2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/jiffies.h>
23 #include <linux/bootmem.h>
24 #include <linux/memblock.h>
25 #include <linux/compiler.h>
26 #include <linux/kernel.h>
27 #include <linux/kmemcheck.h>
28 #include <linux/kasan.h>
29 #include <linux/module.h>
30 #include <linux/suspend.h>
31 #include <linux/pagevec.h>
32 #include <linux/blkdev.h>
33 #include <linux/slab.h>
34 #include <linux/ratelimit.h>
35 #include <linux/oom.h>
36 #include <linux/notifier.h>
37 #include <linux/topology.h>
38 #include <linux/sysctl.h>
39 #include <linux/cpu.h>
40 #include <linux/cpuset.h>
41 #include <linux/memory_hotplug.h>
42 #include <linux/nodemask.h>
43 #include <linux/vmalloc.h>
44 #include <linux/vmstat.h>
45 #include <linux/mempolicy.h>
46 #include <linux/stop_machine.h>
47 #include <linux/sort.h>
48 #include <linux/pfn.h>
49 #include <linux/backing-dev.h>
50 #include <linux/fault-inject.h>
51 #include <linux/page-isolation.h>
52 #include <linux/page_ext.h>
53 #include <linux/debugobjects.h>
54 #include <linux/kmemleak.h>
55 #include <linux/compaction.h>
56 #include <trace/events/kmem.h>
57 #include <linux/prefetch.h>
58 #include <linux/mm_inline.h>
59 #include <linux/migrate.h>
60 #include <linux/page_ext.h>
61 #include <linux/hugetlb.h>
62 #include <linux/sched/rt.h>
63 #include <linux/page_owner.h>
64 #include <linux/kthread.h>
66 #include <asm/sections.h>
67 #include <asm/tlbflush.h>
68 #include <asm/div64.h>
71 /* prevent >1 _updater_ of zone percpu pageset ->high and ->batch fields */
72 static DEFINE_MUTEX(pcp_batch_high_lock
);
73 #define MIN_PERCPU_PAGELIST_FRACTION (8)
75 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
76 DEFINE_PER_CPU(int, numa_node
);
77 EXPORT_PER_CPU_SYMBOL(numa_node
);
80 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
82 * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
83 * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
84 * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem()
85 * defined in <linux/topology.h>.
87 DEFINE_PER_CPU(int, _numa_mem_
); /* Kernel "local memory" node */
88 EXPORT_PER_CPU_SYMBOL(_numa_mem_
);
89 int _node_numa_mem_
[MAX_NUMNODES
];
93 * Array of node states.
95 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
96 [N_POSSIBLE
] = NODE_MASK_ALL
,
97 [N_ONLINE
] = { { [0] = 1UL } },
99 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
100 #ifdef CONFIG_HIGHMEM
101 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
103 #ifdef CONFIG_MOVABLE_NODE
104 [N_MEMORY
] = { { [0] = 1UL } },
106 [N_CPU
] = { { [0] = 1UL } },
109 EXPORT_SYMBOL(node_states
);
111 /* Protect totalram_pages and zone->managed_pages */
112 static DEFINE_SPINLOCK(managed_page_count_lock
);
114 unsigned long totalram_pages __read_mostly
;
115 unsigned long totalreserve_pages __read_mostly
;
116 unsigned long totalcma_pages __read_mostly
;
118 int percpu_pagelist_fraction
;
119 gfp_t gfp_allowed_mask __read_mostly
= GFP_BOOT_MASK
;
122 * A cached value of the page's pageblock's migratetype, used when the page is
123 * put on a pcplist. Used to avoid the pageblock migratetype lookup when
124 * freeing from pcplists in most cases, at the cost of possibly becoming stale.
125 * Also the migratetype set in the page does not necessarily match the pcplist
126 * index, e.g. page might have MIGRATE_CMA set but be on a pcplist with any
127 * other index - this ensures that it will be put on the correct CMA freelist.
129 static inline int get_pcppage_migratetype(struct page
*page
)
134 static inline void set_pcppage_migratetype(struct page
*page
, int migratetype
)
136 page
->index
= migratetype
;
139 #ifdef CONFIG_PM_SLEEP
141 * The following functions are used by the suspend/hibernate code to temporarily
142 * change gfp_allowed_mask in order to avoid using I/O during memory allocations
143 * while devices are suspended. To avoid races with the suspend/hibernate code,
144 * they should always be called with pm_mutex held (gfp_allowed_mask also should
145 * only be modified with pm_mutex held, unless the suspend/hibernate code is
146 * guaranteed not to run in parallel with that modification).
149 static gfp_t saved_gfp_mask
;
151 void pm_restore_gfp_mask(void)
153 WARN_ON(!mutex_is_locked(&pm_mutex
));
154 if (saved_gfp_mask
) {
155 gfp_allowed_mask
= saved_gfp_mask
;
160 void pm_restrict_gfp_mask(void)
162 WARN_ON(!mutex_is_locked(&pm_mutex
));
163 WARN_ON(saved_gfp_mask
);
164 saved_gfp_mask
= gfp_allowed_mask
;
165 gfp_allowed_mask
&= ~(__GFP_IO
| __GFP_FS
);
168 bool pm_suspended_storage(void)
170 if ((gfp_allowed_mask
& (__GFP_IO
| __GFP_FS
)) == (__GFP_IO
| __GFP_FS
))
174 #endif /* CONFIG_PM_SLEEP */
176 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
177 unsigned int pageblock_order __read_mostly
;
180 static void __free_pages_ok(struct page
*page
, unsigned int order
);
183 * results with 256, 32 in the lowmem_reserve sysctl:
184 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
185 * 1G machine -> (16M dma, 784M normal, 224M high)
186 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
187 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
188 * HIGHMEM allocation will leave (224M+784M)/256 of ram reserved in ZONE_DMA
190 * TBD: should special case ZONE_DMA32 machines here - in those we normally
191 * don't need any ZONE_NORMAL reservation
193 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
194 #ifdef CONFIG_ZONE_DMA
197 #ifdef CONFIG_ZONE_DMA32
200 #ifdef CONFIG_HIGHMEM
206 EXPORT_SYMBOL(totalram_pages
);
208 static char * const zone_names
[MAX_NR_ZONES
] = {
209 #ifdef CONFIG_ZONE_DMA
212 #ifdef CONFIG_ZONE_DMA32
216 #ifdef CONFIG_HIGHMEM
220 #ifdef CONFIG_ZONE_DEVICE
225 static void free_compound_page(struct page
*page
);
226 compound_page_dtor
* const compound_page_dtors
[] = {
229 #ifdef CONFIG_HUGETLB_PAGE
234 int min_free_kbytes
= 1024;
235 int user_min_free_kbytes
= -1;
237 static unsigned long __meminitdata nr_kernel_pages
;
238 static unsigned long __meminitdata nr_all_pages
;
239 static unsigned long __meminitdata dma_reserve
;
241 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
242 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
243 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
244 static unsigned long __initdata required_kernelcore
;
245 static unsigned long __initdata required_movablecore
;
246 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
248 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
250 EXPORT_SYMBOL(movable_zone
);
251 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
254 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
255 int nr_online_nodes __read_mostly
= 1;
256 EXPORT_SYMBOL(nr_node_ids
);
257 EXPORT_SYMBOL(nr_online_nodes
);
260 int page_group_by_mobility_disabled __read_mostly
;
262 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
263 static inline void reset_deferred_meminit(pg_data_t
*pgdat
)
265 pgdat
->first_deferred_pfn
= ULONG_MAX
;
268 /* Returns true if the struct page for the pfn is uninitialised */
269 static inline bool __meminit
early_page_uninitialised(unsigned long pfn
)
271 if (pfn
>= NODE_DATA(early_pfn_to_nid(pfn
))->first_deferred_pfn
)
277 static inline bool early_page_nid_uninitialised(unsigned long pfn
, int nid
)
279 if (pfn
>= NODE_DATA(nid
)->first_deferred_pfn
)
286 * Returns false when the remaining initialisation should be deferred until
287 * later in the boot cycle when it can be parallelised.
289 static inline bool update_defer_init(pg_data_t
*pgdat
,
290 unsigned long pfn
, unsigned long zone_end
,
291 unsigned long *nr_initialised
)
293 /* Always populate low zones for address-contrained allocations */
294 if (zone_end
< pgdat_end_pfn(pgdat
))
297 /* Initialise at least 2G of the highest zone */
299 if (*nr_initialised
> (2UL << (30 - PAGE_SHIFT
)) &&
300 (pfn
& (PAGES_PER_SECTION
- 1)) == 0) {
301 pgdat
->first_deferred_pfn
= pfn
;
308 static inline void reset_deferred_meminit(pg_data_t
*pgdat
)
312 static inline bool early_page_uninitialised(unsigned long pfn
)
317 static inline bool early_page_nid_uninitialised(unsigned long pfn
, int nid
)
322 static inline bool update_defer_init(pg_data_t
*pgdat
,
323 unsigned long pfn
, unsigned long zone_end
,
324 unsigned long *nr_initialised
)
331 void set_pageblock_migratetype(struct page
*page
, int migratetype
)
333 if (unlikely(page_group_by_mobility_disabled
&&
334 migratetype
< MIGRATE_PCPTYPES
))
335 migratetype
= MIGRATE_UNMOVABLE
;
337 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
338 PB_migrate
, PB_migrate_end
);
341 #ifdef CONFIG_DEBUG_VM
342 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
346 unsigned long pfn
= page_to_pfn(page
);
347 unsigned long sp
, start_pfn
;
350 seq
= zone_span_seqbegin(zone
);
351 start_pfn
= zone
->zone_start_pfn
;
352 sp
= zone
->spanned_pages
;
353 if (!zone_spans_pfn(zone
, pfn
))
355 } while (zone_span_seqretry(zone
, seq
));
358 pr_err("page 0x%lx outside node %d zone %s [ 0x%lx - 0x%lx ]\n",
359 pfn
, zone_to_nid(zone
), zone
->name
,
360 start_pfn
, start_pfn
+ sp
);
365 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
367 if (!pfn_valid_within(page_to_pfn(page
)))
369 if (zone
!= page_zone(page
))
375 * Temporary debugging check for pages not lying within a given zone.
377 static int bad_range(struct zone
*zone
, struct page
*page
)
379 if (page_outside_zone_boundaries(zone
, page
))
381 if (!page_is_consistent(zone
, page
))
387 static inline int bad_range(struct zone
*zone
, struct page
*page
)
393 static void bad_page(struct page
*page
, const char *reason
,
394 unsigned long bad_flags
)
396 static unsigned long resume
;
397 static unsigned long nr_shown
;
398 static unsigned long nr_unshown
;
400 /* Don't complain about poisoned pages */
401 if (PageHWPoison(page
)) {
402 page_mapcount_reset(page
); /* remove PageBuddy */
407 * Allow a burst of 60 reports, then keep quiet for that minute;
408 * or allow a steady drip of one report per second.
410 if (nr_shown
== 60) {
411 if (time_before(jiffies
, resume
)) {
417 "BUG: Bad page state: %lu messages suppressed\n",
424 resume
= jiffies
+ 60 * HZ
;
426 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
427 current
->comm
, page_to_pfn(page
));
428 dump_page_badflags(page
, reason
, bad_flags
);
433 /* Leave bad fields for debug, except PageBuddy could make trouble */
434 page_mapcount_reset(page
); /* remove PageBuddy */
435 add_taint(TAINT_BAD_PAGE
, LOCKDEP_NOW_UNRELIABLE
);
439 * Higher-order pages are called "compound pages". They are structured thusly:
441 * The first PAGE_SIZE page is called the "head page" and have PG_head set.
443 * The remaining PAGE_SIZE pages are called "tail pages". PageTail() is encoded
444 * in bit 0 of page->compound_head. The rest of bits is pointer to head page.
446 * The first tail page's ->compound_dtor holds the offset in array of compound
447 * page destructors. See compound_page_dtors.
449 * The first tail page's ->compound_order holds the order of allocation.
450 * This usage means that zero-order pages may not be compound.
453 static void free_compound_page(struct page
*page
)
455 __free_pages_ok(page
, compound_order(page
));
458 void prep_compound_page(struct page
*page
, unsigned int order
)
461 int nr_pages
= 1 << order
;
463 set_compound_page_dtor(page
, COMPOUND_PAGE_DTOR
);
464 set_compound_order(page
, order
);
466 for (i
= 1; i
< nr_pages
; i
++) {
467 struct page
*p
= page
+ i
;
468 set_page_count(p
, 0);
469 p
->mapping
= TAIL_MAPPING
;
470 set_compound_head(p
, page
);
474 #ifdef CONFIG_DEBUG_PAGEALLOC
475 unsigned int _debug_guardpage_minorder
;
476 bool _debug_pagealloc_enabled __read_mostly
;
477 bool _debug_guardpage_enabled __read_mostly
;
479 static int __init
early_debug_pagealloc(char *buf
)
484 if (strcmp(buf
, "on") == 0)
485 _debug_pagealloc_enabled
= true;
489 early_param("debug_pagealloc", early_debug_pagealloc
);
491 static bool need_debug_guardpage(void)
493 /* If we don't use debug_pagealloc, we don't need guard page */
494 if (!debug_pagealloc_enabled())
500 static void init_debug_guardpage(void)
502 if (!debug_pagealloc_enabled())
505 _debug_guardpage_enabled
= true;
508 struct page_ext_operations debug_guardpage_ops
= {
509 .need
= need_debug_guardpage
,
510 .init
= init_debug_guardpage
,
513 static int __init
debug_guardpage_minorder_setup(char *buf
)
517 if (kstrtoul(buf
, 10, &res
) < 0 || res
> MAX_ORDER
/ 2) {
518 printk(KERN_ERR
"Bad debug_guardpage_minorder value\n");
521 _debug_guardpage_minorder
= res
;
522 printk(KERN_INFO
"Setting debug_guardpage_minorder to %lu\n", res
);
525 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup
);
527 static inline void set_page_guard(struct zone
*zone
, struct page
*page
,
528 unsigned int order
, int migratetype
)
530 struct page_ext
*page_ext
;
532 if (!debug_guardpage_enabled())
535 page_ext
= lookup_page_ext(page
);
536 __set_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
538 INIT_LIST_HEAD(&page
->lru
);
539 set_page_private(page
, order
);
540 /* Guard pages are not available for any usage */
541 __mod_zone_freepage_state(zone
, -(1 << order
), migratetype
);
544 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
545 unsigned int order
, int migratetype
)
547 struct page_ext
*page_ext
;
549 if (!debug_guardpage_enabled())
552 page_ext
= lookup_page_ext(page
);
553 __clear_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
555 set_page_private(page
, 0);
556 if (!is_migrate_isolate(migratetype
))
557 __mod_zone_freepage_state(zone
, (1 << order
), migratetype
);
560 struct page_ext_operations debug_guardpage_ops
= { NULL
, };
561 static inline void set_page_guard(struct zone
*zone
, struct page
*page
,
562 unsigned int order
, int migratetype
) {}
563 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
564 unsigned int order
, int migratetype
) {}
567 static inline void set_page_order(struct page
*page
, unsigned int order
)
569 set_page_private(page
, order
);
570 __SetPageBuddy(page
);
573 static inline void rmv_page_order(struct page
*page
)
575 __ClearPageBuddy(page
);
576 set_page_private(page
, 0);
580 * This function checks whether a page is free && is the buddy
581 * we can do coalesce a page and its buddy if
582 * (a) the buddy is not in a hole &&
583 * (b) the buddy is in the buddy system &&
584 * (c) a page and its buddy have the same order &&
585 * (d) a page and its buddy are in the same zone.
587 * For recording whether a page is in the buddy system, we set ->_mapcount
588 * PAGE_BUDDY_MAPCOUNT_VALUE.
589 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
590 * serialized by zone->lock.
592 * For recording page's order, we use page_private(page).
594 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
597 if (!pfn_valid_within(page_to_pfn(buddy
)))
600 if (page_is_guard(buddy
) && page_order(buddy
) == order
) {
601 if (page_zone_id(page
) != page_zone_id(buddy
))
604 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
609 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
611 * zone check is done late to avoid uselessly
612 * calculating zone/node ids for pages that could
615 if (page_zone_id(page
) != page_zone_id(buddy
))
618 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
626 * Freeing function for a buddy system allocator.
628 * The concept of a buddy system is to maintain direct-mapped table
629 * (containing bit values) for memory blocks of various "orders".
630 * The bottom level table contains the map for the smallest allocatable
631 * units of memory (here, pages), and each level above it describes
632 * pairs of units from the levels below, hence, "buddies".
633 * At a high level, all that happens here is marking the table entry
634 * at the bottom level available, and propagating the changes upward
635 * as necessary, plus some accounting needed to play nicely with other
636 * parts of the VM system.
637 * At each level, we keep a list of pages, which are heads of continuous
638 * free pages of length of (1 << order) and marked with _mapcount
639 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
641 * So when we are allocating or freeing one, we can derive the state of the
642 * other. That is, if we allocate a small block, and both were
643 * free, the remainder of the region must be split into blocks.
644 * If a block is freed, and its buddy is also free, then this
645 * triggers coalescing into a block of larger size.
650 static inline void __free_one_page(struct page
*page
,
652 struct zone
*zone
, unsigned int order
,
655 unsigned long page_idx
;
656 unsigned long combined_idx
;
657 unsigned long uninitialized_var(buddy_idx
);
659 unsigned int max_order
= MAX_ORDER
;
661 VM_BUG_ON(!zone_is_initialized(zone
));
662 VM_BUG_ON_PAGE(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
, page
);
664 VM_BUG_ON(migratetype
== -1);
665 if (is_migrate_isolate(migratetype
)) {
667 * We restrict max order of merging to prevent merge
668 * between freepages on isolate pageblock and normal
669 * pageblock. Without this, pageblock isolation
670 * could cause incorrect freepage accounting.
672 max_order
= min_t(unsigned int, MAX_ORDER
, pageblock_order
+ 1);
674 __mod_zone_freepage_state(zone
, 1 << order
, migratetype
);
677 page_idx
= pfn
& ((1 << max_order
) - 1);
679 VM_BUG_ON_PAGE(page_idx
& ((1 << order
) - 1), page
);
680 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
682 while (order
< max_order
- 1) {
683 buddy_idx
= __find_buddy_index(page_idx
, order
);
684 buddy
= page
+ (buddy_idx
- page_idx
);
685 if (!page_is_buddy(page
, buddy
, order
))
688 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
689 * merge with it and move up one order.
691 if (page_is_guard(buddy
)) {
692 clear_page_guard(zone
, buddy
, order
, migratetype
);
694 list_del(&buddy
->lru
);
695 zone
->free_area
[order
].nr_free
--;
696 rmv_page_order(buddy
);
698 combined_idx
= buddy_idx
& page_idx
;
699 page
= page
+ (combined_idx
- page_idx
);
700 page_idx
= combined_idx
;
703 set_page_order(page
, order
);
706 * If this is not the largest possible page, check if the buddy
707 * of the next-highest order is free. If it is, it's possible
708 * that pages are being freed that will coalesce soon. In case,
709 * that is happening, add the free page to the tail of the list
710 * so it's less likely to be used soon and more likely to be merged
711 * as a higher order page
713 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
714 struct page
*higher_page
, *higher_buddy
;
715 combined_idx
= buddy_idx
& page_idx
;
716 higher_page
= page
+ (combined_idx
- page_idx
);
717 buddy_idx
= __find_buddy_index(combined_idx
, order
+ 1);
718 higher_buddy
= higher_page
+ (buddy_idx
- combined_idx
);
719 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
720 list_add_tail(&page
->lru
,
721 &zone
->free_area
[order
].free_list
[migratetype
]);
726 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
728 zone
->free_area
[order
].nr_free
++;
731 static inline int free_pages_check(struct page
*page
)
733 const char *bad_reason
= NULL
;
734 unsigned long bad_flags
= 0;
736 if (unlikely(page_mapcount(page
)))
737 bad_reason
= "nonzero mapcount";
738 if (unlikely(page
->mapping
!= NULL
))
739 bad_reason
= "non-NULL mapping";
740 if (unlikely(atomic_read(&page
->_count
) != 0))
741 bad_reason
= "nonzero _count";
742 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
)) {
743 bad_reason
= "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
744 bad_flags
= PAGE_FLAGS_CHECK_AT_FREE
;
747 if (unlikely(page
->mem_cgroup
))
748 bad_reason
= "page still charged to cgroup";
750 if (unlikely(bad_reason
)) {
751 bad_page(page
, bad_reason
, bad_flags
);
754 page_cpupid_reset_last(page
);
755 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
756 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
761 * Frees a number of pages from the PCP lists
762 * Assumes all pages on list are in same zone, and of same order.
763 * count is the number of pages to free.
765 * If the zone was previously in an "all pages pinned" state then look to
766 * see if this freeing clears that state.
768 * And clear the zone's pages_scanned counter, to hold off the "all pages are
769 * pinned" detection logic.
771 static void free_pcppages_bulk(struct zone
*zone
, int count
,
772 struct per_cpu_pages
*pcp
)
777 unsigned long nr_scanned
;
779 spin_lock(&zone
->lock
);
780 nr_scanned
= zone_page_state(zone
, NR_PAGES_SCANNED
);
782 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, -nr_scanned
);
786 struct list_head
*list
;
789 * Remove pages from lists in a round-robin fashion. A
790 * batch_free count is maintained that is incremented when an
791 * empty list is encountered. This is so more pages are freed
792 * off fuller lists instead of spinning excessively around empty
797 if (++migratetype
== MIGRATE_PCPTYPES
)
799 list
= &pcp
->lists
[migratetype
];
800 } while (list_empty(list
));
802 /* This is the only non-empty list. Free them all. */
803 if (batch_free
== MIGRATE_PCPTYPES
)
804 batch_free
= to_free
;
807 int mt
; /* migratetype of the to-be-freed page */
809 page
= list_last_entry(list
, struct page
, lru
);
810 /* must delete as __free_one_page list manipulates */
811 list_del(&page
->lru
);
813 mt
= get_pcppage_migratetype(page
);
814 /* MIGRATE_ISOLATE page should not go to pcplists */
815 VM_BUG_ON_PAGE(is_migrate_isolate(mt
), page
);
816 /* Pageblock could have been isolated meanwhile */
817 if (unlikely(has_isolate_pageblock(zone
)))
818 mt
= get_pageblock_migratetype(page
);
820 __free_one_page(page
, page_to_pfn(page
), zone
, 0, mt
);
821 trace_mm_page_pcpu_drain(page
, 0, mt
);
822 } while (--to_free
&& --batch_free
&& !list_empty(list
));
824 spin_unlock(&zone
->lock
);
827 static void free_one_page(struct zone
*zone
,
828 struct page
*page
, unsigned long pfn
,
832 unsigned long nr_scanned
;
833 spin_lock(&zone
->lock
);
834 nr_scanned
= zone_page_state(zone
, NR_PAGES_SCANNED
);
836 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, -nr_scanned
);
838 if (unlikely(has_isolate_pageblock(zone
) ||
839 is_migrate_isolate(migratetype
))) {
840 migratetype
= get_pfnblock_migratetype(page
, pfn
);
842 __free_one_page(page
, pfn
, zone
, order
, migratetype
);
843 spin_unlock(&zone
->lock
);
846 static int free_tail_pages_check(struct page
*head_page
, struct page
*page
)
851 * We rely page->lru.next never has bit 0 set, unless the page
852 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
854 BUILD_BUG_ON((unsigned long)LIST_POISON1
& 1);
856 if (!IS_ENABLED(CONFIG_DEBUG_VM
)) {
860 if (page
->mapping
!= TAIL_MAPPING
) {
861 bad_page(page
, "corrupted mapping in tail page", 0);
864 if (unlikely(!PageTail(page
))) {
865 bad_page(page
, "PageTail not set", 0);
868 if (unlikely(compound_head(page
) != head_page
)) {
869 bad_page(page
, "compound_head not consistent", 0);
874 page
->mapping
= NULL
;
875 clear_compound_head(page
);
879 static void __meminit
__init_single_page(struct page
*page
, unsigned long pfn
,
880 unsigned long zone
, int nid
)
882 set_page_links(page
, zone
, nid
, pfn
);
883 init_page_count(page
);
884 page_mapcount_reset(page
);
885 page_cpupid_reset_last(page
);
887 INIT_LIST_HEAD(&page
->lru
);
888 #ifdef WANT_PAGE_VIRTUAL
889 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
890 if (!is_highmem_idx(zone
))
891 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
895 static void __meminit
__init_single_pfn(unsigned long pfn
, unsigned long zone
,
898 return __init_single_page(pfn_to_page(pfn
), pfn
, zone
, nid
);
901 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
902 static void init_reserved_page(unsigned long pfn
)
907 if (!early_page_uninitialised(pfn
))
910 nid
= early_pfn_to_nid(pfn
);
911 pgdat
= NODE_DATA(nid
);
913 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
914 struct zone
*zone
= &pgdat
->node_zones
[zid
];
916 if (pfn
>= zone
->zone_start_pfn
&& pfn
< zone_end_pfn(zone
))
919 __init_single_pfn(pfn
, zid
, nid
);
922 static inline void init_reserved_page(unsigned long pfn
)
925 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
928 * Initialised pages do not have PageReserved set. This function is
929 * called for each range allocated by the bootmem allocator and
930 * marks the pages PageReserved. The remaining valid pages are later
931 * sent to the buddy page allocator.
