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 set_compound_head(p
, page
);
473 #ifdef CONFIG_DEBUG_PAGEALLOC
474 unsigned int _debug_guardpage_minorder
;
475 bool _debug_pagealloc_enabled __read_mostly
;
476 bool _debug_guardpage_enabled __read_mostly
;
478 static int __init
early_debug_pagealloc(char *buf
)
483 if (strcmp(buf
, "on") == 0)
484 _debug_pagealloc_enabled
= true;
488 early_param("debug_pagealloc", early_debug_pagealloc
);
490 static bool need_debug_guardpage(void)
492 /* If we don't use debug_pagealloc, we don't need guard page */
493 if (!debug_pagealloc_enabled())
499 static void init_debug_guardpage(void)
501 if (!debug_pagealloc_enabled())
504 _debug_guardpage_enabled
= true;
507 struct page_ext_operations debug_guardpage_ops
= {
508 .need
= need_debug_guardpage
,
509 .init
= init_debug_guardpage
,
512 static int __init
debug_guardpage_minorder_setup(char *buf
)
516 if (kstrtoul(buf
, 10, &res
) < 0 || res
> MAX_ORDER
/ 2) {
517 printk(KERN_ERR
"Bad debug_guardpage_minorder value\n");
520 _debug_guardpage_minorder
= res
;
521 printk(KERN_INFO
"Setting debug_guardpage_minorder to %lu\n", res
);
524 __setup("debug_guardpage_minorder=", debug_guardpage_minorder_setup
);
526 static inline void set_page_guard(struct zone
*zone
, struct page
*page
,
527 unsigned int order
, int migratetype
)
529 struct page_ext
*page_ext
;
531 if (!debug_guardpage_enabled())
534 page_ext
= lookup_page_ext(page
);
535 __set_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
537 INIT_LIST_HEAD(&page
->lru
);
538 set_page_private(page
, order
);
539 /* Guard pages are not available for any usage */
540 __mod_zone_freepage_state(zone
, -(1 << order
), migratetype
);
543 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
544 unsigned int order
, int migratetype
)
546 struct page_ext
*page_ext
;
548 if (!debug_guardpage_enabled())
551 page_ext
= lookup_page_ext(page
);
552 __clear_bit(PAGE_EXT_DEBUG_GUARD
, &page_ext
->flags
);
554 set_page_private(page
, 0);
555 if (!is_migrate_isolate(migratetype
))
556 __mod_zone_freepage_state(zone
, (1 << order
), migratetype
);
559 struct page_ext_operations debug_guardpage_ops
= { NULL
, };
560 static inline void set_page_guard(struct zone
*zone
, struct page
*page
,
561 unsigned int order
, int migratetype
) {}
562 static inline void clear_page_guard(struct zone
*zone
, struct page
*page
,
563 unsigned int order
, int migratetype
) {}
566 static inline void set_page_order(struct page
*page
, unsigned int order
)
568 set_page_private(page
, order
);
569 __SetPageBuddy(page
);
572 static inline void rmv_page_order(struct page
*page
)
574 __ClearPageBuddy(page
);
575 set_page_private(page
, 0);
579 * This function checks whether a page is free && is the buddy
580 * we can do coalesce a page and its buddy if
581 * (a) the buddy is not in a hole &&
582 * (b) the buddy is in the buddy system &&
583 * (c) a page and its buddy have the same order &&
584 * (d) a page and its buddy are in the same zone.
586 * For recording whether a page is in the buddy system, we set ->_mapcount
587 * PAGE_BUDDY_MAPCOUNT_VALUE.
588 * Setting, clearing, and testing _mapcount PAGE_BUDDY_MAPCOUNT_VALUE is
589 * serialized by zone->lock.
591 * For recording page's order, we use page_private(page).
593 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
596 if (!pfn_valid_within(page_to_pfn(buddy
)))
599 if (page_is_guard(buddy
) && page_order(buddy
) == order
) {
600 if (page_zone_id(page
) != page_zone_id(buddy
))
603 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
608 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
610 * zone check is done late to avoid uselessly
611 * calculating zone/node ids for pages that could
614 if (page_zone_id(page
) != page_zone_id(buddy
))
617 VM_BUG_ON_PAGE(page_count(buddy
) != 0, buddy
);
625 * Freeing function for a buddy system allocator.
627 * The concept of a buddy system is to maintain direct-mapped table
628 * (containing bit values) for memory blocks of various "orders".
