages(page
, 1 << order
, 0);
944 static void __free_pages_ok(struct page
*page
, unsigned int order
)
948 unsigned long pfn
= page_to_pfn(page
);
950 if (!free_pages_prepare(page
, order
))
953 migratetype
= get_pfnblock_migratetype(page
, pfn
);
954 local_irq_save(flags
);
955 __count_vm_events(PGFREE
, 1 << order
);
956 set_freepage_migratetype(page
, migratetype
);
957 free_one_page(page_zone(page
), page
, pfn
, order
, migratetype
);
958 local_irq_restore(flags
);
961 static void __init
__free_pages_boot_core(struct page
*page
,
962 unsigned long pfn
, unsigned int order
)
964 unsigned int nr_pages
= 1 << order
;
965 struct page
*p
= page
;
969 for (loop
= 0; loop
< (nr_pages
- 1); loop
++, p
++) {
971 __ClearPageReserved(p
);
972 set_page_count(p
, 0);
974 __ClearPageReserved(p
);
975 set_page_count(p
, 0);
977 page_zone(page
)->managed_pages
+= nr_pages
;
978 set_page_refcounted(page
);
979 __free_pages(page
, order
);
982 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
983 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
984 /* Only safe to use early in boot when initialisation is single-threaded */
985 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata
;
987 int __meminit
early_pfn_to_nid(unsigned long pfn
)
991 /* The system will behave unpredictably otherwise */
992 BUG_ON(system_state
!= SYSTEM_BOOTING
);
994 nid
= __early_pfn_to_nid(pfn
, &early_pfnnid_cache
);
1002 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1003 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1004 struct mminit_pfnnid_cache
*state
)
1008 nid
= __early_pfn_to_nid(pfn
, state
);
1009 if (nid
>= 0 && nid
!= node
)
1014 /* Only safe to use early in boot when initialisation is single-threaded */
1015 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1017 return meminit_pfn_in_nid(pfn
, node
, &early_pfnnid_cache
);
1022 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1026 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1027 struct mminit_pfnnid_cache
*state
)
1034 void __init
__free_pages_bootmem(struct page
*page
, unsigned long pfn
,
1037 if (early_page_uninitialised(pfn
))
1039 return __free_pages_boot_core(page
, pfn
, order
);
1042 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1043 static void __init
deferred_free_range(struct page
*page
,
1044 unsigned long pfn
, int nr_pages
)
1051 /* Free a large naturally-aligned chunk if possible */
1052 if (nr_pages
== MAX_ORDER_NR_PAGES
&&
1053 (pfn
& (MAX_ORDER_NR_PAGES
-1)) == 0) {
1054 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
1055 __free_pages_boot_core(page
, pfn
, MAX_ORDER
-1);
1059 for (i
= 0; i
< nr_pages
; i
++, page
++, pfn
++)
1060 __free_pages_boot_core(page
, pfn
, 0);
1063 static __initdata
DECLARE_RWSEM(pgdat_init_rwsem
);
1065 /* Initialise remaining memory on a node */
1066 static int __init
deferred_init_memmap(void *data
)
1068 pg_data_t
*pgdat
= data
;
1069 int nid
= pgdat
->node_id
;
1070 struct mminit_pfnnid_cache nid_init_state
= { };
1071 unsigned long start
= jiffies
;
1072 unsigned long nr_pages
= 0;
1073 unsigned long walk_start
, walk_end
;
1076 unsigned long first_init_pfn
= pgdat
->first_deferred_pfn
;
1077 const struct cpumask
*cpumask
= cpumask_of_node(pgdat
->node_id
);
1079 if (first_init_pfn
== ULONG_MAX
) {
1080 up_read(&pgdat_init_rwsem
);
1084 /* Bind memory initialisation thread to a local node if possible */
1085 if (!cpumask_empty(cpumask
))
1086 set_cpus_allowed_ptr(current
, cpumask
);
1088 /* Sanity check boundaries */
1089 BUG_ON(pgdat
->first_deferred_pfn
< pgdat
->node_start_pfn
);
1090 BUG_ON(pgdat
->first_deferred_pfn
> pgdat_end_pfn(pgdat
));
1091 pgdat
->first_deferred_pfn
= ULONG_MAX
;
1093 /* Only the highest zone is deferred so find it */
1094 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
1095 zone
= pgdat
->node_zones
+ zid
;
1096 if (first_init_pfn
< zone_end_pfn(zone
))
1100 for_each_mem_pfn_range(i
, nid
, &walk_start
, &walk_end
, NULL
) {
1101 unsigned long pfn
, end_pfn
;
1102 struct page
*page
= NULL
;
1103 struct page
*free_base_page
= NULL
;
1104 unsigned long free_base_pfn
= 0;
1107 end_pfn
= min(walk_end
, zone_end_pfn(zone
));
1108 pfn
= first_init_pfn
;
1109 if (pfn
< walk_start
)
1111 if (pfn
< zone
->zone_start_pfn
)
1112 pfn
= zone
->zone_start_pfn
;
1114 for (; pfn
< end_pfn
; pfn
++) {
1115 if (!pfn_valid_within(pfn
))
1119 * Ensure pfn_valid is checked every
1120 * MAX_ORDER_NR_PAGES for memory holes
1122 if ((pfn
& (MAX_ORDER_NR_PAGES
- 1)) == 0) {
1123 if (!pfn_valid(pfn
)) {
1129 if (!meminit_pfn_in_nid(pfn
, nid
, &nid_init_state
)) {
1134 /* Minimise pfn page lookups and scheduler checks */
1135 if (page
&& (pfn
& (MAX_ORDER_NR_PAGES
- 1)) != 0) {
1138 nr_pages
+= nr_to_free
;
1139 deferred_free_range(free_base_page
,
1140 free_base_pfn
, nr_to_free
);
1141 free_base_page
= NULL
;
1142 free_base_pfn
= nr_to_free
= 0;
1144 page
= pfn_to_page(pfn
);
1149 VM_BUG_ON(page_zone(page
) != zone
);
1153 __init_single_page(page
, pfn
, zid
, nid
);
1154 if (!free_base_page
) {
1155 free_base_page
= page
;
1156 free_base_pfn
= pfn
;
1161 /* Where possible, batch up pages for a single free */
1164 /* Free the current block of pages to allocator */
1165 nr_pages
+= nr_to_free
;
1166 deferred_free_range(free_base_page
, free_base_pfn
,
1168 free_base_page
= NULL
;
1169 free_base_pfn
= nr_to_free
= 0;
1172 first_init_pfn
= max(end_pfn
, first_init_pfn
);
1175 /* Sanity check that the next zone really is unpopulated */
1176 WARN_ON(++zid
< MAX_NR_ZONES
&& populated_zone(++zone
));
1178 pr_info("node %d initialised, %lu pages in %ums\n", nid
, nr_pages
,
1179 jiffies_to_msecs(jiffies
- start
));
1180 up_read(&pgdat_init_rwsem
);
1184 void __init
page_alloc_init_late(void)
1188 for_each_node_state(nid
, N_MEMORY
) {
1189 down_read(&pgdat_init_rwsem
);
1190 kthread_run(deferred_init_memmap
, NODE_DATA(nid
), "pgdatinit%d", nid
);
1193 /* Block until all are initialised */
1194 down_write(&pgdat_init_rwsem
);
1195 up_write(&pgdat_init_rwsem
);
1197 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1200 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1201 void __init
init_cma_reserved_pageblock(struct page
*page
)
1203 unsigned i
= pageblock_nr_pages
;
1204 struct page
*p
= page
;
1207 __ClearPageReserved(p
);
1208 set_page_count(p
, 0);
1211 set_pageblock_migratetype(page
, MIGRATE_CMA
);
1213 if (pageblock_order
>= MAX_ORDER
) {
1214 i
= pageblock_nr_pages
;
1217 set_page_refcounted(p
);
1218 __free_pages(p
, MAX_ORDER
- 1);
1219 p
+= MAX_ORDER_NR_PAGES
;
1220 } while (i
-= MAX_ORDER_NR_PAGES
);
1222 set_page_refcounted(page
);
1223 __free_pages(page
, pageblock_order
);
1226 adjust_managed_page_count(page
, pageblock_nr_pages
);
1231 * The order of subdivision here is critical for the IO subsystem.
1232 * Please do not alter this order without good reasons and regression
1233 * testing. Specifically, as large blocks of memory are subdivided,
1234 * the order in which smaller blocks are delivered depends on the order
1235 * they're subdivided in this function. This is the primary factor
1236 * influencing the order in which pages are delivered to the IO
1237 * subsystem according to empirical testing, and this is also justified
1238 * by considering the behavior of a buddy system containing a single
1239 * large block of memory acted on by a series of small allocations.
1240 * This behavior is a critical factor in sglist merging's success.
1244 static inline void expand(struct zone
*zone
, struct page
*page
,
1245 int low
, int high
, struct free_area
*area
,
1248 unsigned long size
= 1 << high
;
1250 while (high
> low
) {
1254 VM_BUG_ON_PAGE(bad_range(zone
, &page
[size
]), &page
[size
]);
1256 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC
) &&
1257 debug_guardpage_enabled() &&
1258 high
< debug_guardpage_minorder()) {
1260 * Mark as guard pages (or page), that will allow to
1261 * merge back to allocator when buddy will be freed.
1262 * Corresponding page table entries will not be touched,
1263 * pages will stay not present in virtual address space
1265 set_page_guard(zone
, &page
[size
], high
, migratetype
);
1268 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
1270 set_page_order(&page
[size
], high
);
1275 * This page is about to be returned from the page allocator
1277 static inline int check_new_page(struct page
*page
)
1279 const char *bad_reason
= NULL
;
1280 unsigned long bad_flags
= 0;
1282 if (unlikely(page_mapcount(page
)))
1283 bad_reason
= "nonzero mapcount";
1284 if (unlikely(page
->mapping
!= NULL
))
1285 bad_reason
= "non-NULL mapping";
1286 if (unlikely(atomic_read(&page
->_count
) != 0))
1287 bad_reason
= "nonzero _count";
1288 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)) {
1289 bad_reason
= "PAGE_FLAGS_CHECK_AT_PREP flag set";
1290 bad_flags
= PAGE_FLAGS_CHECK_AT_PREP
;
1293 if (unlikely(page
->mem_cgroup
))
1294 bad_reason
= "page still charged to cgroup";
1296 if (unlikely(bad_reason
)) {
1297 bad_page(page
, bad_reason
, bad_flags
);
1303 static int prep_new_page(struct page
*page
, unsigned int order
, gfp_t gfp_flags
,
1308 for (i
= 0; i
< (1 << order
); i
++) {
1309 struct page
*p
= page
+ i
;
1310 if (unlikely(check_new_page(p
)))
1314 set_page_private(page
, 0);
1315 set_page_refcounted(page
);
1317 arch_alloc_page(page
, order
);
1318 kernel_map_pages(page
, 1 << order
, 1);
1319 kasan_alloc_pages(page
, order
);
1321 if (gfp_flags
& __GFP_ZERO
)
1322 for (i
= 0; i
< (1 << order
); i
++)
1323 clear_highpage(page
+ i
);
1325 if (order
&& (gfp_flags
& __GFP_COMP
))
1326 prep_compound_page(page
, order
);
1328 set_page_owner(page
, order
, gfp_flags
);
1331 * page->pfmemalloc is set when ALLOC_NO_WATERMARKS was necessary to
1332 * allocate the page. The expectation is that the caller is taking
1333 * steps that will free more memory. The caller should avoid the page
1334 * being used for !PFMEMALLOC purposes.
1336 page
->pfmemalloc
= !!(alloc_flags
& ALLOC_NO_WATERMARKS
);
1342 * Go through the free lists for the given migratetype and remove
1343 * the smallest available page from the freelists
1346 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
1349 unsigned int current_order
;
1350 struct free_area
*area
;
1353 /* Find a page of the appropriate size in the preferred list */
1354 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
1355 area
= &(zone
->free_area
[current_order
]);
1356 if (list_empty(&area
->free_list
[migratetype
]))
1359 page
= list_entry(area
->free_list
[migratetype
].next
,
1361 list_del(&page
->lru
);
1362 rmv_page_order(page
);
1364 expand(zone
, page
, order
, current_order
, area
, migratetype
);
1365 set_freepage_migratetype(page
, migratetype
);
1374 * This array describes the order lists are fallen back to when
1375 * the free lists for the desirable migrate type are depleted
1377 static int fallbacks
[MIGRATE_TYPES
][4] = {
1378 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
1379 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
1380 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
1382 [MIGRATE_CMA
] = { MIGRATE_RESERVE
}, /* Never used */
1384 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
}, /* Never used */
1385 #ifdef CONFIG_MEMORY_ISOLATION
1386 [MIGRATE_ISOLATE
] = { MIGRATE_RESERVE
}, /* Never used */
1391 static struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1394 return __rmqueue_smallest(zone
, order
, MIGRATE_CMA
);
1397 static inline struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1398 unsigned int order
) { return NULL
; }
1402 * Move the free pages in a range to the free lists of the requested type.
1403 * Note that start_page and end_pages are not aligned on a pageblock
1404 * boundary. If alignment is required, use move_freepages_block()
1406 int move_freepages(struct zone
*zone
,
1407 struct page
*start_page
, struct page
*end_page
,
1411 unsigned long order
;
1412 int pages_moved
= 0;
1414 #ifndef CONFIG_HOLES_IN_ZONE
1416 * page_zone is not safe to call in this context when
1417 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1418 * anyway as we check zone boundaries in move_freepages_block().
1419 * Remove at a later date when no bug reports exist related to
1420 * grouping pages by mobility
1422 VM_BUG_ON(page_zone(start_page
) != page_zone(end_page
));
1425 for (page
= start_page
; page
<= end_page
;) {
1426 /* Make sure we are not inadvertently changing nodes */
1427 VM_BUG_ON_PAGE(page_to_nid(page
) != zone_to_nid(zone
), page
);
1429 if (!pfn_valid_within(page_to_pfn(page
))) {
1434 if (!PageBuddy(page
)) {
1439 order
= page_order(page
);
1440 list_move(&page
->lru
,
1441 &zone
->free_area
[order
].free_list
[migratetype
]);
1442 set_freepage_migratetype(page
, migratetype
);
1444 pages_moved
+= 1 << order
;
1450 int move_freepages_block(struct zone
*zone
, struct page
*page
,
1453 unsigned long start_pfn
, end_pfn
;
1454 struct page
*start_page
, *end_page
;
1456 start_pfn
= page_to_pfn(page
);
1457 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
1458 start_page
= pfn_to_page(start_pfn
);
1459 end_page
= start_page
+ pageblock_nr_pages
- 1;
1460 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
1462 /* Do not cross zone boundaries */
1463 if (!zone_spans_pfn(zone
, start_pfn
))
1465 if (!zone_spans_pfn(zone
, end_pfn
))
1468 return move_freepages(zone
, start_page
, end_page
, migratetype
);
1471 static void change_pageblock_range(struct page
*pageblock_page
,
1472 int start_order
, int migratetype
)
1474 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
1476 while (nr_pageblocks
--) {
1477 set_pageblock_migratetype(pageblock_page
, migratetype
);
1478 pageblock_page
+= pageblock_nr_pages
;
1483 * When we are falling back to another migratetype during allocation, try to
1484 * steal extra free pages from the same pageblocks to satisfy further
1485 * allocations, instead of polluting multiple pageblocks.
1487 * If we are stealing a relatively large buddy page, it is likely there will
1488 * be more free pages in the pageblock, so try to steal them all. For
1489 * reclaimable and unmovable allocations, we steal regardless of page size,
1490 * as fragmentation caused by those allocations polluting movable pageblocks
1491 * is worse than movable allocations stealing from unmovable and reclaimable
1494 static bool can_steal_fallback(unsigned int order
, int start_mt
)
1497 * Leaving this order check is intended, although there is
1498 * relaxed order check in next check. The reason is that
1499 * we can actually steal whole pageblock if this condition met,
1500 * but, below check doesn't guarantee it and that is just heuristic
1501 * so could be changed anytime.
1503 if (order
>= pageblock_order
)
1506 if (order
>= pageblock_order
/ 2 ||
1507 start_mt
== MIGRATE_RECLAIMABLE
||
1508 start_mt
== MIGRATE_UNMOVABLE
||
1509 page_group_by_mobility_disabled
)
1516 * This function implements actual steal behaviour. If order is large enough,
1517 * we can steal whole pageblock. If not, we first move freepages in this
1518 * pageblock and check whether half of pages are moved or not. If half of
1519 * pages are moved, we can change migratetype of pageblock and permanently
1520 * use it's pages as requested migratetype in the future.
1522 static void steal_suitable_fallback(struct zone
*zone
, struct page
*page
,
1525 int current_order
= page_order(page
);
1528 /* Take ownership for orders >= pageblock_order */
1529 if (current_order
>= pageblock_order
) {
1530 change_pageblock_range(page
, current_order
, start_type
);
1534 pages
= move_freepages_block(zone
, page
, start_type
);
1536 /* Claim the whole block if over half of it is free */
1537 if (pages
>= (1 << (pageblock_order
-1)) ||
1538 page_group_by_mobility_disabled
)
1539 set_pageblock_migratetype(page
, start_type
);
1543 * Check whether there is a suitable fallback freepage with requested order.
1544 * If only_stealable is true, this function returns fallback_mt only if
1545 * we can steal other freepages all together. This would help to reduce
1546 * fragmentation due to mixed migratetype pages in one pageblock.
1548 int find_suitable_fallback(struct free_area
*area
, unsigned int order
,
1549 int migratetype
, bool only_stealable
, bool *can_steal
)
1554 if (area
->nr_free
== 0)
1559 fallback_mt
= fallbacks
[migratetype
][i
];
1560 if (fallback_mt
== MIGRATE_RESERVE
)
1563 if (list_empty(&area
->free_list
[fallback_mt
]))
1566 if (can_steal_fallback(order
, migratetype
))
1569 if (!only_stealable
)
1579 /* Remove an element from the buddy allocator from the fallback list */
1580 static inline struct page
*
1581 __rmqueue_fallback(struct zone
*zone
, unsigned int order
, int start_migratetype
)
1583 struct free_area
*area
;
1584 unsigned int current_order
;
1589 /* Find the largest possible block of pages in the other list */
1590 for (current_order
= MAX_ORDER
-1;
1591 current_order
>= order
&& current_order
<= MAX_ORDER
-1;
1593 area
= &(zone
->free_area
[current_order
]);
1594 fallback_mt
= find_suitable_fallback(area
, current_order
,
1595 start_migratetype
, false, &can_steal
);
1596 if (fallback_mt
== -1)
1599 page
= list_entry(area
->free_list
[fallback_mt
].next
,
1602 steal_suitable_fallback(zone
, page
, start_migratetype
);
1604 /* Remove the page from the freelists */
1606 list_del(&page
->lru
);
1607 rmv_page_order(page
);
1609 expand(zone
, page
, order
, current_order
, area
,
1612 * The freepage_migratetype may differ from pageblock's
1613 * migratetype depending on the decisions in
1614 * try_to_steal_freepages(). This is OK as long as it
1615 * does not differ for MIGRATE_CMA pageblocks. For CMA
1616 * we need to make sure unallocated pages flushed from
1617 * pcp lists are returned to the correct freelist.
1619 set_freepage_migratetype(page
, start_migratetype
);
1621 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
1622 start_migratetype
, fallback_mt
);
1631 * Do the hard work of removing an element from the buddy allocator.
1632 * Call me with the zone->lock already held.
1634 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
1640 page
= __rmqueue_smallest(zone
, order
, migratetype
);
1642 if (unlikely(!page
) && migratetype
!= MIGRATE_RESERVE
) {
1643 if (migratetype
== MIGRATE_MOVABLE
)
1644 page
= __rmqueue_cma_fallback(zone
, order
);
1647 page
= __rmqueue_fallback(zone
, order
, migratetype
);
1650 * Use MIGRATE_RESERVE rather than fail an allocation. goto
1651 * is used because __rmqueue_smallest is an inline function
1652 * and we want just one call site
1655 migratetype
= MIGRATE_RESERVE
;
1660 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
1665 * Obtain a specified number of elements from the buddy allocator, all under
1666 * a single hold of the lock, for efficiency. Add them to the supplied list.
1667 * Returns the number of new pages which were placed at *list.