933 void __meminit
reserve_bootmem_region(unsigned long start
, unsigned long end
)
935 unsigned long start_pfn
= PFN_DOWN(start
);
936 unsigned long end_pfn
= PFN_UP(end
);
938 for (; start_pfn
< end_pfn
; start_pfn
++) {
939 if (pfn_valid(start_pfn
)) {
940 struct page
*page
= pfn_to_page(start_pfn
);
942 init_reserved_page(start_pfn
);
944 /* Avoid false-positive PageTail() */
945 INIT_LIST_HEAD(&page
->lru
);
947 SetPageReserved(page
);
952 static bool free_pages_prepare(struct page
*page
, unsigned int order
)
954 bool compound
= PageCompound(page
);
957 VM_BUG_ON_PAGE(PageTail(page
), page
);
958 VM_BUG_ON_PAGE(compound
&& compound_order(page
) != order
, page
);
960 trace_mm_page_free(page
, order
);
961 kmemcheck_free_shadow(page
, order
);
962 kasan_free_pages(page
, order
);
965 page
->mapping
= NULL
;
966 bad
+= free_pages_check(page
);
967 for (i
= 1; i
< (1 << order
); i
++) {
969 bad
+= free_tail_pages_check(page
, page
+ i
);
970 bad
+= free_pages_check(page
+ i
);
975 reset_page_owner(page
, order
);
977 if (!PageHighMem(page
)) {
978 debug_check_no_locks_freed(page_address(page
),
980 debug_check_no_obj_freed(page_address(page
),
983 arch_free_page(page
, order
);
984 kernel_map_pages(page
, 1 << order
, 0);
989 static void __free_pages_ok(struct page
*page
, unsigned int order
)
993 unsigned long pfn
= page_to_pfn(page
);
995 if (!free_pages_prepare(page
, order
))
998 migratetype
= get_pfnblock_migratetype(page
, pfn
);
999 local_irq_save(flags
);
1000 __count_vm_events(PGFREE
, 1 << order
);
1001 free_one_page(page_zone(page
), page
, pfn
, order
, migratetype
);
1002 local_irq_restore(flags
);
1005 static void __init
__free_pages_boot_core(struct page
*page
,
1006 unsigned long pfn
, unsigned int order
)
1008 unsigned int nr_pages
= 1 << order
;
1009 struct page
*p
= page
;
1013 for (loop
= 0; loop
< (nr_pages
- 1); loop
++, p
++) {
1015 __ClearPageReserved(p
);
1016 set_page_count(p
, 0);
1018 __ClearPageReserved(p
);
1019 set_page_count(p
, 0);
1021 page_zone(page
)->managed_pages
+= nr_pages
;
1022 set_page_refcounted(page
);
1023 __free_pages(page
, order
);
1026 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1027 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1029 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata
;
1031 int __meminit
early_pfn_to_nid(unsigned long pfn
)
1033 static DEFINE_SPINLOCK(early_pfn_lock
);
1036 spin_lock(&early_pfn_lock
);
1037 nid
= __early_pfn_to_nid(pfn
, &early_pfnnid_cache
);
1040 spin_unlock(&early_pfn_lock
);
1046 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1047 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1048 struct mminit_pfnnid_cache
*state
)
1052 nid
= __early_pfn_to_nid(pfn
, state
);
1053 if (nid
>= 0 && nid
!= node
)
1058 /* Only safe to use early in boot when initialisation is single-threaded */
1059 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1061 return meminit_pfn_in_nid(pfn
, node
, &early_pfnnid_cache
);
1066 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1070 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1071 struct mminit_pfnnid_cache
*state
)
1078 void __init
__free_pages_bootmem(struct page
*page
, unsigned long pfn
,
1081 if (early_page_uninitialised(pfn
))
1083 return __free_pages_boot_core(page
, pfn
, order
);
1086 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1087 static void __init
deferred_free_range(struct page
*page
,
1088 unsigned long pfn
, int nr_pages
)
1095 /* Free a large naturally-aligned chunk if possible */
1096 if (nr_pages
== MAX_ORDER_NR_PAGES
&&
1097 (pfn
& (MAX_ORDER_NR_PAGES
-1)) == 0) {
1098 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
1099 __free_pages_boot_core(page
, pfn
, MAX_ORDER
-1);
1103 for (i
= 0; i
< nr_pages
; i
++, page
++, pfn
++)
1104 __free_pages_boot_core(page
, pfn
, 0);
1107 /* Completion tracking for deferred_init_memmap() threads */
1108 static atomic_t pgdat_init_n_undone __initdata
;
1109 static __initdata
DECLARE_COMPLETION(pgdat_init_all_done_comp
);
1111 static inline void __init
pgdat_init_report_one_done(void)
1113 if (atomic_dec_and_test(&pgdat_init_n_undone
))
1114 complete(&pgdat_init_all_done_comp
);
1117 /* Initialise remaining memory on a node */
1118 static int __init
deferred_init_memmap(void *data
)
1120 pg_data_t
*pgdat
= data
;
1121 int nid
= pgdat
->node_id
;
1122 struct mminit_pfnnid_cache nid_init_state
= { };
1123 unsigned long start
= jiffies
;
1124 unsigned long nr_pages
= 0;
1125 unsigned long walk_start
, walk_end
;
1128 unsigned long first_init_pfn
= pgdat
->first_deferred_pfn
;
1129 const struct cpumask
*cpumask
= cpumask_of_node(pgdat
->node_id
);
1131 if (first_init_pfn
== ULONG_MAX
) {
1132 pgdat_init_report_one_done();
1136 /* Bind memory initialisation thread to a local node if possible */
1137 if (!cpumask_empty(cpumask
))
1138 set_cpus_allowed_ptr(current
, cpumask
);
1140 /* Sanity check boundaries */
1141 BUG_ON(pgdat
->first_deferred_pfn
< pgdat
->node_start_pfn
);
1142 BUG_ON(pgdat
->first_deferred_pfn
> pgdat_end_pfn(pgdat
));
1143 pgdat
->first_deferred_pfn
= ULONG_MAX
;
1145 /* Only the highest zone is deferred so find it */
1146 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
1147 zone
= pgdat
->node_zones
+ zid
;
1148 if (first_init_pfn
< zone_end_pfn(zone
))
1152 for_each_mem_pfn_range(i
, nid
, &walk_start
, &walk_end
, NULL
) {
1153 unsigned long pfn
, end_pfn
;
1154 struct page
*page
= NULL
;
1155 struct page
*free_base_page
= NULL
;
1156 unsigned long free_base_pfn
= 0;
1159 end_pfn
= min(walk_end
, zone_end_pfn(zone
));
1160 pfn
= first_init_pfn
;
1161 if (pfn
< walk_start
)
1163 if (pfn
< zone
->zone_start_pfn
)
1164 pfn
= zone
->zone_start_pfn
;
1166 for (; pfn
< end_pfn
; pfn
++) {
1167 if (!pfn_valid_within(pfn
))
1171 * Ensure pfn_valid is checked every
1172 * MAX_ORDER_NR_PAGES for memory holes
1174 if ((pfn
& (MAX_ORDER_NR_PAGES
- 1)) == 0) {
1175 if (!pfn_valid(pfn
)) {
1181 if (!meminit_pfn_in_nid(pfn
, nid
, &nid_init_state
)) {
1186 /* Minimise pfn page lookups and scheduler checks */
1187 if (page
&& (pfn
& (MAX_ORDER_NR_PAGES
- 1)) != 0) {
1190 nr_pages
+= nr_to_free
;
1191 deferred_free_range(free_base_page
,
1192 free_base_pfn
, nr_to_free
);
1193 free_base_page
= NULL
;
1194 free_base_pfn
= nr_to_free
= 0;
1196 page
= pfn_to_page(pfn
);
1201 VM_BUG_ON(page_zone(page
) != zone
);
1205 __init_single_page(page
, pfn
, zid
, nid
);
1206 if (!free_base_page
) {
1207 free_base_page
= page
;
1208 free_base_pfn
= pfn
;
1213 /* Where possible, batch up pages for a single free */
1216 /* Free the current block of pages to allocator */
1217 nr_pages
+= nr_to_free
;
1218 deferred_free_range(free_base_page
, free_base_pfn
,
1220 free_base_page
= NULL
;
1221 free_base_pfn
= nr_to_free
= 0;
1224 first_init_pfn
= max(end_pfn
, first_init_pfn
);
1227 /* Sanity check that the next zone really is unpopulated */
1228 WARN_ON(++zid
< MAX_NR_ZONES
&& populated_zone(++zone
));
1230 pr_info("node %d initialised, %lu pages in %ums\n", nid
, nr_pages
,
1231 jiffies_to_msecs(jiffies
- start
));
1233 pgdat_init_report_one_done();
1237 void __init
page_alloc_init_late(void)
1241 /* There will be num_node_state(N_MEMORY) threads */
1242 atomic_set(&pgdat_init_n_undone
, num_node_state(N_MEMORY
));
1243 for_each_node_state(nid
, N_MEMORY
) {
1244 kthread_run(deferred_init_memmap
, NODE_DATA(nid
), "pgdatinit%d", nid
);
1247 /* Block until all are initialised */
1248 wait_for_completion(&pgdat_init_all_done_comp
);
1250 /* Reinit limits that are based on free pages after the kernel is up */
1251 files_maxfiles_init();
1253 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1256 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1257 void __init
init_cma_reserved_pageblock(struct page
*page
)
1259 unsigned i
= pageblock_nr_pages
;
1260 struct page
*p
= page
;
1263 __ClearPageReserved(p
);
1264 set_page_count(p
, 0);
1267 set_pageblock_migratetype(page
, MIGRATE_CMA
);
1269 if (pageblock_order
>= MAX_ORDER
) {
1270 i
= pageblock_nr_pages
;
1273 set_page_refcounted(p
);
1274 __free_pages(p
, MAX_ORDER
- 1);
1275 p
+= MAX_ORDER_NR_PAGES
;
1276 } while (i
-= MAX_ORDER_NR_PAGES
);
1278 set_page_refcounted(page
);
1279 __free_pages(page
, pageblock_order
);
1282 adjust_managed_page_count(page
, pageblock_nr_pages
);
1287 * The order of subdivision here is critical for the IO subsystem.
1288 * Please do not alter this order without good reasons and regression
1289 * testing. Specifically, as large blocks of memory are subdivided,
1290 * the order in which smaller blocks are delivered depends on the order
1291 * they're subdivided in this function. This is the primary factor
1292 * influencing the order in which pages are delivered to the IO
1293 * subsystem according to empirical testing, and this is also justified
1294 * by considering the behavior of a buddy system containing a single
1295 * large block of memory acted on by a series of small allocations.
1296 * This behavior is a critical factor in sglist merging's success.
1300 static inline void expand(struct zone
*zone
, struct page
*page
,
1301 int low
, int high
, struct free_area
*area
,
1304 unsigned long size
= 1 << high
;
1306 while (high
> low
) {
1310 VM_BUG_ON_PAGE(bad_range(zone
, &page
[size
]), &page
[size
]);
1312 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC
) &&
1313 debug_guardpage_enabled() &&
1314 high
< debug_guardpage_minorder()) {
1316 * Mark as guard pages (or page), that will allow to
1317 * merge back to allocator when buddy will be freed.
1318 * Corresponding page table entries will not be touched,
1319 * pages will stay not present in virtual address space
1321 set_page_guard(zone
, &page
[size
], high
, migratetype
);
1324 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
1326 set_page_order(&page
[size
], high
);
1331 * This page is about to be returned from the page allocator
1333 static inline int check_new_page(struct page
*page
)
1335 const char *bad_reason
= NULL
;
1336 unsigned long bad_flags
= 0;
1338 if (unlikely(page_mapcount(page
)))
1339 bad_reason
= "nonzero mapcount";
1340 if (unlikely(page
->mapping
!= NULL
))
1341 bad_reason
= "non-NULL mapping";
1342 if (unlikely(atomic_read(&page
->_count
) != 0))
1343 bad_reason
= "nonzero _count";
1344 if (unlikely(page
->flags
& __PG_HWPOISON
)) {
1345 bad_reason
= "HWPoisoned (hardware-corrupted)";
1346 bad_flags
= __PG_HWPOISON
;
1348 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)) {
1349 bad_reason
= "PAGE_FLAGS_CHECK_AT_PREP flag set";
1350 bad_flags
= PAGE_FLAGS_CHECK_AT_PREP
;
1353 if (unlikely(page
->mem_cgroup
))
1354 bad_reason
= "page still charged to cgroup";
1356 if (unlikely(bad_reason
)) {
1357 bad_page(page
, bad_reason
, bad_flags
);
1363 static int prep_new_page(struct page
*page
, unsigned int order
, gfp_t gfp_flags
,
1368 for (i
= 0; i
< (1 << order
); i
++) {
1369 struct page
*p
= page
+ i
;
1370 if (unlikely(check_new_page(p
)))
1374 set_page_private(page
, 0);
1375 set_page_refcounted(page
);
1377 arch_alloc_page(page
, order
);
1378 kernel_map_pages(page
, 1 << order
, 1);
1379 kasan_alloc_pages(page
, order
);
1381 if (gfp_flags
& __GFP_ZERO
)
1382 for (i
= 0; i
< (1 << order
); i
++)
1383 clear_highpage(page
+ i
);
1385 if (order
&& (gfp_flags
& __GFP_COMP
))
1386 prep_compound_page(page
, order
);
1388 set_page_owner(page
, order
, gfp_flags
);
1391 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1392 * allocate the page. The expectation is that the caller is taking
1393 * steps that will free more memory. The caller should avoid the page
1394 * being used for !PFMEMALLOC purposes.
1396 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
1397 set_page_pfmemalloc(page
);
1399 clear_page_pfmemalloc(page
);
1405 * Go through the free lists for the given migratetype and remove
1406 * the smallest available page from the freelists
1409 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
1412 unsigned int current_order
;
1413 struct free_area
*area
;
1416 /* Find a page of the appropriate size in the preferred list */
1417 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
1418 area
= &(zone
->free_area
[current_order
]);
1419 page
= list_first_entry_or_null(&area
->free_list
[migratetype
],
1423 list_del(&page
->lru
);
1424 rmv_page_order(page
);
1426 expand(zone
, page
, order
, current_order
, area
, migratetype
);
1427 set_pcppage_migratetype(page
, migratetype
);
1436 * This array describes the order lists are fallen back to when
1437 * the free lists for the desirable migrate type are depleted
1439 static int fallbacks
[MIGRATE_TYPES
][4] = {
1440 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1441 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1442 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_TYPES
},
1444 [MIGRATE_CMA
] = { MIGRATE_TYPES
}, /* Never used */
1446 #ifdef CONFIG_MEMORY_ISOLATION
1447 [MIGRATE_ISOLATE
] = { MIGRATE_TYPES
}, /* Never used */
1452 static struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1455 return __rmqueue_smallest(zone
, order
, MIGRATE_CMA
);
1458 static inline struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1459 unsigned int order
) { return NULL
; }
1463 * Move the free pages in a range to the free lists of the requested type.
1464 * Note that start_page and end_pages are not aligned on a pageblock
1465 * boundary. If alignment is required, use move_freepages_block()
1467 int move_freepages(struct zone
*zone
,
1468 struct page
*start_page
, struct page
*end_page
,
1473 int pages_moved
= 0;
1475 #ifndef CONFIG_HOLES_IN_ZONE
1477 * page_zone is not safe to call in this context when
1478 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1479 * anyway as we check zone boundaries in move_freepages_block().
1480 * Remove at a later date when no bug reports exist related to
1481 * grouping pages by mobility
1483 VM_BUG_ON(page_zone(start_page
) != page_zone(end_page
));
1486 for (page
= start_page
; page
<= end_page
;) {
1487 /* Make sure we are not inadvertently changing nodes */
1488 VM_BUG_ON_PAGE(page_to_nid(page
) != zone_to_nid(zone
), page
);
1490 if (!pfn_valid_within(page_to_pfn(page
))) {
1495 if (!PageBuddy(page
)) {
1500 order
= page_order(page
);
1501 list_move(&page
->lru
,
1502 &zone
->free_area
[order
].free_list
[migratetype
]);
1504 pages_moved
+= 1 << order
;
1510 int move_freepages_block(struct zone
*zone
, struct page
*page
,
1513 unsigned long start_pfn
, end_pfn
;
1514 struct page
*start_page
, *end_page
;
1516 start_pfn
= page_to_pfn(page
);
1517 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
1518 start_page
= pfn_to_page(start_pfn
);
1519 end_page
= start_page
+ pageblock_nr_pages
- 1;
1520 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
1522 /* Do not cross zone boundaries */
1523 if (!zone_spans_pfn(zone
, start_pfn
))
1525 if (!zone_spans_pfn(zone
, end_pfn
))
1528 return move_freepages(zone
, start_page
, end_page
, migratetype
);
1531 static void change_pageblock_range(struct page
*pageblock_page
,
1532 int start_order
, int migratetype
)
1534 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
1536 while (nr_pageblocks
--) {
1537 set_pageblock_migratetype(pageblock_page
, migratetype
);
1538 pageblock_page
+= pageblock_nr_pages
;
1543 * When we are falling back to another migratetype during allocation, try to
1544 * steal extra free pages from the same pageblocks to satisfy further
1545 * allocations, instead of polluting multiple pageblocks.
1547 * If we are stealing a relatively large buddy page, it is likely there will
1548 * be more free pages in the pageblock, so try to steal them all. For
1549 * reclaimable and unmovable allocations, we steal regardless of page size,
1550 * as fragmentation caused by those allocations polluting movable pageblocks
1551 * is worse than movable allocations stealing from unmovable and reclaimable
1554 static bool can_steal_fallback(unsigned int order
, int start_mt
)
1557 * Leaving this order check is intended, although there is
1558 * relaxed order check in next check. The reason is that
1559 * we can actually steal whole pageblock if this condition met,
1560 * but, below check doesn't guarantee it and that is just heuristic
1561 * so could be changed anytime.
1563 if (order
>= pageblock_order
)
1566 if (order
>= pageblock_order
/ 2 ||
1567 start_mt
== MIGRATE_RECLAIMABLE
||
1568 start_mt
== MIGRATE_UNMOVABLE
||
1569 page_group_by_mobility_disabled
)
1576 * This function implements actual steal behaviour. If order is large enough,
1577 * we can steal whole pageblock. If not, we first move freepages in this
1578 * pageblock and check whether half of pages are moved or not. If half of
1579 * pages are moved, we can change migratetype of pageblock and permanently
1580 * use it's pages as requested migratetype in the future.
1582 static void steal_suitable_fallback(struct zone
*zone
, struct page
*page
,
1585 unsigned int current_order
= page_order(page
);
1588 /* Take ownership for orders >= pageblock_order */
1589 if (current_order
>= pageblock_order
) {
1590 change_pageblock_range(page
, current_order
, start_type
);
1594 pages
= move_freepages_block(zone
, page
, start_type
);
1596 /* Claim the whole block if over half of it is free */
1597 if (pages
>= (1 << (pageblock_order
-1)) ||
1598 page_group_by_mobility_disabled
)
1599 set_pageblock_migratetype(page
, start_type
);
1603 * Check whether there is a suitable fallback freepage with requested order.
1604 * If only_stealable is true, this function returns fallback_mt only if
1605 * we can steal other freepages all together. This would help to reduce
1606 * fragmentation due to mixed migratetype pages in one pageblock.
1608 int find_suitable_fallback(struct free_area
*area
, unsigned int order
,
1609 int migratetype
, bool only_stealable
, bool *can_steal
)
1614 if (area
->nr_free
== 0)
1619 fallback_mt
= fallbacks
[migratetype
][i
];
1620 if (fallback_mt
== MIGRATE_TYPES
)
1623 if (list_empty(&area
->free_list
[fallback_mt
]))
1626 if (can_steal_fallback(order
, migratetype
))
1629 if (!only_stealable
)
1640 * Reserve a pageblock for exclusive use of high-order atomic allocations if
1641 * there are no empty page blocks that contain a page with a suitable order
1643 static void reserve_highatomic_pageblock(struct page
*page
, struct zone
*zone
,
1644 unsigned int alloc_order
)
1647 unsigned long max_managed
, flags
;
1650 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
1651 * Check is race-prone but harmless.
1653 max_managed
= (zone
->managed_pages
/ 100) + pageblock_nr_pages
;
1654 if (zone
->nr_reserved_highatomic
>= max_managed
)
1657 spin_lock_irqsave(&zone
->lock
, flags
);
1659 /* Recheck the nr_reserved_highatomic limit under the lock */
1660 if (zone
->nr_reserved_highatomic
>= max_managed
)
1664 mt
= get_pageblock_migratetype(page
);
1665 if (mt
!= MIGRATE_HIGHATOMIC
&&
1666 !is_migrate_isolate(mt
) && !is_migrate_cma(mt
)) {
1667 zone
->nr_reserved_highatomic
+= pageblock_nr_pages
;
1668 set_pageblock_migratetype(page
, MIGRATE_HIGHATOMIC
);
1669 move_freepages_block(zone
, page
, MIGRATE_HIGHATOMIC
);
1673 spin_unlock_irqrestore(&zone
->lock
, flags
);
1677 * Used when an allocation is about to fail under memory pressure. This
1678 * potentially hurts the reliability of high-order allocations when under
1679 * intense memory pressure but failed atomic allocations should be easier
1680 * to recover from than an OOM.