629 * The bottom level table contains the map for the smallest allocatable
630 * units of memory (here, pages), and each level above it describes
631 * pairs of units from the levels below, hence, "buddies".
632 * At a high level, all that happens here is marking the table entry
633 * at the bottom level available, and propagating the changes upward
634 * as necessary, plus some accounting needed to play nicely with other
635 * parts of the VM system.
636 * At each level, we keep a list of pages, which are heads of continuous
637 * free pages of length of (1 << order) and marked with _mapcount
638 * PAGE_BUDDY_MAPCOUNT_VALUE. Page's order is recorded in page_private(page)
640 * So when we are allocating or freeing one, we can derive the state of the
641 * other. That is, if we allocate a small block, and both were
642 * free, the remainder of the region must be split into blocks.
643 * If a block is freed, and its buddy is also free, then this
644 * triggers coalescing into a block of larger size.
649 static inline void __free_one_page(struct page
*page
,
651 struct zone
*zone
, unsigned int order
,
654 unsigned long page_idx
;
655 unsigned long combined_idx
;
656 unsigned long uninitialized_var(buddy_idx
);
658 unsigned int max_order
= MAX_ORDER
;
660 VM_BUG_ON(!zone_is_initialized(zone
));
661 VM_BUG_ON_PAGE(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
, page
);
663 VM_BUG_ON(migratetype
== -1);
664 if (is_migrate_isolate(migratetype
)) {
666 * We restrict max order of merging to prevent merge
667 * between freepages on isolate pageblock and normal
668 * pageblock. Without this, pageblock isolation
669 * could cause incorrect freepage accounting.
671 max_order
= min_t(unsigned int, MAX_ORDER
, pageblock_order
+ 1);
673 __mod_zone_freepage_state(zone
, 1 << order
, migratetype
);
676 page_idx
= pfn
& ((1 << max_order
) - 1);
678 VM_BUG_ON_PAGE(page_idx
& ((1 << order
) - 1), page
);
679 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
681 while (order
< max_order
- 1) {
682 buddy_idx
= __find_buddy_index(page_idx
, order
);
683 buddy
= page
+ (buddy_idx
- page_idx
);
684 if (!page_is_buddy(page
, buddy
, order
))
687 * Our buddy is free or it is CONFIG_DEBUG_PAGEALLOC guard page,
688 * merge with it and move up one order.
690 if (page_is_guard(buddy
)) {
691 clear_page_guard(zone
, buddy
, order
, migratetype
);
693 list_del(&buddy
->lru
);
694 zone
->free_area
[order
].nr_free
--;
695 rmv_page_order(buddy
);
697 combined_idx
= buddy_idx
& page_idx
;
698 page
= page
+ (combined_idx
- page_idx
);
699 page_idx
= combined_idx
;
702 set_page_order(page
, order
);
705 * If this is not the largest possible page, check if the buddy
706 * of the next-highest order is free. If it is, it's possible
707 * that pages are being freed that will coalesce soon. In case,
708 * that is happening, add the free page to the tail of the list
709 * so it's less likely to be used soon and more likely to be merged
710 * as a higher order page
712 if ((order
< MAX_ORDER
-2) && pfn_valid_within(page_to_pfn(buddy
))) {
713 struct page
*higher_page
, *higher_buddy
;
714 combined_idx
= buddy_idx
& page_idx
;
715 higher_page
= page
+ (combined_idx
- page_idx
);
716 buddy_idx
= __find_buddy_index(combined_idx
, order
+ 1);
717 higher_buddy
= higher_page
+ (buddy_idx
- combined_idx
);
718 if (page_is_buddy(higher_page
, higher_buddy
, order
+ 1)) {
719 list_add_tail(&page
->lru
,
720 &zone
->free_area
[order
].free_list
[migratetype
]);
725 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
[migratetype
]);
727 zone
->free_area
[order
].nr_free
++;
730 static inline int free_pages_check(struct page
*page
)
732 const char *bad_reason
= NULL
;
733 unsigned long bad_flags
= 0;
735 if (unlikely(page_mapcount(page
)))
736 bad_reason
= "nonzero mapcount";
737 if (unlikely(page
->mapping
!= NULL
))
738 bad_reason
= "non-NULL mapping";
739 if (unlikely(atomic_read(&page
->_count
) != 0))
740 bad_reason
= "nonzero _count";
741 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
)) {
742 bad_reason
= "PAGE_FLAGS_CHECK_AT_FREE flag(s) set";
743 bad_flags
= PAGE_FLAGS_CHECK_AT_FREE
;
746 if (unlikely(page
->mem_cgroup
))
747 bad_reason
= "page still charged to cgroup";
749 if (unlikely(bad_reason
)) {
750 bad_page(page
, bad_reason
, bad_flags
);
753 page_cpupid_reset_last(page
);
754 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
755 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
760 * Frees a number of pages from the PCP lists
761 * Assumes all pages on list are in same zone, and of same order.