1669 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
1670 unsigned long count
, struct list_head
*list
,
1671 int migratetype
, bool cold
)
1675 spin_lock(&zone
->lock
);
1676 for (i
= 0; i
< count
; ++i
) {
1677 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
1678 if (unlikely(page
== NULL
))
1682 * Split buddy pages returned by expand() are received here
1683 * in physical page order. The page is added to the callers and
1684 * list and the list head then moves forward. From the callers
1685 * perspective, the linked list is ordered by page number in
1686 * some conditions. This is useful for IO devices that can
1687 * merge IO requests if the physical pages are ordered
1691 list_add(&page
->lru
, list
);
1693 list_add_tail(&page
->lru
, list
);
1695 if (is_migrate_cma(get_freepage_migratetype(page
)))
1696 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
,
1699 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
1700 spin_unlock(&zone
->lock
);
1706 * Called from the vmstat counter updater to drain pagesets of this
1707 * currently executing processor on remote nodes after they have
1710 * Note that this function must be called with the thread pinned to
1711 * a single processor.
1713 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
1715 unsigned long flags
;
1716 int to_drain
, batch
;
1718 local_irq_save(flags
);
1719 batch
= READ_ONCE(pcp
->batch
);
1720 to_drain
= min(pcp
->count
, batch
);
1722 free_pcppages_bulk(zone
, to_drain
, pcp
);
1723 pcp
->count
-= to_drain
;
1725 local_irq_restore(flags
);
1730 * Drain pcplists of the indicated processor and zone.
1732 * The processor must either be the current processor and the
1733 * thread pinned to the current processor or a processor that
1736 static void drain_pages_zone(unsigned int cpu
, struct zone
*zone
)
1738 unsigned long flags
;
1739 struct per_cpu_pageset
*pset
;
1740 struct per_cpu_pages
*pcp
;
1742 local_irq_save(flags
);
1743 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
1747 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
1750 local_irq_restore(flags
);
1754 * Drain pcplists of all zones on the indicated processor.
1756 * The processor must either be the current processor and the
1757 * thread pinned to the current processor or a processor that
1760 static void drain_pages(unsigned int cpu
)
1764 for_each_populated_zone(zone
) {
1765 drain_pages_zone(cpu
, zone
);
1770 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1772 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
1773 * the single zone's pages.
1775 void drain_local_pages(struct zone
*zone
)
1777 int cpu
= smp_processor_id();
1780 drain_pages_zone(cpu
, zone
);
1786 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1788 * When zone parameter is non-NULL, spill just the single zone's pages.
1790 * Note that this code is protected against sending an IPI to an offline
1791 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1792 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1793 * nothing keeps CPUs from showing up after we populated the cpumask and
1794 * before the call to on_each_cpu_mask().
1796 void drain_all_pages(struct zone
*zone
)
1801 * Allocate in the BSS so we wont require allocation in
1802 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1804 static cpumask_t cpus_with_pcps
;
1807 * We don't care about racing with CPU hotplug event
1808 * as offline notification will cause the notified
1809 * cpu to drain that CPU pcps and on_each_cpu_mask
1810 * disables preemption as part of its processing
1812 for_each_online_cpu(cpu
) {
1813 struct per_cpu_pageset
*pcp
;
1815 bool has_pcps
= false;
1818 pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
1822 for_each_populated_zone(z
) {
1823 pcp
= per_cpu_ptr(z
->pageset
, cpu
);
1824 if (pcp
->pcp
.count
) {
1832 cpumask_set_cpu(cpu
, &cpus_with_pcps
);
1834 cpumask_clear_cpu(cpu
, &cpus_with_pcps
);
1836 on_each_cpu_mask(&cpus_with_pcps
, (smp_call_func_t
) drain_local_pages
,
1840 #ifdef CONFIG_HIBERNATION
1842 void mark_free_pages(struct zone
*zone
)
1844 unsigned long pfn
, max_zone_pfn
;
1845 unsigned long flags
;
1846 unsigned int order
, t
;
1847 struct list_head
*curr
;
1849 if (zone_is_empty(zone
))
1852 spin_lock_irqsave(&zone
->lock
, flags
);
1854 max_zone_pfn
= zone_end_pfn(zone
);
1855 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
1856 if (pfn_valid(pfn
)) {
1857 struct page
*page
= pfn_to_page(pfn
);
1859 if (!swsusp_page_is_forbidden(page
))
1860 swsusp_unset_page_free(page
);
1863 for_each_migratetype_order(order
, t
) {
1864 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
1867 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
1868 for (i
= 0; i
< (1UL << order
); i
++)
1869 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
1872 spin_unlock_irqrestore(&zone
->lock
, flags
);
1874 #endif /* CONFIG_PM */
1877 * Free a 0-order page
1878 * cold == true ? free a cold page : free a hot page
1880 void free_hot_cold_page(struct page
*page
, bool cold
)
1882 struct zone
*zone
= page_zone(page
);
1883 struct per_cpu_pages
*pcp
;
1884 unsigned long flags
;
1885 unsigned long pfn
= page_to_pfn(page
);
1888 if (!free_pages_prepare(page
, 0))
1891 migratetype
= get_pfnblock_migratetype(page
, pfn
);
1892 set_freepage_migratetype(page
, migratetype
);
1893 local_irq_save(flags
);
1894 __count_vm_event(PGFREE
);
1897 * We only track unmovable, reclaimable and movable on pcp lists.
1898 * Free ISOLATE pages back to the allocator because they are being
1899 * offlined but treat RESERVE as movable pages so we can get those
1900 * areas back if necessary. Otherwise, we may have to free
1901 * excessively into the page allocator
1903 if (migratetype
>= MIGRATE_PCPTYPES
) {
1904 if (unlikely(is_migrate_isolate(migratetype
))) {
1905 free_one_page(zone
, page
, pfn
, 0, migratetype
);
1908 migratetype
= MIGRATE_MOVABLE
;
1911 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
1913 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
1915 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
1917 if (pcp
->count
>= pcp
->high
) {
1918 unsigned long batch
= READ_ONCE(pcp
->batch
);
1919 free_pcppages_bulk(zone
, batch
, pcp
);
1920 pcp
->count
-= batch
;
1924 local_irq_restore(flags
);
1928 * Free a list of 0-order pages
1930 void free_hot_cold_page_list(struct list_head
*list
, bool cold
)
1932 struct page
*page
, *next
;
1934 list_for_each_entry_safe(page
, next
, list
, lru
) {
1935 trace_mm_page_free_batched(page
, cold
);
1936 free_hot_cold_page(page
, cold
);
1941 * split_page takes a non-compound higher-order page, and splits it into
1942 * n (1<<order) sub-pages: page[0..n]
1943 * Each sub-page must be freed individually.
1945 * Note: this is probably too low level an operation for use in drivers.
1946 * Please consult with lkml before using this in your driver.
1948 void split_page(struct page
*page
, unsigned int order
)
1952 VM_BUG_ON_PAGE(PageCompound(page
), page
);
1953 VM_BUG_ON_PAGE(!page_count(page
), page
);
1955 #ifdef CONFIG_KMEMCHECK
1957 * Split shadow pages too, because free(page[0]) would
1958 * otherwise free the whole shadow.
1960 if (kmemcheck_page_is_tracked(page
))
1961 split_page(virt_to_page(page
[0].shadow
), order
);
1964 set_page_owner(page
, 0, 0);
1965 for (i
= 1; i
< (1 << order
); i
++) {
1966 set_page_refcounted(page
+ i
);
1967 set_page_owner(page
+ i
, 0, 0);
1970 EXPORT_SYMBOL_GPL(split_page
);
1972 int __isolate_free_page(struct page
*page
, unsigned int order
)
1974 unsigned long watermark
;
1978 BUG_ON(!PageBuddy(page
));
1980 zone
= page_zone(page
);
1981 mt
= get_pageblock_migratetype(page
);
1983 if (!is_migrate_isolate(mt
)) {
1984 /* Obey watermarks as if the page was being allocated */
1985 watermark
= low_wmark_pages(zone
) + (1 << order
);
1986 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
1989 __mod_zone_freepage_state(zone
, -(1UL << order
), mt
);
1992 /* Remove page from free list */
1993 list_del(&page
->lru
);
1994 zone
->free_area
[order
].nr_free
--;
1995 rmv_page_order(page
);
1997 /* Set the pageblock if the isolated page is at least a pageblock */
1998 if (order
>= pageblock_order
- 1) {
1999 struct page
*endpage
= page
+ (1 << order
) - 1;
2000 for (; page
< endpage
; page
+= pageblock_nr_pages
) {
2001 int mt
= get_pageblock_migratetype(page
);
2002 if (!is_migrate_isolate(mt
) && !is_migrate_cma(mt
))
2003 set_pageblock_migratetype(page
,
2008 set_page_owner(page
, order
, 0);
2009 return 1UL << order
;
2013 * Similar to split_page except the page is already free. As this is only
2014 * being used for migration, the migratetype of the block also changes.
2015 * As this is called with interrupts disabled, the caller is responsible
2016 * for calling arch_alloc_page() and kernel_map_page() after interrupts
2019 * Note: this is probably too low level an operation for use in drivers.
2020 * Please consult with lkml before using this in your driver.
2022 int split_free_page(struct page
*page
)
2027 order
= page_order(page
);
2029 nr_pages
= __isolate_free_page(page
, order
);
2033 /* Split into individual pages */
2034 set_page_refcounted(page
);
2035 split_page(page
, order
);
2040 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2043 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
2044 struct zone
*zone
, unsigned int order
,
2045 gfp_t gfp_flags
, int migratetype
)
2047 unsigned long flags
;
2049 bool cold
= ((gfp_flags
& __GFP_COLD
) != 0);
2051 if (likely(order
== 0)) {
2052 struct per_cpu_pages
*pcp
;
2053 struct list_head
*list
;
2055 local_irq_save(flags
);
2056 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2057 list
= &pcp
->lists
[migratetype
];
2058 if (list_empty(list
)) {
2059 pcp
->count
+= rmqueue_bulk(zone
, 0,
2062 if (unlikely(list_empty(list
)))
2067 page
= list_entry(list
->prev
, struct page
, lru
);
2069 page
= list_entry(list
->next
, struct page
, lru
);
2071 list_del(&page
->lru
);
2074 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
2076 * __GFP_NOFAIL is not to be used in new code.
2078 * All __GFP_NOFAIL callers should be fixed so that they
2079 * properly detect and handle allocation failures.
2081 * We most definitely don't want callers attempting to
2082 * allocate greater than order-1 page units with
2085 WARN_ON_ONCE(order
> 1);
2087 spin_lock_irqsave(&zone
->lock
, flags
);
2088 page
= __rmqueue(zone
, order
, migratetype
);
2089 spin_unlock(&zone
->lock
);
2092 __mod_zone_freepage_state(zone
, -(1 << order
),
2093 get_freepage_migratetype(page
));
2096 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
, -(1 << order
));
2097 if (atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]) <= 0 &&
2098 !test_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
))
2099 set_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
);
2101 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
2102 zone_statistics(preferred_zone
, zone
, gfp_flags
);
2103 local_irq_restore(flags
);
2105 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
2109 local_irq_restore(flags
);
2113 #ifdef CONFIG_FAIL_PAGE_ALLOC
2116 struct fault_attr attr
;
2118 u32 ignore_gfp_highmem
;
2119 u32 ignore_gfp_wait
;
2121 } fail_page_alloc
= {
2122 .attr
= FAULT_ATTR_INITIALIZER
,
2123 .ignore_gfp_wait
= 1,
2124 .ignore_gfp_highmem
= 1,
2128 static int __init
setup_fail_page_alloc(char *str
)
2130 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
2132 __setup("fail_page_alloc=", setup_fail_page_alloc
);
2134 static bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2136 if (order
< fail_page_alloc
.min_order
)
2138 if (gfp_mask
& __GFP_NOFAIL
)
2140 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
2142 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
2145 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
2148 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2150 static int __init
fail_page_alloc_debugfs(void)
2152 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
2155 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
2156 &fail_page_alloc
.attr
);
2158 return PTR_ERR(dir
);
2160 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
2161 &fail_page_alloc
.ignore_gfp_wait
))
2163 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
2164 &fail_page_alloc
.ignore_gfp_highmem
))
2166 if (!debugfs_create_u32("min-order", mode
, dir
,
2167 &fail_page_alloc
.min_order
))
2172 debugfs_remove_recursive(dir
);
2177 late_initcall(fail_page_alloc_debugfs
);
2179 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2181 #else /* CONFIG_FAIL_PAGE_ALLOC */
2183 static inline bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2188 #endif /* CONFIG_FAIL_PAGE_ALLOC */
2191 * Return true if free pages are above 'mark'. This takes into account the order
2192 * of the allocation.
2194 static bool __zone_watermark_ok(struct zone
*z
, unsigned int order
,
2195 unsigned long mark
, int classzone_idx
, int alloc_flags
,
2198 /* free_pages may go negative - that's OK */
2203 free_pages
-= (1 << order
) - 1;
2204 if (alloc_flags
& ALLOC_HIGH
)
2206 if (alloc_flags
& ALLOC_HARDER
)
2209 /* If allocation can't use CMA areas don't use free CMA pages */
2210 if (!(alloc_flags
& ALLOC_CMA
))
2211 free_cma
= zone_page_state(z
, NR_FREE_CMA_PAGES
);
2214 if (free_pages
- free_cma
<= min
+ z
->lowmem_reserve
[classzone_idx
])
2216 for (o
= 0; o
< order
; o
++) {
2217 /* At the next order, this order's pages become unavailable */
2218 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
2220 /* Require fewer higher order pages to be free */
2223 if (free_pages
<= min
)
2229 bool zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
2230 int classzone_idx
, int alloc_flags
)
2232 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
2233 zone_page_state(z
, NR_FREE_PAGES
));
2236 bool zone_watermark_ok_safe(struct zone
*z
, unsigned int order
,
2237 unsigned long mark
, int classzone_idx
, int alloc_flags
)
2239 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
2241 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
2242 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
2244 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
2250 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
2251 * skip over zones that are not allowed by the cpuset, or that have
2252 * been recently (in last second) found to be nearly full. See further
2253 * comments in mmzone.h. Reduces cache footprint of zonelist scans
2254 * that have to skip over a lot of full or unallowed zones.
2256 * If the zonelist cache is present in the passed zonelist, then
2257 * returns a pointer to the allowed node mask (either the current
2258 * tasks mems_allowed, or node_states[N_MEMORY].)
2260 * If the zonelist cache is not available for this zonelist, does
2261 * nothing and returns NULL.
2263 * If the fullzones BITMAP in the zonelist cache is stale (more than
2264 * a second since last zap'd) then we zap it out (clear its bits.)
2266 * We hold off even calling zlc_setup, until after we've checked the
2267 * first zone in the zonelist, on the theory that most allocations will
2268 * be satisfied from that first zone, so best to examine that zone as
2269 * quickly as we can.
2271 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
2273 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
2274 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
2276 zlc
= zonelist
->zlcache_ptr
;
2280 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
2281 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2282 zlc
->last_full_zap
= jiffies
;
2285 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
2286 &cpuset_current_mems_allowed
:
2287 &node_states
[N_MEMORY
];
2288 return allowednodes
;
2292 * Given 'z' scanning a zonelist, run a couple of quick checks to see
2293 * if it is worth looking at further for free memory:
2294 * 1) Check that the zone isn't thought to be full (doesn't have its
2295 * bit set in the zonelist_cache fullzones BITMAP).
2296 * 2) Check that the zones node (obtained from the zonelist_cache
2297 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
2298 * Return true (non-zero) if zone is worth looking at further, or
2299 * else return false (zero) if it is not.
2301 * This check -ignores- the distinction between various watermarks,
2302 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
2303 * found to be full for any variation of these watermarks, it will
2304 * be considered full for up to one second by all requests, unless
2305 * we are so low on memory on all allowed nodes that we are forced
2306 * into the second scan of the zonelist.
2308 * In the second scan we ignore this zonelist cache and exactly
2309 * apply the watermarks to all zones, even it is slower to do so.
2310 * We are low on memory in the second scan, and should leave no stone
2311 * unturned looking for a free page.
2313 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
2314 nodemask_t
*allowednodes
)
2316 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
2317 int i
; /* index of *z in zonelist zones */
2318 int n
; /* node that zone *z is on */
2320 zlc
= zonelist
->zlcache_ptr
;
2324 i
= z
- zonelist
->_zonerefs
;
2327 /* This zone is worth trying if it is allowed but not full */
2328 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
2332 * Given 'z' scanning a zonelist, set the corresponding bit in
2333 * zlc->fullzones, so that subsequent attempts to allocate a page
2334 * from that zone don't waste time re-examining it.
2336 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
2338 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
2339 int i
; /* index of *z in zonelist zones */
2341 zlc
= zonelist
->zlcache_ptr
;
2345 i
= z
- zonelist
->_zonerefs
;
2347 set_bit(i
, zlc
->fullzones
);
2351 * clear all zones full, called after direct reclaim makes progress so that
2352 * a zone that was recently full is not skipped over for up to a second
2354 static void zlc_clear_zones_full(struct zonelist
*zonelist
)
2356 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
2358 zlc
= zonelist
->zlcache_ptr
;
2362 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2365 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
2367 return local_zone
->node
== zone
->node
;
2370 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2372 return node_distance(zone_to_nid(local_zone
), zone_to_nid(zone
)) <
2376 #else /* CONFIG_NUMA */
2378 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
2383 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
2384 nodemask_t
*allowednodes
)
2389 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
2393 static void zlc_clear_zones_full(struct zonelist
*zonelist
)
2397 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
2402 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2407 #endif /* CONFIG_NUMA */
2409 static void reset_alloc_batches(struct zone
*preferred_zone
)
2411 struct zone
*zone
= preferred_zone
->zone_pgdat
->node_zones
;
2414 mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
2415 high_wmark_pages(zone
) - low_wmark_pages(zone
) -
2416 atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]));
2417 clear_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
);
2418 } while (zone
++ != preferred_zone
);
2422 * get_page_from_freelist goes through the zonelist trying to allocate
2425 static struct page
*
2426 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
, int alloc_flags
,
2427 const struct alloc_context
*ac
)
2429 struct zonelist
*zonelist
= ac
->zonelist
;
2431 struct page
*page
= NULL
;
2433 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
2434 int zlc_active
= 0; /* set if using zonelist_cache */
2435 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
2436 bool consider_zone_dirty
= (alloc_flags
& ALLOC_WMARK_LOW
) &&
2437 (gfp_mask
& __GFP_WRITE
);
2438 int nr_fair_skipped
= 0;
2439 bool zonelist_rescan
;
2442 zonelist_rescan
= false;
2445 * Scan zonelist, looking for a zone with enough free.
2446 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2448 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
, ac
->high_zoneidx
,
2452 if (IS_ENABLED(CONFIG_NUMA
) && zlc_active
&&
2453 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
2455 if (cpusets_enabled() &&
2456 (alloc_flags
& ALLOC_CPUSET
) &&
2457 !cpuset_zone_allowed(zone
, gfp_mask
))
2460 * Distribute pages in proportion to the individual
2461 * zone size to ensure fair page aging. The zone a
2462 * page was allocated in should have no effect on the
2463 * time the page has in memory before being reclaimed.
2465 if (alloc_flags
& ALLOC_FAIR
) {
2466 if (!zone_local(ac
->preferred_zone
, zone
))
2468 if (test_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
)) {
2474 * When allocating a page cache page for writing, we
2475 * want to get it from a zone that is within its dirty
2476 * limit, such that no single zone holds more than its
2477 * proportional share of globally allowed dirty pages.
2478 * The dirty limits take into account the zone's
2479 * lowmem reserves and high watermark so that kswapd
2480 * should be able to balance it without having to
2481 * write pages from its LRU list.
2483 * This may look like it could increase pressure on
2484 * lower zones by failing allocations in higher zones
2485 * before they are full. But the pages that do spill
2486 * over are limited as the lower zones are protected
2487 * by this very same mechanism. It should not become
2488 * a practical burden to them.
2490 * XXX: For now, allow allocations to potentially
2491 * exceed the per-zone dirty limit in the slowpath
2492 * (ALLOC_WMARK_LOW unset) before going into reclaim,
2493 * which is important when on a NUMA setup the allowed
2494 * zones are together not big enough to reach the
2495 * global limit. The proper fix for these situations
2496 * will require awareness of zones in the
2497 * dirty-throttling and the flusher threads.