1682 static void unreserve_highatomic_pageblock(const struct alloc_context
*ac
)
1684 struct zonelist
*zonelist
= ac
->zonelist
;
1685 unsigned long flags
;
1691 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
, ac
->high_zoneidx
,
1693 /* Preserve at least one pageblock */
1694 if (zone
->nr_reserved_highatomic
<= pageblock_nr_pages
)
1697 spin_lock_irqsave(&zone
->lock
, flags
);
1698 for (order
= 0; order
< MAX_ORDER
; order
++) {
1699 struct free_area
*area
= &(zone
->free_area
[order
]);
1701 page
= list_first_entry_or_null(
1702 &area
->free_list
[MIGRATE_HIGHATOMIC
],
1708 * It should never happen but changes to locking could
1709 * inadvertently allow a per-cpu drain to add pages
1710 * to MIGRATE_HIGHATOMIC while unreserving so be safe
1711 * and watch for underflows.
1713 zone
->nr_reserved_highatomic
-= min(pageblock_nr_pages
,
1714 zone
->nr_reserved_highatomic
);
1717 * Convert to ac->migratetype and avoid the normal
1718 * pageblock stealing heuristics. Minimally, the caller
1719 * is doing the work and needs the pages. More
1720 * importantly, if the block was always converted to
1721 * MIGRATE_UNMOVABLE or another type then the number
1722 * of pageblocks that cannot be completely freed
1725 set_pageblock_migratetype(page
, ac
->migratetype
);
1726 move_freepages_block(zone
, page
, ac
->migratetype
);
1727 spin_unlock_irqrestore(&zone
->lock
, flags
);
1730 spin_unlock_irqrestore(&zone
->lock
, flags
);
1734 /* Remove an element from the buddy allocator from the fallback list */
1735 static inline struct page
*
1736 __rmqueue_fallback(struct zone
*zone
, unsigned int order
, int start_migratetype
)
1738 struct free_area
*area
;
1739 unsigned int current_order
;
1744 /* Find the largest possible block of pages in the other list */
1745 for (current_order
= MAX_ORDER
-1;
1746 current_order
>= order
&& current_order
<= MAX_ORDER
-1;
1748 area
= &(zone
->free_area
[current_order
]);
1749 fallback_mt
= find_suitable_fallback(area
, current_order
,
1750 start_migratetype
, false, &can_steal
);
1751 if (fallback_mt
== -1)
1754 page
= list_first_entry(&area
->free_list
[fallback_mt
],
1757 steal_suitable_fallback(zone
, page
, start_migratetype
);
1759 /* Remove the page from the freelists */
1761 list_del(&page
->lru
);
1762 rmv_page_order(page
);
1764 expand(zone
, page
, order
, current_order
, area
,
1767 * The pcppage_migratetype may differ from pageblock's
1768 * migratetype depending on the decisions in
1769 * find_suitable_fallback(). This is OK as long as it does not
1770 * differ for MIGRATE_CMA pageblocks. Those can be used as
1771 * fallback only via special __rmqueue_cma_fallback() function
1773 set_pcppage_migratetype(page
, start_migratetype
);
1775 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
1776 start_migratetype
, fallback_mt
);
1785 * Do the hard work of removing an element from the buddy allocator.
1786 * Call me with the zone->lock already held.
1788 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
1793 page
= __rmqueue_smallest(zone
, order
, migratetype
);
1794 if (unlikely(!page
)) {
1795 if (migratetype
== MIGRATE_MOVABLE
)
1796 page
= __rmqueue_cma_fallback(zone
, order
);
1799 page
= __rmqueue_fallback(zone
, order
, migratetype
);
1802 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
1807 * Obtain a specified number of elements from the buddy allocator, all under
1808 * a single hold of the lock, for efficiency. Add them to the supplied list.
1809 * Returns the number of new pages which were placed at *list.
1811 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
1812 unsigned long count
, struct list_head
*list
,
1813 int migratetype
, bool cold
)
1817 spin_lock(&zone
->lock
);
1818 for (i
= 0; i
< count
; ++i
) {
1819 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
1820 if (unlikely(page
== NULL
))
1824 * Split buddy pages returned by expand() are received here
1825 * in physical page order. The page is added to the callers and
1826 * list and the list head then moves forward. From the callers
1827 * perspective, the linked list is ordered by page number in
1828 * some conditions. This is useful for IO devices that can
1829 * merge IO requests if the physical pages are ordered
1833 list_add(&page
->lru
, list
);
1835 list_add_tail(&page
->lru
, list
);
1837 if (is_migrate_cma(get_pcppage_migratetype(page
)))
1838 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
,
1841 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
1842 spin_unlock(&zone
->lock
);
1848 * Called from the vmstat counter updater to drain pagesets of this
1849 * currently executing processor on remote nodes after they have
1852 * Note that this function must be called with the thread pinned to
1853 * a single processor.
1855 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
1857 unsigned long flags
;
1858 int to_drain
, batch
;
1860 local_irq_save(flags
);
1861 batch
= READ_ONCE(pcp
->batch
);
1862 to_drain
= min(pcp
->count
, batch
);
1864 free_pcppages_bulk(zone
, to_drain
, pcp
);
1865 pcp
->count
-= to_drain
;
1867 local_irq_restore(flags
);
1872 * Drain pcplists of the indicated processor and zone.
1874 * The processor must either be the current processor and the
1875 * thread pinned to the current processor or a processor that
1878 static void drain_pages_zone(unsigned int cpu
, struct zone
*zone
)
1880 unsigned long flags
;
1881 struct per_cpu_pageset
*pset
;
1882 struct per_cpu_pages
*pcp
;
1884 local_irq_save(flags
);
1885 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
1889 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
1892 local_irq_restore(flags
);
1896 * Drain pcplists of all zones on the indicated processor.
1898 * The processor must either be the current processor and the
1899 * thread pinned to the current processor or a processor that
1902 static void drain_pages(unsigned int cpu
)
1906 for_each_populated_zone(zone
) {
1907 drain_pages_zone(cpu
, zone
);
1912 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1914 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
1915 * the single zone's pages.
1917 void drain_local_pages(struct zone
*zone
)
1919 int cpu
= smp_processor_id();
1922 drain_pages_zone(cpu
, zone
);
1928 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1930 * When zone parameter is non-NULL, spill just the single zone's pages.
1932 * Note that this code is protected against sending an IPI to an offline
1933 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1934 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1935 * nothing keeps CPUs from showing up after we populated the cpumask and
1936 * before the call to on_each_cpu_mask().
1938 void drain_all_pages(struct zone
*zone
)
1943 * Allocate in the BSS so we wont require allocation in
1944 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1946 static cpumask_t cpus_with_pcps
;
1949 * We don't care about racing with CPU hotplug event
1950 * as offline notification will cause the notified
1951 * cpu to drain that CPU pcps and on_each_cpu_mask
1952 * disables preemption as part of its processing
1954 for_each_online_cpu(cpu
) {
1955 struct per_cpu_pageset
*pcp
;
1957 bool has_pcps
= false;
1960 pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
1964 for_each_populated_zone(z
) {
1965 pcp
= per_cpu_ptr(z
->pageset
, cpu
);
1966 if (pcp
->pcp
.count
) {
1974 cpumask_set_cpu(cpu
, &cpus_with_pcps
);
1976 cpumask_clear_cpu(cpu
, &cpus_with_pcps
);
1978 on_each_cpu_mask(&cpus_with_pcps
, (smp_call_func_t
) drain_local_pages
,
1982 #ifdef CONFIG_HIBERNATION
1984 void mark_free_pages(struct zone
*zone
)
1986 unsigned long pfn
, max_zone_pfn
;
1987 unsigned long flags
;
1988 unsigned int order
, t
;
1991 if (zone_is_empty(zone
))
1994 spin_lock_irqsave(&zone
->lock
, flags
);
1996 max_zone_pfn
= zone_end_pfn(zone
);
1997 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
1998 if (pfn_valid(pfn
)) {
1999 page
= pfn_to_page(pfn
);
2000 if (!swsusp_page_is_forbidden(page
))
2001 swsusp_unset_page_free(page
);
2004 for_each_migratetype_order(order
, t
) {
2005 list_for_each_entry(page
,
2006 &zone
->free_area
[order
].free_list
[t
], lru
) {
2009 pfn
= page_to_pfn(page
);
2010 for (i
= 0; i
< (1UL << order
); i
++)
2011 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
2014 spin_unlock_irqrestore(&zone
->lock
, flags
);
2016 #endif /* CONFIG_PM */
2019 * Free a 0-order page
2020 * cold == true ? free a cold page : free a hot page
2022 void free_hot_cold_page(struct page
*page
, bool cold
)
2024 struct zone
*zone
= page_zone(page
);
2025 struct per_cpu_pages
*pcp
;
2026 unsigned long flags
;
2027 unsigned long pfn
= page_to_pfn(page
);
2030 if (!free_pages_prepare(page
, 0))
2033 migratetype
= get_pfnblock_migratetype(page
, pfn
);
2034 set_pcppage_migratetype(page
, migratetype
);
2035 local_irq_save(flags
);
2036 __count_vm_event(PGFREE
);
2039 * We only track unmovable, reclaimable and movable on pcp lists.
2040 * Free ISOLATE pages back to the allocator because they are being
2041 * offlined but treat RESERVE as movable pages so we can get those
2042 * areas back if necessary. Otherwise, we may have to free
2043 * excessively into the page allocator
2045 if (migratetype
>= MIGRATE_PCPTYPES
) {
2046 if (unlikely(is_migrate_isolate(migratetype
))) {
2047 free_one_page(zone
, page
, pfn
, 0, migratetype
);
2050 migratetype
= MIGRATE_MOVABLE
;
2053 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2055 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
2057 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
2059 if (pcp
->count
>= pcp
->high
) {
2060 unsigned long batch
= READ_ONCE(pcp
->batch
);
2061 free_pcppages_bulk(zone
, batch
, pcp
);
2062 pcp
->count
-= batch
;
2066 local_irq_restore(flags
);
2070 * Free a list of 0-order pages
2072 void free_hot_cold_page_list(struct list_head
*list
, bool cold
)
2074 struct page
*page
, *next
;
2076 list_for_each_entry_safe(page
, next
, list
, lru
) {
2077 trace_mm_page_free_batched(page
, cold
);
2078 free_hot_cold_page(page
, cold
);
2083 * split_page takes a non-compound higher-order page, and splits it into
2084 * n (1<<order) sub-pages: page[0..n]
2085 * Each sub-page must be freed individually.
2087 * Note: this is probably too low level an operation for use in drivers.
2088 * Please consult with lkml before using this in your driver.
2090 void split_page(struct page
*page
, unsigned int order
)
2095 VM_BUG_ON_PAGE(PageCompound(page
), page
);
2096 VM_BUG_ON_PAGE(!page_count(page
), page
);
2098 #ifdef CONFIG_KMEMCHECK
2100 * Split shadow pages too, because free(page[0]) would
2101 * otherwise free the whole shadow.
2103 if (kmemcheck_page_is_tracked(page
))
2104 split_page(virt_to_page(page
[0].shadow
), order
);
2107 gfp_mask
= get_page_owner_gfp(page
);
2108 set_page_owner(page
, 0, gfp_mask
);
2109 for (i
= 1; i
< (1 << order
); i
++) {
2110 set_page_refcounted(page
+ i
);
2111 set_page_owner(page
+ i
, 0, gfp_mask
);
2114 EXPORT_SYMBOL_GPL(split_page
);
2116 int __isolate_free_page(struct page
*page
, unsigned int order
)
2118 unsigned long watermark
;
2122 BUG_ON(!PageBuddy(page
));
2124 zone
= page_zone(page
);
2125 mt
= get_pageblock_migratetype(page
);
2127 if (!is_migrate_isolate(mt
)) {
2128 /* Obey watermarks as if the page was being allocated */
2129 watermark
= low_wmark_pages(zone
) + (1 << order
);
2130 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
2133 __mod_zone_freepage_state(zone
, -(1UL << order
), mt
);
2136 /* Remove page from free list */
2137 list_del(&page
->lru
);
2138 zone
->free_area
[order
].nr_free
--;
2139 rmv_page_order(page
);
2141 set_page_owner(page
, order
, __GFP_MOVABLE
);
2143 /* Set the pageblock if the isolated page is at least a pageblock */
2144 if (order
>= pageblock_order
- 1) {
2145 struct page
*endpage
= page
+ (1 << order
) - 1;
2146 for (; page
< endpage
; page
+= pageblock_nr_pages
) {
2147 int mt
= get_pageblock_migratetype(page
);
2148 if (!is_migrate_isolate(mt
) && !is_migrate_cma(mt
))
2149 set_pageblock_migratetype(page
,
2155 return 1UL << order
;
2159 * Similar to split_page except the page is already free. As this is only
2160 * being used for migration, the migratetype of the block also changes.
2161 * As this is called with interrupts disabled, the caller is responsible
2162 * for calling arch_alloc_page() and kernel_map_page() after interrupts
2165 * Note: this is probably too low level an operation for use in drivers.
2166 * Please consult with lkml before using this in your driver.
2168 int split_free_page(struct page
*page
)
2173 order
= page_order(page
);
2175 nr_pages
= __isolate_free_page(page
, order
);
2179 /* Split into individual pages */
2180 set_page_refcounted(page
);
2181 split_page(page
, order
);
2186 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2189 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
2190 struct zone
*zone
, unsigned int order
,
2191 gfp_t gfp_flags
, int alloc_flags
, int migratetype
)
2193 unsigned long flags
;
2195 bool cold
= ((gfp_flags
& __GFP_COLD
) != 0);
2197 if (likely(order
== 0)) {
2198 struct per_cpu_pages
*pcp
;
2199 struct list_head
*list
;
2201 local_irq_save(flags
);
2202 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2203 list
= &pcp
->lists
[migratetype
];
2204 if (list_empty(list
)) {
2205 pcp
->count
+= rmqueue_bulk(zone
, 0,
2208 if (unlikely(list_empty(list
)))
2213 page
= list_last_entry(list
, struct page
, lru
);
2215 page
= list_first_entry(list
, struct page
, lru
);
2217 list_del(&page
->lru
);
2220 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
2222 * __GFP_NOFAIL is not to be used in new code.
2224 * All __GFP_NOFAIL callers should be fixed so that they
2225 * properly detect and handle allocation failures.
2227 * We most definitely don't want callers attempting to
2228 * allocate greater than order-1 page units with
2231 WARN_ON_ONCE(order
> 1);
2233 spin_lock_irqsave(&zone
->lock
, flags
);
2236 if (alloc_flags
& ALLOC_HARDER
) {
2237 page
= __rmqueue_smallest(zone
, order
, MIGRATE_HIGHATOMIC
);
2239 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
2242 page
= __rmqueue(zone
, order
, migratetype
);
2243 spin_unlock(&zone
->lock
);
2246 __mod_zone_freepage_state(zone
, -(1 << order
),
2247 get_pcppage_migratetype(page
));
2250 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
, -(1 << order
));
2251 if (atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]) <= 0 &&
2252 !test_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
))
2253 set_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
);
2255 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
2256 zone_statistics(preferred_zone
, zone
, gfp_flags
);
2257 local_irq_restore(flags
);
2259 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
2263 local_irq_restore(flags
);
2267 #ifdef CONFIG_FAIL_PAGE_ALLOC
2270 struct fault_attr attr
;
2272 bool ignore_gfp_highmem
;
2273 bool ignore_gfp_reclaim
;
2275 } fail_page_alloc
= {
2276 .attr
= FAULT_ATTR_INITIALIZER
,
2277 .ignore_gfp_reclaim
= true,
2278 .ignore_gfp_highmem
= true,
2282 static int __init
setup_fail_page_alloc(char *str
)
2284 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
2286 __setup("fail_page_alloc=", setup_fail_page_alloc
);
2288 static bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2290 if (order
< fail_page_alloc
.min_order
)
2292 if (gfp_mask
& __GFP_NOFAIL
)
2294 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
2296 if (fail_page_alloc
.ignore_gfp_reclaim
&&
2297 (gfp_mask
& __GFP_DIRECT_RECLAIM
))
2300 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
2303 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2305 static int __init
fail_page_alloc_debugfs(void)
2307 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
2310 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
2311 &fail_page_alloc
.attr
);
2313 return PTR_ERR(dir
);
2315 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
2316 &fail_page_alloc
.ignore_gfp_reclaim
))
2318 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
2319 &fail_page_alloc
.ignore_gfp_highmem
))
2321 if (!debugfs_create_u32("min-order", mode
, dir
,
2322 &fail_page_alloc
.min_order
))
2327 debugfs_remove_recursive(dir
);
2332 late_initcall(fail_page_alloc_debugfs
);
2334 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2336 #else /* CONFIG_FAIL_PAGE_ALLOC */
2338 static inline bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2343 #endif /* CONFIG_FAIL_PAGE_ALLOC */
2346 * Return true if free base pages are above 'mark'. For high-order checks it
2347 * will return true of the order-0 watermark is reached and there is at least
2348 * one free page of a suitable size. Checking now avoids taking the zone lock
2349 * to check in the allocation paths if no pages are free.
2351 static bool __zone_watermark_ok(struct zone
*z
, unsigned int order
,
2352 unsigned long mark
, int classzone_idx
, int alloc_flags
,
2357 const int alloc_harder
= (alloc_flags
& ALLOC_HARDER
);
2359 /* free_pages may go negative - that's OK */
2360 free_pages
-= (1 << order
) - 1;
2362 if (alloc_flags
& ALLOC_HIGH
)
2366 * If the caller does not have rights to ALLOC_HARDER then subtract
2367 * the high-atomic reserves. This will over-estimate the size of the
2368 * atomic reserve but it avoids a search.
2370 if (likely(!alloc_harder
))
2371 free_pages
-= z
->nr_reserved_highatomic
;
2376 /* If allocation can't use CMA areas don't use free CMA pages */
2377 if (!(alloc_flags
& ALLOC_CMA
))
2378 free_pages
-= zone_page_state(z
, NR_FREE_CMA_PAGES
);
2382 * Check watermarks for an order-0 allocation request. If these
2383 * are not met, then a high-order request also cannot go ahead
2384 * even if a suitable page happened to be free.
2386 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
2389 /* If this is an order-0 request then the watermark is fine */
2393 /* For a high-order request, check at least one suitable page is free */
2394 for (o
= order
; o
< MAX_ORDER
; o
++) {
2395 struct free_area
*area
= &z
->free_area
[o
];
2404 for (mt
= 0; mt
< MIGRATE_PCPTYPES
; mt
++) {
2405 if (!list_empty(&area
->free_list
[mt
]))
2410 if ((alloc_flags
& ALLOC_CMA
) &&
2411 !list_empty(&area
->free_list
[MIGRATE_CMA
])) {
2419 bool zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
2420 int classzone_idx
, int alloc_flags
)
2422 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
2423 zone_page_state(z
, NR_FREE_PAGES
));
2426 bool zone_watermark_ok_safe(struct zone
*z
, unsigned int order
,
2427 unsigned long mark
, int classzone_idx
)
2429 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
2431 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
2432 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
2434 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, 0,
2439 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
2441 return local_zone
->node
== zone
->node
;
2444 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2446 return node_distance(zone_to_nid(local_zone
), zone_to_nid(zone
)) <
2449 #else /* CONFIG_NUMA */
2450 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
2455 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2459 #endif /* CONFIG_NUMA */
2461 static void reset_alloc_batches(struct zone
*preferred_zone
)
2463 struct zone
*zone
= preferred_zone
->zone_pgdat
->node_zones
;
2466 mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
2467 high_wmark_pages(zone
) - low_wmark_pages(zone
) -
2468 atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]));
2469 clear_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
);
2470 } while (zone
++ != preferred_zone
);
2474 * get_page_from_freelist goes through the zonelist trying to allocate
2477 static struct page
*
2478 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
, int alloc_flags
,
2479 const struct alloc_context
*ac
)
2481 struct zonelist
*zonelist
= ac
->zonelist
;
2483 struct page
*page
= NULL
;
2485 int nr_fair_skipped
= 0;
2486 bool zonelist_rescan
;
2489 zonelist_rescan
= false;
2492 * Scan zonelist, looking for a zone with enough free.
2493 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2495 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
, ac
->high_zoneidx
,
2499 if (cpusets_enabled() &&
2500 (alloc_flags
& ALLOC_CPUSET
) &&
2501 !cpuset_zone_allowed(zone
, gfp_mask
))
2504 * Distribute pages in proportion to the individual
2505 * zone size to ensure fair page aging. The zone a
2506 * page was allocated in should have no effect on the
2507 * time the page has in memory before being reclaimed.
2509 if (alloc_flags
& ALLOC_FAIR
) {
2510 if (!zone_local(ac
->preferred_zone
, zone
))
2512 if (test_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
)) {
2518 * When allocating a page cache page for writing, we
2519 * want to get it from a zone that is within its dirty
2520 * limit, such that no single zone holds more than its
2521 * proportional share of globally allowed dirty pages.
2522 * The dirty limits take into account the zone's
2523 * lowmem reserves and high watermark so that kswapd
2524 * should be able to balance it without having to
2525 * write pages from its LRU list.
2527 * This may look like it could increase pressure on
2528 * lower zones by failing allocations in higher zones
2529 * before they are full. But the pages that do spill
2530 * over are limited as the lower zones are protected
2531 * by this very same mechanism. It should not become
2532 * a practical burden to them.
2534 * XXX: For now, allow allocations to potentially
2535 * exceed the per-zone dirty limit in the slowpath
2536 * (spread_dirty_pages unset) before going into reclaim,
2537 * which is important when on a NUMA setup the allowed
2538 * zones are together not big enough to reach the
2539 * global limit. The proper fix for these situations
2540 * will require awareness of zones in the
2541 * dirty-throttling and the flusher threads.
2543 if (ac
->spread_dirty_pages
&& !zone_dirty_ok(zone
))
2546 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
2547 if (!zone_watermark_ok(zone
, order
, mark
,
2548 ac
->classzone_idx
, alloc_flags
)) {
2551 /* Checked here to keep the fast path fast */
2552 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
2553 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
2556 if (zone_reclaim_mode
== 0 ||
2557 !zone_allows_reclaim(ac
->preferred_zone
, zone
))
2560 ret
= zone_reclaim(zone
, gfp_mask
, order
);
2562 case ZONE_RECLAIM_NOSCAN
:
2565 case ZONE_RECLAIM_FULL
:
2566 /* scanned but unreclaimable */
2569 /* did we reclaim enough */
2570 if (zone_watermark_ok(zone
, order
, mark
,
2571 ac
->classzone_idx
, alloc_flags
))
2579 page
= buffered_rmqueue(ac
->preferred_zone
, zone
, order
,
2580 gfp_mask
, alloc_flags
, ac
->migratetype
);
2582 if (prep_new_page(page
, order
, gfp_mask
, alloc_flags
))
2586 * If this is a high-order atomic allocation then check
2587 * if the pageblock should be reserved for the future
2589 if (unlikely(order
&& (alloc_flags
& ALLOC_HARDER
)))
2590 reserve_highatomic_pageblock(page
, zone
, order
);
2597 * The first pass makes sure allocations are spread fairly within the
2598 * local node. However, the local node might have free pages left
2599 * after the fairness batches are exhausted, and remote zones haven't
2600 * even been considered yet. Try once more without fairness, and
2601 * include remote zones now, before entering the slowpath and waking
2602 * kswapd: prefer spilling to a remote zone over swapping locally.