762 * count is the number of pages to free.
764 * If the zone was previously in an "all pages pinned" state then look to
765 * see if this freeing clears that state.
767 * And clear the zone's pages_scanned counter, to hold off the "all pages are
768 * pinned" detection logic.
770 static void free_pcppages_bulk(struct zone
*zone
, int count
,
771 struct per_cpu_pages
*pcp
)
776 unsigned long nr_scanned
;
778 spin_lock(&zone
->lock
);
779 nr_scanned
= zone_page_state(zone
, NR_PAGES_SCANNED
);
781 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, -nr_scanned
);
785 struct list_head
*list
;
788 * Remove pages from lists in a round-robin fashion. A
789 * batch_free count is maintained that is incremented when an
790 * empty list is encountered. This is so more pages are freed
791 * off fuller lists instead of spinning excessively around empty
796 if (++migratetype
== MIGRATE_PCPTYPES
)
798 list
= &pcp
->lists
[migratetype
];
799 } while (list_empty(list
));
801 /* This is the only non-empty list. Free them all. */
802 if (batch_free
== MIGRATE_PCPTYPES
)
803 batch_free
= to_free
;
806 int mt
; /* migratetype of the to-be-freed page */
808 page
= list_entry(list
->prev
, struct page
, lru
);
809 /* must delete as __free_one_page list manipulates */
810 list_del(&page
->lru
);
812 mt
= get_pcppage_migratetype(page
);
813 /* MIGRATE_ISOLATE page should not go to pcplists */
814 VM_BUG_ON_PAGE(is_migrate_isolate(mt
), page
);
815 /* Pageblock could have been isolated meanwhile */
816 if (unlikely(has_isolate_pageblock(zone
)))
817 mt
= get_pageblock_migratetype(page
);
819 __free_one_page(page
, page_to_pfn(page
), zone
, 0, mt
);
820 trace_mm_page_pcpu_drain(page
, 0, mt
);
821 } while (--to_free
&& --batch_free
&& !list_empty(list
));
823 spin_unlock(&zone
->lock
);
826 static void free_one_page(struct zone
*zone
,
827 struct page
*page
, unsigned long pfn
,
831 unsigned long nr_scanned
;
832 spin_lock(&zone
->lock
);
833 nr_scanned
= zone_page_state(zone
, NR_PAGES_SCANNED
);
835 __mod_zone_page_state(zone
, NR_PAGES_SCANNED
, -nr_scanned
);
837 if (unlikely(has_isolate_pageblock(zone
) ||
838 is_migrate_isolate(migratetype
))) {
839 migratetype
= get_pfnblock_migratetype(page
, pfn
);
841 __free_one_page(page
, pfn
, zone
, order
, migratetype
);
842 spin_unlock(&zone
->lock
);
845 static int free_tail_pages_check(struct page
*head_page
, struct page
*page
)
850 * We rely page->lru.next never has bit 0 set, unless the page
851 * is PageTail(). Let's make sure that's true even for poisoned ->lru.
853 BUILD_BUG_ON((unsigned long)LIST_POISON1
& 1);
855 if (!IS_ENABLED(CONFIG_DEBUG_VM
)) {
859 if (unlikely(!PageTail(page
))) {
860 bad_page(page
, "PageTail not set", 0);
863 if (unlikely(compound_head(page
) != head_page
)) {
864 bad_page(page
, "compound_head not consistent", 0);
869 clear_compound_head(page
);
873 static void __meminit
__init_single_page(struct page
*page
, unsigned long pfn
,
874 unsigned long zone
, int nid
)
876 set_page_links(page
, zone
, nid
, pfn
);
877 init_page_count(page
);
878 page_mapcount_reset(page
);
879 page_cpupid_reset_last(page
);
881 INIT_LIST_HEAD(&page
->lru
);
882 #ifdef WANT_PAGE_VIRTUAL
883 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
884 if (!is_highmem_idx(zone
))
885 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
889 static void __meminit
__init_single_pfn(unsigned long pfn
, unsigned long zone
,
892 return __init_single_page(pfn_to_page(pfn
), pfn
, zone
, nid
);
895 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
896 static void init_reserved_page(unsigned long pfn
)
901 if (!early_page_uninitialised(pfn
))
904 nid
= early_pfn_to_nid(pfn
);
905 pgdat
= NODE_DATA(nid
);
907 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
908 struct zone
*zone
= &pgdat
->node_zones
[zid
];
910 if (pfn
>= zone
->zone_start_pfn
&& pfn
< zone_end_pfn(zone
))
913 __init_single_pfn(pfn
, zid
, nid
);
916 static inline void init_reserved_page(unsigned long pfn
)
919 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
922 * Initialised pages do not have PageReserved set. This function is
923 * called for each range allocated by the bootmem allocator and
924 * marks the pages PageReserved. The remaining valid pages are later
925 * sent to the buddy page allocator.