2499 if (consider_zone_dirty
&& !zone_dirty_ok(zone
))
2502 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
2503 if (!zone_watermark_ok(zone
, order
, mark
,
2504 ac
->classzone_idx
, alloc_flags
)) {
2507 /* Checked here to keep the fast path fast */
2508 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
2509 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
2512 if (IS_ENABLED(CONFIG_NUMA
) &&
2513 !did_zlc_setup
&& nr_online_nodes
> 1) {
2515 * we do zlc_setup if there are multiple nodes
2516 * and before considering the first zone allowed
2519 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
2524 if (zone_reclaim_mode
== 0 ||
2525 !zone_allows_reclaim(ac
->preferred_zone
, zone
))
2526 goto this_zone_full
;
2529 * As we may have just activated ZLC, check if the first
2530 * eligible zone has failed zone_reclaim recently.
2532 if (IS_ENABLED(CONFIG_NUMA
) && zlc_active
&&
2533 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
2536 ret
= zone_reclaim(zone
, gfp_mask
, order
);
2538 case ZONE_RECLAIM_NOSCAN
:
2541 case ZONE_RECLAIM_FULL
:
2542 /* scanned but unreclaimable */
2545 /* did we reclaim enough */
2546 if (zone_watermark_ok(zone
, order
, mark
,
2547 ac
->classzone_idx
, alloc_flags
))
2551 * Failed to reclaim enough to meet watermark.
2552 * Only mark the zone full if checking the min
2553 * watermark or if we failed to reclaim just
2554 * 1<<order pages or else the page allocator
2555 * fastpath will prematurely mark zones full
2556 * when the watermark is between the low and
2559 if (((alloc_flags
& ALLOC_WMARK_MASK
) == ALLOC_WMARK_MIN
) ||
2560 ret
== ZONE_RECLAIM_SOME
)
2561 goto this_zone_full
;
2568 page
= buffered_rmqueue(ac
->preferred_zone
, zone
, order
,
2569 gfp_mask
, ac
->migratetype
);
2571 if (prep_new_page(page
, order
, gfp_mask
, alloc_flags
))
2576 if (IS_ENABLED(CONFIG_NUMA
) && zlc_active
)
2577 zlc_mark_zone_full(zonelist
, z
);
2581 * The first pass makes sure allocations are spread fairly within the
2582 * local node. However, the local node might have free pages left
2583 * after the fairness batches are exhausted, and remote zones haven't
2584 * even been considered yet. Try once more without fairness, and
2585 * include remote zones now, before entering the slowpath and waking
2586 * kswapd: prefer spilling to a remote zone over swapping locally.
2588 if (alloc_flags
& ALLOC_FAIR
) {
2589 alloc_flags
&= ~ALLOC_FAIR
;
2590 if (nr_fair_skipped
) {
2591 zonelist_rescan
= true;
2592 reset_alloc_batches(ac
->preferred_zone
);
2594 if (nr_online_nodes
> 1)
2595 zonelist_rescan
= true;
2598 if (unlikely(IS_ENABLED(CONFIG_NUMA
) && zlc_active
)) {
2599 /* Disable zlc cache for second zonelist scan */
2601 zonelist_rescan
= true;
2604 if (zonelist_rescan
)
2611 * Large machines with many possible nodes should not always dump per-node
2612 * meminfo in irq context.
2614 static inline bool should_suppress_show_mem(void)
2619 ret
= in_interrupt();
2624 static DEFINE_RATELIMIT_STATE(nopage_rs
,
2625 DEFAULT_RATELIMIT_INTERVAL
,
2626 DEFAULT_RATELIMIT_BURST
);
2628 void warn_alloc_failed(gfp_t gfp_mask
, int order
, const char *fmt
, ...)
2630 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
2632 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
) ||
2633 debug_guardpage_minorder() > 0)
2637 * This documents exceptions given to allocations in certain
2638 * contexts that are allowed to allocate outside current's set
2641 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2642 if (test_thread_flag(TIF_MEMDIE
) ||
2643 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
2644 filter
&= ~SHOW_MEM_FILTER_NODES
;
2645 if (in_interrupt() || !(gfp_mask
& __GFP_WAIT
))
2646 filter
&= ~SHOW_MEM_FILTER_NODES
;
2649 struct va_format vaf
;
2652 va_start(args
, fmt
);
2657 pr_warn("%pV", &vaf
);
2662 pr_warn("%s: page allocation failure: order:%d, mode:0x%x\n",
2663 current
->comm
, order
, gfp_mask
);
2666 if (!should_suppress_show_mem())
2670 static inline struct page
*
2671 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
2672 const struct alloc_context
*ac
, unsigned long *did_some_progress
)
2676 *did_some_progress
= 0;
2679 * Acquire the oom lock. If that fails, somebody else is
2680 * making progress for us.
2682 if (!mutex_trylock(&oom_lock
)) {
2683 *did_some_progress
= 1;
2684 schedule_timeout_uninterruptible(1);
2689 * Go through the zonelist yet one more time, keep very high watermark
2690 * here, this is only to catch a parallel oom killing, we must fail if
2691 * we're still under heavy pressure.
2693 page
= get_page_from_freelist(gfp_mask
| __GFP_HARDWALL
, order
,
2694 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
, ac
);
2698 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2699 /* Coredumps can quickly deplete all memory reserves */
2700 if (current
->flags
& PF_DUMPCORE
)
2702 /* The OOM killer will not help higher order allocs */
2703 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2705 /* The OOM killer does not needlessly kill tasks for lowmem */
2706 if (ac
->high_zoneidx
< ZONE_NORMAL
)
2708 /* The OOM killer does not compensate for IO-less reclaim */
2709 if (!(gfp_mask
& __GFP_FS
)) {
2711 * XXX: Page reclaim didn't yield anything,
2712 * and the OOM killer can't be invoked, but
2713 * keep looping as per tradition.
2715 *did_some_progress
= 1;
2718 if (pm_suspended_storage())
2720 /* The OOM killer may not free memory on a specific node */
2721 if (gfp_mask
& __GFP_THISNODE
)
2724 /* Exhausted what can be done so it's blamo time */
2725 if (out_of_memory(ac
->zonelist
, gfp_mask
, order
, ac
->nodemask
, false)
2726 || WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
))
2727 *did_some_progress
= 1;
2729 mutex_unlock(&oom_lock
);
2733 #ifdef CONFIG_COMPACTION
2734 /* Try memory compaction for high-order allocations before reclaim */
2735 static struct page
*
2736 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2737 int alloc_flags
, const struct alloc_context
*ac
,
2738 enum migrate_mode mode
, int *contended_compaction
,
2739 bool *deferred_compaction
)
2741 unsigned long compact_result
;
2747 current
->flags
|= PF_MEMALLOC
;
2748 compact_result
= try_to_compact_pages(gfp_mask
, order
, alloc_flags
, ac
,
2749 mode
, contended_compaction
);
2750 current
->flags
&= ~PF_MEMALLOC
;
2752 switch (compact_result
) {
2753 case COMPACT_DEFERRED
:
2754 *deferred_compaction
= true;
2756 case COMPACT_SKIPPED
:
2763 * At least in one zone compaction wasn't deferred or skipped, so let's
2764 * count a compaction stall
2766 count_vm_event(COMPACTSTALL
);
2768 page
= get_page_from_freelist(gfp_mask
, order
,
2769 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
2772 struct zone
*zone
= page_zone(page
);
2774 zone
->compact_blockskip_flush
= false;
2775 compaction_defer_reset(zone
, order
, true);
2776 count_vm_event(COMPACTSUCCESS
);
2781 * It's bad if compaction run occurs and fails. The most likely reason
2782 * is that pages exist, but not enough to satisfy watermarks.
2784 count_vm_event(COMPACTFAIL
);
2791 static inline struct page
*
2792 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2793 int alloc_flags
, const struct alloc_context
*ac
,
2794 enum migrate_mode mode
, int *contended_compaction
,
2795 bool *deferred_compaction
)
2799 #endif /* CONFIG_COMPACTION */
2801 /* Perform direct synchronous page reclaim */
2803 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
,
2804 const struct alloc_context
*ac
)
2806 struct reclaim_state reclaim_state
;
2811 /* We now go into synchronous reclaim */
2812 cpuset_memory_pressure_bump();
2813 current
->flags
|= PF_MEMALLOC
;
2814 lockdep_set_current_reclaim_state(gfp_mask
);
2815 reclaim_state
.reclaimed_slab
= 0;
2816 current
->reclaim_state
= &reclaim_state
;
2818 progress
= try_to_free_pages(ac
->zonelist
, order
, gfp_mask
,
2821 current
->reclaim_state
= NULL
;
2822 lockdep_clear_current_reclaim_state();
2823 current
->flags
&= ~PF_MEMALLOC
;
2830 /* The really slow allocator path where we enter direct reclaim */
2831 static inline struct page
*
2832 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
2833 int alloc_flags
, const struct alloc_context
*ac
,
2834 unsigned long *did_some_progress
)
2836 struct page
*page
= NULL
;
2837 bool drained
= false;
2839 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, ac
);
2840 if (unlikely(!(*did_some_progress
)))
2843 /* After successful reclaim, reconsider all zones for allocation */
2844 if (IS_ENABLED(CONFIG_NUMA
))
2845 zlc_clear_zones_full(ac
->zonelist
);
2848 page
= get_page_from_freelist(gfp_mask
, order
,
2849 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
2852 * If an allocation failed after direct reclaim, it could be because
2853 * pages are pinned on the per-cpu lists. Drain them and try again
2855 if (!page
&& !drained
) {
2856 drain_all_pages(NULL
);
2865 * This is called in the allocator slow-path if the allocation request is of
2866 * sufficient urgency to ignore watermarks and take other desperate measures
2868 static inline struct page
*
2869 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
2870 const struct alloc_context
*ac
)
2875 page
= get_page_from_freelist(gfp_mask
, order
,
2876 ALLOC_NO_WATERMARKS
, ac
);
2878 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
2879 wait_iff_congested(ac
->preferred_zone
, BLK_RW_ASYNC
,
2881 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
2886 static void wake_all_kswapds(unsigned int order
, const struct alloc_context
*ac
)
2891 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
,
2892 ac
->high_zoneidx
, ac
->nodemask
)
2893 wakeup_kswapd(zone
, order
, zone_idx(ac
->preferred_zone
));
2897 gfp_to_alloc_flags(gfp_t gfp_mask
)
2899 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
2900 const bool atomic
= !(gfp_mask
& (__GFP_WAIT
| __GFP_NO_KSWAPD
));
2902 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2903 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
2906 * The caller may dip into page reserves a bit more if the caller
2907 * cannot run direct reclaim, or if the caller has realtime scheduling
2908 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2909 * set both ALLOC_HARDER (atomic == true) and ALLOC_HIGH (__GFP_HIGH).
2911 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
2915 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2916 * if it can't schedule.
2918 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2919 alloc_flags
|= ALLOC_HARDER
;
2921 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
2922 * comment for __cpuset_node_allowed().
2924 alloc_flags
&= ~ALLOC_CPUSET
;
2925 } else if (unlikely(rt_task(current
)) && !in_interrupt())
2926 alloc_flags
|= ALLOC_HARDER
;
2928 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
2929 if (gfp_mask
& __GFP_MEMALLOC
)
2930 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2931 else if (in_serving_softirq() && (current
->flags
& PF_MEMALLOC
))
2932 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2933 else if (!in_interrupt() &&
2934 ((current
->flags
& PF_MEMALLOC
) ||
2935 unlikely(test_thread_flag(TIF_MEMDIE
))))
2936 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2939 if (gfpflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
2940 alloc_flags
|= ALLOC_CMA
;
2945 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask
)
2947 return !!(gfp_to_alloc_flags(gfp_mask
) & ALLOC_NO_WATERMARKS
);
2950 static inline struct page
*
2951 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
2952 struct alloc_context
*ac
)
2954 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
2955 struct page
*page
= NULL
;
2957 unsigned long pages_reclaimed
= 0;
2958 unsigned long did_some_progress
;
2959 enum migrate_mode migration_mode
= MIGRATE_ASYNC
;
2960 bool deferred_compaction
= false;
2961 int contended_compaction
= COMPACT_CONTENDED_NONE
;
2964 * In the slowpath, we sanity check order to avoid ever trying to
2965 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2966 * be using allocators in order of preference for an area that is
2969 if (order
>= MAX_ORDER
) {
2970 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
2975 * If this allocation cannot block and it is for a specific node, then
2976 * fail early. There's no need to wakeup kswapd or retry for a
2977 * speculative node-specific allocation.
2979 if (IS_ENABLED(CONFIG_NUMA
) && (gfp_mask
& __GFP_THISNODE
) && !wait
)
2983 if (!(gfp_mask
& __GFP_NO_KSWAPD
))
2984 wake_all_kswapds(order
, ac
);
2987 * OK, we're below the kswapd watermark and have kicked background
2988 * reclaim. Now things get more complex, so set up alloc_flags according
2989 * to how we want to proceed.
2991 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
2994 * Find the true preferred zone if the allocation is unconstrained by
2997 if (!(alloc_flags
& ALLOC_CPUSET
) && !ac
->nodemask
) {
2998 struct zoneref
*preferred_zoneref
;
2999 preferred_zoneref
= first_zones_zonelist(ac
->zonelist
,
3000 ac
->high_zoneidx
, NULL
, &ac
->preferred_zone
);
3001 ac
->classzone_idx
= zonelist_zone_idx(preferred_zoneref
);
3004 /* This is the last chance, in general, before the goto nopage. */
3005 page
= get_page_from_freelist(gfp_mask
, order
,
3006 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
3010 /* Allocate without watermarks if the context allows */
3011 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
3013 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3014 * the allocation is high priority and these type of
3015 * allocations are system rather than user orientated
3017 ac
->zonelist
= node_zonelist(numa_node_id(), gfp_mask
);
3019 page
= __alloc_pages_high_priority(gfp_mask
, order
, ac
);
3026 /* Atomic allocations - we can't balance anything */
3029 * All existing users of the deprecated __GFP_NOFAIL are
3030 * blockable, so warn of any new users that actually allow this
3031 * type of allocation to fail.
3033 WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
);
3037 /* Avoid recursion of direct reclaim */
3038 if (current
->flags
& PF_MEMALLOC
)
3041 /* Avoid allocations with no watermarks from looping endlessly */
3042 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
3046 * Try direct compaction. The first pass is asynchronous. Subsequent
3047 * attempts after direct reclaim are synchronous
3049 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
, ac
,
3051 &contended_compaction
,
3052 &deferred_compaction
);
3056 /* Checks for THP-specific high-order allocations */
3057 if ((gfp_mask
& GFP_TRANSHUGE
) == GFP_TRANSHUGE
) {
3059 * If compaction is deferred for high-order allocations, it is
3060 * because sync compaction recently failed. If this is the case
3061 * and the caller requested a THP allocation, we do not want
3062 * to heavily disrupt the system, so we fail the allocation
3063 * instead of entering direct reclaim.
3065 if (deferred_compaction
)
3069 * In all zones where compaction was attempted (and not
3070 * deferred or skipped), lock contention has been detected.
3071 * For THP allocation we do not want to disrupt the others
3072 * so we fallback to base pages instead.
3074 if (contended_compaction
== COMPACT_CONTENDED_LOCK
)
3078 * If compaction was aborted due to need_resched(), we do not
3079 * want to further increase allocation latency, unless it is
3080 * khugepaged trying to collapse.
3082 if (contended_compaction
== COMPACT_CONTENDED_SCHED
3083 && !(current
->flags
& PF_KTHREAD
))
3088 * It can become very expensive to allocate transparent hugepages at
3089 * fault, so use asynchronous memory compaction for THP unless it is
3090 * khugepaged trying to collapse.
3092 if ((gfp_mask
& GFP_TRANSHUGE
) != GFP_TRANSHUGE
||
3093 (current
->flags
& PF_KTHREAD
))
3094 migration_mode
= MIGRATE_SYNC_LIGHT
;
3096 /* Try direct reclaim and then allocating */
3097 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
, alloc_flags
, ac
,
3098 &did_some_progress
);
3102 /* Do not loop if specifically requested */
3103 if (gfp_mask
& __GFP_NORETRY
)
3106 /* Keep reclaiming pages as long as there is reasonable progress */
3107 pages_reclaimed
+= did_some_progress
;
3108 if ((did_some_progress
&& order
<= PAGE_ALLOC_COSTLY_ORDER
) ||
3109 ((gfp_mask
& __GFP_REPEAT
) && pages_reclaimed
< (1 << order
))) {
3110 /* Wait for some write requests to complete then retry */
3111 wait_iff_congested(ac
->preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
3115 /* Reclaim has failed us, start killing things */
3116 page
= __alloc_pages_may_oom(gfp_mask
, order
, ac
, &did_some_progress
);
3120 /* Retry as long as the OOM killer is making progress */
3121 if (did_some_progress
)
3126 * High-order allocations do not necessarily loop after
3127 * direct reclaim and reclaim/compaction depends on compaction
3128 * being called after reclaim so call directly if necessary
3130 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
,
3132 &contended_compaction
,
3133 &deferred_compaction
);
3137 warn_alloc_failed(gfp_mask
, order
, NULL
);
3143 * This is the 'heart' of the zoned buddy allocator.
3146 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
3147 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
3149 struct zoneref
*preferred_zoneref
;
3150 struct page
*page
= NULL
;
3151 unsigned int cpuset_mems_cookie
;
3152 int alloc_flags
= ALLOC_WMARK_LOW
|ALLOC_CPUSET
|ALLOC_FAIR
;
3153 gfp_t alloc_mask
; /* The gfp_t that was actually used for allocation */
3154 struct alloc_context ac
= {
3155 .high_zoneidx
= gfp_zone(gfp_mask
),
3156 .nodemask
= nodemask
,
3157 .migratetype
= gfpflags_to_migratetype(gfp_mask
),
3160 gfp_mask
&= gfp_allowed_mask
;
3162 lockdep_trace_alloc(gfp_mask
);
3164 might_sleep_if(gfp_mask
& __GFP_WAIT
);
3166 if (should_fail_alloc_page(gfp_mask
, order
))
3170 * Check the zones suitable for the gfp_mask contain at least one
3171 * valid zone. It's possible to have an empty zonelist as a result
3172 * of __GFP_THISNODE and a memoryless node
3174 if (unlikely(!zonelist
->_zonerefs
->zone
))
3177 if (IS_ENABLED(CONFIG_CMA
) && ac
.migratetype
== MIGRATE_MOVABLE
)
3178 alloc_flags
|= ALLOC_CMA
;
3181 cpuset_mems_cookie
= read_mems_allowed_begin();
3183 /* We set it here, as __alloc_pages_slowpath might have changed it */
3184 ac
.zonelist
= zonelist
;
3185 /* The preferred zone is used for statistics later */
3186 preferred_zoneref
= first_zones_zonelist(ac
.zonelist
, ac
.high_zoneidx
,
3187 ac
.nodemask
? : &cpuset_current_mems_allowed
,
3188 &ac
.preferred_zone
);
3189 if (!ac
.preferred_zone
)
3191 ac
.classzone_idx
= zonelist_zone_idx(preferred_zoneref
);
3193 /* First allocation attempt */
3194 alloc_mask
= gfp_mask
|__GFP_HARDWALL
;
3195 page
= get_page_from_freelist(alloc_mask
, order
, alloc_flags
, &ac
);
3196 if (unlikely(!page
)) {
3198 * Runtime PM, block IO and its error handling path
3199 * can deadlock because I/O on the device might not
3202 alloc_mask
= memalloc_noio_flags(gfp_mask
);
3204 page
= __alloc_pages_slowpath(alloc_mask
, order
, &ac
);
3207 if (kmemcheck_enabled
&& page
)
3208 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
3210 trace_mm_page_alloc(page
, order
, alloc_mask
, ac
.migratetype
);
3214 * When updating a task's mems_allowed, it is possible to race with
3215 * parallel threads in such a way that an allocation can fail while
3216 * the mask is being updated. If a page allocation is about to fail,
3217 * check if the cpuset changed during allocation and if so, retry.
3219 if (unlikely(!page
&& read_mems_allowed_retry(cpuset_mems_cookie
)))
3224 EXPORT_SYMBOL(__alloc_pages_nodemask
);
3227 * Common helper functions.