2604 if (alloc_flags
& ALLOC_FAIR
) {
2605 alloc_flags
&= ~ALLOC_FAIR
;
2606 if (nr_fair_skipped
) {
2607 zonelist_rescan
= true;
2608 reset_alloc_batches(ac
->preferred_zone
);
2610 if (nr_online_nodes
> 1)
2611 zonelist_rescan
= true;
2614 if (zonelist_rescan
)
2621 * Large machines with many possible nodes should not always dump per-node
2622 * meminfo in irq context.
2624 static inline bool should_suppress_show_mem(void)
2629 ret
= in_interrupt();
2634 static DEFINE_RATELIMIT_STATE(nopage_rs
,
2635 DEFAULT_RATELIMIT_INTERVAL
,
2636 DEFAULT_RATELIMIT_BURST
);
2638 void warn_alloc_failed(gfp_t gfp_mask
, unsigned int order
, const char *fmt
, ...)
2640 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
2642 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
) ||
2643 debug_guardpage_minorder() > 0)
2647 * This documents exceptions given to allocations in certain
2648 * contexts that are allowed to allocate outside current's set
2651 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2652 if (test_thread_flag(TIF_MEMDIE
) ||
2653 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
2654 filter
&= ~SHOW_MEM_FILTER_NODES
;
2655 if (in_interrupt() || !(gfp_mask
& __GFP_DIRECT_RECLAIM
))
2656 filter
&= ~SHOW_MEM_FILTER_NODES
;
2659 struct va_format vaf
;
2662 va_start(args
, fmt
);
2667 pr_warn("%pV", &vaf
);
2672 pr_warn("%s: page allocation failure: order:%u, mode:0x%x\n",
2673 current
->comm
, order
, gfp_mask
);
2676 if (!should_suppress_show_mem())
2680 static inline struct page
*
2681 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
2682 const struct alloc_context
*ac
, unsigned long *did_some_progress
)
2684 struct oom_control oc
= {
2685 .zonelist
= ac
->zonelist
,
2686 .nodemask
= ac
->nodemask
,
2687 .gfp_mask
= gfp_mask
,
2692 *did_some_progress
= 0;
2695 * Acquire the oom lock. If that fails, somebody else is
2696 * making progress for us.
2698 if (!mutex_trylock(&oom_lock
)) {
2699 *did_some_progress
= 1;
2700 schedule_timeout_uninterruptible(1);
2705 * Go through the zonelist yet one more time, keep very high watermark
2706 * here, this is only to catch a parallel oom killing, we must fail if
2707 * we're still under heavy pressure.
2709 page
= get_page_from_freelist(gfp_mask
| __GFP_HARDWALL
, order
,
2710 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
, ac
);
2714 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2715 /* Coredumps can quickly deplete all memory reserves */
2716 if (current
->flags
& PF_DUMPCORE
)
2718 /* The OOM killer will not help higher order allocs */
2719 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2721 /* The OOM killer does not needlessly kill tasks for lowmem */
2722 if (ac
->high_zoneidx
< ZONE_NORMAL
)
2724 /* The OOM killer does not compensate for IO-less reclaim */
2725 if (!(gfp_mask
& __GFP_FS
)) {
2727 * XXX: Page reclaim didn't yield anything,
2728 * and the OOM killer can't be invoked, but
2729 * keep looping as per tradition.
2731 *did_some_progress
= 1;
2734 if (pm_suspended_storage())
2736 /* The OOM killer may not free memory on a specific node */
2737 if (gfp_mask
& __GFP_THISNODE
)
2740 /* Exhausted what can be done so it's blamo time */
2741 if (out_of_memory(&oc
) || WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
)) {
2742 *did_some_progress
= 1;
2744 if (gfp_mask
& __GFP_NOFAIL
) {
2745 page
= get_page_from_freelist(gfp_mask
, order
,
2746 ALLOC_NO_WATERMARKS
|ALLOC_CPUSET
, ac
);
2748 * fallback to ignore cpuset restriction if our nodes
2752 page
= get_page_from_freelist(gfp_mask
, order
,
2753 ALLOC_NO_WATERMARKS
, ac
);
2757 mutex_unlock(&oom_lock
);
2761 #ifdef CONFIG_COMPACTION
2762 /* Try memory compaction for high-order allocations before reclaim */
2763 static struct page
*
2764 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2765 int alloc_flags
, const struct alloc_context
*ac
,
2766 enum migrate_mode mode
, int *contended_compaction
,
2767 bool *deferred_compaction
)
2769 unsigned long compact_result
;
2775 current
->flags
|= PF_MEMALLOC
;
2776 compact_result
= try_to_compact_pages(gfp_mask
, order
, alloc_flags
, ac
,
2777 mode
, contended_compaction
);
2778 current
->flags
&= ~PF_MEMALLOC
;
2780 switch (compact_result
) {
2781 case COMPACT_DEFERRED
:
2782 *deferred_compaction
= true;
2784 case COMPACT_SKIPPED
:
2791 * At least in one zone compaction wasn't deferred or skipped, so let's
2792 * count a compaction stall
2794 count_vm_event(COMPACTSTALL
);
2796 page
= get_page_from_freelist(gfp_mask
, order
,
2797 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
2800 struct zone
*zone
= page_zone(page
);
2802 zone
->compact_blockskip_flush
= false;
2803 compaction_defer_reset(zone
, order
, true);
2804 count_vm_event(COMPACTSUCCESS
);
2809 * It's bad if compaction run occurs and fails. The most likely reason
2810 * is that pages exist, but not enough to satisfy watermarks.
2812 count_vm_event(COMPACTFAIL
);
2819 static inline struct page
*
2820 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2821 int alloc_flags
, const struct alloc_context
*ac
,
2822 enum migrate_mode mode
, int *contended_compaction
,
2823 bool *deferred_compaction
)
2827 #endif /* CONFIG_COMPACTION */
2829 /* Perform direct synchronous page reclaim */
2831 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
,
2832 const struct alloc_context
*ac
)
2834 struct reclaim_state reclaim_state
;
2839 /* We now go into synchronous reclaim */
2840 cpuset_memory_pressure_bump();
2841 current
->flags
|= PF_MEMALLOC
;
2842 lockdep_set_current_reclaim_state(gfp_mask
);
2843 reclaim_state
.reclaimed_slab
= 0;
2844 current
->reclaim_state
= &reclaim_state
;
2846 progress
= try_to_free_pages(ac
->zonelist
, order
, gfp_mask
,
2849 current
->reclaim_state
= NULL
;
2850 lockdep_clear_current_reclaim_state();
2851 current
->flags
&= ~PF_MEMALLOC
;
2858 /* The really slow allocator path where we enter direct reclaim */
2859 static inline struct page
*
2860 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
2861 int alloc_flags
, const struct alloc_context
*ac
,
2862 unsigned long *did_some_progress
)
2864 struct page
*page
= NULL
;
2865 bool drained
= false;
2867 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, ac
);
2868 if (unlikely(!(*did_some_progress
)))
2872 page
= get_page_from_freelist(gfp_mask
, order
,
2873 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
2876 * If an allocation failed after direct reclaim, it could be because
2877 * pages are pinned on the per-cpu lists or in high alloc reserves.
2878 * Shrink them them and try again
2880 if (!page
&& !drained
) {
2881 unreserve_highatomic_pageblock(ac
);
2882 drain_all_pages(NULL
);
2890 static void wake_all_kswapds(unsigned int order
, const struct alloc_context
*ac
)
2895 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
,
2896 ac
->high_zoneidx
, ac
->nodemask
)
2897 wakeup_kswapd(zone
, order
, zone_idx(ac
->preferred_zone
));
2901 gfp_to_alloc_flags(gfp_t gfp_mask
)
2903 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
2905 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2906 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
2909 * The caller may dip into page reserves a bit more if the caller
2910 * cannot run direct reclaim, or if the caller has realtime scheduling
2911 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2912 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
2914 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
2916 if (gfp_mask
& __GFP_ATOMIC
) {
2918 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2919 * if it can't schedule.
2921 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2922 alloc_flags
|= ALLOC_HARDER
;
2924 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
2925 * comment for __cpuset_node_allowed().
2927 alloc_flags
&= ~ALLOC_CPUSET
;
2928 } else if (unlikely(rt_task(current
)) && !in_interrupt())
2929 alloc_flags
|= ALLOC_HARDER
;
2931 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
2932 if (gfp_mask
& __GFP_MEMALLOC
)
2933 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2934 else if (in_serving_softirq() && (current
->flags
& PF_MEMALLOC
))
2935 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2936 else if (!in_interrupt() &&
2937 ((current
->flags
& PF_MEMALLOC
) ||
2938 unlikely(test_thread_flag(TIF_MEMDIE
))))
2939 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2942 if (gfpflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
2943 alloc_flags
|= ALLOC_CMA
;
2948 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask
)
2950 return !!(gfp_to_alloc_flags(gfp_mask
) & ALLOC_NO_WATERMARKS
);
2953 static inline bool is_thp_gfp_mask(gfp_t gfp_mask
)
2955 return (gfp_mask
& (GFP_TRANSHUGE
| __GFP_KSWAPD_RECLAIM
)) == GFP_TRANSHUGE
;
2958 static inline struct page
*
2959 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
2960 struct alloc_context
*ac
)
2962 bool can_direct_reclaim
= gfp_mask
& __GFP_DIRECT_RECLAIM
;
2963 struct page
*page
= NULL
;
2965 unsigned long pages_reclaimed
= 0;
2966 unsigned long did_some_progress
;
2967 enum migrate_mode migration_mode
= MIGRATE_ASYNC
;
2968 bool deferred_compaction
= false;
2969 int contended_compaction
= COMPACT_CONTENDED_NONE
;
2972 * In the slowpath, we sanity check order to avoid ever trying to
2973 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2974 * be using allocators in order of preference for an area that is
2977 if (order
>= MAX_ORDER
) {
2978 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
2983 * We also sanity check to catch abuse of atomic reserves being used by
2984 * callers that are not in atomic context.
2986 if (WARN_ON_ONCE((gfp_mask
& (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)) ==
2987 (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)))
2988 gfp_mask
&= ~__GFP_ATOMIC
;
2991 * If this allocation cannot block and it is for a specific node, then
2992 * fail early. There's no need to wakeup kswapd or retry for a
2993 * speculative node-specific allocation.
2995 if (IS_ENABLED(CONFIG_NUMA
) && (gfp_mask
& __GFP_THISNODE
) && !can_direct_reclaim
)
2999 if (gfp_mask
& __GFP_KSWAPD_RECLAIM
)
3000 wake_all_kswapds(order
, ac
);
3003 * OK, we're below the kswapd watermark and have kicked background
3004 * reclaim. Now things get more complex, so set up alloc_flags according
3005 * to how we want to proceed.
3007 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
3010 * Find the true preferred zone if the allocation is unconstrained by
3013 if (!(alloc_flags
& ALLOC_CPUSET
) && !ac
->nodemask
) {
3014 struct zoneref
*preferred_zoneref
;
3015 preferred_zoneref
= first_zones_zonelist(ac
->zonelist
,
3016 ac
->high_zoneidx
, NULL
, &ac
->preferred_zone
);
3017 ac
->classzone_idx
= zonelist_zone_idx(preferred_zoneref
);
3020 /* This is the last chance, in general, before the goto nopage. */
3021 page
= get_page_from_freelist(gfp_mask
, order
,
3022 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
3026 /* Allocate without watermarks if the context allows */
3027 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
3029 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3030 * the allocation is high priority and these type of
3031 * allocations are system rather than user orientated
3033 ac
->zonelist
= node_zonelist(numa_node_id(), gfp_mask
);
3034 page
= get_page_from_freelist(gfp_mask
, order
,
3035 ALLOC_NO_WATERMARKS
, ac
);
3040 /* Caller is not willing to reclaim, we can't balance anything */
3041 if (!can_direct_reclaim
) {
3043 * All existing users of the __GFP_NOFAIL are blockable, so warn
3044 * of any new users that actually allow this type of allocation
3047 WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
);
3051 /* Avoid recursion of direct reclaim */
3052 if (current
->flags
& PF_MEMALLOC
) {
3054 * __GFP_NOFAIL request from this context is rather bizarre
3055 * because we cannot reclaim anything and only can loop waiting
3056 * for somebody to do a work for us.
3058 if (WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
)) {
3065 /* Avoid allocations with no watermarks from looping endlessly */
3066 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
3070 * Try direct compaction. The first pass is asynchronous. Subsequent
3071 * attempts after direct reclaim are synchronous
3073 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
, ac
,
3075 &contended_compaction
,
3076 &deferred_compaction
);
3080 /* Checks for THP-specific high-order allocations */
3081 if (is_thp_gfp_mask(gfp_mask
)) {
3083 * If compaction is deferred for high-order allocations, it is
3084 * because sync compaction recently failed. If this is the case
3085 * and the caller requested a THP allocation, we do not want
3086 * to heavily disrupt the system, so we fail the allocation
3087 * instead of entering direct reclaim.
3089 if (deferred_compaction
)
3093 * In all zones where compaction was attempted (and not
3094 * deferred or skipped), lock contention has been detected.
3095 * For THP allocation we do not want to disrupt the others
3096 * so we fallback to base pages instead.
3098 if (contended_compaction
== COMPACT_CONTENDED_LOCK
)
3102 * If compaction was aborted due to need_resched(), we do not
3103 * want to further increase allocation latency, unless it is
3104 * khugepaged trying to collapse.
3106 if (contended_compaction
== COMPACT_CONTENDED_SCHED
3107 && !(current
->flags
& PF_KTHREAD
))
3112 * It can become very expensive to allocate transparent hugepages at
3113 * fault, so use asynchronous memory compaction for THP unless it is
3114 * khugepaged trying to collapse.
3116 if (!is_thp_gfp_mask(gfp_mask
) || (current
->flags
& PF_KTHREAD
))
3117 migration_mode
= MIGRATE_SYNC_LIGHT
;
3119 /* Try direct reclaim and then allocating */
3120 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
, alloc_flags
, ac
,
3121 &did_some_progress
);
3125 /* Do not loop if specifically requested */
3126 if (gfp_mask
& __GFP_NORETRY
)
3129 /* Keep reclaiming pages as long as there is reasonable progress */
3130 pages_reclaimed
+= did_some_progress
;
3131 if ((did_some_progress
&& order
<= PAGE_ALLOC_COSTLY_ORDER
) ||
3132 ((gfp_mask
& __GFP_REPEAT
) && pages_reclaimed
< (1 << order
))) {
3133 /* Wait for some write requests to complete then retry */
3134 wait_iff_congested(ac
->preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
3138 /* Reclaim has failed us, start killing things */
3139 page
= __alloc_pages_may_oom(gfp_mask
, order
, ac
, &did_some_progress
);
3143 /* Retry as long as the OOM killer is making progress */
3144 if (did_some_progress
)
3149 * High-order allocations do not necessarily loop after
3150 * direct reclaim and reclaim/compaction depends on compaction
3151 * being called after reclaim so call directly if necessary
3153 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
,
3155 &contended_compaction
,
3156 &deferred_compaction
);
3160 warn_alloc_failed(gfp_mask
, order
, NULL
);
3166 * This is the 'heart' of the zoned buddy allocator.
3169 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
3170 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
3172 struct zoneref
*preferred_zoneref
;
3173 struct page
*page
= NULL
;
3174 unsigned int cpuset_mems_cookie
;
3175 int alloc_flags
= ALLOC_WMARK_LOW
|ALLOC_CPUSET
|ALLOC_FAIR
;
3176 gfp_t alloc_mask
; /* The gfp_t that was actually used for allocation */
3177 struct alloc_context ac
= {
3178 .high_zoneidx
= gfp_zone(gfp_mask
),
3179 .nodemask
= nodemask
,
3180 .migratetype
= gfpflags_to_migratetype(gfp_mask
),
3183 gfp_mask
&= gfp_allowed_mask
;
3185 lockdep_trace_alloc(gfp_mask
);
3187 might_sleep_if(gfp_mask
& __GFP_DIRECT_RECLAIM
);
3189 if (should_fail_alloc_page(gfp_mask
, order
))
3193 * Check the zones suitable for the gfp_mask contain at least one
3194 * valid zone. It's possible to have an empty zonelist as a result
3195 * of __GFP_THISNODE and a memoryless node
3197 if (unlikely(!zonelist
->_zonerefs
->zone
))
3200 if (IS_ENABLED(CONFIG_CMA
) && ac
.migratetype
== MIGRATE_MOVABLE
)
3201 alloc_flags
|= ALLOC_CMA
;
3204 cpuset_mems_cookie
= read_mems_allowed_begin();
3206 /* We set it here, as __alloc_pages_slowpath might have changed it */
3207 ac
.zonelist
= zonelist
;
3209 /* Dirty zone balancing only done in the fast path */
3210 ac
.spread_dirty_pages
= (gfp_mask
& __GFP_WRITE
);
3212 /* The preferred zone is used for statistics later */
3213 preferred_zoneref
= first_zones_zonelist(ac
.zonelist
, ac
.high_zoneidx
,
3214 ac
.nodemask
? : &cpuset_current_mems_allowed
,
3215 &ac
.preferred_zone
);
3216 if (!ac
.preferred_zone
)
3218 ac
.classzone_idx
= zonelist_zone_idx(preferred_zoneref
);
3220 /* First allocation attempt */
3221 alloc_mask
= gfp_mask
|__GFP_HARDWALL
;
3222 page
= get_page_from_freelist(alloc_mask
, order
, alloc_flags
, &ac
);
3223 if (unlikely(!page
)) {
3225 * Runtime PM, block IO and its error handling path
3226 * can deadlock because I/O on the device might not
3229 alloc_mask
= memalloc_noio_flags(gfp_mask
);
3230 ac
.spread_dirty_pages
= false;
3232 page
= __alloc_pages_slowpath(alloc_mask
, order
, &ac
);
3235 if (kmemcheck_enabled
&& page
)
3236 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
3238 trace_mm_page_alloc(page
, order
, alloc_mask
, ac
.migratetype
);
3242 * When updating a task's mems_allowed, it is possible to race with
3243 * parallel threads in such a way that an allocation can fail while
3244 * the mask is being updated. If a page allocation is about to fail,
3245 * check if the cpuset changed during allocation and if so, retry.
3247 if (unlikely(!page
&& read_mems_allowed_retry(cpuset_mems_cookie
)))
3252 EXPORT_SYMBOL(__alloc_pages_nodemask
);
3255 * Common helper functions.
3257 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
3262 * __get_free_pages() returns a 32-bit address, which cannot represent
3265 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
3267 page
= alloc_pages(gfp_mask
, order
);
3270 return (unsigned long) page_address(page
);
3272 EXPORT_SYMBOL(__get_free_pages
);
3274 unsigned long get_zeroed_page(gfp_t gfp_mask
)
3276 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
3278 EXPORT_SYMBOL(get_zeroed_page
);
3280 void __free_pages(struct page
*page
, unsigned int order
)
3282 if (put_page_testzero(page
)) {
3284 free_hot_cold_page(page
, false);
3286 __free_pages_ok(page
, order
);
3290 EXPORT_SYMBOL(__free_pages
);
3292 void free_pages(unsigned long addr
, unsigned int order
)
3295 VM_BUG_ON(!virt_addr_valid((void *)addr
));
3296 __free_pages(virt_to_page((void *)addr
), order
);
3300 EXPORT_SYMBOL(free_pages
);
3304 * An arbitrary-length arbitrary-offset area of memory which resides
3305 * within a 0 or higher order page. Multiple fragments within that page
3306 * are individually refcounted, in the page's reference counter.
3308 * The page_frag functions below provide a simple allocation framework for
3309 * page fragments. This is used by the network stack and network device
3310 * drivers to provide a backing region of memory for use as either an
3311 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3313 static struct page
*__page_frag_refill(struct page_frag_cache
*nc
,
3316 struct page
*page
= NULL
;
3317 gfp_t gfp
= gfp_mask
;
3319 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3320 gfp_mask
|= __GFP_COMP
| __GFP_NOWARN
| __GFP_NORETRY
|
3322 page
= alloc_pages_node(NUMA_NO_NODE
, gfp_mask
,
3323 PAGE_FRAG_CACHE_MAX_ORDER
);
3324 nc
->size
= page
? PAGE_FRAG_CACHE_MAX_SIZE
: PAGE_SIZE
;
3326 if (unlikely(!page
))
3327 page
= alloc_pages_node(NUMA_NO_NODE
, gfp
, 0);
3329 nc
->va
= page
? page_address(page
) : NULL
;
3334 void *__alloc_page_frag(struct page_frag_cache
*nc
,
3335 unsigned int fragsz
, gfp_t gfp_mask
)
3337 unsigned int size
= PAGE_SIZE
;
3341 if (unlikely(!nc
->va
)) {
3343 page
= __page_frag_refill(nc
, gfp_mask
);
3347 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3348 /* if size can vary use size else just use PAGE_SIZE */
3351 /* Even if we own the page, we do not use atomic_set().
3352 * This would break get_page_unless_zero() users.
3354 atomic_add(size
- 1, &page
->_count
);
3356 /* reset page count bias and offset to start of new frag */
3357 nc
->pfmemalloc
= page_is_pfmemalloc(page
);
3358 nc
->pagecnt_bias
= size
;
3362 offset
= nc
->offset
- fragsz
;
3363 if (unlikely(offset
< 0)) {
3364 page
= virt_to_page(nc
->va
);
3366 if (!atomic_sub_and_test(nc
->pagecnt_bias
, &page
->_count
))
3369 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3370 /* if size can vary use size else just use PAGE_SIZE */
3373 /* OK, page count is 0, we can safely set it */
3374 atomic_set(&page
->_count
, size
);
3376 /* reset page count bias and offset to start of new frag */
3377 nc
->pagecnt_bias
= size
;
3378 offset
= size
- fragsz
;
3382 nc
->offset
= offset
;
3384 return nc
->va
+ offset
;
3386 EXPORT_SYMBOL(__alloc_page_frag
);
3389 * Frees a page fragment allocated out of either a compound or order 0 page.
3391 void __free_page_frag(void *addr
)
3393 struct page
*page
= virt_to_head_page(addr
);
3395 if (unlikely(put_page_testzero(page
)))
3396 __free_pages_ok(page
, compound_order(page
));
3398 EXPORT_SYMBOL(__free_page_frag
);
3401 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
3402 * of the current memory cgroup if __GFP_ACCOUNT is set, other than that it is
3403 * equivalent to alloc_pages.
3405 * It should be used when the caller would like to use kmalloc, but since the
3406 * allocation is large, it has to fall back to the page allocator.
3408 struct page
*alloc_kmem_pages(gfp_t gfp_mask
, unsigned int order
)
3412 page
= alloc_pages(gfp_mask
, order
);
3413 if (page
&& memcg_kmem_charge(page
, gfp_mask
, order
) != 0) {
3414 __free_pages(page
, order
);
3420 struct page
*alloc_kmem_pages_node(int nid
, gfp_t gfp_mask
, unsigned int order
)
3424 page
= alloc_pages_node(nid
, gfp_mask
, order
);
3425 if (page
&& memcg_kmem_charge(page
, gfp_mask
, order
) != 0) {
3426 __free_pages(page
, order
);
3433 * __free_kmem_pages and free_kmem_pages will free pages allocated with
3436 void __free_kmem_pages(struct page
*page
, unsigned int order
)
3438 memcg_kmem_uncharge(page
, order
);
3439 __free_pages(page
, order
);
3442 void free_kmem_pages(unsigned long addr
, unsigned int order
)
3445 VM_BUG_ON(!virt_addr_valid((void *)addr
));
3446 __free_kmem_pages(virt_to_page((void *)addr
), order
);
3450 static void *make_alloc_exact(unsigned long addr
, unsigned int order
,
3454 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
3455 unsigned long used
= addr
+ PAGE_ALIGN(size
);
3457 split_page(virt_to_page((void *)addr
), order
);
3458 while (used
< alloc_end
) {
3463 return (void *)addr
;
3467 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3468 * @size: the number of bytes to allocate
3469 * @gfp_mask: GFP flags for the allocation
3471 * This function is similar to alloc_pages(), except that it allocates the
3472 * minimum number of pages to satisfy the request. alloc_pages() can only
3473 * allocate memory in power-of-two pages.