927 void __meminit
reserve_bootmem_region(unsigned long start
, unsigned long end
)
929 unsigned long start_pfn
= PFN_DOWN(start
);
930 unsigned long end_pfn
= PFN_UP(end
);
932 for (; start_pfn
< end_pfn
; start_pfn
++) {
933 if (pfn_valid(start_pfn
)) {
934 struct page
*page
= pfn_to_page(start_pfn
);
936 init_reserved_page(start_pfn
);
938 /* Avoid false-positive PageTail() */
939 INIT_LIST_HEAD(&page
->lru
);
941 SetPageReserved(page
);
946 static bool free_pages_prepare(struct page
*page
, unsigned int order
)
948 bool compound
= PageCompound(page
);
951 VM_BUG_ON_PAGE(PageTail(page
), page
);
952 VM_BUG_ON_PAGE(compound
&& compound_order(page
) != order
, page
);
954 trace_mm_page_free(page
, order
);
955 kmemcheck_free_shadow(page
, order
);
956 kasan_free_pages(page
, order
);
959 page
->mapping
= NULL
;
960 bad
+= free_pages_check(page
);
961 for (i
= 1; i
< (1 << order
); i
++) {
963 bad
+= free_tail_pages_check(page
, page
+ i
);
964 bad
+= free_pages_check(page
+ i
);
969 reset_page_owner(page
, order
);
971 if (!PageHighMem(page
)) {
972 debug_check_no_locks_freed(page_address(page
),
974 debug_check_no_obj_freed(page_address(page
),
977 arch_free_page(page
, order
);
978 kernel_map_pages(page
, 1 << order
, 0);
983 static void __free_pages_ok(struct page
*page
, unsigned int order
)
987 unsigned long pfn
= page_to_pfn(page
);
989 if (!free_pages_prepare(page
, order
))
992 migratetype
= get_pfnblock_migratetype(page
, pfn
);
993 local_irq_save(flags
);
994 __count_vm_events(PGFREE
, 1 << order
);
995 free_one_page(page_zone(page
), page
, pfn
, order
, migratetype
);
996 local_irq_restore(flags
);
999 static void __init
__free_pages_boot_core(struct page
*page
,
1000 unsigned long pfn
, unsigned int order
)
1002 unsigned int nr_pages
= 1 << order
;
1003 struct page
*p
= page
;
1007 for (loop
= 0; loop
< (nr_pages
- 1); loop
++, p
++) {
1009 __ClearPageReserved(p
);
1010 set_page_count(p
, 0);
1012 __ClearPageReserved(p
);
1013 set_page_count(p
, 0);
1015 page_zone(page
)->managed_pages
+= nr_pages
;
1016 set_page_refcounted(page
);
1017 __free_pages(page
, order
);
1020 #if defined(CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID) || \
1021 defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP)
1023 static struct mminit_pfnnid_cache early_pfnnid_cache __meminitdata
;
1025 int __meminit
early_pfn_to_nid(unsigned long pfn
)
1027 static DEFINE_SPINLOCK(early_pfn_lock
);
1030 spin_lock(&early_pfn_lock
);
1031 nid
= __early_pfn_to_nid(pfn
, &early_pfnnid_cache
);
1034 spin_unlock(&early_pfn_lock
);
1040 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
1041 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1042 struct mminit_pfnnid_cache
*state
)
1046 nid
= __early_pfn_to_nid(pfn
, state
);
1047 if (nid
>= 0 && nid
!= node
)
1052 /* Only safe to use early in boot when initialisation is single-threaded */
1053 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1055 return meminit_pfn_in_nid(pfn
, node
, &early_pfnnid_cache
);
1060 static inline bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
1064 static inline bool __meminit
meminit_pfn_in_nid(unsigned long pfn
, int node
,
1065 struct mminit_pfnnid_cache
*state
)
1072 void __init