3229 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
3234 * __get_free_pages() returns a 32-bit address, which cannot represent
3237 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
3239 page
= alloc_pages(gfp_mask
, order
);
3242 return (unsigned long) page_address(page
);
3244 EXPORT_SYMBOL(__get_free_pages
);
3246 unsigned long get_zeroed_page(gfp_t gfp_mask
)
3248 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
3250 EXPORT_SYMBOL(get_zeroed_page
);
3252 void __free_pages(struct page
*page
, unsigned int order
)
3254 if (put_page_testzero(page
)) {
3256 free_hot_cold_page(page
, false);
3258 __free_pages_ok(page
, order
);
3262 EXPORT_SYMBOL(__free_pages
);
3264 void free_pages(unsigned long addr
, unsigned int order
)
3267 VM_BUG_ON(!virt_addr_valid((void *)addr
));
3268 __free_pages(virt_to_page((void *)addr
), order
);
3272 EXPORT_SYMBOL(free_pages
);
3276 * An arbitrary-length arbitrary-offset area of memory which resides
3277 * within a 0 or higher order page. Multiple fragments within that page
3278 * are individually refcounted, in the page's reference counter.
3280 * The page_frag functions below provide a simple allocation framework for
3281 * page fragments. This is used by the network stack and network device
3282 * drivers to provide a backing region of memory for use as either an
3283 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3285 static struct page
*__page_frag_refill(struct page_frag_cache
*nc
,
3288 struct page
*page
= NULL
;
3289 gfp_t gfp
= gfp_mask
;
3291 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3292 gfp_mask
|= __GFP_COMP
| __GFP_NOWARN
| __GFP_NORETRY
|
3294 page
= alloc_pages_node(NUMA_NO_NODE
, gfp_mask
,
3295 PAGE_FRAG_CACHE_MAX_ORDER
);
3296 nc
->size
= page
? PAGE_FRAG_CACHE_MAX_SIZE
: PAGE_SIZE
;
3298 if (unlikely(!page
))
3299 page
= alloc_pages_node(NUMA_NO_NODE
, gfp
, 0);
3301 nc
->va
= page
? page_address(page
) : NULL
;
3306 void *__alloc_page_frag(struct page_frag_cache
*nc
,
3307 unsigned int fragsz
, gfp_t gfp_mask
)
3309 unsigned int size
= PAGE_SIZE
;
3313 if (unlikely(!nc
->va
)) {
3315 page
= __page_frag_refill(nc
, gfp_mask
);
3319 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3320 /* if size can vary use size else just use PAGE_SIZE */
3323 /* Even if we own the page, we do not use atomic_set().
3324 * This would break get_page_unless_zero() users.
3326 atomic_add(size
- 1, &page
->_count
);
3328 /* reset page count bias and offset to start of new frag */
3329 nc
->pfmemalloc
= page
->pfmemalloc
;
3330 nc
->pagecnt_bias
= size
;
3334 offset
= nc
->offset
- fragsz
;
3335 if (unlikely(offset
< 0)) {
3336 page
= virt_to_page(nc
->va
);
3338 if (!atomic_sub_and_test(nc
->pagecnt_bias
, &page
->_count
))
3341 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3342 /* if size can vary use size else just use PAGE_SIZE */
3345 /* OK, page count is 0, we can safely set it */
3346 atomic_set(&page
->_count
, size
);
3348 /* reset page count bias and offset to start of new frag */
3349 nc
->pagecnt_bias
= size
;
3350 offset
= size
- fragsz
;
3354 nc
->offset
= offset
;
3356 return nc
->va
+ offset
;
3358 EXPORT_SYMBOL(__alloc_page_frag
);
3361 * Frees a page fragment allocated out of either a compound or order 0 page.
3363 void __free_page_frag(void *addr
)
3365 struct page
*page
= virt_to_head_page(addr
);
3367 if (unlikely(put_page_testzero(page
)))
3368 __free_pages_ok(page
, compound_order(page
));
3370 EXPORT_SYMBOL(__free_page_frag
);
3373 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
3374 * of the current memory cgroup.
3376 * It should be used when the caller would like to use kmalloc, but since the
3377 * allocation is large, it has to fall back to the page allocator.
3379 struct page
*alloc_kmem_pages(gfp_t gfp_mask
, unsigned int order
)
3382 struct mem_cgroup
*memcg
= NULL
;
3384 if (!memcg_kmem_newpage_charge(gfp_mask
, &memcg
, order
))
3386 page
= alloc_pages(gfp_mask
, order
);
3387 memcg_kmem_commit_charge(page
, memcg
, order
);
3391 struct page
*alloc_kmem_pages_node(int nid
, gfp_t gfp_mask
, unsigned int order
)
3394 struct mem_cgroup
*memcg
= NULL
;
3396 if (!memcg_kmem_newpage_charge(gfp_mask
, &memcg
, order
))
3398 page
= alloc_pages_node(nid
, gfp_mask
, order
);
3399 memcg_kmem_commit_charge(page
, memcg
, order
);
3404 * __free_kmem_pages and free_kmem_pages will free pages allocated with
3407 void __free_kmem_pages(struct page
*page
, unsigned int order
)
3409 memcg_kmem_uncharge_pages(page
, order
);
3410 __free_pages(page
, order
);
3413 void free_kmem_pages(unsigned long addr
, unsigned int order
)
3416 VM_BUG_ON(!virt_addr_valid((void *)addr
));
3417 __free_kmem_pages(virt_to_page((void *)addr
), order
);
3421 static void *make_alloc_exact(unsigned long addr
, unsigned order
, size_t size
)
3424 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
3425 unsigned long used
= addr
+ PAGE_ALIGN(size
);
3427 split_page(virt_to_page((void *)addr
), order
);
3428 while (used
< alloc_end
) {
3433 return (void *)addr
;
3437 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3438 * @size: the number of bytes to allocate
3439 * @gfp_mask: GFP flags for the allocation
3441 * This function is similar to alloc_pages(), except that it allocates the
3442 * minimum number of pages to satisfy the request. alloc_pages() can only
3443 * allocate memory in power-of-two pages.
3445 * This function is also limited by MAX_ORDER.
3447 * Memory allocated by this function must be released by free_pages_exact().
3449 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
3451 unsigned int order
= get_order(size
);
3454 addr
= __get_free_pages(gfp_mask
, order
);
3455 return make_alloc_exact(addr
, order
, size
);
3457 EXPORT_SYMBOL(alloc_pages_exact
);
3460 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3462 * @nid: the preferred node ID where memory should be allocated
3463 * @size: the number of bytes to allocate
3464 * @gfp_mask: GFP flags for the allocation
3466 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3468 * Note this is not alloc_pages_exact_node() which allocates on a specific node,
3471 void * __meminit
alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
3473 unsigned order
= get_order(size
);
3474 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
3477 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
3481 * free_pages_exact - release memory allocated via alloc_pages_exact()
3482 * @virt: the value returned by alloc_pages_exact.
3483 * @size: size of allocation, same value as passed to alloc_pages_exact().
3485 * Release the memory allocated by a previous call to alloc_pages_exact.
3487 void free_pages_exact(void *virt
, size_t size
)
3489 unsigned long addr
= (unsigned long)virt
;
3490 unsigned long end
= addr
+ PAGE_ALIGN(size
);
3492 while (addr
< end
) {
3497 EXPORT_SYMBOL(free_pages_exact
);
3500 * nr_free_zone_pages - count number of pages beyond high watermark
3501 * @offset: The zone index of the highest zone
3503 * nr_free_zone_pages() counts the number of counts pages which are beyond the
3504 * high watermark within all zones at or below a given zone index. For each
3505 * zone, the number of pages is calculated as:
3506 * managed_pages - high_pages
3508 static unsigned long nr_free_zone_pages(int offset
)
3513 /* Just pick one node, since fallback list is circular */
3514 unsigned long sum
= 0;
3516 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
3518 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
3519 unsigned long size
= zone
->managed_pages
;
3520 unsigned long high
= high_wmark_pages(zone
);
3529 * nr_free_buffer_pages - count number of pages beyond high watermark
3531 * nr_free_buffer_pages() counts the number of pages which are beyond the high
3532 * watermark within ZONE_DMA and ZONE_NORMAL.
3534 unsigned long nr_free_buffer_pages(void)
3536 return nr_free_zone_pages(gfp_zone(GFP_USER
));
3538 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
3541 * nr_free_pagecache_pages - count number of pages beyond high watermark
3543 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3544 * high watermark within all zones.
3546 unsigned long nr_free_pagecache_pages(void)
3548 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
3551 static inline void show_node(struct zone
*zone
)
3553 if (IS_ENABLED(CONFIG_NUMA
))
3554 printk("Node %d ", zone_to_nid(zone
));
3557 void si_meminfo(struct sysinfo
*val
)
3559 val
->totalram
= totalram_pages
;
3560 val
->sharedram
= global_page_state(NR_SHMEM
);
3561 val
->freeram
= global_page_state(NR_FREE_PAGES
);
3562 val
->bufferram
= nr_blockdev_pages();
3563 val
->totalhigh
= totalhigh_pages
;
3564 val
->freehigh
= nr_free_highpages();
3565 val
->mem_unit
= PAGE_SIZE
;
3568 EXPORT_SYMBOL(si_meminfo
);
3571 void si_meminfo_node(struct sysinfo
*val
, int nid
)
3573 int zone_type
; /* needs to be signed */
3574 unsigned long managed_pages
= 0;
3575 pg_data_t
*pgdat
= NODE_DATA(nid
);
3577 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++)
3578 managed_pages
+= pgdat
->node_zones
[zone_type
].managed_pages
;
3579 val
->totalram
= managed_pages
;
3580 val
->sharedram
= node_page_state(nid
, NR_SHMEM
);
3581 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
3582 #ifdef CONFIG_HIGHMEM
3583 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].managed_pages
;
3584 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
3590 val
->mem_unit
= PAGE_SIZE
;
3595 * Determine whether the node should be displayed or not, depending on whether
3596 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
3598 bool skip_free_areas_node(unsigned int flags
, int nid
)
3601 unsigned int cpuset_mems_cookie
;
3603 if (!(flags
& SHOW_MEM_FILTER_NODES
))
3607 cpuset_mems_cookie
= read_mems_allowed_begin();
3608 ret
= !node_isset(nid
, cpuset_current_mems_allowed
);
3609 } while (read_mems_allowed_retry(cpuset_mems_cookie
));
3614 #define K(x) ((x) << (PAGE_SHIFT-10))
3616 static void show_migration_types(unsigned char type
)
3618 static const char types
[MIGRATE_TYPES
] = {
3619 [MIGRATE_UNMOVABLE
] = 'U',
3620 [MIGRATE_RECLAIMABLE
] = 'E',
3621 [MIGRATE_MOVABLE
] = 'M',
3622 [MIGRATE_RESERVE
] = 'R',
3624 [MIGRATE_CMA
] = 'C',
3626 #ifdef CONFIG_MEMORY_ISOLATION
3627 [MIGRATE_ISOLATE
] = 'I',
3630 char tmp
[MIGRATE_TYPES
+ 1];
3634 for (i
= 0; i
< MIGRATE_TYPES
; i
++) {
3635 if (type
& (1 << i
))
3640 printk("(%s) ", tmp
);
3644 * Show free area list (used inside shift_scroll-lock stuff)
3645 * We also calculate the percentage fragmentation. We do this by counting the
3646 * memory on each free list with the exception of the first item on the list.
3649 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
3652 void show_free_areas(unsigned int filter
)
3654 unsigned long free_pcp
= 0;
3658 for_each_populated_zone(zone
) {
3659 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3662 for_each_online_cpu(cpu
)
3663 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
3666 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3667 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3668 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
3669 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3670 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3671 " free:%lu free_pcp:%lu free_cma:%lu\n",
3672 global_page_state(NR_ACTIVE_ANON
),
3673 global_page_state(NR_INACTIVE_ANON
),
3674 global_page_state(NR_ISOLATED_ANON
),
3675 global_page_state(NR_ACTIVE_FILE
),
3676 global_page_state(NR_INACTIVE_FILE
),
3677 global_page_state(NR_ISOLATED_FILE
),
3678 global_page_state(NR_UNEVICTABLE
),
3679 global_page_state(NR_FILE_DIRTY
),
3680 global_page_state(NR_WRITEBACK
),
3681 global_page_state(NR_UNSTABLE_NFS
),
3682 global_page_state(NR_SLAB_RECLAIMABLE
),
3683 global_page_state(NR_SLAB_UNRECLAIMABLE
),
3684 global_page_state(NR_FILE_MAPPED
),
3685 global_page_state(NR_SHMEM
),
3686 global_page_state(NR_PAGETABLE
),
3687 global_page_state(NR_BOUNCE
),
3688 global_page_state(NR_FREE_PAGES
),
3690 global_page_state(NR_FREE_CMA_PAGES
));
3692 for_each_populated_zone(zone
) {
3695 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3699 for_each_online_cpu(cpu
)
3700 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
3708 " active_anon:%lukB"
3709 " inactive_anon:%lukB"
3710 " active_file:%lukB"
3711 " inactive_file:%lukB"
3712 " unevictable:%lukB"
3713 " isolated(anon):%lukB"
3714 " isolated(file):%lukB"
3722 " slab_reclaimable:%lukB"
3723 " slab_unreclaimable:%lukB"
3724 " kernel_stack:%lukB"
3731 " writeback_tmp:%lukB"
3732 " pages_scanned:%lu"
3733 " all_unreclaimable? %s"
3736 K(zone_page_state(zone
, NR_FREE_PAGES
)),
3737 K(min_wmark_pages(zone
)),
3738 K(low_wmark_pages(zone
)),
3739 K(high_wmark_pages(zone
)),
3740 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
3741 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
3742 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
3743 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
3744 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
3745 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
3746 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
3747 K(zone
->present_pages
),
3748 K(zone
->managed_pages
),
3749 K(zone_page_state(zone
, NR_MLOCK
)),
3750 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
3751 K(zone_page_state(zone
, NR_WRITEBACK
)),
3752 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
3753 K(zone_page_state(zone
, NR_SHMEM
)),
3754 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
3755 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
3756 zone_page_state(zone
, NR_KERNEL_STACK
) *
3758 K(zone_page_state(zone
, NR_PAGETABLE
)),
3759 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
3760 K(zone_page_state(zone
, NR_BOUNCE
)),
3762 K(this_cpu_read(zone
->pageset
->pcp
.count
)),
3763 K(zone_page_state(zone
, NR_FREE_CMA_PAGES
)),
3764 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
3765 K(zone_page_state(zone
, NR_PAGES_SCANNED
)),
3766 (!zone_reclaimable(zone
) ? "yes" : "no")
3768 printk("lowmem_reserve[]:");
3769 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3770 printk(" %ld", zone
->lowmem_reserve
[i
]);
3774 for_each_populated_zone(zone
) {
3775 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
3776 unsigned char types
[MAX_ORDER
];
3778 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3781 printk("%s: ", zone
->name
);
3783 spin_lock_irqsave(&zone
->lock
, flags
);
3784 for (order
= 0; order
< MAX_ORDER
; order
++) {
3785 struct free_area
*area
= &zone
->free_area
[order
];
3788 nr
[order
] = area
->nr_free
;
3789 total
+= nr
[order
] << order
;
3792 for (type
= 0; type
< MIGRATE_TYPES
; type
++) {
3793 if (!list_empty(&area
->free_list
[type
]))
3794 types
[order
] |= 1 << type
;
3797 spin_unlock_irqrestore(&zone
->lock
, flags
);
3798 for (order
= 0; order
< MAX_ORDER
; order
++) {
3799 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
3801 show_migration_types(types
[order
]);
3803 printk("= %lukB\n", K(total
));
3806 hugetlb_show_meminfo();
3808 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
3810 show_swap_cache_info();
3813 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
3815 zoneref
->zone
= zone
;
3816 zoneref
->zone_idx
= zone_idx(zone
);
3820 * Builds allocation fallback zone lists.
3822 * Add all populated zones of a node to the zonelist.
3824 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
3828 enum zone_type zone_type
= MAX_NR_ZONES
;
3832 zone
= pgdat
->node_zones
+ zone_type
;
3833 if (populated_zone(zone
)) {
3834 zoneref_set_zone(zone
,
3835 &zonelist
->_zonerefs
[nr_zones
++]);
3836 check_highest_zone(zone_type
);
3838 } while (zone_type
);
3846 * 0 = automatic detection of better ordering.
3847 * 1 = order by ([node] distance, -zonetype)
3848 * 2 = order by (-zonetype, [node] distance)
3850 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3851 * the same zonelist. So only NUMA can configure this param.
3853 #define ZONELIST_ORDER_DEFAULT 0
3854 #define ZONELIST_ORDER_NODE 1
3855 #define ZONELIST_ORDER_ZONE 2
3857 /* zonelist order in the kernel.
3858 * set_zonelist_order() will set this to NODE or ZONE.
3860 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3861 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
3865 /* The value user specified ....changed by config */
3866 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3867 /* string for sysctl */
3868 #define NUMA_ZONELIST_ORDER_LEN 16
3869 char numa_zonelist_order
[16] = "default";
3872 * interface for configure zonelist ordering.
3873 * command line option "numa_zonelist_order"
3874 * = "[dD]efault - default, automatic configuration.
3875 * = "[nN]ode - order by node locality, then by zone within node
3876 * = "[zZ]one - order by zone, then by locality within zone
3879 static int __parse_numa_zonelist_order(char *s
)
3881 if (*s
== 'd' || *s
== 'D') {
3882 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3883 } else if (*s
== 'n' || *s
== 'N') {
3884 user_zonelist_order
= ZONELIST_ORDER_NODE
;
3885 } else if (*s
== 'z' || *s
== 'Z') {
3886 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
3889 "Ignoring invalid numa_zonelist_order value: "
3896 static __init
int setup_numa_zonelist_order(char *s
)
3903 ret
= __parse_numa_zonelist_order(s
);
3905 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
3909 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
3912 * sysctl handler for numa_zonelist_order
3914 int numa_zonelist_order_handler(struct ctl_table
*table
, int write
,
3915 void __user
*buffer
, size_t *length
,
3918 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
3920 static DEFINE_MUTEX(zl_order_mutex
);
3922 mutex_lock(&zl_order_mutex
);
3924 if (strlen((char *)table
->data
) >= NUMA_ZONELIST_ORDER_LEN
) {
3928 strcpy(saved_string
, (char *)table
->data
);
3930 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
3934 int oldval
= user_zonelist_order
;
3936 ret
= __parse_numa_zonelist_order((char *)table
->data
);
3939 * bogus value. restore saved string
3941 strncpy((char *)table
->data
, saved_string
,
3942 NUMA_ZONELIST_ORDER_LEN
);
3943 user_zonelist_order
= oldval
;
3944 } else if (oldval
!= user_zonelist_order
) {
3945 mutex_lock(&zonelists_mutex
);
3946 build_all_zonelists(NULL
, NULL
);
3947 mutex_unlock(&zonelists_mutex
);
3951 mutex_unlock(&zl_order_mutex
);
3956 #define MAX_NODE_LOAD (nr_online_nodes)
3957 static int node_load
[MAX_NUMNODES
];
3960 * find_next_best_node - find the next node that should appear in a given node's fallback list
3961 * @node: node whose fallback list we're appending
3962 * @used_node_mask: nodemask_t of already used nodes
3964 * We use a number of factors to determine which is the next node that should
3965 * appear on a given node's fallback list. The node should not have appeared
3966 * already in @node's fallback list, and it should be the next closest node
3967 * according to the distance array (which contains arbitrary distance values
3968 * from each node to each node in the system), and should also prefer nodes
3969 * with no CPUs, since presumably they'll have very little allocation pressure
3970 * on them otherwise.
3971 * It returns -1 if no node is found.
3973 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
3976 int min_val
= INT_MAX
;
3977 int best_node
= NUMA_NO_NODE
;
3978 const struct cpumask
*tmp
= cpumask_of_node(0);
3980 /* Use the local node if we haven't already */
3981 if (!node_isset(node
, *used_node_mask
)) {
3982 node_set(node
, *used_node_mask
);
3986 for_each_node_state(n
, N_MEMORY
) {
3988 /* Don't want a node to appear more than once */
3989 if (node_isset(n
, *used_node_mask
))
3992 /* Use the distance array to find the distance */
3993 val
= node_distance(node
, n
);
3995 /* Penalize nodes under us ("prefer the next node") */
3998 /* Give preference to headless and unused nodes */
3999 tmp
= cpumask_of_node(n
);
4000 if (!cpumask_empty(tmp
))
4001 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
4003 /* Slight preference for less loaded node */
4004 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
4005 val
+= node_load
[n
];
4007 if (val
< min_val
) {
4014 node_set(best_node
, *used_node_mask
);
4021 * Build zonelists ordered by node and zones within node.