3475 * This function is also limited by MAX_ORDER.
3477 * Memory allocated by this function must be released by free_pages_exact().
3479 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
3481 unsigned int order
= get_order(size
);
3484 addr
= __get_free_pages(gfp_mask
, order
);
3485 return make_alloc_exact(addr
, order
, size
);
3487 EXPORT_SYMBOL(alloc_pages_exact
);
3490 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3492 * @nid: the preferred node ID where memory should be allocated
3493 * @size: the number of bytes to allocate
3494 * @gfp_mask: GFP flags for the allocation
3496 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3499 void * __meminit
alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
3501 unsigned int order
= get_order(size
);
3502 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
3505 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
3509 * free_pages_exact - release memory allocated via alloc_pages_exact()
3510 * @virt: the value returned by alloc_pages_exact.
3511 * @size: size of allocation, same value as passed to alloc_pages_exact().
3513 * Release the memory allocated by a previous call to alloc_pages_exact.
3515 void free_pages_exact(void *virt
, size_t size
)
3517 unsigned long addr
= (unsigned long)virt
;
3518 unsigned long end
= addr
+ PAGE_ALIGN(size
);
3520 while (addr
< end
) {
3525 EXPORT_SYMBOL(free_pages_exact
);
3528 * nr_free_zone_pages - count number of pages beyond high watermark
3529 * @offset: The zone index of the highest zone
3531 * nr_free_zone_pages() counts the number of counts pages which are beyond the
3532 * high watermark within all zones at or below a given zone index. For each
3533 * zone, the number of pages is calculated as:
3534 * managed_pages - high_pages
3536 static unsigned long nr_free_zone_pages(int offset
)
3541 /* Just pick one node, since fallback list is circular */
3542 unsigned long sum
= 0;
3544 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
3546 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
3547 unsigned long size
= zone
->managed_pages
;
3548 unsigned long high
= high_wmark_pages(zone
);
3557 * nr_free_buffer_pages - count number of pages beyond high watermark
3559 * nr_free_buffer_pages() counts the number of pages which are beyond the high
3560 * watermark within ZONE_DMA and ZONE_NORMAL.
3562 unsigned long nr_free_buffer_pages(void)
3564 return nr_free_zone_pages(gfp_zone(GFP_USER
));
3566 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
3569 * nr_free_pagecache_pages - count number of pages beyond high watermark
3571 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3572 * high watermark within all zones.
3574 unsigned long nr_free_pagecache_pages(void)
3576 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
3579 static inline void show_node(struct zone
*zone
)
3581 if (IS_ENABLED(CONFIG_NUMA
))
3582 printk("Node %d ", zone_to_nid(zone
));
3585 void si_meminfo(struct sysinfo
*val
)
3587 val
->totalram
= totalram_pages
;
3588 val
->sharedram
= global_page_state(NR_SHMEM
);
3589 val
->freeram
= global_page_state(NR_FREE_PAGES
);
3590 val
->bufferram
= nr_blockdev_pages();
3591 val
->totalhigh
= totalhigh_pages
;
3592 val
->freehigh
= nr_free_highpages();
3593 val
->mem_unit
= PAGE_SIZE
;
3596 EXPORT_SYMBOL(si_meminfo
);
3599 void si_meminfo_node(struct sysinfo
*val
, int nid
)
3601 int zone_type
; /* needs to be signed */
3602 unsigned long managed_pages
= 0;
3603 pg_data_t
*pgdat
= NODE_DATA(nid
);
3605 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++)
3606 managed_pages
+= pgdat
->node_zones
[zone_type
].managed_pages
;
3607 val
->totalram
= managed_pages
;
3608 val
->sharedram
= node_page_state(nid
, NR_SHMEM
);
3609 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
3610 #ifdef CONFIG_HIGHMEM
3611 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].managed_pages
;
3612 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
3618 val
->mem_unit
= PAGE_SIZE
;
3623 * Determine whether the node should be displayed or not, depending on whether
3624 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
3626 bool skip_free_areas_node(unsigned int flags
, int nid
)
3629 unsigned int cpuset_mems_cookie
;
3631 if (!(flags
& SHOW_MEM_FILTER_NODES
))
3635 cpuset_mems_cookie
= read_mems_allowed_begin();
3636 ret
= !node_isset(nid
, cpuset_current_mems_allowed
);
3637 } while (read_mems_allowed_retry(cpuset_mems_cookie
));
3642 #define K(x) ((x) << (PAGE_SHIFT-10))
3644 static void show_migration_types(unsigned char type
)
3646 static const char types
[MIGRATE_TYPES
] = {
3647 [MIGRATE_UNMOVABLE
] = 'U',
3648 [MIGRATE_MOVABLE
] = 'M',
3649 [MIGRATE_RECLAIMABLE
] = 'E',
3650 [MIGRATE_HIGHATOMIC
] = 'H',
3652 [MIGRATE_CMA
] = 'C',
3654 #ifdef CONFIG_MEMORY_ISOLATION
3655 [MIGRATE_ISOLATE
] = 'I',
3658 char tmp
[MIGRATE_TYPES
+ 1];
3662 for (i
= 0; i
< MIGRATE_TYPES
; i
++) {
3663 if (type
& (1 << i
))
3668 printk("(%s) ", tmp
);
3672 * Show free area list (used inside shift_scroll-lock stuff)
3673 * We also calculate the percentage fragmentation. We do this by counting the
3674 * memory on each free list with the exception of the first item on the list.
3677 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
3680 void show_free_areas(unsigned int filter
)
3682 unsigned long free_pcp
= 0;
3686 for_each_populated_zone(zone
) {
3687 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3690 for_each_online_cpu(cpu
)
3691 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
3694 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3695 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3696 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
3697 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3698 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3699 " free:%lu free_pcp:%lu free_cma:%lu\n",
3700 global_page_state(NR_ACTIVE_ANON
),
3701 global_page_state(NR_INACTIVE_ANON
),
3702 global_page_state(NR_ISOLATED_ANON
),
3703 global_page_state(NR_ACTIVE_FILE
),
3704 global_page_state(NR_INACTIVE_FILE
),
3705 global_page_state(NR_ISOLATED_FILE
),
3706 global_page_state(NR_UNEVICTABLE
),
3707 global_page_state(NR_FILE_DIRTY
),
3708 global_page_state(NR_WRITEBACK
),
3709 global_page_state(NR_UNSTABLE_NFS
),
3710 global_page_state(NR_SLAB_RECLAIMABLE
),
3711 global_page_state(NR_SLAB_UNRECLAIMABLE
),
3712 global_page_state(NR_FILE_MAPPED
),
3713 global_page_state(NR_SHMEM
),
3714 global_page_state(NR_PAGETABLE
),
3715 global_page_state(NR_BOUNCE
),
3716 global_page_state(NR_FREE_PAGES
),
3718 global_page_state(NR_FREE_CMA_PAGES
));
3720 for_each_populated_zone(zone
) {
3723 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3727 for_each_online_cpu(cpu
)
3728 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
3736 " active_anon:%lukB"
3737 " inactive_anon:%lukB"
3738 " active_file:%lukB"
3739 " inactive_file:%lukB"
3740 " unevictable:%lukB"
3741 " isolated(anon):%lukB"
3742 " isolated(file):%lukB"
3750 " slab_reclaimable:%lukB"
3751 " slab_unreclaimable:%lukB"
3752 " kernel_stack:%lukB"
3759 " writeback_tmp:%lukB"
3760 " pages_scanned:%lu"
3761 " all_unreclaimable? %s"
3764 K(zone_page_state(zone
, NR_FREE_PAGES
)),
3765 K(min_wmark_pages(zone
)),
3766 K(low_wmark_pages(zone
)),
3767 K(high_wmark_pages(zone
)),
3768 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
3769 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
3770 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
3771 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
3772 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
3773 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
3774 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
3775 K(zone
->present_pages
),
3776 K(zone
->managed_pages
),
3777 K(zone_page_state(zone
, NR_MLOCK
)),
3778 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
3779 K(zone_page_state(zone
, NR_WRITEBACK
)),
3780 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
3781 K(zone_page_state(zone
, NR_SHMEM
)),
3782 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
3783 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
3784 zone_page_state(zone
, NR_KERNEL_STACK
) *
3786 K(zone_page_state(zone
, NR_PAGETABLE
)),
3787 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
3788 K(zone_page_state(zone
, NR_BOUNCE
)),
3790 K(this_cpu_read(zone
->pageset
->pcp
.count
)),
3791 K(zone_page_state(zone
, NR_FREE_CMA_PAGES
)),
3792 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
3793 K(zone_page_state(zone
, NR_PAGES_SCANNED
)),
3794 (!zone_reclaimable(zone
) ? "yes" : "no")
3796 printk("lowmem_reserve[]:");
3797 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3798 printk(" %ld", zone
->lowmem_reserve
[i
]);
3802 for_each_populated_zone(zone
) {
3804 unsigned long nr
[MAX_ORDER
], flags
, total
= 0;
3805 unsigned char types
[MAX_ORDER
];
3807 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3810 printk("%s: ", zone
->name
);
3812 spin_lock_irqsave(&zone
->lock
, flags
);
3813 for (order
= 0; order
< MAX_ORDER
; order
++) {
3814 struct free_area
*area
= &zone
->free_area
[order
];
3817 nr
[order
] = area
->nr_free
;
3818 total
+= nr
[order
] << order
;
3821 for (type
= 0; type
< MIGRATE_TYPES
; type
++) {
3822 if (!list_empty(&area
->free_list
[type
]))
3823 types
[order
] |= 1 << type
;
3826 spin_unlock_irqrestore(&zone
->lock
, flags
);
3827 for (order
= 0; order
< MAX_ORDER
; order
++) {
3828 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
3830 show_migration_types(types
[order
]);
3832 printk("= %lukB\n", K(total
));
3835 hugetlb_show_meminfo();
3837 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
3839 show_swap_cache_info();
3842 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
3844 zoneref
->zone
= zone
;
3845 zoneref
->zone_idx
= zone_idx(zone
);
3849 * Builds allocation fallback zone lists.
3851 * Add all populated zones of a node to the zonelist.
3853 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
3857 enum zone_type zone_type
= MAX_NR_ZONES
;
3861 zone
= pgdat
->node_zones
+ zone_type
;
3862 if (populated_zone(zone
)) {
3863 zoneref_set_zone(zone
,
3864 &zonelist
->_zonerefs
[nr_zones
++]);
3865 check_highest_zone(zone_type
);
3867 } while (zone_type
);
3875 * 0 = automatic detection of better ordering.
3876 * 1 = order by ([node] distance, -zonetype)
3877 * 2 = order by (-zonetype, [node] distance)
3879 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3880 * the same zonelist. So only NUMA can configure this param.
3882 #define ZONELIST_ORDER_DEFAULT 0
3883 #define ZONELIST_ORDER_NODE 1
3884 #define ZONELIST_ORDER_ZONE 2
3886 /* zonelist order in the kernel.
3887 * set_zonelist_order() will set this to NODE or ZONE.
3889 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3890 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
3894 /* The value user specified ....changed by config */
3895 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3896 /* string for sysctl */
3897 #define NUMA_ZONELIST_ORDER_LEN 16
3898 char numa_zonelist_order
[16] = "default";
3901 * interface for configure zonelist ordering.
3902 * command line option "numa_zonelist_order"
3903 * = "[dD]efault - default, automatic configuration.
3904 * = "[nN]ode - order by node locality, then by zone within node
3905 * = "[zZ]one - order by zone, then by locality within zone
3908 static int __parse_numa_zonelist_order(char *s
)
3910 if (*s
== 'd' || *s
== 'D') {
3911 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3912 } else if (*s
== 'n' || *s
== 'N') {
3913 user_zonelist_order
= ZONELIST_ORDER_NODE
;
3914 } else if (*s
== 'z' || *s
== 'Z') {
3915 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
3918 "Ignoring invalid numa_zonelist_order value: "
3925 static __init
int setup_numa_zonelist_order(char *s
)
3932 ret
= __parse_numa_zonelist_order(s
);
3934 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
3938 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
3941 * sysctl handler for numa_zonelist_order
3943 int numa_zonelist_order_handler(struct ctl_table
*table
, int write
,
3944 void __user
*buffer
, size_t *length
,
3947 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
3949 static DEFINE_MUTEX(zl_order_mutex
);
3951 mutex_lock(&zl_order_mutex
);
3953 if (strlen((char *)table
->data
) >= NUMA_ZONELIST_ORDER_LEN
) {
3957 strcpy(saved_string
, (char *)table
->data
);
3959 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
3963 int oldval
= user_zonelist_order
;
3965 ret
= __parse_numa_zonelist_order((char *)table
->data
);
3968 * bogus value. restore saved string
3970 strncpy((char *)table
->data
, saved_string
,
3971 NUMA_ZONELIST_ORDER_LEN
);
3972 user_zonelist_order
= oldval
;
3973 } else if (oldval
!= user_zonelist_order
) {
3974 mutex_lock(&zonelists_mutex
);
3975 build_all_zonelists(NULL
, NULL
);
3976 mutex_unlock(&zonelists_mutex
);
3980 mutex_unlock(&zl_order_mutex
);
3985 #define MAX_NODE_LOAD (nr_online_nodes)
3986 static int node_load
[MAX_NUMNODES
];
3989 * find_next_best_node - find the next node that should appear in a given node's fallback list
3990 * @node: node whose fallback list we're appending
3991 * @used_node_mask: nodemask_t of already used nodes
3993 * We use a number of factors to determine which is the next node that should
3994 * appear on a given node's fallback list. The node should not have appeared
3995 * already in @node's fallback list, and it should be the next closest node
3996 * according to the distance array (which contains arbitrary distance values
3997 * from each node to each node in the system), and should also prefer nodes
3998 * with no CPUs, since presumably they'll have very little allocation pressure
3999 * on them otherwise.
4000 * It returns -1 if no node is found.
4002 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
4005 int min_val
= INT_MAX
;
4006 int best_node
= NUMA_NO_NODE
;
4007 const struct cpumask
*tmp
= cpumask_of_node(0);
4009 /* Use the local node if we haven't already */
4010 if (!node_isset(node
, *used_node_mask
)) {
4011 node_set(node
, *used_node_mask
);
4015 for_each_node_state(n
, N_MEMORY
) {
4017 /* Don't want a node to appear more than once */
4018 if (node_isset(n
, *used_node_mask
))
4021 /* Use the distance array to find the distance */
4022 val
= node_distance(node
, n
);
4024 /* Penalize nodes under us ("prefer the next node") */
4027 /* Give preference to headless and unused nodes */
4028 tmp
= cpumask_of_node(n
);
4029 if (!cpumask_empty(tmp
))
4030 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
4032 /* Slight preference for less loaded node */
4033 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
4034 val
+= node_load
[n
];
4036 if (val
< min_val
) {
4043 node_set(best_node
, *used_node_mask
);
4050 * Build zonelists ordered by node and zones within node.
4051 * This results in maximum locality--normal zone overflows into local
4052 * DMA zone, if any--but risks exhausting DMA zone.
4054 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
4057 struct zonelist
*zonelist
;
4059 zonelist
= &pgdat
->node_zonelists
[0];
4060 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
4062 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4063 zonelist
->_zonerefs
[j
].zone
= NULL
;
4064 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4068 * Build gfp_thisnode zonelists
4070 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
4073 struct zonelist
*zonelist
;
4075 zonelist
= &pgdat
->node_zonelists
[1];
4076 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4077 zonelist
->_zonerefs
[j
].zone
= NULL
;
4078 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4082 * Build zonelists ordered by zone and nodes within zones.
4083 * This results in conserving DMA zone[s] until all Normal memory is
4084 * exhausted, but results in overflowing to remote node while memory
4085 * may still exist in local DMA zone.
4087 static int node_order
[MAX_NUMNODES
];
4089 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
4092 int zone_type
; /* needs to be signed */
4094 struct zonelist
*zonelist
;
4096 zonelist
= &pgdat
->node_zonelists
[0];
4098 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
4099 for (j
= 0; j
< nr_nodes
; j
++) {
4100 node
= node_order
[j
];
4101 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
4102 if (populated_zone(z
)) {
4104 &zonelist
->_zonerefs
[pos
++]);
4105 check_highest_zone(zone_type
);
4109 zonelist
->_zonerefs
[pos
].zone
= NULL
;
4110 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
4113 #if defined(CONFIG_64BIT)
4115 * Devices that require DMA32/DMA are relatively rare and do not justify a
4116 * penalty to every machine in case the specialised case applies. Default
4117 * to Node-ordering on 64-bit NUMA machines
4119 static int default_zonelist_order(void)
4121 return ZONELIST_ORDER_NODE
;
4125 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4126 * by the kernel. If processes running on node 0 deplete the low memory zone
4127 * then reclaim will occur more frequency increasing stalls and potentially
4128 * be easier to OOM if a large percentage of the zone is under writeback or
4129 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4130 * Hence, default to zone ordering on 32-bit.
4132 static int default_zonelist_order(void)
4134 return ZONELIST_ORDER_ZONE
;
4136 #endif /* CONFIG_64BIT */
4138 static void set_zonelist_order(void)
4140 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
4141 current_zonelist_order
= default_zonelist_order();
4143 current_zonelist_order
= user_zonelist_order
;
4146 static void build_zonelists(pg_data_t
*pgdat
)
4149 nodemask_t used_mask
;
4150 int local_node
, prev_node
;
4151 struct zonelist
*zonelist
;
4152 unsigned int order
= current_zonelist_order
;
4154 /* initialize zonelists */
4155 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
4156 zonelist
= pgdat
->node_zonelists
+ i
;
4157 zonelist
->_zonerefs
[0].zone
= NULL
;
4158 zonelist
->_zonerefs
[0].zone_idx
= 0;
4161 /* NUMA-aware ordering of nodes */
4162 local_node
= pgdat
->node_id
;
4163 load
= nr_online_nodes
;
4164 prev_node
= local_node
;
4165 nodes_clear(used_mask
);
4167 memset(node_order
, 0, sizeof(node_order
));
4170 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
4172 * We don't want to pressure a particular node.
4173 * So adding penalty to the first node in same
4174 * distance group to make it round-robin.
4176 if (node_distance(local_node
, node
) !=
4177 node_distance(local_node
, prev_node
))
4178 node_load
[node
] = load
;
4182 if (order
== ZONELIST_ORDER_NODE
)
4183 build_zonelists_in_node_order(pgdat
, node
);
4185 node_order
[i
++] = node
; /* remember order */
4188 if (order
== ZONELIST_ORDER_ZONE
) {
4189 /* calculate node order -- i.e., DMA last! */
4190 build_zonelists_in_zone_order(pgdat
, i
);
4193 build_thisnode_zonelists(pgdat
);
4196 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4198 * Return node id of node used for "local" allocations.
4199 * I.e., first node id of first zone in arg node's generic zonelist.
4200 * Used for initializing percpu 'numa_mem', which is used primarily
4201 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4203 int local_memory_node(int node
)
4207 (void)first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
4208 gfp_zone(GFP_KERNEL
),
4215 #else /* CONFIG_NUMA */
4217 static void set_zonelist_order(void)
4219 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
4222 static void build_zonelists(pg_data_t
*pgdat
)
4224 int node
, local_node
;
4226 struct zonelist
*zonelist
;
4228 local_node
= pgdat
->node_id
;
4230 zonelist
= &pgdat
->node_zonelists
[0];
4231 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4234 * Now we build the zonelist so that it contains the zones
4235 * of all the other nodes.
4236 * We don't want to pressure a particular node, so when
4237 * building the zones for node N, we make sure that the
4238 * zones coming right after the local ones are those from
4239 * node N+1 (modulo N)
4241 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
4242 if (!node_online(node
))
4244 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4246 for (node
= 0; node
< local_node
; node
++) {
4247 if (!node_online(node
))
4249 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4252 zonelist
->_zonerefs
[j
].zone
= NULL
;
4253 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4256 #endif /* CONFIG_NUMA */
4259 * Boot pageset table. One per cpu which is going to be used for all
4260 * zones and all nodes. The parameters will be set in such a way
4261 * that an item put on a list will immediately be handed over to
4262 * the buddy list. This is safe since pageset manipulation is done
4263 * with interrupts disabled.
4265 * The boot_pagesets must be kept even after bootup is complete for
4266 * unused processors and/or zones. They do play a role for bootstrapping
4267 * hotplugged processors.
4269 * zoneinfo_show() and maybe other functions do
4270 * not check if the processor is online before following the pageset pointer.
4271 * Other parts of the kernel may not check if the zone is available.
4273 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
4274 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
4275 static void setup_zone_pageset(struct zone
*zone
);
4278 * Global mutex to protect against size modification of zonelists
4279 * as well as to serialize pageset setup for the new populated zone.
4281 DEFINE_MUTEX(zonelists_mutex
);
4283 /* return values int ....just for stop_machine() */
4284 static int __build_all_zonelists(void *data
)
4288 pg_data_t
*self
= data
;
4291 memset(node_load
, 0, sizeof(node_load
));
4294 if (self
&& !node_online(self
->node_id
)) {
4295 build_zonelists(self
);
4298 for_each_online_node(nid
) {
4299 pg_data_t
*pgdat
= NODE_DATA(nid
);
4301 build_zonelists(pgdat
);
4305 * Initialize the boot_pagesets that are going to be used
4306 * for bootstrapping processors. The real pagesets for
4307 * each zone will be allocated later when the per cpu
4308 * allocator is available.
4310 * boot_pagesets are used also for bootstrapping offline
4311 * cpus if the system is already booted because the pagesets
4312 * are needed to initialize allocators on a specific cpu too.
4313 * F.e. the percpu allocator needs the page allocator which
4314 * needs the percpu allocator in order to allocate its pagesets
4315 * (a chicken-egg dilemma).
4317 for_each_possible_cpu(cpu
) {
4318 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
4320 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4322 * We now know the "local memory node" for each node--
4323 * i.e., the node of the first zone in the generic zonelist.
4324 * Set up numa_mem percpu variable for on-line cpus. During
4325 * boot, only the boot cpu should be on-line; we'll init the
4326 * secondary cpus' numa_mem as they come on-line. During
4327 * node/memory hotplug, we'll fixup all on-line cpus.
4329 if (cpu_online(cpu
))
4330 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
4337 static noinline
void __init
4338 build_all_zonelists_init(void)
4340 __build_all_zonelists(NULL
);
4341 mminit_verify_zonelist();
4342 cpuset_init_current_mems_allowed();
4346 * Called with zonelists_mutex held always
4347 * unless system_state == SYSTEM_BOOTING.