4022 * This results in maximum locality--normal zone overflows into local
4023 * DMA zone, if any--but risks exhausting DMA zone.
4025 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
4028 struct zonelist
*zonelist
;
4030 zonelist
= &pgdat
->node_zonelists
[0];
4031 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
4033 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4034 zonelist
->_zonerefs
[j
].zone
= NULL
;
4035 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4039 * Build gfp_thisnode zonelists
4041 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
4044 struct zonelist
*zonelist
;
4046 zonelist
= &pgdat
->node_zonelists
[1];
4047 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4048 zonelist
->_zonerefs
[j
].zone
= NULL
;
4049 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4053 * Build zonelists ordered by zone and nodes within zones.
4054 * This results in conserving DMA zone[s] until all Normal memory is
4055 * exhausted, but results in overflowing to remote node while memory
4056 * may still exist in local DMA zone.
4058 static int node_order
[MAX_NUMNODES
];
4060 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
4063 int zone_type
; /* needs to be signed */
4065 struct zonelist
*zonelist
;
4067 zonelist
= &pgdat
->node_zonelists
[0];
4069 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
4070 for (j
= 0; j
< nr_nodes
; j
++) {
4071 node
= node_order
[j
];
4072 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
4073 if (populated_zone(z
)) {
4075 &zonelist
->_zonerefs
[pos
++]);
4076 check_highest_zone(zone_type
);
4080 zonelist
->_zonerefs
[pos
].zone
= NULL
;
4081 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
4084 #if defined(CONFIG_64BIT)
4086 * Devices that require DMA32/DMA are relatively rare and do not justify a
4087 * penalty to every machine in case the specialised case applies. Default
4088 * to Node-ordering on 64-bit NUMA machines
4090 static int default_zonelist_order(void)
4092 return ZONELIST_ORDER_NODE
;
4096 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4097 * by the kernel. If processes running on node 0 deplete the low memory zone
4098 * then reclaim will occur more frequency increasing stalls and potentially
4099 * be easier to OOM if a large percentage of the zone is under writeback or
4100 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4101 * Hence, default to zone ordering on 32-bit.
4103 static int default_zonelist_order(void)
4105 return ZONELIST_ORDER_ZONE
;
4107 #endif /* CONFIG_64BIT */
4109 static void set_zonelist_order(void)
4111 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
4112 current_zonelist_order
= default_zonelist_order();
4114 current_zonelist_order
= user_zonelist_order
;
4117 static void build_zonelists(pg_data_t
*pgdat
)
4121 nodemask_t used_mask
;
4122 int local_node
, prev_node
;
4123 struct zonelist
*zonelist
;
4124 int order
= current_zonelist_order
;
4126 /* initialize zonelists */
4127 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
4128 zonelist
= pgdat
->node_zonelists
+ i
;
4129 zonelist
->_zonerefs
[0].zone
= NULL
;
4130 zonelist
->_zonerefs
[0].zone_idx
= 0;
4133 /* NUMA-aware ordering of nodes */
4134 local_node
= pgdat
->node_id
;
4135 load
= nr_online_nodes
;
4136 prev_node
= local_node
;
4137 nodes_clear(used_mask
);
4139 memset(node_order
, 0, sizeof(node_order
));
4142 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
4144 * We don't want to pressure a particular node.
4145 * So adding penalty to the first node in same
4146 * distance group to make it round-robin.
4148 if (node_distance(local_node
, node
) !=
4149 node_distance(local_node
, prev_node
))
4150 node_load
[node
] = load
;
4154 if (order
== ZONELIST_ORDER_NODE
)
4155 build_zonelists_in_node_order(pgdat
, node
);
4157 node_order
[j
++] = node
; /* remember order */
4160 if (order
== ZONELIST_ORDER_ZONE
) {
4161 /* calculate node order -- i.e., DMA last! */
4162 build_zonelists_in_zone_order(pgdat
, j
);
4165 build_thisnode_zonelists(pgdat
);
4168 /* Construct the zonelist performance cache - see further mmzone.h */
4169 static void build_zonelist_cache(pg_data_t
*pgdat
)
4171 struct zonelist
*zonelist
;
4172 struct zonelist_cache
*zlc
;
4175 zonelist
= &pgdat
->node_zonelists
[0];
4176 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
4177 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
4178 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
4179 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
4182 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4184 * Return node id of node used for "local" allocations.
4185 * I.e., first node id of first zone in arg node's generic zonelist.
4186 * Used for initializing percpu 'numa_mem', which is used primarily
4187 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4189 int local_memory_node(int node
)
4193 (void)first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
4194 gfp_zone(GFP_KERNEL
),
4201 #else /* CONFIG_NUMA */
4203 static void set_zonelist_order(void)
4205 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
4208 static void build_zonelists(pg_data_t
*pgdat
)
4210 int node
, local_node
;
4212 struct zonelist
*zonelist
;
4214 local_node
= pgdat
->node_id
;
4216 zonelist
= &pgdat
->node_zonelists
[0];
4217 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4220 * Now we build the zonelist so that it contains the zones
4221 * of all the other nodes.
4222 * We don't want to pressure a particular node, so when
4223 * building the zones for node N, we make sure that the
4224 * zones coming right after the local ones are those from
4225 * node N+1 (modulo N)
4227 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
4228 if (!node_online(node
))
4230 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4232 for (node
= 0; node
< local_node
; node
++) {
4233 if (!node_online(node
))
4235 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4238 zonelist
->_zonerefs
[j
].zone
= NULL
;
4239 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4242 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
4243 static void build_zonelist_cache(pg_data_t
*pgdat
)
4245 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
4248 #endif /* CONFIG_NUMA */
4251 * Boot pageset table. One per cpu which is going to be used for all
4252 * zones and all nodes. The parameters will be set in such a way
4253 * that an item put on a list will immediately be handed over to
4254 * the buddy list. This is safe since pageset manipulation is done
4255 * with interrupts disabled.
4257 * The boot_pagesets must be kept even after bootup is complete for
4258 * unused processors and/or zones. They do play a role for bootstrapping
4259 * hotplugged processors.
4261 * zoneinfo_show() and maybe other functions do
4262 * not check if the processor is online before following the pageset pointer.
4263 * Other parts of the kernel may not check if the zone is available.
4265 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
4266 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
4267 static void setup_zone_pageset(struct zone
*zone
);
4270 * Global mutex to protect against size modification of zonelists
4271 * as well as to serialize pageset setup for the new populated zone.
4273 DEFINE_MUTEX(zonelists_mutex
);
4275 /* return values int ....just for stop_machine() */
4276 static int __build_all_zonelists(void *data
)
4280 pg_data_t
*self
= data
;
4283 memset(node_load
, 0, sizeof(node_load
));
4286 if (self
&& !node_online(self
->node_id
)) {
4287 build_zonelists(self
);
4288 build_zonelist_cache(self
);
4291 for_each_online_node(nid
) {
4292 pg_data_t
*pgdat
= NODE_DATA(nid
);
4294 build_zonelists(pgdat
);
4295 build_zonelist_cache(pgdat
);
4299 * Initialize the boot_pagesets that are going to be used
4300 * for bootstrapping processors. The real pagesets for
4301 * each zone will be allocated later when the per cpu
4302 * allocator is available.
4304 * boot_pagesets are used also for bootstrapping offline
4305 * cpus if the system is already booted because the pagesets
4306 * are needed to initialize allocators on a specific cpu too.
4307 * F.e. the percpu allocator needs the page allocator which
4308 * needs the percpu allocator in order to allocate its pagesets
4309 * (a chicken-egg dilemma).
4311 for_each_possible_cpu(cpu
) {
4312 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
4314 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4316 * We now know the "local memory node" for each node--
4317 * i.e., the node of the first zone in the generic zonelist.
4318 * Set up numa_mem percpu variable for on-line cpus. During
4319 * boot, only the boot cpu should be on-line; we'll init the
4320 * secondary cpus' numa_mem as they come on-line. During
4321 * node/memory hotplug, we'll fixup all on-line cpus.
4323 if (cpu_online(cpu
))
4324 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
4331 static noinline
void __init
4332 build_all_zonelists_init(void)
4334 __build_all_zonelists(NULL
);
4335 mminit_verify_zonelist();
4336 cpuset_init_current_mems_allowed();
4340 * Called with zonelists_mutex held always
4341 * unless system_state == SYSTEM_BOOTING.
4343 * __ref due to (1) call of __meminit annotated setup_zone_pageset
4344 * [we're only called with non-NULL zone through __meminit paths] and
4345 * (2) call of __init annotated helper build_all_zonelists_init
4346 * [protected by SYSTEM_BOOTING].
4348 void __ref
build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
)
4350 set_zonelist_order();
4352 if (system_state
== SYSTEM_BOOTING
) {
4353 build_all_zonelists_init();
4355 #ifdef CONFIG_MEMORY_HOTPLUG
4357 setup_zone_pageset(zone
);
4359 /* we have to stop all cpus to guarantee there is no user
4361 stop_machine(__build_all_zonelists
, pgdat
, NULL
);
4362 /* cpuset refresh routine should be here */
4364 vm_total_pages
= nr_free_pagecache_pages();
4366 * Disable grouping by mobility if the number of pages in the
4367 * system is too low to allow the mechanism to work. It would be
4368 * more accurate, but expensive to check per-zone. This check is
4369 * made on memory-hotadd so a system can start with mobility
4370 * disabled and enable it later
4372 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
4373 page_group_by_mobility_disabled
= 1;
4375 page_group_by_mobility_disabled
= 0;
4377 pr_info("Built %i zonelists in %s order, mobility grouping %s. "
4378 "Total pages: %ld\n",
4380 zonelist_order_name
[current_zonelist_order
],
4381 page_group_by_mobility_disabled
? "off" : "on",
4384 pr_info("Policy zone: %s\n", zone_names
[policy_zone
]);
4389 * Helper functions to size the waitqueue hash table.
4390 * Essentially these want to choose hash table sizes sufficiently
4391 * large so that collisions trying to wait on pages are rare.
4392 * But in fact, the number of active page waitqueues on typical
4393 * systems is ridiculously low, less than 200. So this is even
4394 * conservative, even though it seems large.
4396 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
4397 * waitqueues, i.e. the size of the waitq table given the number of pages.
4399 #define PAGES_PER_WAITQUEUE 256
4401 #ifndef CONFIG_MEMORY_HOTPLUG
4402 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
4404 unsigned long size
= 1;
4406 pages
/= PAGES_PER_WAITQUEUE
;
4408 while (size
< pages
)
4412 * Once we have dozens or even hundreds of threads sleeping
4413 * on IO we've got bigger problems than wait queue collision.
4414 * Limit the size of the wait table to a reasonable size.
4416 size
= min(size
, 4096UL);
4418 return max(size
, 4UL);
4422 * A zone's size might be changed by hot-add, so it is not possible to determine
4423 * a suitable size for its wait_table. So we use the maximum size now.
4425 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
4427 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
4428 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4429 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
4431 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4432 * or more by the traditional way. (See above). It equals:
4434 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
4435 * ia64(16K page size) : = ( 8G + 4M)byte.
4436 * powerpc (64K page size) : = (32G +16M)byte.
4438 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
4445 * This is an integer logarithm so that shifts can be used later
4446 * to extract the more random high bits from the multiplicative
4447 * hash function before the remainder is taken.
4449 static inline unsigned long wait_table_bits(unsigned long size
)
4455 * Check if a pageblock contains reserved pages
4457 static int pageblock_is_reserved(unsigned long start_pfn
, unsigned long end_pfn
)
4461 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
4462 if (!pfn_valid_within(pfn
) || PageReserved(pfn_to_page(pfn
)))
4469 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
4470 * of blocks reserved is based on min_wmark_pages(zone). The memory within
4471 * the reserve will tend to store contiguous free pages. Setting min_free_kbytes
4472 * higher will lead to a bigger reserve which will get freed as contiguous
4473 * blocks as reclaim kicks in
4475 static void setup_zone_migrate_reserve(struct zone
*zone
)
4477 unsigned long start_pfn
, pfn
, end_pfn
, block_end_pfn
;
4479 unsigned long block_migratetype
;
4484 * Get the start pfn, end pfn and the number of blocks to reserve
4485 * We have to be careful to be aligned to pageblock_nr_pages to
4486 * make sure that we always check pfn_valid for the first page in
4489 start_pfn
= zone
->zone_start_pfn
;
4490 end_pfn
= zone_end_pfn(zone
);
4491 start_pfn
= roundup(start_pfn
, pageblock_nr_pages
);
4492 reserve
= roundup(min_wmark_pages(zone
), pageblock_nr_pages
) >>
4496 * Reserve blocks are generally in place to help high-order atomic
4497 * allocations that are short-lived. A min_free_kbytes value that
4498 * would result in more than 2 reserve blocks for atomic allocations
4499 * is assumed to be in place to help anti-fragmentation for the
4500 * future allocation of hugepages at runtime.
4502 reserve
= min(2, reserve
);
4503 old_reserve
= zone
->nr_migrate_reserve_block
;
4505 /* When memory hot-add, we almost always need to do nothing */
4506 if (reserve
== old_reserve
)
4508 zone
->nr_migrate_reserve_block
= reserve
;
4510 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
4511 if (!early_page_nid_uninitialised(pfn
, zone_to_nid(zone
)))
4514 if (!pfn_valid(pfn
))
4516 page
= pfn_to_page(pfn
);
4518 /* Watch out for overlapping nodes */
4519 if (page_to_nid(page
) != zone_to_nid(zone
))
4522 block_migratetype
= get_pageblock_migratetype(page
);
4524 /* Only test what is necessary when the reserves are not met */
4527 * Blocks with reserved pages will never free, skip
4530 block_end_pfn
= min(pfn
+ pageblock_nr_pages
, end_pfn
);
4531 if (pageblock_is_reserved(pfn
, block_end_pfn
))
4534 /* If this block is reserved, account for it */
4535 if (block_migratetype
== MIGRATE_RESERVE
) {
4540 /* Suitable for reserving if this block is movable */
4541 if (block_migratetype
== MIGRATE_MOVABLE
) {
4542 set_pageblock_migratetype(page
,
4544 move_freepages_block(zone
, page
,
4549 } else if (!old_reserve
) {
4551 * At boot time we don't need to scan the whole zone
4552 * for turning off MIGRATE_RESERVE.
4558 * If the reserve is met and this is a previous reserved block,
4561 if (block_migratetype
== MIGRATE_RESERVE
) {
4562 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4563 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
4569 * Initially all pages are reserved - free ones are freed
4570 * up by free_all_bootmem() once the early boot process is
4571 * done. Non-atomic initialization, single-pass.
4573 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
4574 unsigned long start_pfn
, enum memmap_context context
)
4576 pg_data_t
*pgdat
= NODE_DATA(nid
);
4577 unsigned long end_pfn
= start_pfn
+ size
;
4580 unsigned long nr_initialised
= 0;
4582 if (highest_memmap_pfn
< end_pfn
- 1)
4583 highest_memmap_pfn
= end_pfn
- 1;
4585 z
= &pgdat
->node_zones
[zone
];
4586 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
4588 * There can be holes in boot-time mem_map[]s
4589 * handed to this function. They do not
4590 * exist on hotplugged memory.
4592 if (context
== MEMMAP_EARLY
) {
4593 if (!early_pfn_valid(pfn
))
4595 if (!early_pfn_in_nid(pfn
, nid
))
4597 if (!update_defer_init(pgdat
, pfn
, end_pfn
,
4603 * Mark the block movable so that blocks are reserved for
4604 * movable at startup. This will force kernel allocations
4605 * to reserve their blocks rather than leaking throughout
4606 * the address space during boot when many long-lived
4607 * kernel allocations are made. Later some blocks near
4608 * the start are marked MIGRATE_RESERVE by
4609 * setup_zone_migrate_reserve()
4611 * bitmap is created for zone's valid pfn range. but memmap
4612 * can be created for invalid pages (for alignment)
4613 * check here not to call set_pageblock_migratetype() against
4616 if (!(pfn
& (pageblock_nr_pages
- 1))) {
4617 struct page
*page
= pfn_to_page(pfn
);
4619 __init_single_page(page
, pfn
, zone
, nid
);
4620 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4622 __init_single_pfn(pfn
, zone
, nid
);
4627 static void __meminit
zone_init_free_lists(struct zone
*zone
)
4629 unsigned int order
, t
;
4630 for_each_migratetype_order(order
, t
) {
4631 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
4632 zone
->free_area
[order
].nr_free
= 0;
4636 #ifndef __HAVE_ARCH_MEMMAP_INIT
4637 #define memmap_init(size, nid, zone, start_pfn) \
4638 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4641 static int zone_batchsize(struct zone
*zone
)
4647 * The per-cpu-pages pools are set to around 1000th of the
4648 * size of the zone. But no more than 1/2 of a meg.
4650 * OK, so we don't know how big the cache is. So guess.
4652 batch
= zone
->managed_pages
/ 1024;
4653 if (batch
* PAGE_SIZE
> 512 * 1024)
4654 batch
= (512 * 1024) / PAGE_SIZE
;
4655 batch
/= 4; /* We effectively *= 4 below */
4660 * Clamp the batch to a 2^n - 1 value. Having a power
4661 * of 2 value was found to be more likely to have
4662 * suboptimal cache aliasing properties in some cases.
4664 * For example if 2 tasks are alternately allocating
4665 * batches of pages, one task can end up with a lot
4666 * of pages of one half of the possible page colors
4667 * and the other with pages of the other colors.
4669 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
4674 /* The deferral and batching of frees should be suppressed under NOMMU
4677 * The problem is that NOMMU needs to be able to allocate large chunks
4678 * of contiguous memory as there's no hardware page translation to
4679 * assemble apparent contiguous memory from discontiguous pages.
4681 * Queueing large contiguous runs of pages for batching, however,
4682 * causes the pages to actually be freed in smaller chunks. As there
4683 * can be a significant delay between the individual batches being
4684 * recycled, this leads to the once large chunks of space being
4685 * fragmented and becoming unavailable for high-order allocations.
4692 * pcp->high and pcp->batch values are related and dependent on one another:
4693 * ->batch must never be higher then ->high.
4694 * The following function updates them in a safe manner without read side
4697 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4698 * those fields changing asynchronously (acording the the above rule).
4700 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4701 * outside of boot time (or some other assurance that no concurrent updaters
4704 static void pageset_update(struct per_cpu_pages
*pcp
, unsigned long high
,
4705 unsigned long batch
)
4707 /* start with a fail safe value for batch */
4711 /* Update high, then batch, in order */
4718 /* a companion to pageset_set_high() */
4719 static void pageset_set_batch(struct per_cpu_pageset
*p
, unsigned long batch
)
4721 pageset_update(&p
->pcp
, 6 * batch
, max(1UL, 1 * batch
));
4724 static void pageset_init(struct per_cpu_pageset
*p
)
4726 struct per_cpu_pages
*pcp
;
4729 memset(p
, 0, sizeof(*p
));
4733 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
4734 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
4737 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
4740 pageset_set_batch(p
, batch
);
4744 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
4745 * to the value high for the pageset p.
4747 static void pageset_set_high(struct per_cpu_pageset
*p
,
4750 unsigned long batch
= max(1UL, high
/ 4);
4751 if ((high
/ 4) > (PAGE_SHIFT
* 8))
4752 batch
= PAGE_SHIFT
* 8;
4754 pageset_update(&p
->pcp
, high
, batch
);
4757 static void pageset_set_high_and_batch(struct zone
*zone
,
4758 struct per_cpu_pageset
*pcp
)
4760 if (percpu_pagelist_fraction
)
4761 pageset_set_high(pcp
,
4762 (zone
->managed_pages
/
4763 percpu_pagelist_fraction
));
4765 pageset_set_batch(pcp
, zone_batchsize(zone
));
4768 static void __meminit
zone_pageset_init(struct zone
*zone
, int cpu
)
4770 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
4773 pageset_set_high_and_batch(zone
, pcp
);
4776 static void __meminit
setup_zone_pageset(struct zone
*zone
)
4779 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
4780 for_each_possible_cpu(cpu
)
4781 zone_pageset_init(zone
, cpu
);
4785 * Allocate per cpu pagesets and initialize them.
4786 * Before this call only boot pagesets were available.
4788 void __init
setup_per_cpu_pageset(void)
4792 for_each_populated_zone(zone
)
4793 setup_zone_pageset(zone
);
4796 static noinline __init_refok
4797 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
4803 * The per-page waitqueue mechanism uses hashed waitqueues
4806 zone
->wait_table_hash_nr_entries
=
4807 wait_table_hash_nr_entries(zone_size_pages
);
4808 zone
->wait_table_bits
=
4809 wait_table_bits(zone
->wait_table_hash_nr_entries
);
4810 alloc_size
= zone
->wait_table_hash_nr_entries
4811 * sizeof(wait_queue_head_t
);
4813 if (!slab_is_available()) {
4814 zone
->wait_table
= (wait_queue_head_t
*)
4815 memblock_virt_alloc_node_nopanic(
4816 alloc_size
, zone
->zone_pgdat
->node_id
);
4819 * This case means that a zone whose size was 0 gets new memory
4820 * via memory hot-add.
4821 * But it may be the case that a new node was hot-added. In
4822 * this case vmalloc() will not be able to use this new node's
4823 * memory - this wait_table must be initialized to use this new
4824 * node itself as well.
4825 * To use this new node's memory, further consideration will be
4828 zone
->wait_table
= vmalloc(alloc_size
);
4830 if (!zone
->wait_table
)
4833 for (i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
4834 init_waitqueue_head(zone
->wait_table
+ i
);
4839 static __meminit
void zone_pcp_init(struct zone
*zone
)
4842 * per cpu subsystem is not up at this point. The following code
4843 * relies on the ability of the linker to provide the
4844 * offset of a (static) per cpu variable into the per cpu area.
4846 zone
->pageset
= &boot_pageset
;
4848 if (populated_zone(zone
))
4849 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
4850 zone
->name
, zone
->present_pages
,
4851 zone_batchsize(zone
));
4854 int __meminit
init_currently_empty_zone(struct zone
*zone
,
4855 unsigned long zone_start_pfn
,
4857 enum memmap_context context
)
4859 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
4861 ret
= zone_wait_table_init(zone
, size
);
4864 pgdat
->nr_zones
= zone_idx(zone
) + 1;
4866 zone
->zone_start_pfn
= zone_start_pfn
;
4868 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
4869 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4871 (unsigned long)zone_idx(zone
),
4872 zone_start_pfn
, (zone_start_pfn
+ size
));
4874 zone_init_free_lists(zone
);
4879 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4880 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4883 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4885 int __meminit
__early_pfn_to_nid(unsigned long pfn
,
4886 struct mminit_pfnnid_cache
*state
)
4888 unsigned long start_pfn
, end_pfn
;
4891 if (state
->last_start
<= pfn
&& pfn
< state
->last_end
)
4892 return state
->last_nid
;
4894 nid
= memblock_search_pfn_nid(pfn
, &start_pfn
, &end_pfn
);
4896 state
->last_start
= start_pfn
;
4897 state
->last_end
= end_pfn
;
4898 state
->last_nid
= nid
;
4903 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4906 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
4907 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4908 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
4910 * If an architecture guarantees that all ranges registered contain no holes
4911 * and may be freed, this this function may be used instead of calling
4912 * memblock_free_early_nid() manually.
4914 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
4916 unsigned long start_pfn
, end_pfn
;
4919 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
4920 start_pfn
= min(start_pfn
, max_low_pfn
);
4921 end_pfn
= min(end_pfn
, max_low_pfn
);
4923 if (start_pfn
< end_pfn
)
4924 memblock_free_early_nid(PFN_PHYS(start_pfn
),
4925 (end_pfn
- start_pfn
) << PAGE_SHIFT
,
4931 * sparse_memory_present_with_active_regions - Call memory_present for each active range
4932 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4934 * If an architecture guarantees that all ranges registered contain no holes and may
4935 * be freed, this function may be used instead of calling memory_present() manually.
4937 void __init
sparse_memory_present_with_active_regions(int nid
)
4939 unsigned long start_pfn
, end_pfn
;
4942 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
4943 memory_present(this_nid
, start_pfn
, end_pfn
);
4947 * get_pfn_range_for_nid - Return the start and end page frames for a node
4948 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4949 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4950 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4952 * It returns the start and end page frame of a node based on information
4953 * provided by memblock_set_node(). If called for a node
4954 * with no available memory, a warning is printed and the start and end
4957 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
4958 unsigned long *start_pfn
, unsigned long *end_pfn
)
4960 unsigned long this_start_pfn
, this_end_pfn
;
4966 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
4967 *start_pfn
= min(*start_pfn
, this_start_pfn
);
4968 *end_pfn
= max(*end_pfn
, this_end_pfn
);
4971 if (*start_pfn
== -1UL)
4976 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4977 * assumption is made that zones within a node are ordered in monotonic
4978 * increasing memory addresses so that the "highest" populated zone is used
4980 static void __init
find_usable_zone_for_movable(void)
4983 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
4984 if (zone_index
== ZONE_MOVABLE
)
4987 if (arch_zone_highest_possible_pfn
[zone_index
] >
4988 arch_zone_lowest_possible_pfn
[zone_index
])
4992 VM_BUG_ON(zone_index
== -1);
4993 movable_zone
= zone_index
;
4997 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4998 * because it is sized independent of architecture. Unlike the other zones,
4999 * the starting point for ZONE_MOVABLE is not fixed. It may be different
5000 * in each node depending on the size of each node and how evenly kernelcore
5001 * is distributed. This helper function adjusts the zone ranges
5002 * provided by the architecture for a given node by using the end of the
5003 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
5004 * zones within a node are in order of monotonic increases memory addresses
5006 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
5007 unsigned long zone_type
,
5008 unsigned long node_start_pfn
,
5009 unsigned long node_end_pfn
,
5010 unsigned long *zone_start_pfn
,
5011 unsigned long *zone_end_pfn
)
5013 /* Only adjust if ZONE_MOVABLE is on this node */
5014 if (zone_movable_pfn
[nid
]) {
5015 /* Size ZONE_MOVABLE */
5016 if (zone_type
== ZONE_MOVABLE
) {
5017 *zone_start_pfn
= zone_movable_pfn
[nid
];
5018 *zone_end_pfn
= min(node_end_pfn
,
5019 arch_zone_highest_possible_pfn
[movable_zone
]);
5021 /* Adjust for ZONE_MOVABLE starting within this range */
5022 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
5023 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
5024 *zone_end_pfn
= zone_movable_pfn
[nid
];
5026 /* Check if this whole range is within ZONE_MOVABLE */
5027 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
5028 *zone_start_pfn
= *zone_end_pfn
;
5033 * Return the number of pages a zone spans in a node, including holes
5034 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
5036 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5037 unsigned long zone_type
,
5038 unsigned long node_start_pfn
,
5039 unsigned long node_end_pfn
,
5040 unsigned long *ignored
)
5042 unsigned long zone_start_pfn
, zone_end_pfn
;
5044 /* Get the start and end of the zone */
5045 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
5046 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
5047 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5048 node_start_pfn
, node_end_pfn
,
5049 &zone_start_pfn
, &zone_end_pfn
);
5051 /* Check that this node has pages within the zone's required range */
5052 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
5055 /* Move the zone boundaries inside the node if necessary */
5056 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
5057 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
5059 /* Return the spanned pages */
5060 return zone_end_pfn
- zone_start_pfn
;
5064 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
5065 * then all holes in the requested range will be accounted for.
5067 unsigned long __meminit
__absent_pages_in_range(int nid
,
5068 unsigned long range_start_pfn
,
5069 unsigned long range_end_pfn
)
5071 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
5072 unsigned long start_pfn
, end_pfn
;
5075 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5076 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
5077 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
5078 nr_absent
-= end_pfn
- start_pfn
;
5084 * absent_pages_in_range - Return number of page frames in holes within a range
5085 * @start_pfn: The start PFN to start searching for holes
5086 * @end_pfn: The end PFN to stop searching for holes
5088 * It returns the number of pages frames in memory holes within a range.
5090 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
5091 unsigned long end_pfn
)
5093 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
5096 /* Return the number of page frames in holes in a zone on a node */
5097 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5098 unsigned long zone_type
,
5099 unsigned long node_start_pfn
,
5100 unsigned long node_end_pfn
,
5101 unsigned long *ignored
)
5103 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
5104 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
5105 unsigned long zone_start_pfn
, zone_end_pfn
;
5107 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
5108 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
5110 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5111 node_start_pfn
, node_end_pfn
,
5112 &zone_start_pfn
, &zone_end_pfn
);
5113 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
5116 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5117 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5118 unsigned long zone_type
,
5119 unsigned long node_start_pfn
,
5120 unsigned long node_end_pfn
,
5121 unsigned long *zones_size
)
5123 return zones_size
[zone_type
];
5126 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5127 unsigned long zone_type
,
5128 unsigned long node_start_pfn
,
5129 unsigned long node_end_pfn
,
5130 unsigned long *zholes_size
)
5135 return zholes_size
[zone_type
];
5138 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5140 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
5141 unsigned long node_start_pfn
,
5142 unsigned long node_end_pfn
,
5143 unsigned long *zones_size
,
5144 unsigned long *zholes_size
)
5146 unsigned long realtotalpages
= 0, totalpages
= 0;
5149 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5150 struct zone
*zone
= pgdat
->node_zones
+ i
;
5151 unsigned long size
, real_size
;
5153 size
= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
5157 real_size
= size
- zone_absent_pages_in_node(pgdat
->node_id
, i
,
5158 node_start_pfn
, node_end_pfn
,
5160 zone
->spanned_pages
= size
;
5161 zone
->present_pages
= real_size
;
5164 realtotalpages
+= real_size
;
5167 pgdat
->node_spanned_pages
= totalpages
;
5168 pgdat
->node_present_pages
= realtotalpages
;
5169 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
5173 #ifndef CONFIG_SPARSEMEM
5175 * Calculate the size of the zone->blockflags rounded to an unsigned long
5176 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5177 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
5178 * round what is now in bits to nearest long in bits, then return it in
5181 static unsigned long __init
usemap_size(unsigned long zone_start_pfn
, unsigned long zonesize
)
5183 unsigned long usemapsize
;
5185 zonesize
+= zone_start_pfn
& (pageblock_nr_pages
-1);
5186 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
5187 usemapsize
= usemapsize
>> pageblock_order
;
5188 usemapsize
*= NR_PAGEBLOCK_BITS
;
5189 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
5191 return usemapsize
/ 8;
5194 static void __init
setup_usemap(struct pglist_data
*pgdat
,
5196 unsigned long zone_start_pfn
,
5197 unsigned long zonesize
)
5199 unsigned long usemapsize
= usemap_size(zone_start_pfn
, zonesize
);
5200 zone
->pageblock_flags
= NULL
;
5202 zone
->pageblock_flags
=
5203 memblock_virt_alloc_node_nopanic(usemapsize
,
5207 static inline void setup_usemap(struct pglist_data
*pgdat
, struct zone
*zone
,
5208 unsigned long zone_start_pfn
, unsigned long zonesize
) {}
5209 #endif /* CONFIG_SPARSEMEM */
5211 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
5213 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
5214 void __paginginit
set_pageblock_order(void)
5218 /* Check that pageblock_nr_pages has not already been setup */
5219 if (pageblock_order
)
5222 if (HPAGE_SHIFT
> PAGE_SHIFT
)
5223 order
= HUGETLB_PAGE_ORDER
;
5225 order
= MAX_ORDER
- 1;
5228 * Assume the largest contiguous order of interest is a huge page.
5229 * This value may be variable depending on boot parameters on IA64 and
5232 pageblock_order
= order
;
5234 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5237 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
5238 * is unused as pageblock_order is set at compile-time. See
5239 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5242 void __paginginit
set_pageblock_order(void)
5246 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5248 static unsigned long __paginginit
calc_memmap_size(unsigned long spanned_pages
,
5249 unsigned long present_pages
)
5251 unsigned long pages
= spanned_pages
;
5254 * Provide a more accurate estimation if there are holes within
5255 * the zone and SPARSEMEM is in use. If there are holes within the
5256 * zone, each populated memory region may cost us one or two extra
5257 * memmap pages due to alignment because memmap pages for each
5258 * populated regions may not naturally algined on page boundary.
5259 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5261 if (spanned_pages
> present_pages
+ (present_pages
>> 4) &&
5262 IS_ENABLED(CONFIG_SPARSEMEM
))
5263 pages
= present_pages
;
5265 return PAGE_ALIGN(pages
* sizeof(struct page
)) >> PAGE_SHIFT
;
5269 * Set up the zone data structures:
5270 * - mark all pages reserved
5271 * - mark all memory queues empty
5272 * - clear the memory bitmaps
5274 * NOTE: pgdat should get zeroed by caller.
5276 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
5277 unsigned long node_start_pfn
, unsigned long node_end_pfn
)
5280 int nid
= pgdat
->node_id
;
5281 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
5284 pgdat_resize_init(pgdat
);
5285 #ifdef CONFIG_NUMA_BALANCING
5286 spin_lock_init(&pgdat
->numabalancing_migrate_lock
);
5287 pgdat
->numabalancing_migrate_nr_pages
= 0;
5288 pgdat
->numabalancing_migrate_next_window
= jiffies
;
5290 init_waitqueue_head(&pgdat
->kswapd_wait
);
5291 init_waitqueue_head(&pgdat
->pfmemalloc_wait
);
5292 pgdat_page_ext_init(pgdat
);
5294 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5295 struct zone
*zone
= pgdat
->node_zones
+ j
;
5296 unsigned long size
, realsize
, freesize
, memmap_pages
;
5298 size
= zone
->spanned_pages
;
5299 realsize
= freesize
= zone
->present_pages
;
5302 * Adjust freesize so that it accounts for how much memory
5303 * is used by this zone for memmap. This affects the watermark
5304 * and per-cpu initialisations
5306 memmap_pages
= calc_memmap_size(size
, realsize
);
5307 if (!is_highmem_idx(j
)) {
5308 if (freesize
>= memmap_pages
) {
5309 freesize
-= memmap_pages
;
5312 " %s zone: %lu pages used for memmap\n",
5313 zone_names
[j
], memmap_pages
);
5316 " %s zone: %lu pages exceeds freesize %lu\n",
5317 zone_names
[j
], memmap_pages
, freesize
);
5320 /* Account for reserved pages */
5321 if (j
== 0 && freesize
> dma_reserve
) {
5322 freesize
-= dma_reserve
;
5323 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
5324 zone_names
[0], dma_reserve
);
5327 if (!is_highmem_idx(j
))
5328 nr_kernel_pages
+= freesize
;
5329 /* Charge for highmem memmap if there are enough kernel pages */
5330 else if (nr_kernel_pages
> memmap_pages
* 2)
5331 nr_kernel_pages
-= memmap_pages
;
5332 nr_all_pages
+= freesize
;
5335 * Set an approximate value for lowmem here, it will be adjusted
5336 * when the bootmem allocator frees pages into the buddy system.
5337 * And all highmem pages will be managed by the buddy system.
5339 zone
->managed_pages
= is_highmem_idx(j
) ? realsize
: freesize
;
5342 zone
->min_unmapped_pages
= (freesize
*sysctl_min_unmapped_ratio
)
5344 zone
->min_slab_pages
= (freesize
* sysctl_min_slab_ratio
) / 100;
5346 zone
->name
= zone_names
[j
];
5347 spin_lock_init(&zone
->lock
);
5348 spin_lock_init(&zone
->lru_lock
);
5349 zone_seqlock_init(zone
);
5350 zone
->zone_pgdat
= pgdat
;
5351 zone_pcp_init(zone
);
5353 /* For bootup, initialized properly in watermark setup */
5354 mod_zone_page_state(zone
, NR_ALLOC_BATCH
, zone
->managed_pages
);
5356 lruvec_init(&zone
->lruvec
);
5360 set_pageblock_order();
5361 setup_usemap(pgdat
, zone
, zone_start_pfn
, size
);
5362 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
5363 size
, MEMMAP_EARLY
);
5365 memmap_init(size
, nid
, j
, zone_start_pfn
);
5366 zone_start_pfn
+= size
;
5370 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
5372 /* Skip empty nodes */
5373 if (!pgdat
->node_spanned_pages
)
5376 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5377 /* ia64 gets its own node_mem_map, before this, without bootmem */
5378 if (!pgdat
->node_mem_map
) {
5379 unsigned long size
, start
, end
;
5383 * The zone's endpoints aren't required to be MAX_ORDER
5384 * aligned but the node_mem_map endpoints must be in order
5385 * for the buddy allocator to function correctly.
5387 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
5388 end
= pgdat_end_pfn(pgdat
);
5389 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
5390 size
= (end
- start
) * sizeof(struct page
);
5391 map
= alloc_remap(pgdat
->node_id
, size
);
5393 map
= memblock_virt_alloc_node_nopanic(size
,
5395 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
5397 #ifndef CONFIG_NEED_MULTIPLE_NODES
5399 * With no DISCONTIG, the global mem_map is just set as node 0's
5401 if (pgdat
== NODE_DATA(0)) {
5402 mem_map
= NODE_DATA(0)->node_mem_map
;
5403 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5404 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
5405 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
5406 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5409 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
5412 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
5413 unsigned long node_start_pfn
, unsigned long *zholes_size
)
5415 pg_data_t
*pgdat
= NODE_DATA(nid
);
5416 unsigned long start_pfn
= 0;
5417 unsigned long end_pfn
= 0;
5419 /* pg_data_t should be reset to zero when it's allocated */
5420 WARN_ON(pgdat
->nr_zones
|| pgdat
->classzone_idx
);
5422 reset_deferred_meminit(pgdat
);
5423 pgdat
->node_id
= nid
;
5424 pgdat
->node_start_pfn
= node_start_pfn
;
5425 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5426 get_pfn_range_for_nid(nid
, &start_pfn
, &end_pfn
);
5427 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid
,
5428 (u64
)start_pfn
<< PAGE_SHIFT
, ((u64
)end_pfn
<< PAGE_SHIFT
) - 1);
5430 calculate_node_totalpages(pgdat
, start_pfn
, end_pfn
,
5431 zones_size
, zholes_size
);
5433 alloc_node_mem_map(pgdat
);
5434 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5435 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5436 nid
, (unsigned long)pgdat
,
5437 (unsigned long)pgdat
->node_mem_map
);
5440 free_area_init_core(pgdat
, start_pfn
, end_pfn
);
5443 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5445 #if MAX_NUMNODES > 1
5447 * Figure out the number of possible node ids.
5449 void __init
setup_nr_node_ids(void)
5452 unsigned int highest
= 0;
5454 for_each_node_mask(node
, node_possible_map
)
5456 nr_node_ids
= highest
+ 1;
5461 * node_map_pfn_alignment - determine the maximum internode alignment
5463 * This function should be called after node map is populated and sorted.
5464 * It calculates the maximum power of two alignment which can distinguish
5467 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5468 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
5469 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
5470 * shifted, 1GiB is enough and this function will indicate so.
5472 * This is used to test whether pfn -> nid mapping of the chosen memory
5473 * model has fine enough granularity to avoid incorrect mapping for the
5474 * populated node map.
5476 * Returns the determined alignment in pfn's. 0 if there is no alignment
5477 * requirement (single node).
5479 unsigned long __init
node_map_pfn_alignment(void)
5481 unsigned long accl_mask
= 0, last_end
= 0;
5482 unsigned long start
, end
, mask
;
5486 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
5487 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
5494 * Start with a mask granular enough to pin-point to the
5495 * start pfn and tick off bits one-by-one until it becomes
5496 * too coarse to separate the current node from the last.
5498 mask
= ~((1 << __ffs(start
)) - 1);
5499 while (mask
&& last_end
<= (start
& (mask
<< 1)))
5502 /* accumulate all internode masks */
5506 /* convert mask to number of pages */
5507 return ~accl_mask
+ 1;
5510 /* Find the lowest pfn for a node */
5511 static unsigned long __init
find_min_pfn_for_node(int nid
)
5513 unsigned long min_pfn
= ULONG_MAX
;
5514 unsigned long start_pfn
;
5517 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
5518 min_pfn
= min(min_pfn
, start_pfn
);
5520 if (min_pfn
== ULONG_MAX
) {
5522 "Could not find start_pfn for node %d\n", nid
);
5530 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5532 * It returns the minimum PFN based on information provided via
5533 * memblock_set_node().