4349 * __ref due to (1) call of __meminit annotated setup_zone_pageset
4350 * [we're only called with non-NULL zone through __meminit paths] and
4351 * (2) call of __init annotated helper build_all_zonelists_init
4352 * [protected by SYSTEM_BOOTING].
4354 void __ref
build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
)
4356 set_zonelist_order();
4358 if (system_state
== SYSTEM_BOOTING
) {
4359 build_all_zonelists_init();
4361 #ifdef CONFIG_MEMORY_HOTPLUG
4363 setup_zone_pageset(zone
);
4365 /* we have to stop all cpus to guarantee there is no user
4367 stop_machine(__build_all_zonelists
, pgdat
, NULL
);
4368 /* cpuset refresh routine should be here */
4370 vm_total_pages
= nr_free_pagecache_pages();
4372 * Disable grouping by mobility if the number of pages in the
4373 * system is too low to allow the mechanism to work. It would be
4374 * more accurate, but expensive to check per-zone. This check is
4375 * made on memory-hotadd so a system can start with mobility
4376 * disabled and enable it later
4378 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
4379 page_group_by_mobility_disabled
= 1;
4381 page_group_by_mobility_disabled
= 0;
4383 pr_info("Built %i zonelists in %s order, mobility grouping %s. "
4384 "Total pages: %ld\n",
4386 zonelist_order_name
[current_zonelist_order
],
4387 page_group_by_mobility_disabled
? "off" : "on",
4390 pr_info("Policy zone: %s\n", zone_names
[policy_zone
]);
4395 * Helper functions to size the waitqueue hash table.
4396 * Essentially these want to choose hash table sizes sufficiently
4397 * large so that collisions trying to wait on pages are rare.
4398 * But in fact, the number of active page waitqueues on typical
4399 * systems is ridiculously low, less than 200. So this is even
4400 * conservative, even though it seems large.
4402 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
4403 * waitqueues, i.e. the size of the waitq table given the number of pages.
4405 #define PAGES_PER_WAITQUEUE 256
4407 #ifndef CONFIG_MEMORY_HOTPLUG
4408 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
4410 unsigned long size
= 1;
4412 pages
/= PAGES_PER_WAITQUEUE
;
4414 while (size
< pages
)
4418 * Once we have dozens or even hundreds of threads sleeping
4419 * on IO we've got bigger problems than wait queue collision.
4420 * Limit the size of the wait table to a reasonable size.
4422 size
= min(size
, 4096UL);
4424 return max(size
, 4UL);
4428 * A zone's size might be changed by hot-add, so it is not possible to determine
4429 * a suitable size for its wait_table. So we use the maximum size now.
4431 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
4433 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
4434 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4435 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
4437 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4438 * or more by the traditional way. (See above). It equals:
4440 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
4441 * ia64(16K page size) : = ( 8G + 4M)byte.
4442 * powerpc (64K page size) : = (32G +16M)byte.
4444 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
4451 * This is an integer logarithm so that shifts can be used later
4452 * to extract the more random high bits from the multiplicative
4453 * hash function before the remainder is taken.
4455 static inline unsigned long wait_table_bits(unsigned long size
)
4461 * Initially all pages are reserved - free ones are freed
4462 * up by free_all_bootmem() once the early boot process is
4463 * done. Non-atomic initialization, single-pass.
4465 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
4466 unsigned long start_pfn
, enum memmap_context context
)
4468 pg_data_t
*pgdat
= NODE_DATA(nid
);
4469 unsigned long end_pfn
= start_pfn
+ size
;
4472 unsigned long nr_initialised
= 0;
4474 if (highest_memmap_pfn
< end_pfn
- 1)
4475 highest_memmap_pfn
= end_pfn
- 1;
4477 z
= &pgdat
->node_zones
[zone
];
4478 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
4480 * There can be holes in boot-time mem_map[]s
4481 * handed to this function. They do not
4482 * exist on hotplugged memory.
4484 if (context
== MEMMAP_EARLY
) {
4485 if (!early_pfn_valid(pfn
))
4487 if (!early_pfn_in_nid(pfn
, nid
))
4489 if (!update_defer_init(pgdat
, pfn
, end_pfn
,
4495 * Mark the block movable so that blocks are reserved for
4496 * movable at startup. This will force kernel allocations
4497 * to reserve their blocks rather than leaking throughout
4498 * the address space during boot when many long-lived
4499 * kernel allocations are made.
4501 * bitmap is created for zone's valid pfn range. but memmap
4502 * can be created for invalid pages (for alignment)
4503 * check here not to call set_pageblock_migratetype() against
4506 if (!(pfn
& (pageblock_nr_pages
- 1))) {
4507 struct page
*page
= pfn_to_page(pfn
);
4509 __init_single_page(page
, pfn
, zone
, nid
);
4510 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4512 __init_single_pfn(pfn
, zone
, nid
);
4517 static void __meminit
zone_init_free_lists(struct zone
*zone
)
4519 unsigned int order
, t
;
4520 for_each_migratetype_order(order
, t
) {
4521 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
4522 zone
->free_area
[order
].nr_free
= 0;
4526 #ifndef __HAVE_ARCH_MEMMAP_INIT
4527 #define memmap_init(size, nid, zone, start_pfn) \
4528 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4531 static int zone_batchsize(struct zone
*zone
)
4537 * The per-cpu-pages pools are set to around 1000th of the
4538 * size of the zone. But no more than 1/2 of a meg.
4540 * OK, so we don't know how big the cache is. So guess.
4542 batch
= zone
->managed_pages
/ 1024;
4543 if (batch
* PAGE_SIZE
> 512 * 1024)
4544 batch
= (512 * 1024) / PAGE_SIZE
;
4545 batch
/= 4; /* We effectively *= 4 below */
4550 * Clamp the batch to a 2^n - 1 value. Having a power
4551 * of 2 value was found to be more likely to have
4552 * suboptimal cache aliasing properties in some cases.
4554 * For example if 2 tasks are alternately allocating
4555 * batches of pages, one task can end up with a lot
4556 * of pages of one half of the possible page colors
4557 * and the other with pages of the other colors.
4559 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
4564 /* The deferral and batching of frees should be suppressed under NOMMU
4567 * The problem is that NOMMU needs to be able to allocate large chunks
4568 * of contiguous memory as there's no hardware page translation to
4569 * assemble apparent contiguous memory from discontiguous pages.
4571 * Queueing large contiguous runs of pages for batching, however,
4572 * causes the pages to actually be freed in smaller chunks. As there
4573 * can be a significant delay between the individual batches being
4574 * recycled, this leads to the once large chunks of space being
4575 * fragmented and becoming unavailable for high-order allocations.
4582 * pcp->high and pcp->batch values are related and dependent on one another:
4583 * ->batch must never be higher then ->high.
4584 * The following function updates them in a safe manner without read side
4587 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4588 * those fields changing asynchronously (acording the the above rule).
4590 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4591 * outside of boot time (or some other assurance that no concurrent updaters
4594 static void pageset_update(struct per_cpu_pages
*pcp
, unsigned long high
,
4595 unsigned long batch
)
4597 /* start with a fail safe value for batch */
4601 /* Update high, then batch, in order */
4608 /* a companion to pageset_set_high() */
4609 static void pageset_set_batch(struct per_cpu_pageset
*p
, unsigned long batch
)
4611 pageset_update(&p
->pcp
, 6 * batch
, max(1UL, 1 * batch
));
4614 static void pageset_init(struct per_cpu_pageset
*p
)
4616 struct per_cpu_pages
*pcp
;
4619 memset(p
, 0, sizeof(*p
));
4623 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
4624 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
4627 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
4630 pageset_set_batch(p
, batch
);
4634 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
4635 * to the value high for the pageset p.
4637 static void pageset_set_high(struct per_cpu_pageset
*p
,
4640 unsigned long batch
= max(1UL, high
/ 4);
4641 if ((high
/ 4) > (PAGE_SHIFT
* 8))
4642 batch
= PAGE_SHIFT
* 8;
4644 pageset_update(&p
->pcp
, high
, batch
);
4647 static void pageset_set_high_and_batch(struct zone
*zone
,
4648 struct per_cpu_pageset
*pcp
)
4650 if (percpu_pagelist_fraction
)
4651 pageset_set_high(pcp
,
4652 (zone
->managed_pages
/
4653 percpu_pagelist_fraction
));
4655 pageset_set_batch(pcp
, zone_batchsize(zone
));
4658 static void __meminit
zone_pageset_init(struct zone
*zone
, int cpu
)
4660 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
4663 pageset_set_high_and_batch(zone
, pcp
);
4666 static void __meminit
setup_zone_pageset(struct zone
*zone
)
4669 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
4670 for_each_possible_cpu(cpu
)
4671 zone_pageset_init(zone
, cpu
);
4675 * Allocate per cpu pagesets and initialize them.
4676 * Before this call only boot pagesets were available.
4678 void __init
setup_per_cpu_pageset(void)
4682 for_each_populated_zone(zone
)
4683 setup_zone_pageset(zone
);
4686 static noinline __init_refok
4687 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
4693 * The per-page waitqueue mechanism uses hashed waitqueues
4696 zone
->wait_table_hash_nr_entries
=
4697 wait_table_hash_nr_entries(zone_size_pages
);
4698 zone
->wait_table_bits
=
4699 wait_table_bits(zone
->wait_table_hash_nr_entries
);
4700 alloc_size
= zone
->wait_table_hash_nr_entries
4701 * sizeof(wait_queue_head_t
);
4703 if (!slab_is_available()) {
4704 zone
->wait_table
= (wait_queue_head_t
*)
4705 memblock_virt_alloc_node_nopanic(
4706 alloc_size
, zone
->zone_pgdat
->node_id
);
4709 * This case means that a zone whose size was 0 gets new memory
4710 * via memory hot-add.
4711 * But it may be the case that a new node was hot-added. In
4712 * this case vmalloc() will not be able to use this new node's
4713 * memory - this wait_table must be initialized to use this new
4714 * node itself as well.
4715 * To use this new node's memory, further consideration will be
4718 zone
->wait_table
= vmalloc(alloc_size
);
4720 if (!zone
->wait_table
)
4723 for (i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
4724 init_waitqueue_head(zone
->wait_table
+ i
);
4729 static __meminit
void zone_pcp_init(struct zone
*zone
)
4732 * per cpu subsystem is not up at this point. The following code
4733 * relies on the ability of the linker to provide the
4734 * offset of a (static) per cpu variable into the per cpu area.
4736 zone
->pageset
= &boot_pageset
;
4738 if (populated_zone(zone
))
4739 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
4740 zone
->name
, zone
->present_pages
,
4741 zone_batchsize(zone
));
4744 int __meminit
init_currently_empty_zone(struct zone
*zone
,
4745 unsigned long zone_start_pfn
,
4748 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
4750 ret
= zone_wait_table_init(zone
, size
);
4753 pgdat
->nr_zones
= zone_idx(zone
) + 1;
4755 zone
->zone_start_pfn
= zone_start_pfn
;
4757 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
4758 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4760 (unsigned long)zone_idx(zone
),
4761 zone_start_pfn
, (zone_start_pfn
+ size
));
4763 zone_init_free_lists(zone
);
4768 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4769 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4772 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4774 int __meminit
__early_pfn_to_nid(unsigned long pfn
,
4775 struct mminit_pfnnid_cache
*state
)
4777 unsigned long start_pfn
, end_pfn
;
4780 if (state
->last_start
<= pfn
&& pfn
< state
->last_end
)
4781 return state
->last_nid
;
4783 nid
= memblock_search_pfn_nid(pfn
, &start_pfn
, &end_pfn
);
4785 state
->last_start
= start_pfn
;
4786 state
->last_end
= end_pfn
;
4787 state
->last_nid
= nid
;
4792 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4795 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
4796 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4797 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
4799 * If an architecture guarantees that all ranges registered contain no holes
4800 * and may be freed, this this function may be used instead of calling
4801 * memblock_free_early_nid() manually.
4803 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
4805 unsigned long start_pfn
, end_pfn
;
4808 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
4809 start_pfn
= min(start_pfn
, max_low_pfn
);
4810 end_pfn
= min(end_pfn
, max_low_pfn
);
4812 if (start_pfn
< end_pfn
)
4813 memblock_free_early_nid(PFN_PHYS(start_pfn
),
4814 (end_pfn
- start_pfn
) << PAGE_SHIFT
,
4820 * sparse_memory_present_with_active_regions - Call memory_present for each active range
4821 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4823 * If an architecture guarantees that all ranges registered contain no holes and may
4824 * be freed, this function may be used instead of calling memory_present() manually.
4826 void __init
sparse_memory_present_with_active_regions(int nid
)
4828 unsigned long start_pfn
, end_pfn
;
4831 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
4832 memory_present(this_nid
, start_pfn
, end_pfn
);
4836 * get_pfn_range_for_nid - Return the start and end page frames for a node
4837 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4838 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4839 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4841 * It returns the start and end page frame of a node based on information
4842 * provided by memblock_set_node(). If called for a node
4843 * with no available memory, a warning is printed and the start and end
4846 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
4847 unsigned long *start_pfn
, unsigned long *end_pfn
)
4849 unsigned long this_start_pfn
, this_end_pfn
;
4855 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
4856 *start_pfn
= min(*start_pfn
, this_start_pfn
);
4857 *end_pfn
= max(*end_pfn
, this_end_pfn
);
4860 if (*start_pfn
== -1UL)
4865 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4866 * assumption is made that zones within a node are ordered in monotonic
4867 * increasing memory addresses so that the "highest" populated zone is used
4869 static void __init
find_usable_zone_for_movable(void)
4872 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
4873 if (zone_index
== ZONE_MOVABLE
)
4876 if (arch_zone_highest_possible_pfn
[zone_index
] >
4877 arch_zone_lowest_possible_pfn
[zone_index
])
4881 VM_BUG_ON(zone_index
== -1);
4882 movable_zone
= zone_index
;
4886 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4887 * because it is sized independent of architecture. Unlike the other zones,
4888 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4889 * in each node depending on the size of each node and how evenly kernelcore
4890 * is distributed. This helper function adjusts the zone ranges
4891 * provided by the architecture for a given node by using the end of the
4892 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4893 * zones within a node are in order of monotonic increases memory addresses
4895 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
4896 unsigned long zone_type
,
4897 unsigned long node_start_pfn
,
4898 unsigned long node_end_pfn
,
4899 unsigned long *zone_start_pfn
,
4900 unsigned long *zone_end_pfn
)
4902 /* Only adjust if ZONE_MOVABLE is on this node */
4903 if (zone_movable_pfn
[nid
]) {
4904 /* Size ZONE_MOVABLE */
4905 if (zone_type
== ZONE_MOVABLE
) {
4906 *zone_start_pfn
= zone_movable_pfn
[nid
];
4907 *zone_end_pfn
= min(node_end_pfn
,
4908 arch_zone_highest_possible_pfn
[movable_zone
]);
4910 /* Adjust for ZONE_MOVABLE starting within this range */
4911 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
4912 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
4913 *zone_end_pfn
= zone_movable_pfn
[nid
];
4915 /* Check if this whole range is within ZONE_MOVABLE */
4916 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
4917 *zone_start_pfn
= *zone_end_pfn
;
4922 * Return the number of pages a zone spans in a node, including holes
4923 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4925 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
4926 unsigned long zone_type
,
4927 unsigned long node_start_pfn
,
4928 unsigned long node_end_pfn
,
4929 unsigned long *ignored
)
4931 unsigned long zone_start_pfn
, zone_end_pfn
;
4933 /* When hotadd a new node from cpu_up(), the node should be empty */
4934 if (!node_start_pfn
&& !node_end_pfn
)
4937 /* Get the start and end of the zone */
4938 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
4939 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
4940 adjust_zone_range_for_zone_movable(nid
, zone_type
,
4941 node_start_pfn
, node_end_pfn
,
4942 &zone_start_pfn
, &zone_end_pfn
);
4944 /* Check that this node has pages within the zone's required range */
4945 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
4948 /* Move the zone boundaries inside the node if necessary */
4949 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
4950 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
4952 /* Return the spanned pages */
4953 return zone_end_pfn
- zone_start_pfn
;
4957 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
4958 * then all holes in the requested range will be accounted for.
4960 unsigned long __meminit
__absent_pages_in_range(int nid
,
4961 unsigned long range_start_pfn
,
4962 unsigned long range_end_pfn
)
4964 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
4965 unsigned long start_pfn
, end_pfn
;
4968 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
4969 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
4970 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
4971 nr_absent
-= end_pfn
- start_pfn
;
4977 * absent_pages_in_range - Return number of page frames in holes within a range
4978 * @start_pfn: The start PFN to start searching for holes
4979 * @end_pfn: The end PFN to stop searching for holes
4981 * It returns the number of pages frames in memory holes within a range.
4983 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
4984 unsigned long end_pfn
)
4986 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
4989 /* Return the number of page frames in holes in a zone on a node */
4990 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
4991 unsigned long zone_type
,
4992 unsigned long node_start_pfn
,
4993 unsigned long node_end_pfn
,
4994 unsigned long *ignored
)
4996 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
4997 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
4998 unsigned long zone_start_pfn
, zone_end_pfn
;
5000 /* When hotadd a new node from cpu_up(), the node should be empty */
5001 if (!node_start_pfn
&& !node_end_pfn
)
5004 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
5005 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
5007 adjust_zone_range_for_zone_movable(nid
, zone_type
,
5008 node_start_pfn
, node_end_pfn
,
5009 &zone_start_pfn
, &zone_end_pfn
);
5010 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
5013 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5014 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
5015 unsigned long zone_type
,
5016 unsigned long node_start_pfn
,
5017 unsigned long node_end_pfn
,
5018 unsigned long *zones_size
)
5020 return zones_size
[zone_type
];
5023 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5024 unsigned long zone_type
,
5025 unsigned long node_start_pfn
,
5026 unsigned long node_end_pfn
,
5027 unsigned long *zholes_size
)
5032 return zholes_size
[zone_type
];
5035 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5037 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
5038 unsigned long node_start_pfn
,
5039 unsigned long node_end_pfn
,
5040 unsigned long *zones_size
,
5041 unsigned long *zholes_size
)
5043 unsigned long realtotalpages
= 0, totalpages
= 0;
5046 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5047 struct zone
*zone
= pgdat
->node_zones
+ i
;
5048 unsigned long size
, real_size
;
5050 size
= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
5054 real_size
= size
- zone_absent_pages_in_node(pgdat
->node_id
, i
,
5055 node_start_pfn
, node_end_pfn
,
5057 zone
->spanned_pages
= size
;
5058 zone
->present_pages
= real_size
;
5061 realtotalpages
+= real_size
;
5064 pgdat
->node_spanned_pages
= totalpages
;
5065 pgdat
->node_present_pages
= realtotalpages
;
5066 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
5070 #ifndef CONFIG_SPARSEMEM
5072 * Calculate the size of the zone->blockflags rounded to an unsigned long
5073 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5074 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
5075 * round what is now in bits to nearest long in bits, then return it in
5078 static unsigned long __init
usemap_size(unsigned long zone_start_pfn
, unsigned long zonesize
)
5080 unsigned long usemapsize
;
5082 zonesize
+= zone_start_pfn
& (pageblock_nr_pages
-1);
5083 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
5084 usemapsize
= usemapsize
>> pageblock_order
;
5085 usemapsize
*= NR_PAGEBLOCK_BITS
;
5086 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
5088 return usemapsize
/ 8;
5091 static void __init
setup_usemap(struct pglist_data
*pgdat
,
5093 unsigned long zone_start_pfn
,
5094 unsigned long zonesize
)
5096 unsigned long usemapsize
= usemap_size(zone_start_pfn
, zonesize
);
5097 zone
->pageblock_flags
= NULL
;
5099 zone
->pageblock_flags
=
5100 memblock_virt_alloc_node_nopanic(usemapsize
,
5104 static inline void setup_usemap(struct pglist_data
*pgdat
, struct zone
*zone
,
5105 unsigned long zone_start_pfn
, unsigned long zonesize
) {}
5106 #endif /* CONFIG_SPARSEMEM */
5108 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
5110 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
5111 void __paginginit
set_pageblock_order(void)
5115 /* Check that pageblock_nr_pages has not already been setup */
5116 if (pageblock_order
)
5119 if (HPAGE_SHIFT
> PAGE_SHIFT
)
5120 order
= HUGETLB_PAGE_ORDER
;
5122 order
= MAX_ORDER
- 1;
5125 * Assume the largest contiguous order of interest is a huge page.
5126 * This value may be variable depending on boot parameters on IA64 and
5129 pageblock_order
= order
;
5131 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5134 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
5135 * is unused as pageblock_order is set at compile-time. See
5136 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5139 void __paginginit
set_pageblock_order(void)
5143 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5145 static unsigned long __paginginit
calc_memmap_size(unsigned long spanned_pages
,
5146 unsigned long present_pages
)
5148 unsigned long pages
= spanned_pages
;
5151 * Provide a more accurate estimation if there are holes within
5152 * the zone and SPARSEMEM is in use. If there are holes within the
5153 * zone, each populated memory region may cost us one or two extra
5154 * memmap pages due to alignment because memmap pages for each
5155 * populated regions may not naturally algined on page boundary.
5156 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5158 if (spanned_pages
> present_pages
+ (present_pages
>> 4) &&
5159 IS_ENABLED(CONFIG_SPARSEMEM
))
5160 pages
= present_pages
;
5162 return PAGE_ALIGN(pages
* sizeof(struct page
)) >> PAGE_SHIFT
;
5166 * Set up the zone data structures:
5167 * - mark all pages reserved
5168 * - mark all memory queues empty
5169 * - clear the memory bitmaps
5171 * NOTE: pgdat should get zeroed by caller.
5173 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
)
5176 int nid
= pgdat
->node_id
;
5177 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
5180 pgdat_resize_init(pgdat
);
5181 #ifdef CONFIG_NUMA_BALANCING
5182 spin_lock_init(&pgdat
->numabalancing_migrate_lock
);
5183 pgdat
->numabalancing_migrate_nr_pages
= 0;
5184 pgdat
->numabalancing_migrate_next_window
= jiffies
;
5186 init_waitqueue_head(&pgdat
->kswapd_wait
);
5187 init_waitqueue_head(&pgdat
->pfmemalloc_wait
);
5188 pgdat_page_ext_init(pgdat
);
5190 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5191 struct zone
*zone
= pgdat
->node_zones
+ j
;
5192 unsigned long size
, realsize
, freesize
, memmap_pages
;
5194 size
= zone
->spanned_pages
;
5195 realsize
= freesize
= zone
->present_pages
;
5198 * Adjust freesize so that it accounts for how much memory
5199 * is used by this zone for memmap. This affects the watermark
5200 * and per-cpu initialisations
5202 memmap_pages
= calc_memmap_size(size
, realsize
);
5203 if (!is_highmem_idx(j
)) {
5204 if (freesize
>= memmap_pages
) {
5205 freesize
-= memmap_pages
;
5208 " %s zone: %lu pages used for memmap\n",
5209 zone_names
[j
], memmap_pages
);
5212 " %s zone: %lu pages exceeds freesize %lu\n",
5213 zone_names
[j
], memmap_pages
, freesize
);
5216 /* Account for reserved pages */
5217 if (j
== 0 && freesize
> dma_reserve
) {
5218 freesize
-= dma_reserve
;
5219 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
5220 zone_names
[0], dma_reserve
);
5223 if (!is_highmem_idx(j
))
5224 nr_kernel_pages
+= freesize
;
5225 /* Charge for highmem memmap if there are enough kernel pages */
5226 else if (nr_kernel_pages
> memmap_pages
* 2)
5227 nr_kernel_pages
-= memmap_pages
;
5228 nr_all_pages
+= freesize
;
5231 * Set an approximate value for lowmem here, it will be adjusted
5232 * when the bootmem allocator frees pages into the buddy system.