5535 unsigned long __init
find_min_pfn_with_active_regions(void)
5537 return find_min_pfn_for_node(MAX_NUMNODES
);
5541 * early_calculate_totalpages()
5542 * Sum pages in active regions for movable zone.
5543 * Populate N_MEMORY for calculating usable_nodes.
5545 static unsigned long __init
early_calculate_totalpages(void)
5547 unsigned long totalpages
= 0;
5548 unsigned long start_pfn
, end_pfn
;
5551 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
5552 unsigned long pages
= end_pfn
- start_pfn
;
5554 totalpages
+= pages
;
5556 node_set_state(nid
, N_MEMORY
);
5562 * Find the PFN the Movable zone begins in each node. Kernel memory
5563 * is spread evenly between nodes as long as the nodes have enough
5564 * memory. When they don't, some nodes will have more kernelcore than
5567 static void __init
find_zone_movable_pfns_for_nodes(void)
5570 unsigned long usable_startpfn
;
5571 unsigned long kernelcore_node
, kernelcore_remaining
;
5572 /* save the state before borrow the nodemask */
5573 nodemask_t saved_node_state
= node_states
[N_MEMORY
];
5574 unsigned long totalpages
= early_calculate_totalpages();
5575 int usable_nodes
= nodes_weight(node_states
[N_MEMORY
]);
5576 struct memblock_region
*r
;
5578 /* Need to find movable_zone earlier when movable_node is specified. */
5579 find_usable_zone_for_movable();
5582 * If movable_node is specified, ignore kernelcore and movablecore
5585 if (movable_node_is_enabled()) {
5586 for_each_memblock(memory
, r
) {
5587 if (!memblock_is_hotpluggable(r
))
5592 usable_startpfn
= PFN_DOWN(r
->base
);
5593 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
5594 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
5602 * If movablecore=nn[KMG] was specified, calculate what size of
5603 * kernelcore that corresponds so that memory usable for
5604 * any allocation type is evenly spread. If both kernelcore
5605 * and movablecore are specified, then the value of kernelcore
5606 * will be used for required_kernelcore if it's greater than
5607 * what movablecore would have allowed.
5609 if (required_movablecore
) {
5610 unsigned long corepages
;
5613 * Round-up so that ZONE_MOVABLE is at least as large as what
5614 * was requested by the user
5616 required_movablecore
=
5617 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
5618 corepages
= totalpages
- required_movablecore
;
5620 required_kernelcore
= max(required_kernelcore
, corepages
);
5623 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
5624 if (!required_kernelcore
)
5627 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5628 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
5631 /* Spread kernelcore memory as evenly as possible throughout nodes */
5632 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5633 for_each_node_state(nid
, N_MEMORY
) {
5634 unsigned long start_pfn
, end_pfn
;
5637 * Recalculate kernelcore_node if the division per node
5638 * now exceeds what is necessary to satisfy the requested
5639 * amount of memory for the kernel
5641 if (required_kernelcore
< kernelcore_node
)
5642 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5645 * As the map is walked, we track how much memory is usable
5646 * by the kernel using kernelcore_remaining. When it is
5647 * 0, the rest of the node is usable by ZONE_MOVABLE
5649 kernelcore_remaining
= kernelcore_node
;
5651 /* Go through each range of PFNs within this node */
5652 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5653 unsigned long size_pages
;
5655 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
5656 if (start_pfn
>= end_pfn
)
5659 /* Account for what is only usable for kernelcore */
5660 if (start_pfn
< usable_startpfn
) {
5661 unsigned long kernel_pages
;
5662 kernel_pages
= min(end_pfn
, usable_startpfn
)
5665 kernelcore_remaining
-= min(kernel_pages
,
5666 kernelcore_remaining
);
5667 required_kernelcore
-= min(kernel_pages
,
5668 required_kernelcore
);
5670 /* Continue if range is now fully accounted */
5671 if (end_pfn
<= usable_startpfn
) {
5674 * Push zone_movable_pfn to the end so
5675 * that if we have to rebalance
5676 * kernelcore across nodes, we will
5677 * not double account here
5679 zone_movable_pfn
[nid
] = end_pfn
;
5682 start_pfn
= usable_startpfn
;
5686 * The usable PFN range for ZONE_MOVABLE is from
5687 * start_pfn->end_pfn. Calculate size_pages as the
5688 * number of pages used as kernelcore
5690 size_pages
= end_pfn
- start_pfn
;
5691 if (size_pages
> kernelcore_remaining
)
5692 size_pages
= kernelcore_remaining
;
5693 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
5696 * Some kernelcore has been met, update counts and
5697 * break if the kernelcore for this node has been
5700 required_kernelcore
-= min(required_kernelcore
,
5702 kernelcore_remaining
-= size_pages
;
5703 if (!kernelcore_remaining
)
5709 * If there is still required_kernelcore, we do another pass with one
5710 * less node in the count. This will push zone_movable_pfn[nid] further
5711 * along on the nodes that still have memory until kernelcore is
5715 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
5719 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5720 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
5721 zone_movable_pfn
[nid
] =
5722 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
5725 /* restore the node_state */
5726 node_states
[N_MEMORY
] = saved_node_state
;
5729 /* Any regular or high memory on that node ? */
5730 static void check_for_memory(pg_data_t
*pgdat
, int nid
)
5732 enum zone_type zone_type
;
5734 if (N_MEMORY
== N_NORMAL_MEMORY
)
5737 for (zone_type
= 0; zone_type
<= ZONE_MOVABLE
- 1; zone_type
++) {
5738 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
5739 if (populated_zone(zone
)) {
5740 node_set_state(nid
, N_HIGH_MEMORY
);
5741 if (N_NORMAL_MEMORY
!= N_HIGH_MEMORY
&&
5742 zone_type
<= ZONE_NORMAL
)
5743 node_set_state(nid
, N_NORMAL_MEMORY
);
5750 * free_area_init_nodes - Initialise all pg_data_t and zone data
5751 * @max_zone_pfn: an array of max PFNs for each zone
5753 * This will call free_area_init_node() for each active node in the system.
5754 * Using the page ranges provided by memblock_set_node(), the size of each
5755 * zone in each node and their holes is calculated. If the maximum PFN
5756 * between two adjacent zones match, it is assumed that the zone is empty.
5757 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5758 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5759 * starts where the previous one ended. For example, ZONE_DMA32 starts
5760 * at arch_max_dma_pfn.
5762 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
5764 unsigned long start_pfn
, end_pfn
;
5767 /* Record where the zone boundaries are */
5768 memset(arch_zone_lowest_possible_pfn
, 0,
5769 sizeof(arch_zone_lowest_possible_pfn
));
5770 memset(arch_zone_highest_possible_pfn
, 0,
5771 sizeof(arch_zone_highest_possible_pfn
));
5772 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
5773 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
5774 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
5775 if (i
== ZONE_MOVABLE
)
5777 arch_zone_lowest_possible_pfn
[i
] =
5778 arch_zone_highest_possible_pfn
[i
-1];
5779 arch_zone_highest_possible_pfn
[i
] =
5780 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
5782 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
5783 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
5785 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5786 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
5787 find_zone_movable_pfns_for_nodes();
5789 /* Print out the zone ranges */
5790 pr_info("Zone ranges:\n");
5791 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5792 if (i
== ZONE_MOVABLE
)
5794 pr_info(" %-8s ", zone_names
[i
]);
5795 if (arch_zone_lowest_possible_pfn
[i
] ==
5796 arch_zone_highest_possible_pfn
[i
])
5799 pr_cont("[mem %#018Lx-%#018Lx]\n",
5800 (u64
)arch_zone_lowest_possible_pfn
[i
]
5802 ((u64
)arch_zone_highest_possible_pfn
[i
]
5803 << PAGE_SHIFT
) - 1);
5806 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5807 pr_info("Movable zone start for each node\n");
5808 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
5809 if (zone_movable_pfn
[i
])
5810 pr_info(" Node %d: %#018Lx\n", i
,
5811 (u64
)zone_movable_pfn
[i
] << PAGE_SHIFT
);
5814 /* Print out the early node map */
5815 pr_info("Early memory node ranges\n");
5816 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
5817 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid
,
5818 (u64
)start_pfn
<< PAGE_SHIFT
,
5819 ((u64
)end_pfn
<< PAGE_SHIFT
) - 1);
5821 /* Initialise every node */
5822 mminit_verify_pageflags_layout();
5823 setup_nr_node_ids();
5824 for_each_online_node(nid
) {
5825 pg_data_t
*pgdat
= NODE_DATA(nid
);
5826 free_area_init_node(nid
, NULL
,
5827 find_min_pfn_for_node(nid
), NULL
);
5829 /* Any memory on that node */
5830 if (pgdat
->node_present_pages
)
5831 node_set_state(nid
, N_MEMORY
);
5832 check_for_memory(pgdat
, nid
);
5836 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
5838 unsigned long long coremem
;
5842 coremem
= memparse(p
, &p
);
5843 *core
= coremem
>> PAGE_SHIFT
;
5845 /* Paranoid check that UL is enough for the coremem value */
5846 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
5852 * kernelcore=size sets the amount of memory for use for allocations that
5853 * cannot be reclaimed or migrated.
5855 static int __init
cmdline_parse_kernelcore(char *p
)
5857 return cmdline_parse_core(p
, &required_kernelcore
);
5861 * movablecore=size sets the amount of memory for use for allocations that
5862 * can be reclaimed or migrated.
5864 static int __init
cmdline_parse_movablecore(char *p
)
5866 return cmdline_parse_core(p
, &required_movablecore
);
5869 early_param("kernelcore", cmdline_parse_kernelcore
);
5870 early_param("movablecore", cmdline_parse_movablecore
);
5872 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5874 void adjust_managed_page_count(struct page
*page
, long count
)
5876 spin_lock(&managed_page_count_lock
);
5877 page_zone(page
)->managed_pages
+= count
;
5878 totalram_pages
+= count
;
5879 #ifdef CONFIG_HIGHMEM
5880 if (PageHighMem(page
))
5881 totalhigh_pages
+= count
;
5883 spin_unlock(&managed_page_count_lock
);
5885 EXPORT_SYMBOL(adjust_managed_page_count
);
5887 unsigned long free_reserved_area(void *start
, void *end
, int poison
, char *s
)
5890 unsigned long pages
= 0;
5892 start
= (void *)PAGE_ALIGN((unsigned long)start
);
5893 end
= (void *)((unsigned long)end
& PAGE_MASK
);
5894 for (pos
= start
; pos
< end
; pos
+= PAGE_SIZE
, pages
++) {
5895 if ((unsigned int)poison
<= 0xFF)
5896 memset(pos
, poison
, PAGE_SIZE
);
5897 free_reserved_page(virt_to_page(pos
));
5901 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
5902 s
, pages
<< (PAGE_SHIFT
- 10), start
, end
);
5906 EXPORT_SYMBOL(free_reserved_area
);
5908 #ifdef CONFIG_HIGHMEM
5909 void free_highmem_page(struct page
*page
)
5911 __free_reserved_page(page
);
5913 page_zone(page
)->managed_pages
++;
5919 void __init
mem_init_print_info(const char *str
)
5921 unsigned long physpages
, codesize
, datasize
, rosize
, bss_size
;
5922 unsigned long init_code_size
, init_data_size
;
5924 physpages
= get_num_physpages();
5925 codesize
= _etext
- _stext
;
5926 datasize
= _edata
- _sdata
;
5927 rosize
= __end_rodata
- __start_rodata
;
5928 bss_size
= __bss_stop
- __bss_start
;
5929 init_data_size
= __init_end
- __init_begin
;
5930 init_code_size
= _einittext
- _sinittext
;
5933 * Detect special cases and adjust section sizes accordingly:
5934 * 1) .init.* may be embedded into .data sections
5935 * 2) .init.text.* may be out of [__init_begin, __init_end],
5936 * please refer to arch/tile/kernel/vmlinux.lds.S.
5937 * 3) .rodata.* may be embedded into .text or .data sections.
5939 #define adj_init_size(start, end, size, pos, adj) \
5941 if (start <= pos && pos < end && size > adj) \
5945 adj_init_size(__init_begin
, __init_end
, init_data_size
,
5946 _sinittext
, init_code_size
);
5947 adj_init_size(_stext
, _etext
, codesize
, _sinittext
, init_code_size
);
5948 adj_init_size(_sdata
, _edata
, datasize
, __init_begin
, init_data_size
);
5949 adj_init_size(_stext
, _etext
, codesize
, __start_rodata
, rosize
);
5950 adj_init_size(_sdata
, _edata
, datasize
, __start_rodata
, rosize
);
5952 #undef adj_init_size
5954 pr_info("Memory: %luK/%luK available "
5955 "(%luK kernel code, %luK rwdata, %luK rodata, "
5956 "%luK init, %luK bss, %luK reserved, %luK cma-reserved"
5957 #ifdef CONFIG_HIGHMEM
5961 nr_free_pages() << (PAGE_SHIFT
-10), physpages
<< (PAGE_SHIFT
-10),
5962 codesize
>> 10, datasize
>> 10, rosize
>> 10,
5963 (init_data_size
+ init_code_size
) >> 10, bss_size
>> 10,
5964 (physpages
- totalram_pages
- totalcma_pages
) << (PAGE_SHIFT
-10),
5965 totalcma_pages
<< (PAGE_SHIFT
-10),
5966 #ifdef CONFIG_HIGHMEM
5967 totalhigh_pages
<< (PAGE_SHIFT
-10),
5969 str
? ", " : "", str
? str
: "");
5973 * set_dma_reserve - set the specified number of pages reserved in the first zone
5974 * @new_dma_reserve: The number of pages to mark reserved
5976 * The per-cpu batchsize and zone watermarks are determined by present_pages.
5977 * In the DMA zone, a significant percentage may be consumed by kernel image
5978 * and other unfreeable allocations which can skew the watermarks badly. This
5979 * function may optionally be used to account for unfreeable pages in the
5980 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5981 * smaller per-cpu batchsize.
5983 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
5985 dma_reserve
= new_dma_reserve
;
5988 void __init
free_area_init(unsigned long *zones_size
)
5990 free_area_init_node(0, zones_size
,
5991 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
5994 static int page_alloc_cpu_notify(struct notifier_block
*self
,
5995 unsigned long action
, void *hcpu
)
5997 int cpu
= (unsigned long)hcpu
;
5999 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
6000 lru_add_drain_cpu(cpu
);
6004 * Spill the event counters of the dead processor
6005 * into the current processors event counters.
6006 * This artificially elevates the count of the current
6009 vm_events_fold_cpu(cpu
);
6012 * Zero the differential counters of the dead processor
6013 * so that the vm statistics are consistent.
6015 * This is only okay since the processor is dead and cannot
6016 * race with what we are doing.
6018 cpu_vm_stats_fold(cpu
);
6023 void __init
page_alloc_init(void)
6025 hotcpu_notifier(page_alloc_cpu_notify
, 0);
6029 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
6030 * or min_free_kbytes changes.
6032 static void calculate_totalreserve_pages(void)
6034 struct pglist_data
*pgdat
;
6035 unsigned long reserve_pages
= 0;
6036 enum zone_type i
, j
;
6038 for_each_online_pgdat(pgdat
) {
6039 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
6040 struct zone
*zone
= pgdat
->node_zones
+ i
;
6043 /* Find valid and maximum lowmem_reserve in the zone */
6044 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
6045 if (zone
->lowmem_reserve
[j
] > max
)
6046 max
= zone
->lowmem_reserve
[j
];
6049 /* we treat the high watermark as reserved pages. */
6050 max
+= high_wmark_pages(zone
);
6052 if (max
> zone
->managed_pages
)
6053 max
= zone
->managed_pages
;
6054 reserve_pages
+= max
;
6056 * Lowmem reserves are not available to
6057 * GFP_HIGHUSER page cache allocations and
6058 * kswapd tries to balance zones to their high
6059 * watermark. As a result, neither should be
6060 * regarded as dirtyable memory, to prevent a
6061 * situation where reclaim has to clean pages
6062 * in order to balance the zones.
6064 zone
->dirty_balance_reserve
= max
;
6067 dirty_balance_reserve
= reserve_pages
;
6068 totalreserve_pages
= reserve_pages
;
6072 * setup_per_zone_lowmem_reserve - called whenever
6073 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
6074 * has a correct pages reserved value, so an adequate number of
6075 * pages are left in the zone after a successful __alloc_pages().
6077 static void setup_per_zone_lowmem_reserve(void)
6079 struct pglist_data
*pgdat
;
6080 enum zone_type j
, idx
;
6082 for_each_online_pgdat(pgdat
) {
6083 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
6084 struct zone
*zone
= pgdat
->node_zones
+ j
;
6085 unsigned long managed_pages
= zone
->managed_pages
;
6087 zone
->lowmem_reserve
[j
] = 0;
6091 struct zone
*lower_zone
;
6095 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
6096 sysctl_lowmem_reserve_ratio
[idx
] = 1;
6098 lower_zone
= pgdat
->node_zones
+ idx
;
6099 lower_zone
->lowmem_reserve
[j
] = managed_pages
/
6100 sysctl_lowmem_reserve_ratio
[idx
];
6101 managed_pages
+= lower_zone
->managed_pages
;
6106 /* update totalreserve_pages */
6107 calculate_totalreserve_pages();
6110 static void __setup_per_zone_wmarks(void)
6112 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
6113 unsigned long lowmem_pages
= 0;
6115 unsigned long flags
;
6117 /* Calculate total number of !ZONE_HIGHMEM pages */
6118 for_each_zone(zone
) {
6119 if (!is_highmem(zone
))
6120 lowmem_pages
+= zone
->managed_pages
;
6123 for_each_zone(zone
) {
6126 spin_lock_irqsave(&zone
->lock
, flags
);
6127 tmp
= (u64
)pages_min
* zone
->managed_pages
;
6128 do_div(tmp
, lowmem_pages
);
6129 if (is_highmem(zone
)) {
6131 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6132 * need highmem pages, so cap pages_min to a small
6135 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
6136 * deltas control asynch page reclaim, and so should
6137 * not be capped for highmem.
6139 unsigned long min_pages
;
6141 min_pages
= zone
->managed_pages
/ 1024;
6142 min_pages
= clamp(min_pages
, SWAP_CLUSTER_MAX
, 128UL);
6143 zone
->watermark
[WMARK_MIN
] = min_pages
;
6146 * If it's a lowmem zone, reserve a number of pages
6147 * proportionate to the zone's size.
6149 zone
->watermark
[WMARK_MIN
] = tmp
;
6152 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + (tmp
>> 2);
6153 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + (tmp
>> 1);
6155 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
6156 high_wmark_pages(zone
) - low_wmark_pages(zone
) -
6157 atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]));
6159 setup_zone_migrate_reserve(zone
);
6160 spin_unlock_irqrestore(&zone
->lock
, flags
);
6163 /* update totalreserve_pages */
6164 calculate_totalreserve_pages();
6168 * setup_per_zone_wmarks - called when min_free_kbytes changes
6169 * or when memory is hot-{added|removed}
6171 * Ensures that the watermark[min,low,high] values for each zone are set
6172 * correctly with respect to min_free_kbytes.
6174 void setup_per_zone_wmarks(void)
6176 mutex_lock(&zonelists_mutex
);
6177 __setup_per_zone_wmarks();
6178 mutex_unlock(&zonelists_mutex
);
6182 * The inactive anon list should be small enough that the VM never has to
6183 * do too much work, but large enough that each inactive page has a chance
6184 * to be referenced again before it is swapped out.
6186 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
6187 * INACTIVE_ANON pages on this zone's LRU, maintained by the
6188 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
6189 * the anonymous pages are kept on the inactive list.
6192 * memory ratio inactive anon
6193 * -------------------------------------
6202 static void __meminit
calculate_zone_inactive_ratio(struct zone
*zone
)
6204 unsigned int gb
, ratio
;
6206 /* Zone size in gigabytes */
6207 gb
= zone
->managed_pages
>> (30 - PAGE_SHIFT
);
6209 ratio
= int_sqrt(10 * gb
);
6213 zone
->inactive_ratio
= ratio
;
6216 static void __meminit
setup_per_zone_inactive_ratio(void)
6221 calculate_zone_inactive_ratio(zone
);
6225 * Initialise min_free_kbytes.
6227 * For small machines we want it small (128k min). For large machines
6228 * we want it large (64MB max). But it is not linear, because network
6229 * bandwidth does not increase linearly with machine size. We use
6231 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
6232 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6248 int __meminit
init_per_zone_wmark_min(void)
6250 unsigned long lowmem_kbytes
;
6251 int new_min_free_kbytes
;
6253 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
6254 new_min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
6256 if (new_min_free_kbytes
> user_min_free_kbytes
) {
6257 min_free_kbytes
= new_min_free_kbytes
;
6258 if (min_free_kbytes
< 128)
6259 min_free_kbytes
= 128;
6260 if (min_free_kbytes
> 65536)
6261 min_free_kbytes
= 65536;
6263 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6264 new_min_free_kbytes
, user_min_free_kbytes
);
6266 setup_per_zone_wmarks();
6267 refresh_zone_stat_thresholds();
6268 setup_per_zone_lowmem_reserve();
6269 setup_per_zone_inactive_ratio();
6272 module_init(init_per_zone_wmark_min
)
6275 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
6276 * that we can call two helper functions whenever min_free_kbytes
6279 int min_free_kbytes_sysctl_handler(struct ctl_table
*table
, int write
,
6280 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6284 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6289 user_min_free_kbytes
= min_free_kbytes
;
6290 setup_per_zone_wmarks();
6296 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6297 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6302 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6307 zone
->min_unmapped_pages
= (zone
->managed_pages
*
6308 sysctl_min_unmapped_ratio
) / 100;
6312 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6313 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6318 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6323 zone
->min_slab_pages
= (zone
->managed_pages
*
6324 sysctl_min_slab_ratio
) / 100;
6330 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6331 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6332 * whenever sysctl_lowmem_reserve_ratio changes.
6334 * The reserve ratio obviously has absolutely no relation with the
6335 * minimum watermarks. The lowmem reserve ratio can only make sense
6336 * if in function of the boot time zone sizes.
6338 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6339 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6341 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6342 setup_per_zone_lowmem_reserve();
6347 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
6348 * cpu. It is the fraction of total pages in each zone that a hot per cpu
6349 * pagelist can have before it gets flushed back to buddy allocator.
6351 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table
*table
, int write
,
6352 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6355 int old_percpu_pagelist_fraction
;
6358 mutex_lock(&pcp_batch_high_lock
);
6359 old_percpu_pagelist_fraction
= percpu_pagelist_fraction
;
6361 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6362 if (!write
|| ret
< 0)
6365 /* Sanity checking to avoid pcp imbalance */
6366 if (percpu_pagelist_fraction
&&
6367 percpu_pagelist_fraction
< MIN_PERCPU_PAGELIST_FRACTION
) {
6368 percpu_pagelist_fraction
= old_percpu_pagelist_fraction
;
6374 if (percpu_pagelist_fraction
== old_percpu_pagelist_fraction
)
6377 for_each_populated_zone(zone
) {
6380 for_each_possible_cpu(cpu
)
6381 pageset_set_high_and_batch(zone
,
6382 per_cpu_ptr(zone
->pageset
, cpu
));
6385 mutex_unlock(&pcp_batch_high_lock
);
6390 int hashdist
= HASHDIST_DEFAULT
;
6392 static int __init
set_hashdist(char *str
)
6396 hashdist
= simple_strtoul(str
, &str
, 0);
6399 __setup("hashdist=", set_hashdist
);
6403 * allocate a large system hash table from bootmem
6404 * - it is assumed that the hash table must contain an exact power-of-2
6405 * quantity of entries
6406 * - limit is the number of hash buckets, not the total allocation size
6408 void *__init
alloc_large_system_hash(const char *tablename
,
6409 unsigned long bucketsize
,
6410 unsigned long numentries
,
6413 unsigned int *_hash_shift
,
6414 unsigned int *_hash_mask
,
6415 unsigned long low_limit
,
6416 unsigned long high_limit
)
6418 unsigned long long max
= high_limit
;
6419 unsigned long log2qty
, size
;
6422 /* allow the kernel cmdline to have a say */
6424 /* round applicable memory size up to nearest megabyte */
6425 numentries
= nr_kernel_pages
;
6427 /* It isn't necessary when PAGE_SIZE >= 1MB */
6428 if (PAGE_SHIFT
< 20)
6429 numentries
= round_up(numentries
, (1<<20)/PAGE_SIZE
);
6431 /* limit to 1 bucket per 2^scale bytes of low memory */
6432 if (scale
> PAGE_SHIFT
)
6433 numentries
>>= (scale
- PAGE_SHIFT
);
6435 numentries
<<= (PAGE_SHIFT
- scale
);
6437 /* Make sure we've got at least a 0-order allocation.. */
6438 if (unlikely(flags
& HASH_SMALL
)) {
6439 /* Makes no sense without HASH_EARLY */
6440 WARN_ON(!(flags
& HASH_EARLY
));
6441 if (!(numentries
>> *_hash_shift
)) {
6442 numentries
= 1UL << *_hash_shift
;
6443 BUG_ON(!numentries
);
6445 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
6446 numentries
= PAGE_SIZE
/ bucketsize
;
6448 numentries
= roundup_pow_of_two(numentries
);
6450 /* limit allocation size to 1/16 total memory by default */
6452 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
6453 do_div(max
, bucketsize
);
6455 max
= min(max
, 0x80000000ULL
);
6457 if (numentries
< low_limit
)
6458 numentries
= low_limit
;
6459 if (numentries
> max
)
6462 log2qty
= ilog2(numentries
);
6465 size
= bucketsize
<< log2qty
;
6466 if (flags
& HASH_EARLY
)
6467 table
= memblock_virt_alloc_nopanic(size
, 0);
6469 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
6472 * If bucketsize is not a power-of-two, we may free
6473 * some pages at the end of hash table which
6474 * alloc_pages_exact() automatically does
6476 if (get_order(size
) < MAX_ORDER
) {
6477 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
6478 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
6481 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
6484 panic("Failed to allocate %s hash table\n", tablename
);
6486 printk(KERN_INFO
"%s hash table entries: %ld (order: %d, %lu bytes)\n",
6489 ilog2(size
) - PAGE_SHIFT
,
6493 *_hash_shift
= log2qty
;
6495 *_hash_mask
= (1 << log2qty
) - 1;
6500 /* Return a pointer to the bitmap storing bits affecting a block of pages */
6501 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
6504 #ifdef CONFIG_SPARSEMEM
6505 return __pfn_to_section(pfn
)->pageblock_flags
;
6507 return zone
->pageblock_flags
;
6508 #endif /* CONFIG_SPARSEMEM */
6511 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
6513 #ifdef CONFIG_SPARSEMEM
6514 pfn
&= (PAGES_PER_SECTION
-1);
6515 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
6517 pfn
= pfn
- round_down(zone
->zone_start_pfn
, pageblock_nr_pages
);
6518 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
6519 #endif /* CONFIG_SPARSEMEM */
6523 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
6524 * @page: The page within the block of interest
6525 * @pfn: The target page frame number
6526 * @end_bitidx: The last bit of interest to retrieve
6527 * @mask: mask of bits that the caller is interested in
6529 * Return: pageblock_bits flags
6531 unsigned long get_pfnblock_flags_mask(struct page
*page
, unsigned long pfn
,
6532 unsigned long end_bitidx
,
6536 unsigned long *bitmap
;
6537 unsigned long bitidx
, word_bitidx
;
6540 zone
= page_zone(page
);
6541 bitmap
= get_pageblock_bitmap(zone
, pfn
);
6542 bitidx
= pfn_to_bitidx(zone
, pfn
);
6543 word_bitidx
= bitidx
/ BITS_PER_LONG
;
6544 bitidx
&= (BITS_PER_LONG
-1);
6546 word
= bitmap
[word_bitidx
];
6547 bitidx
+= end_bitidx
;
6548 return (word
>> (BITS_PER_LONG
- bitidx
- 1)) & mask
;
6552 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
6553 * @page: The page within the block of interest
6554 * @flags: The flags to set
6555 * @pfn: The target page frame number
6556 * @end_bitidx: The last bit of interest
6557 * @mask: mask of bits that the caller is interested in
6559 void set_pfnblock_flags_mask(struct page
*page
, unsigned long flags
,
6561 unsigned long end_bitidx
,
6565 unsigned long *bitmap
;
6566 unsigned long bitidx
, word_bitidx
;
6567 unsigned long old_word
, word
;
6569 BUILD_BUG_ON(NR_PAGEBLOCK_BITS
!= 4);
6571 zone
= page_zone(page
);
6572 bitmap
= get_pageblock_bitmap(zone
, pfn
);
6573 bitidx
= pfn_to_bitidx(zone
, pfn
);
6574 word_bitidx
= bitidx
/ BITS_PER_LONG
;
6575 bitidx
&= (BITS_PER_LONG
-1);
6577 VM_BUG_ON_PAGE(!zone_spans_pfn(zone
, pfn
), page
);
6579 bitidx
+= end_bitidx
;
6580 mask
<<= (BITS_PER_LONG
- bitidx
- 1);
6581 flags
<<= (BITS_PER_LONG
- bitidx
- 1);
6583 word
= READ_ONCE(bitmap
[word_bitidx
]);
6585 old_word
= cmpxchg(&bitmap
[word_bitidx
], word
, (word
& ~mask
) | flags
);
6586 if (word
== old_word
)
6593 * This function checks whether pageblock includes unmovable pages or not.
6594 * If @count is not zero, it is okay to include less @count unmovable pages
6596 * PageLRU check without isolation or lru_lock could race so that
6597 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6598 * expect this function should be exact.
6600 bool has_unmovable_pages(struct zone
*zone
, struct page
*page
, int count
,
6601 bool skip_hwpoisoned_pages
)
6603 unsigned long pfn
, iter
, found
;
6607 * For avoiding noise data, lru_add_drain_all() should be called
6608 * If ZONE_MOVABLE, the zone never contains unmovable pages
6610 if (zone_idx(zone
) == ZONE_MOVABLE
)
6612 mt
= get_pageblock_migratetype(page
);
6613 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
6616 pfn
= page_to_pfn(page
);
6617 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
6618 unsigned long check
= pfn
+ iter
;
6620 if (!pfn_valid_within(check
))
6623 page
= pfn_to_page(check
);
6626 * Hugepages are not in LRU lists, but they're movable.
6627 * We need not scan over tail pages bacause we don't
6628 * handle each tail page individually in migration.
6630 if (PageHuge(page
)) {
6631 iter
= round_up(iter
+ 1, 1<<compound_order(page
)) - 1;
6636 * We can't use page_count without pin a page
6637 * because another CPU can free compound page.
6638 * This check already skips compound tails of THP
6639 * because their page->_count is zero at all time.
6641 if (!atomic_read(&page
->_count
)) {
6642 if (PageBuddy(page
))
6643 iter
+= (1 << page_order(page
)) - 1;
6648 * The HWPoisoned page may be not in buddy system, and
6649 * page_count() is not 0.
6651 if (skip_hwpoisoned_pages
&& PageHWPoison(page
))
6657 * If there are RECLAIMABLE pages, we need to check
6658 * it. But now, memory offline itself doesn't call
6659 * shrink_node_slabs() and it still to be fixed.
6662 * If the page is not RAM, page_count()should be 0.
6663 * we don't need more check. This is an _used_ not-movable page.
6665 * The problematic thing here is PG_reserved pages. PG_reserved
6666 * is set to both of a memory hole page and a _used_ kernel
6675 bool is_pageblock_removable_nolock(struct page
*page
)
6681 * We have to be careful here because we are iterating over memory
6682 * sections which are not zone aware so we might end up outside of
6683 * the zone but still within the section.
6684 * We have to take care about the node as well. If the node is offline
6685 * its NODE_DATA will be NULL - see page_zone.
6687 if (!node_online(page_to_nid(page
)))
6690 zone
= page_zone(page
);
6691 pfn
= page_to_pfn(page
);
6692 if (!zone_spans_pfn(zone
, pfn
))
6695 return !has_unmovable_pages(zone
, page
, 0, true);
6700 static unsigned long pfn_max_align_down(unsigned long pfn
)
6702 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
6703 pageblock_nr_pages
) - 1);
6706 static unsigned long pfn_max_align_up(unsigned long pfn
)
6708 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
6709 pageblock_nr_pages
));
6712 /* [start, end) must belong to a single zone. */
6713 static int __alloc_contig_migrate_range(struct compact_control
*cc
,
6714 unsigned long start
, unsigned long end
)
6716 /* This function is based on compact_zone() from compaction.c. */
6717 unsigned long nr_reclaimed
;
6718 unsigned long pfn
= start
;
6719 unsigned int tries
= 0;
6724 while (pfn
< end
|| !list_empty(&cc
->migratepages
)) {
6725 if (fatal_signal_pending(current
)) {
6730 if (list_empty(&cc
->migratepages
)) {
6731 cc
->nr_migratepages
= 0;
6732 pfn
= isolate_migratepages_range(cc
, pfn
, end
);
6738 } else if (++tries
== 5) {
6739 ret
= ret
< 0 ? ret
: -EBUSY
;
6743 nr_reclaimed
= reclaim_clean_pages_from_list(cc
->zone
,
6745 cc
->nr_migratepages
-= nr_reclaimed
;
6747 ret
= migrate_pages(&cc
->migratepages
, alloc_migrate_target
,
6748 NULL
, 0, cc
->mode
, MR_CMA
);
6751 putback_movable_pages(&cc
->migratepages
);
6758 * alloc_contig_range() -- tries to allocate given range of pages
6759 * @start: start PFN to allocate
6760 * @end: one-past-the-last PFN to allocate
6761 * @migratetype: migratetype of the underlaying pageblocks (either
6762 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6763 * in range must have the same migratetype and it must
6764 * be either of the two.
6766 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6767 * aligned, however it's the caller's responsibility to guarantee that
6768 * we are the only thread that changes migrate type of pageblocks the
6771 * The PFN range must belong to a single zone.
6773 * Returns zero on success or negative error code. On success all
6774 * pages which PFN is in [start, end) are allocated for the caller and
6775 * need to be freed with free_contig_range().
6777 int alloc_contig_range(unsigned long start
, unsigned long end
,
6778 unsigned migratetype
)
6780 unsigned long outer_start
, outer_end
;
6783 struct compact_control cc
= {
6784 .nr_migratepages
= 0,
6786 .zone
= page_zone(pfn_to_page(start
)),
6787 .mode
= MIGRATE_SYNC
,
6788 .ignore_skip_hint
= true,
6790 INIT_LIST_HEAD(&cc
.migratepages
);
6793 * What we do here is we mark all pageblocks in range as
6794 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6795 * have different sizes, and due to the way page allocator
6796 * work, we align the range to biggest of the two pages so
6797 * that page allocator won't try to merge buddies from
6798 * different pageblocks and change MIGRATE_ISOLATE to some
6799 * other migration type.
6801 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6802 * migrate the pages from an unaligned range (ie. pages that
6803 * we are interested in). This will put all the pages in
6804 * range back to page allocator as MIGRATE_ISOLATE.
6806 * When this is done, we take the pages in range from page
6807 * allocator removing them from the buddy system. This way
6808 * page allocator will never consider using them.
6810 * This lets us mark the pageblocks back as
6811 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6812 * aligned range but not in the unaligned, original range are
6813 * put back to page allocator so that buddy can use them.
6816 ret
= start_isolate_page_range(pfn_max_align_down(start
),
6817 pfn_max_align_up(end
), migratetype
,
6822 ret
= __alloc_contig_migrate_range(&cc
, start
, end
);
6827 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6828 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6829 * more, all pages in [start, end) are free in page allocator.
6830 * What we are going to do is to allocate all pages from
6831 * [start, end) (that is remove them from page allocator).
6833 * The only problem is that pages at the beginning and at the
6834 * end of interesting range may be not aligned with pages that
6835 * page allocator holds, ie. they can be part of higher order
6836 * pages. Because of this, we reserve the bigger range and
6837 * once this is done free the pages we are not interested in.
6839 * We don't have to hold zone->lock here because the pages are
6840 * isolated thus they won't get removed from buddy.
6843 lru_add_drain_all();
6844 drain_all_pages(cc
.zone
);
6847 outer_start
= start
;
6848 while (!PageBuddy(pfn_to_page(outer_start
))) {
6849 if (++order
>= MAX_ORDER
) {
6853 outer_start
&= ~0UL << order
;
6856 /* Make sure the range is really isolated. */
6857 if (test_pages_isolated(outer_start
, end
, false)) {
6858 pr_info("%s: [%lx, %lx) PFNs busy\n",
6859 __func__
, outer_start
, end
);
6864 /* Grab isolated pages from freelists. */
6865 outer_end
= isolate_freepages_range(&cc
, outer_start
, end
);
6871 /* Free head and tail (if any) */
6872 if (start
!= outer_start
)
6873 free_contig_range(outer_start
, start
- outer_start
);
6874 if (end
!= outer_end
)
6875 free_contig_range(end
, outer_end
- end
);
6878 undo_isolate_page_range(pfn_max_align_down(start
),
6879 pfn_max_align_up(end
), migratetype
);
6883 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
6885 unsigned int count
= 0;
6887 for (; nr_pages
--; pfn
++) {
6888 struct page
*page
= pfn_to_page(pfn
);
6890 count
+= page_count(page
) != 1;
6893 WARN(count
!= 0, "%d pages are still in use!\n", count
);
6897 #ifdef CONFIG_MEMORY_HOTPLUG
6899 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6900 * page high values need to be recalulated.
6902 void __meminit
zone_pcp_update(struct zone
*zone
)
6905 mutex_lock(&pcp_batch_high_lock
);
6906 for_each_possible_cpu(cpu
)
6907 pageset_set_high_and_batch(zone
,
6908 per_cpu_ptr(zone
->pageset
, cpu
));
6909 mutex_unlock(&pcp_batch_high_lock
);
6913 void zone_pcp_reset(struct zone
*zone
)
6915 unsigned long flags
;
6917 struct per_cpu_pageset
*pset
;
6919 /* avoid races with drain_pages() */
6920 local_irq_save(flags
);
6921 if (zone
->pageset
!= &boot_pageset
) {
6922 for_each_online_cpu(cpu
) {
6923 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
6924 drain_zonestat(zone
, pset
);
6926 free_percpu(zone
->pageset
);
6927 zone
->pageset
= &boot_pageset
;
6929 local_irq_restore(flags
);
6932 #ifdef CONFIG_MEMORY_HOTREMOVE
6934 * All pages in the range must be isolated before calling this.
6937 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
6941 unsigned int order
, i
;
6943 unsigned long flags
;
6944 /* find the first valid pfn */
6945 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
6950 zone
= page_zone(pfn_to_page(pfn
));
6951 spin_lock_irqsave(&zone
->lock
, flags
);
6953 while (pfn
< end_pfn
) {
6954 if (!pfn_valid(pfn
)) {
6958 page
= pfn_to_page(pfn
);
6960 * The HWPoisoned page may be not in buddy system, and
6961 * page_count() is not 0.
6963 if (unlikely(!PageBuddy(page
) && PageHWPoison(page
))) {
6965 SetPageReserved(page
);
6969 BUG_ON(page_count(page
));
6970 BUG_ON(!PageBuddy(page
));
6971 order
= page_order(page
);
6972 #ifdef CONFIG_DEBUG_VM
6973 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
6974 pfn
, 1 << order
, end_pfn
);
6976 list_del(&page
->lru
);
6977 rmv_page_order(page
);
6978 zone
->free_area
[order
].nr_free
--;
6979 for (i
= 0; i
< (1 << order
); i
++)
6980 SetPageReserved((page
+i
));
6981 pfn
+= (1 << order
);
6983 spin_unlock_irqrestore(&zone
->lock
, flags
);
6987 #ifdef CONFIG_MEMORY_FAILURE
6988 bool is_free_buddy_page(struct page
*page
)
6990 struct zone
*zone
= page_zone(page
);
6991 unsigned long pfn
= page_to_pfn(page
);
6992 unsigned long flags
;
6995 spin_lock_irqsave(&zone
->lock
, flags
);
6996 for (order
= 0; order
< MAX_ORDER
; order
++) {
6997 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
6999 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
7002 spin_unlock_irqrestore(&zone
->lock
, flags
);
7004 return order
< MAX_ORDER
;