5233 * And all highmem pages will be managed by the buddy system.
5235 zone
->managed_pages
= is_highmem_idx(j
) ? realsize
: freesize
;
5238 zone
->min_unmapped_pages
= (freesize
*sysctl_min_unmapped_ratio
)
5240 zone
->min_slab_pages
= (freesize
* sysctl_min_slab_ratio
) / 100;
5242 zone
->name
= zone_names
[j
];
5243 spin_lock_init(&zone
->lock
);
5244 spin_lock_init(&zone
->lru_lock
);
5245 zone_seqlock_init(zone
);
5246 zone
->zone_pgdat
= pgdat
;
5247 zone_pcp_init(zone
);
5249 /* For bootup, initialized properly in watermark setup */
5250 mod_zone_page_state(zone
, NR_ALLOC_BATCH
, zone
->managed_pages
);
5252 lruvec_init(&zone
->lruvec
);
5256 set_pageblock_order();
5257 setup_usemap(pgdat
, zone
, zone_start_pfn
, size
);
5258 ret
= init_currently_empty_zone(zone
, zone_start_pfn
, size
);
5260 memmap_init(size
, nid
, j
, zone_start_pfn
);
5261 zone_start_pfn
+= size
;
5265 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
5267 unsigned long __maybe_unused start
= 0;
5268 unsigned long __maybe_unused offset
= 0;
5270 /* Skip empty nodes */
5271 if (!pgdat
->node_spanned_pages
)
5274 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5275 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
5276 offset
= pgdat
->node_start_pfn
- start
;
5277 /* ia64 gets its own node_mem_map, before this, without bootmem */
5278 if (!pgdat
->node_mem_map
) {
5279 unsigned long size
, end
;
5283 * The zone's endpoints aren't required to be MAX_ORDER
5284 * aligned but the node_mem_map endpoints must be in order
5285 * for the buddy allocator to function correctly.
5287 end
= pgdat_end_pfn(pgdat
);
5288 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
5289 size
= (end
- start
) * sizeof(struct page
);
5290 map
= alloc_remap(pgdat
->node_id
, size
);
5292 map
= memblock_virt_alloc_node_nopanic(size
,
5294 pgdat
->node_mem_map
= map
+ offset
;
5296 #ifndef CONFIG_NEED_MULTIPLE_NODES
5298 * With no DISCONTIG, the global mem_map is just set as node 0's
5300 if (pgdat
== NODE_DATA(0)) {
5301 mem_map
= NODE_DATA(0)->node_mem_map
;
5302 #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
5303 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
5305 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5308 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
5311 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
5312 unsigned long node_start_pfn
, unsigned long *zholes_size
)
5314 pg_data_t
*pgdat
= NODE_DATA(nid
);
5315 unsigned long start_pfn
= 0;
5316 unsigned long end_pfn
= 0;
5318 /* pg_data_t should be reset to zero when it's allocated */
5319 WARN_ON(pgdat
->nr_zones
|| pgdat
->classzone_idx
);
5321 reset_deferred_meminit(pgdat
);
5322 pgdat
->node_id
= nid
;
5323 pgdat
->node_start_pfn
= node_start_pfn
;
5324 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5325 get_pfn_range_for_nid(nid
, &start_pfn
, &end_pfn
);
5326 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid
,
5327 (u64
)start_pfn
<< PAGE_SHIFT
,
5328 end_pfn
? ((u64
)end_pfn
<< PAGE_SHIFT
) - 1 : 0);
5330 calculate_node_totalpages(pgdat
, start_pfn
, end_pfn
,
5331 zones_size
, zholes_size
);
5333 alloc_node_mem_map(pgdat
);
5334 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5335 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5336 nid
, (unsigned long)pgdat
,
5337 (unsigned long)pgdat
->node_mem_map
);
5340 free_area_init_core(pgdat
);
5343 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5345 #if MAX_NUMNODES > 1
5347 * Figure out the number of possible node ids.
5349 void __init
setup_nr_node_ids(void)
5351 unsigned int highest
;
5353 highest
= find_last_bit(node_possible_map
.bits
, MAX_NUMNODES
);
5354 nr_node_ids
= highest
+ 1;
5359 * node_map_pfn_alignment - determine the maximum internode alignment
5361 * This function should be called after node map is populated and sorted.
5362 * It calculates the maximum power of two alignment which can distinguish
5365 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5366 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
5367 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
5368 * shifted, 1GiB is enough and this function will indicate so.
5370 * This is used to test whether pfn -> nid mapping of the chosen memory
5371 * model has fine enough granularity to avoid incorrect mapping for the
5372 * populated node map.
5374 * Returns the determined alignment in pfn's. 0 if there is no alignment
5375 * requirement (single node).
5377 unsigned long __init
node_map_pfn_alignment(void)
5379 unsigned long accl_mask
= 0, last_end
= 0;
5380 unsigned long start
, end
, mask
;
5384 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
5385 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
5392 * Start with a mask granular enough to pin-point to the
5393 * start pfn and tick off bits one-by-one until it becomes
5394 * too coarse to separate the current node from the last.
5396 mask
= ~((1 << __ffs(start
)) - 1);
5397 while (mask
&& last_end
<= (start
& (mask
<< 1)))
5400 /* accumulate all internode masks */
5404 /* convert mask to number of pages */
5405 return ~accl_mask
+ 1;
5408 /* Find the lowest pfn for a node */
5409 static unsigned long __init
find_min_pfn_for_node(int nid
)
5411 unsigned long min_pfn
= ULONG_MAX
;
5412 unsigned long start_pfn
;
5415 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
5416 min_pfn
= min(min_pfn
, start_pfn
);
5418 if (min_pfn
== ULONG_MAX
) {
5420 "Could not find start_pfn for node %d\n", nid
);
5428 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5430 * It returns the minimum PFN based on information provided via
5431 * memblock_set_node().
5433 unsigned long __init
find_min_pfn_with_active_regions(void)
5435 return find_min_pfn_for_node(MAX_NUMNODES
);
5439 * early_calculate_totalpages()
5440 * Sum pages in active regions for movable zone.
5441 * Populate N_MEMORY for calculating usable_nodes.
5443 static unsigned long __init
early_calculate_totalpages(void)
5445 unsigned long totalpages
= 0;
5446 unsigned long start_pfn
, end_pfn
;
5449 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
5450 unsigned long pages
= end_pfn
- start_pfn
;
5452 totalpages
+= pages
;
5454 node_set_state(nid
, N_MEMORY
);
5460 * Find the PFN the Movable zone begins in each node. Kernel memory
5461 * is spread evenly between nodes as long as the nodes have enough
5462 * memory. When they don't, some nodes will have more kernelcore than
5465 static void __init
find_zone_movable_pfns_for_nodes(void)
5468 unsigned long usable_startpfn
;
5469 unsigned long kernelcore_node
, kernelcore_remaining
;
5470 /* save the state before borrow the nodemask */
5471 nodemask_t saved_node_state
= node_states
[N_MEMORY
];
5472 unsigned long totalpages
= early_calculate_totalpages();
5473 int usable_nodes
= nodes_weight(node_states
[N_MEMORY
]);
5474 struct memblock_region
*r
;
5476 /* Need to find movable_zone earlier when movable_node is specified. */
5477 find_usable_zone_for_movable();
5480 * If movable_node is specified, ignore kernelcore and movablecore
5483 if (movable_node_is_enabled()) {
5484 for_each_memblock(memory
, r
) {
5485 if (!memblock_is_hotpluggable(r
))
5490 usable_startpfn
= PFN_DOWN(r
->base
);
5491 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
5492 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
5500 * If movablecore=nn[KMG] was specified, calculate what size of
5501 * kernelcore that corresponds so that memory usable for
5502 * any allocation type is evenly spread. If both kernelcore
5503 * and movablecore are specified, then the value of kernelcore
5504 * will be used for required_kernelcore if it's greater than
5505 * what movablecore would have allowed.
5507 if (required_movablecore
) {
5508 unsigned long corepages
;
5511 * Round-up so that ZONE_MOVABLE is at least as large as what
5512 * was requested by the user
5514 required_movablecore
=
5515 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
5516 required_movablecore
= min(totalpages
, required_movablecore
);
5517 corepages
= totalpages
- required_movablecore
;
5519 required_kernelcore
= max(required_kernelcore
, corepages
);
5523 * If kernelcore was not specified or kernelcore size is larger
5524 * than totalpages, there is no ZONE_MOVABLE.
5526 if (!required_kernelcore
|| required_kernelcore
>= totalpages
)
5529 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5530 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
5533 /* Spread kernelcore memory as evenly as possible throughout nodes */
5534 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5535 for_each_node_state(nid
, N_MEMORY
) {
5536 unsigned long start_pfn
, end_pfn
;
5539 * Recalculate kernelcore_node if the division per node
5540 * now exceeds what is necessary to satisfy the requested
5541 * amount of memory for the kernel
5543 if (required_kernelcore
< kernelcore_node
)
5544 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5547 * As the map is walked, we track how much memory is usable
5548 * by the kernel using kernelcore_remaining. When it is
5549 * 0, the rest of the node is usable by ZONE_MOVABLE
5551 kernelcore_remaining
= kernelcore_node
;
5553 /* Go through each range of PFNs within this node */
5554 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5555 unsigned long size_pages
;
5557 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
5558 if (start_pfn
>= end_pfn
)
5561 /* Account for what is only usable for kernelcore */
5562 if (start_pfn
< usable_startpfn
) {
5563 unsigned long kernel_pages
;
5564 kernel_pages
= min(end_pfn
, usable_startpfn
)
5567 kernelcore_remaining
-= min(kernel_pages
,
5568 kernelcore_remaining
);
5569 required_kernelcore
-= min(kernel_pages
,
5570 required_kernelcore
);
5572 /* Continue if range is now fully accounted */
5573 if (end_pfn
<= usable_startpfn
) {
5576 * Push zone_movable_pfn to the end so
5577 * that if we have to rebalance
5578 * kernelcore across nodes, we will
5579 * not double account here
5581 zone_movable_pfn
[nid
] = end_pfn
;
5584 start_pfn
= usable_startpfn
;
5588 * The usable PFN range for ZONE_MOVABLE is from
5589 * start_pfn->end_pfn. Calculate size_pages as the
5590 * number of pages used as kernelcore
5592 size_pages
= end_pfn
- start_pfn
;
5593 if (size_pages
> kernelcore_remaining
)
5594 size_pages
= kernelcore_remaining
;
5595 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
5598 * Some kernelcore has been met, update counts and
5599 * break if the kernelcore for this node has been
5602 required_kernelcore
-= min(required_kernelcore
,
5604 kernelcore_remaining
-= size_pages
;
5605 if (!kernelcore_remaining
)
5611 * If there is still required_kernelcore, we do another pass with one
5612 * less node in the count. This will push zone_movable_pfn[nid] further
5613 * along on the nodes that still have memory until kernelcore is
5617 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
5621 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5622 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
5623 zone_movable_pfn
[nid
] =
5624 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
5627 /* restore the node_state */
5628 node_states
[N_MEMORY
] = saved_node_state
;
5631 /* Any regular or high memory on that node ? */
5632 static void check_for_memory(pg_data_t
*pgdat
, int nid
)
5634 enum zone_type zone_type
;
5636 if (N_MEMORY
== N_NORMAL_MEMORY
)
5639 for (zone_type
= 0; zone_type
<= ZONE_MOVABLE
- 1; zone_type
++) {
5640 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
5641 if (populated_zone(zone
)) {
5642 node_set_state(nid
, N_HIGH_MEMORY
);
5643 if (N_NORMAL_MEMORY
!= N_HIGH_MEMORY
&&
5644 zone_type
<= ZONE_NORMAL
)
5645 node_set_state(nid
, N_NORMAL_MEMORY
);
5652 * free_area_init_nodes - Initialise all pg_data_t and zone data
5653 * @max_zone_pfn: an array of max PFNs for each zone
5655 * This will call free_area_init_node() for each active node in the system.
5656 * Using the page ranges provided by memblock_set_node(), the size of each
5657 * zone in each node and their holes is calculated. If the maximum PFN
5658 * between two adjacent zones match, it is assumed that the zone is empty.
5659 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5660 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5661 * starts where the previous one ended. For example, ZONE_DMA32 starts
5662 * at arch_max_dma_pfn.
5664 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
5666 unsigned long start_pfn
, end_pfn
;
5669 /* Record where the zone boundaries are */
5670 memset(arch_zone_lowest_possible_pfn
, 0,
5671 sizeof(arch_zone_lowest_possible_pfn
));
5672 memset(arch_zone_highest_possible_pfn
, 0,
5673 sizeof(arch_zone_highest_possible_pfn
));
5674 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
5675 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
5676 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
5677 if (i
== ZONE_MOVABLE
)
5679 arch_zone_lowest_possible_pfn
[i
] =
5680 arch_zone_highest_possible_pfn
[i
-1];
5681 arch_zone_highest_possible_pfn
[i
] =
5682 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
5684 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
5685 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
5687 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5688 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
5689 find_zone_movable_pfns_for_nodes();
5691 /* Print out the zone ranges */
5692 pr_info("Zone ranges:\n");
5693 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5694 if (i
== ZONE_MOVABLE
)
5696 pr_info(" %-8s ", zone_names
[i
]);
5697 if (arch_zone_lowest_possible_pfn
[i
] ==
5698 arch_zone_highest_possible_pfn
[i
])
5701 pr_cont("[mem %#018Lx-%#018Lx]\n",
5702 (u64
)arch_zone_lowest_possible_pfn
[i
]
5704 ((u64
)arch_zone_highest_possible_pfn
[i
]
5705 << PAGE_SHIFT
) - 1);
5708 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5709 pr_info("Movable zone start for each node\n");
5710 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
5711 if (zone_movable_pfn
[i
])
5712 pr_info(" Node %d: %#018Lx\n", i
,
5713 (u64
)zone_movable_pfn
[i
] << PAGE_SHIFT
);
5716 /* Print out the early node map */
5717 pr_info("Early memory node ranges\n");
5718 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
5719 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid
,
5720 (u64
)start_pfn
<< PAGE_SHIFT
,
5721 ((u64
)end_pfn
<< PAGE_SHIFT
) - 1);
5723 /* Initialise every node */
5724 mminit_verify_pageflags_layout();
5725 setup_nr_node_ids();
5726 for_each_online_node(nid
) {
5727 pg_data_t
*pgdat
= NODE_DATA(nid
);
5728 free_area_init_node(nid
, NULL
,
5729 find_min_pfn_for_node(nid
), NULL
);
5731 /* Any memory on that node */
5732 if (pgdat
->node_present_pages
)
5733 node_set_state(nid
, N_MEMORY
);
5734 check_for_memory(pgdat
, nid
);
5738 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
5740 unsigned long long coremem
;
5744 coremem
= memparse(p
, &p
);
5745 *core
= coremem
>> PAGE_SHIFT
;
5747 /* Paranoid check that UL is enough for the coremem value */
5748 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
5754 * kernelcore=size sets the amount of memory for use for allocations that
5755 * cannot be reclaimed or migrated.
5757 static int __init
cmdline_parse_kernelcore(char *p
)
5759 return cmdline_parse_core(p
, &required_kernelcore
);
5763 * movablecore=size sets the amount of memory for use for allocations that
5764 * can be reclaimed or migrated.
5766 static int __init
cmdline_parse_movablecore(char *p
)
5768 return cmdline_parse_core(p
, &required_movablecore
);
5771 early_param("kernelcore", cmdline_parse_kernelcore
);
5772 early_param("movablecore", cmdline_parse_movablecore
);
5774 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5776 void adjust_managed_page_count(struct page
*page
, long count
)
5778 spin_lock(&managed_page_count_lock
);
5779 page_zone(page
)->managed_pages
+= count
;
5780 totalram_pages
+= count
;
5781 #ifdef CONFIG_HIGHMEM
5782 if (PageHighMem(page
))
5783 totalhigh_pages
+= count
;
5785 spin_unlock(&managed_page_count_lock
);
5787 EXPORT_SYMBOL(adjust_managed_page_count
);
5789 unsigned long free_reserved_area(void *start
, void *end
, int poison
, char *s
)
5792 unsigned long pages
= 0;
5794 start
= (void *)PAGE_ALIGN((unsigned long)start
);
5795 end
= (void *)((unsigned long)end
& PAGE_MASK
);
5796 for (pos
= start
; pos
< end
; pos
+= PAGE_SIZE
, pages
++) {
5797 if ((unsigned int)poison
<= 0xFF)
5798 memset(pos
, poison
, PAGE_SIZE
);
5799 free_reserved_page(virt_to_page(pos
));
5803 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
5804 s
, pages
<< (PAGE_SHIFT
- 10), start
, end
);
5808 EXPORT_SYMBOL(free_reserved_area
);
5810 #ifdef CONFIG_HIGHMEM
5811 void free_highmem_page(struct page
*page
)
5813 __free_reserved_page(page
);
5815 page_zone(page
)->managed_pages
++;
5821 void __init
mem_init_print_info(const char *str
)
5823 unsigned long physpages
, codesize
, datasize
, rosize
, bss_size
;
5824 unsigned long init_code_size
, init_data_size
;
5826 physpages
= get_num_physpages();
5827 codesize
= _etext
- _stext
;
5828 datasize
= _edata
- _sdata
;
5829 rosize
= __end_rodata
- __start_rodata
;
5830 bss_size
= __bss_stop
- __bss_start
;
5831 init_data_size
= __init_end
- __init_begin
;
5832 init_code_size
= _einittext
- _sinittext
;
5835 * Detect special cases and adjust section sizes accordingly:
5836 * 1) .init.* may be embedded into .data sections
5837 * 2) .init.text.* may be out of [__init_begin, __init_end],
5838 * please refer to arch/tile/kernel/vmlinux.lds.S.
5839 * 3) .rodata.* may be embedded into .text or .data sections.
5841 #define adj_init_size(start, end, size, pos, adj) \
5843 if (start <= pos && pos < end && size > adj) \
5847 adj_init_size(__init_begin
, __init_end
, init_data_size
,
5848 _sinittext
, init_code_size
);
5849 adj_init_size(_stext
, _etext
, codesize
, _sinittext
, init_code_size
);
5850 adj_init_size(_sdata
, _edata
, datasize
, __init_begin
, init_data_size
);
5851 adj_init_size(_stext
, _etext
, codesize
, __start_rodata
, rosize
);
5852 adj_init_size(_sdata
, _edata
, datasize
, __start_rodata
, rosize
);
5854 #undef adj_init_size
5856 pr_info("Memory: %luK/%luK available "
5857 "(%luK kernel code, %luK rwdata, %luK rodata, "
5858 "%luK init, %luK bss, %luK reserved, %luK cma-reserved"
5859 #ifdef CONFIG_HIGHMEM
5863 nr_free_pages() << (PAGE_SHIFT
-10), physpages
<< (PAGE_SHIFT
-10),
5864 codesize
>> 10, datasize
>> 10, rosize
>> 10,
5865 (init_data_size
+ init_code_size
) >> 10, bss_size
>> 10,
5866 (physpages
- totalram_pages
- totalcma_pages
) << (PAGE_SHIFT
-10),
5867 totalcma_pages
<< (PAGE_SHIFT
-10),
5868 #ifdef CONFIG_HIGHMEM
5869 totalhigh_pages
<< (PAGE_SHIFT
-10),
5871 str
? ", " : "", str
? str
: "");
5875 * set_dma_reserve - set the specified number of pages reserved in the first zone
5876 * @new_dma_reserve: The number of pages to mark reserved
5878 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
5879 * In the DMA zone, a significant percentage may be consumed by kernel image
5880 * and other unfreeable allocations which can skew the watermarks badly. This
5881 * function may optionally be used to account for unfreeable pages in the
5882 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5883 * smaller per-cpu batchsize.
5885 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
5887 dma_reserve
= new_dma_reserve
;
5890 void __init
free_area_init(unsigned long *zones_size
)
5892 free_area_init_node(0, zones_size
,
5893 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
5896 static int page_alloc_cpu_notify(struct notifier_block
*self
,
5897 unsigned long action
, void *hcpu
)
5899 int cpu
= (unsigned long)hcpu
;
5901 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
5902 lru_add_drain_cpu(cpu
);
5906 * Spill the event counters of the dead processor
5907 * into the current processors event counters.
5908 * This artificially elevates the count of the current
5911 vm_events_fold_cpu(cpu
);
5914 * Zero the differential counters of the dead processor
5915 * so that the vm statistics are consistent.
5917 * This is only okay since the processor is dead and cannot
5918 * race with what we are doing.
5920 cpu_vm_stats_fold(cpu
);
5925 void __init
page_alloc_init(void)
5927 hotcpu_notifier(page_alloc_cpu_notify
, 0);
5931 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
5932 * or min_free_kbytes changes.
5934 static void calculate_totalreserve_pages(void)
5936 struct pglist_data
*pgdat
;
5937 unsigned long reserve_pages
= 0;
5938 enum zone_type i
, j
;
5940 for_each_online_pgdat(pgdat
) {
5941 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5942 struct zone
*zone
= pgdat
->node_zones
+ i
;
5945 /* Find valid and maximum lowmem_reserve in the zone */
5946 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
5947 if (zone
->lowmem_reserve
[j
] > max
)
5948 max
= zone
->lowmem_reserve
[j
];
5951 /* we treat the high watermark as reserved pages. */
5952 max
+= high_wmark_pages(zone
);
5954 if (max
> zone
->managed_pages
)
5955 max
= zone
->managed_pages
;
5957 zone
->totalreserve_pages
= max
;
5959 reserve_pages
+= max
;
5962 totalreserve_pages
= reserve_pages
;
5966 * setup_per_zone_lowmem_reserve - called whenever
5967 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
5968 * has a correct pages reserved value, so an adequate number of
5969 * pages are left in the zone after a successful __alloc_pages().
5971 static void setup_per_zone_lowmem_reserve(void)
5973 struct pglist_data
*pgdat
;
5974 enum zone_type j
, idx
;
5976 for_each_online_pgdat(pgdat
) {
5977 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5978 struct zone
*zone
= pgdat
->node_zones
+ j
;
5979 unsigned long managed_pages
= zone
->managed_pages
;
5981 zone
->lowmem_reserve
[j
] = 0;
5985 struct zone
*lower_zone
;
5989 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
5990 sysctl_lowmem_reserve_ratio
[idx
] = 1;
5992 lower_zone
= pgdat
->node_zones
+ idx
;
5993 lower_zone
->lowmem_reserve
[j
] = managed_pages
/
5994 sysctl_lowmem_reserve_ratio
[idx
];
5995 managed_pages
+= lower_zone
->managed_pages
;
6000 /* update totalreserve_pages */
6001 calculate_totalreserve_pages();
6004 static void __setup_per_zone_wmarks(void)
6006 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
6007 unsigned long lowmem_pages
= 0;
6009 unsigned long flags
;
6011 /* Calculate total number of !ZONE_HIGHMEM pages */
6012 for_each_zone(zone
) {
6013 if (!is_highmem(zone
))
6014 lowmem_pages
+= zone
->managed_pages
;
6017 for_each_zone(zone
) {
6020 spin_lock_irqsave(&zone
->lock
, flags
);
6021 tmp
= (u64
)pages_min
* zone
->managed_pages
;
6022 do_div(tmp
, lowmem_pages
);
6023 if (is_highmem(zone
)) {
6025 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6026 * need highmem pages, so cap pages_min to a small
6029 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
6030 * deltas control asynch page reclaim, and so should
6031 * not be capped for highmem.
6033 unsigned long min_pages
;
6035 min_pages
= zone
->managed_pages
/ 1024;
6036 min_pages
= clamp(min_pages
, SWAP_CLUSTER_MAX
, 128UL);
6037 zone
->watermark
[WMARK_MIN
] = min_pages
;
6040 * If it's a lowmem zone, reserve a number of pages
6041 * proportionate to the zone's size.
6043 zone
->watermark
[WMARK_MIN
] = tmp
;
6046 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + (tmp
>> 2);
6047 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + (tmp
>> 1);
6049 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
6050 high_wmark_pages(zone
) - low_wmark_pages(zone
) -
6051 atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]));
6053 spin_unlock_irqrestore(&zone
->lock
, flags
);
6056 /* update totalreserve_pages */
6057 calculate_totalreserve_pages();
6061 * setup_per_zone_wmarks - called when min_free_kbytes changes
6062 * or when memory is hot-{added|removed}
6064 * Ensures that the watermark[min,low,high] values for each zone are set
6065 * correctly with respect to min_free_kbytes.
6067 void setup_per_zone_wmarks(void)
6069 mutex_lock(&zonelists_mutex
);
6070 __setup_per_zone_wmarks();
6071 mutex_unlock(&zonelists_mutex
);
6075 * The inactive anon list should be small enough that the VM never has to
6076 * do too much work, but large enough that each inactive page has a chance
6077 * to be referenced again before it is swapped out.
6079 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
6080 * INACTIVE_ANON pages on this zone's LRU, maintained by the
6081 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
6082 * the anonymous pages are kept on the inactive list.
6085 * memory ratio inactive anon
6086 * -------------------------------------
6095 static void __meminit
calculate_zone_inactive_ratio(struct zone
*zone
)
6097 unsigned int gb
, ratio
;
6099 /* Zone size in gigabytes */
6100 gb
= zone
->managed_pages
>> (30 - PAGE_SHIFT
);
6102 ratio
= int_sqrt(10 * gb
);
6106 zone
->inactive_ratio
= ratio
;
6109 static void __meminit
setup_per_zone_inactive_ratio(void)
6114 calculate_zone_inactive_ratio(zone
);
6118 * Initialise min_free_kbytes.
6120 * For small machines we want it small (128k min). For large machines
6121 * we want it large (64MB max). But it is not linear, because network
6122 * bandwidth does not increase linearly with machine size. We use
6124 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
6125 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6141 int __meminit
init_per_zone_wmark_min(void)
6143 unsigned long lowmem_kbytes
;
6144 int new_min_free_kbytes
;
6146 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
6147 new_min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
6149 if (new_min_free_kbytes
> user_min_free_kbytes
) {
6150 min_free_kbytes
= new_min_free_kbytes
;
6151 if (min_free_kbytes
< 128)
6152 min_free_kbytes
= 128;
6153 if (min_free_kbytes
> 65536)
6154 min_free_kbytes
= 65536;
6156 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6157 new_min_free_kbytes
, user_min_free_kbytes
);
6159 setup_per_zone_wmarks();
6160 refresh_zone_stat_thresholds();
6161 setup_per_zone_lowmem_reserve();
6162 setup_per_zone_inactive_ratio();
6165 module_init(init_per_zone_wmark_min
)
6168 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
6169 * that we can call two helper functions whenever min_free_kbytes
6172 int min_free_kbytes_sysctl_handler(struct ctl_table
*table
, int write
,
6173 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6177 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6182 user_min_free_kbytes
= min_free_kbytes
;
6183 setup_per_zone_wmarks();
6189 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6190 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6195 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6200 zone
->min_unmapped_pages
= (zone
->managed_pages
*
6201 sysctl_min_unmapped_ratio
) / 100;
6205 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6206 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6211 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6216 zone
->min_slab_pages
= (zone
->managed_pages
*
6217 sysctl_min_slab_ratio
) / 100;
6223 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6224 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6225 * whenever sysctl_lowmem_reserve_ratio changes.
6227 * The reserve ratio obviously has absolutely no relation with the
6228 * minimum watermarks. The lowmem reserve ratio can only make sense
6229 * if in function of the boot time zone sizes.
6231 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6232 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6234 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6235 setup_per_zone_lowmem_reserve();
6240 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
6241 * cpu. It is the fraction of total pages in each zone that a hot per cpu
6242 * pagelist can have before it gets flushed back to buddy allocator.
6244 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table
*table
, int write
,
6245 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6248 int old_percpu_pagelist_fraction
;
6251 mutex_lock(&pcp_batch_high_lock
);
6252 old_percpu_pagelist_fraction
= percpu_pagelist_fraction
;
6254 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6255 if (!write
|| ret
< 0)
6258 /* Sanity checking to avoid pcp imbalance */
6259 if (percpu_pagelist_fraction
&&
6260 percpu_pagelist_fraction
< MIN_PERCPU_PAGELIST_FRACTION
) {
6261 percpu_pagelist_fraction
= old_percpu_pagelist_fraction
;
6267 if (percpu_pagelist_fraction
== old_percpu_pagelist_fraction
)
6270 for_each_populated_zone(zone
) {
6273 for_each_possible_cpu(cpu
)
6274 pageset_set_high_and_batch(zone
,
6275 per_cpu_ptr(zone
->pageset
, cpu
));
6278 mutex_unlock(&pcp_batch_high_lock
);
6283 int hashdist
= HASHDIST_DEFAULT
;
6285 static int __init
set_hashdist(char *str
)
6289 hashdist
= simple_strtoul(str
, &str
, 0);
6292 __setup("hashdist=", set_hashdist
);
6296 * allocate a large system hash table from bootmem
6297 * - it is assumed that the hash table must contain an exact power-of-2
6298 * quantity of entries
6299 * - limit is the number of hash buckets, not the total allocation size
6301 void *__init
alloc_large_system_hash(const char *tablename
,
6302 unsigned long bucketsize
,
6303 unsigned long numentries
,
6306 unsigned int *_hash_shift
,
6307 unsigned int *_hash_mask
,
6308 unsigned long low_limit
,
6309 unsigned long high_limit
)
6311 unsigned long long max
= high_limit
;
6312 unsigned long log2qty
, size
;
6315 /* allow the kernel cmdline to have a say */
6317 /* round applicable memory size up to nearest megabyte */
6318 numentries
= nr_kernel_pages
;
6320 /* It isn't necessary when PAGE_SIZE >= 1MB */
6321 if (PAGE_SHIFT
< 20)
6322 numentries
= round_up(numentries
, (1<<20)/PAGE_SIZE
);
6324 /* limit to 1 bucket per 2^scale bytes of low memory */
6325 if (scale
> PAGE_SHIFT
)
6326 numentries
>>= (scale
- PAGE_SHIFT
);
6328 numentries
<<= (PAGE_SHIFT
- scale
);
6330 /* Make sure we've got at least a 0-order allocation.. */
6331 if (unlikely(flags
& HASH_SMALL
)) {
6332 /* Makes no sense without HASH_EARLY */
6333 WARN_ON(!(flags
& HASH_EARLY
));
6334 if (!(numentries
>> *_hash_shift
)) {
6335 numentries
= 1UL << *_hash_shift
;
6336 BUG_ON(!numentries
);
6338 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
6339 numentries
= PAGE_SIZE
/ bucketsize
;
6341 numentries
= roundup_pow_of_two(numentries
);
6343 /* limit allocation size to 1/16 total memory by default */
6345 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
6346 do_div(max
, bucketsize
);
6348 max
= min(max
, 0x80000000ULL
);
6350 if (numentries
< low_limit
)
6351 numentries
= low_limit
;
6352 if (numentries
> max
)
6355 log2qty
= ilog2(numentries
);
6358 size
= bucketsize
<< log2qty
;
6359 if (flags
& HASH_EARLY
)
6360 table
= memblock_virt_alloc_nopanic(size
, 0);
6362 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
6365 * If bucketsize is not a power-of-two, we may free
6366 * some pages at the end of hash table which
6367 * alloc_pages_exact() automatically does
6369 if (get_order(size
) < MAX_ORDER
) {
6370 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
6371 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
6374 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
6377 panic("Failed to allocate %s hash table\n", tablename
);
6379 printk(KERN_INFO
"%s hash table entries: %ld (order: %d, %lu bytes)\n",
6382 ilog2(size
) - PAGE_SHIFT
,
6386 *_hash_shift
= log2qty
;
6388 *_hash_mask
= (1 << log2qty
) - 1;
6393 /* Return a pointer to the bitmap storing bits affecting a block of pages */
6394 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
6397 #ifdef CONFIG_SPARSEMEM
6398 return __pfn_to_section(pfn
)->pageblock_flags
;
6400 return zone
->pageblock_flags
;
6401 #endif /* CONFIG_SPARSEMEM */
6404 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
6406 #ifdef CONFIG_SPARSEMEM
6407 pfn
&= (PAGES_PER_SECTION
-1);
6408 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
6410 pfn
= pfn
- round_down(zone
->zone_start_pfn
, pageblock_nr_pages
);
6411 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
6412 #endif /* CONFIG_SPARSEMEM */
6416 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
6417 * @page: The page within the block of interest
6418 * @pfn: The target page frame number
6419 * @end_bitidx: The last bit of interest to retrieve
6420 * @mask: mask of bits that the caller is interested in
6422 * Return: pageblock_bits flags
6424 unsigned long get_pfnblock_flags_mask(struct page
*page
, unsigned long pfn
,
6425 unsigned long end_bitidx
,
6429 unsigned long *bitmap
;
6430 unsigned long bitidx
, word_bitidx
;
6433 zone
= page_zone(page
);
6434 bitmap
= get_pageblock_bitmap(zone
, pfn
);
6435 bitidx
= pfn_to_bitidx(zone
, pfn
);
6436 word_bitidx
= bitidx
/ BITS_PER_LONG
;
6437 bitidx
&= (BITS_PER_LONG
-1);
6439 word
= bitmap
[word_bitidx
];
6440 bitidx
+= end_bitidx
;
6441 return (word
>> (BITS_PER_LONG
- bitidx
- 1)) & mask
;
6445 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
6446 * @page: The page within the block of interest
6447 * @flags: The flags to set
6448 * @pfn: The target page frame number
6449 * @end_bitidx: The last bit of interest
6450 * @mask: mask of bits that the caller is interested in
6452 void set_pfnblock_flags_mask(struct page
*page
, unsigned long flags
,
6454 unsigned long end_bitidx
,
6458 unsigned long *bitmap
;
6459 unsigned long bitidx
, word_bitidx
;
6460 unsigned long old_word
, word
;
6462 BUILD_BUG_ON(NR_PAGEBLOCK_BITS
!= 4);
6464 zone
= page_zone(page
);
6465 bitmap
= get_pageblock_bitmap(zone
, pfn
);
6466 bitidx
= pfn_to_bitidx(zone
, pfn
);
6467 word_bitidx
= bitidx
/ BITS_PER_LONG
;
6468 bitidx
&= (BITS_PER_LONG
-1);
6470 VM_BUG_ON_PAGE(!zone_spans_pfn(zone
, pfn
), page
);
6472 bitidx
+= end_bitidx
;
6473 mask
<<= (BITS_PER_LONG
- bitidx
- 1);
6474 flags
<<= (BITS_PER_LONG
- bitidx
- 1);
6476 word
= READ_ONCE(bitmap
[word_bitidx
]);
6478 old_word
= cmpxchg(&bitmap
[word_bitidx
], word
, (word
& ~mask
) | flags
);
6479 if (word
== old_word
)
6486 * This function checks whether pageblock includes unmovable pages or not.
6487 * If @count is not zero, it is okay to include less @count unmovable pages
6489 * PageLRU check without isolation or lru_lock could race so that
6490 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6491 * expect this function should be exact.
6493 bool has_unmovable_pages(struct zone
*zone
, struct page
*page
, int count
,
6494 bool skip_hwpoisoned_pages
)
6496 unsigned long pfn
, iter
, found
;
6500 * For avoiding noise data, lru_add_drain_all() should be called
6501 * If ZONE_MOVABLE, the zone never contains unmovable pages
6503 if (zone_idx(zone
) == ZONE_MOVABLE
)
6505 mt
= get_pageblock_migratetype(page
);
6506 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
6509 pfn
= page_to_pfn(page
);
6510 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
6511 unsigned long check
= pfn
+ iter
;
6513 if (!pfn_valid_within(check
))
6516 page
= pfn_to_page(check
);
6519 * Hugepages are not in LRU lists, but they're movable.
6520 * We need not scan over tail pages bacause we don't
6521 * handle each tail page individually in migration.
6523 if (PageHuge(page
)) {
6524 iter
= round_up(iter
+ 1, 1<<compound_order(page
)) - 1;
6529 * We can't use page_count without pin a page
6530 * because another CPU can free compound page.
6531 * This check already skips compound tails of THP
6532 * because their page->_count is zero at all time.
6534 if (!atomic_read(&page
->_count
)) {
6535 if (PageBuddy(page
))
6536 iter
+= (1 << page_order(page
)) - 1;
6541 * The HWPoisoned page may be not in buddy system, and
6542 * page_count() is not 0.
6544 if (skip_hwpoisoned_pages
&& PageHWPoison(page
))
6550 * If there are RECLAIMABLE pages, we need to check
6551 * it. But now, memory offline itself doesn't call
6552 * shrink_node_slabs() and it still to be fixed.
6555 * If the page is not RAM, page_count()should be 0.
6556 * we don't need more check. This is an _used_ not-movable page.
6558 * The problematic thing here is PG_reserved pages. PG_reserved
6559 * is set to both of a memory hole page and a _used_ kernel
6568 bool is_pageblock_removable_nolock(struct page
*page
)
6574 * We have to be careful here because we are iterating over memory
6575 * sections which are not zone aware so we might end up outside of
6576 * the zone but still within the section.
6577 * We have to take care about the node as well. If the node is offline
6578 * its NODE_DATA will be NULL - see page_zone.
6580 if (!node_online(page_to_nid(page
)))
6583 zone
= page_zone(page
);
6584 pfn
= page_to_pfn(page
);
6585 if (!zone_spans_pfn(zone
, pfn
))
6588 return !has_unmovable_pages(zone
, page
, 0, true);
6593 static unsigned long pfn_max_align_down(unsigned long pfn
)
6595 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
6596 pageblock_nr_pages
) - 1);
6599 static unsigned long pfn_max_align_up(unsigned long pfn
)
6601 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
6602 pageblock_nr_pages
));
6605 /* [start, end) must belong to a single zone. */
6606 static int __alloc_contig_migrate_range(struct compact_control
*cc
,
6607 unsigned long start
, unsigned long end
)
6609 /* This function is based on compact_zone() from compaction.c. */
6610 unsigned long nr_reclaimed
;
6611 unsigned long pfn
= start
;
6612 unsigned int tries
= 0;
6617 while (pfn
< end
|| !list_empty(&cc
->migratepages
)) {
6618 if (fatal_signal_pending(current
)) {
6623 if (list_empty(&cc
->migratepages
)) {
6624 cc
->nr_migratepages
= 0;
6625 pfn
= isolate_migratepages_range(cc
, pfn
, end
);
6631 } else if (++tries
== 5) {
6632 ret
= ret
< 0 ? ret
: -EBUSY
;
6636 nr_reclaimed
= reclaim_clean_pages_from_list(cc
->zone
,
6638 cc
->nr_migratepages
-= nr_reclaimed
;
6640 ret
= migrate_pages(&cc
->migratepages
, alloc_migrate_target
,
6641 NULL
, 0, cc
->mode
, MR_CMA
);
6644 putback_movable_pages(&cc
->migratepages
);
6651 * alloc_contig_range() -- tries to allocate given range of pages
6652 * @start: start PFN to allocate
6653 * @end: one-past-the-last PFN to allocate
6654 * @migratetype: migratetype of the underlaying pageblocks (either
6655 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6656 * in range must have the same migratetype and it must
6657 * be either of the two.
6659 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6660 * aligned, however it's the caller's responsibility to guarantee that
6661 * we are the only thread that changes migrate type of pageblocks the
6664 * The PFN range must belong to a single zone.
6666 * Returns zero on success or negative error code. On success all
6667 * pages which PFN is in [start, end) are allocated for the caller and
6668 * need to be freed with free_contig_range().
6670 int alloc_contig_range(unsigned long start
, unsigned long end
,
6671 unsigned migratetype
)
6673 unsigned long outer_start
, outer_end
;
6677 struct compact_control cc
= {
6678 .nr_migratepages
= 0,
6680 .zone
= page_zone(pfn_to_page(start
)),
6681 .mode
= MIGRATE_SYNC
,
6682 .ignore_skip_hint
= true,
6684 INIT_LIST_HEAD(&cc
.migratepages
);
6687 * What we do here is we mark all pageblocks in range as
6688 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6689 * have different sizes, and due to the way page allocator
6690 * work, we align the range to biggest of the two pages so
6691 * that page allocator won't try to merge buddies from
6692 * different pageblocks and change MIGRATE_ISOLATE to some
6693 * other migration type.
6695 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6696 * migrate the pages from an unaligned range (ie. pages that
6697 * we are interested in). This will put all the pages in
6698 * range back to page allocator as MIGRATE_ISOLATE.
6700 * When this is done, we take the pages in range from page
6701 * allocator removing them from the buddy system. This way
6702 * page allocator will never consider using them.
6704 * This lets us mark the pageblocks back as
6705 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6706 * aligned range but not in the unaligned, original range are
6707 * put back to page allocator so that buddy can use them.
6710 ret
= start_isolate_page_range(pfn_max_align_down(start
),
6711 pfn_max_align_up(end
), migratetype
,
6717 * In case of -EBUSY, we'd like to know which page causes problem.
6718 * So, just fall through. We will check it in test_pages_isolated().
6720 ret
= __alloc_contig_migrate_range(&cc
, start
, end
);
6721 if (ret
&& ret
!= -EBUSY
)
6725 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6726 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6727 * more, all pages in [start, end) are free in page allocator.
6728 * What we are going to do is to allocate all pages from
6729 * [start, end) (that is remove them from page allocator).
6731 * The only problem is that pages at the beginning and at the
6732 * end of interesting range may be not aligned with pages that
6733 * page allocator holds, ie. they can be part of higher order
6734 * pages. Because of this, we reserve the bigger range and
6735 * once this is done free the pages we are not interested in.
6737 * We don't have to hold zone->lock here because the pages are
6738 * isolated thus they won't get removed from buddy.
6741 lru_add_drain_all();
6742 drain_all_pages(cc
.zone
);
6745 outer_start
= start
;
6746 while (!PageBuddy(pfn_to_page(outer_start
))) {
6747 if (++order
>= MAX_ORDER
) {
6748 outer_start
= start
;
6751 outer_start
&= ~0UL << order
;
6754 if (outer_start
!= start
) {
6755 order
= page_order(pfn_to_page(outer_start
));
6758 * outer_start page could be small order buddy page and
6759 * it doesn't include start page. Adjust outer_start
6760 * in this case to report failed page properly
6761 * on tracepoint in test_pages_isolated()
6763 if (outer_start
+ (1UL << order
) <= start
)
6764 outer_start
= start
;
6767 /* Make sure the range is really isolated. */
6768 if (test_pages_isolated(outer_start
, end
, false)) {
6769 pr_info("%s: [%lx, %lx) PFNs busy\n",
6770 __func__
, outer_start
, end
);
6775 /* Grab isolated pages from freelists. */
6776 outer_end
= isolate_freepages_range(&cc
, outer_start
, end
);
6782 /* Free head and tail (if any) */
6783 if (start
!= outer_start
)
6784 free_contig_range(outer_start
, start
- outer_start
);
6785 if (end
!= outer_end
)
6786 free_contig_range(end
, outer_end
- end
);
6789 undo_isolate_page_range(pfn_max_align_down(start
),
6790 pfn_max_align_up(end
), migratetype
);
6794 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
6796 unsigned int count
= 0;
6798 for (; nr_pages
--; pfn
++) {
6799 struct page
*page
= pfn_to_page(pfn
);
6801 count
+= page_count(page
) != 1;
6804 WARN(count
!= 0, "%d pages are still in use!\n", count
);
6808 #ifdef CONFIG_MEMORY_HOTPLUG
6810 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6811 * page high values need to be recalulated.
6813 void __meminit
zone_pcp_update(struct zone
*zone
)
6816 mutex_lock(&pcp_batch_high_lock
);
6817 for_each_possible_cpu(cpu
)
6818 pageset_set_high_and_batch(zone
,
6819 per_cpu_ptr(zone
->pageset
, cpu
));
6820 mutex_unlock(&pcp_batch_high_lock
);
6824 void zone_pcp_reset(struct zone
*zone
)
6826 unsigned long flags
;
6828 struct per_cpu_pageset
*pset
;
6830 /* avoid races with drain_pages() */
6831 local_irq_save(flags
);
6832 if (zone
->pageset
!= &boot_pageset
) {
6833 for_each_online_cpu(cpu
) {
6834 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
6835 drain_zonestat(zone
, pset
);
6837 free_percpu(zone
->pageset
);
6838 zone
->pageset
= &boot_pageset
;
6840 local_irq_restore(flags
);
6843 #ifdef CONFIG_MEMORY_HOTREMOVE
6845 * All pages in the range must be isolated before calling this.
6848 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
6852 unsigned int order
, i
;
6854 unsigned long flags
;
6855 /* find the first valid pfn */
6856 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
6861 zone
= page_zone(pfn_to_page(pfn
));
6862 spin_lock_irqsave(&zone
->lock
, flags
);
6864 while (pfn
< end_pfn
) {
6865 if (!pfn_valid(pfn
)) {
6869 page
= pfn_to_page(pfn
);
6871 * The HWPoisoned page may be not in buddy system, and
6872 * page_count() is not 0.
6874 if (unlikely(!PageBuddy(page
) && PageHWPoison(page
))) {
6876 SetPageReserved(page
);
6880 BUG_ON(page_count(page
));
6881 BUG_ON(!PageBuddy(page
));
6882 order
= page_order(page
);
6883 #ifdef CONFIG_DEBUG_VM
6884 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
6885 pfn
, 1 << order
, end_pfn
);
6887 list_del(&page
->lru
);
6888 rmv_page_order(page
);
6889 zone
->free_area
[order
].nr_free
--;
6890 for (i
= 0; i
< (1 << order
); i
++)
6891 SetPageReserved((page
+i
));
6892 pfn
+= (1 << order
);
6894 spin_unlock_irqrestore(&zone
->lock
, flags
);
6898 #ifdef CONFIG_MEMORY_FAILURE
6899 bool is_free_buddy_page(struct page
*page
)
6901 struct zone
*zone
= page_zone(page
);
6902 unsigned long pfn
= page_to_pfn(page
);
6903 unsigned long flags
;
6906 spin_lock_irqsave(&zone
->lock
, flags
);
6907 for (order
= 0; order
< MAX_ORDER
; order
++) {
6908 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
6910 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
6913 spin_unlock_irqrestore(&zone
->lock
, flags
);
6915 return order
< MAX_ORDER
;