__free_pages_bootmem(struct page
*page
, unsigned long pfn
,
1075 if (early_page_uninitialised(pfn
))
1077 return __free_pages_boot_core(page
, pfn
, order
);
1080 #ifdef CONFIG_DEFERRED_STRUCT_PAGE_INIT
1081 static void __init
deferred_free_range(struct page
*page
,
1082 unsigned long pfn
, int nr_pages
)
1089 /* Free a large naturally-aligned chunk if possible */
1090 if (nr_pages
== MAX_ORDER_NR_PAGES
&&
1091 (pfn
& (MAX_ORDER_NR_PAGES
-1)) == 0) {
1092 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
1093 __free_pages_boot_core(page
, pfn
, MAX_ORDER
-1);
1097 for (i
= 0; i
< nr_pages
; i
++, page
++, pfn
++)
1098 __free_pages_boot_core(page
, pfn
, 0);
1101 /* Completion tracking for deferred_init_memmap() threads */
1102 static atomic_t pgdat_init_n_undone __initdata
;
1103 static __initdata
DECLARE_COMPLETION(pgdat_init_all_done_comp
);
1105 static inline void __init
pgdat_init_report_one_done(void)
1107 if (atomic_dec_and_test(&pgdat_init_n_undone
))
1108 complete(&pgdat_init_all_done_comp
);
1111 /* Initialise remaining memory on a node */
1112 static int __init
deferred_init_memmap(void *data
)
1114 pg_data_t
*pgdat
= data
;
1115 int nid
= pgdat
->node_id
;
1116 struct mminit_pfnnid_cache nid_init_state
= { };
1117 unsigned long start
= jiffies
;
1118 unsigned long nr_pages
= 0;
1119 unsigned long walk_start
, walk_end
;
1122 unsigned long first_init_pfn
= pgdat
->first_deferred_pfn
;
1123 const struct cpumask
*cpumask
= cpumask_of_node(pgdat
->node_id
);
1125 if (first_init_pfn
== ULONG_MAX
) {
1126 pgdat_init_report_one_done();
1130 /* Bind memory initialisation thread to a local node if possible */
1131 if (!cpumask_empty(cpumask
))
1132 set_cpus_allowed_ptr(current
, cpumask
);
1134 /* Sanity check boundaries */
1135 BUG_ON(pgdat
->first_deferred_pfn
< pgdat
->node_start_pfn
);
1136 BUG_ON(pgdat
->first_deferred_pfn
> pgdat_end_pfn(pgdat
));
1137 pgdat
->first_deferred_pfn
= ULONG_MAX
;
1139 /* Only the highest zone is deferred so find it */
1140 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
1141 zone
= pgdat
->node_zones
+ zid
;
1142 if (first_init_pfn
< zone_end_pfn(zone
))
1146 for_each_mem_pfn_range(i
, nid
, &walk_start
, &walk_end
, NULL
) {
1147 unsigned long pfn
, end_pfn
;
1148 struct page
*page
= NULL
;
1149 struct page
*free_base_page
= NULL
;
1150 unsigned long free_base_pfn
= 0;
1153 end_pfn
= min(walk_end
, zone_end_pfn(zone
));
1154 pfn
= first_init_pfn
;
1155 if (pfn
< walk_start
)
1157 if (pfn
< zone
->zone_start_pfn
)
1158 pfn
= zone
->zone_start_pfn
;
1160 for (; pfn
< end_pfn
; pfn
++) {
1161 if (!pfn_valid_within(pfn
))
1165 * Ensure pfn_valid is checked every
1166 * MAX_ORDER_NR_PAGES for memory holes
1168 if ((pfn
& (MAX_ORDER_NR_PAGES
- 1)) == 0) {
1169 if (!pfn_valid(pfn
)) {
1175 if (!meminit_pfn_in_nid(pfn
, nid
, &nid_init_state
)) {
1180 /* Minimise pfn page lookups and scheduler checks */
1181 if (page
&& (pfn
& (MAX_ORDER_NR_PAGES
- 1)) != 0) {
1184 nr_pages
+= nr_to_free
;
1185 deferred_free_range(free_base_page
,
1186 free_base_pfn
, nr_to_free
);
1187 free_base_page
= NULL
;
1188 free_base_pfn
= nr_to_free
= 0;
1190 page
= pfn_to_page(pfn
);
1195 VM_BUG_ON(page_zone(page
) != zone
);
1199 __init_single_page(page
, pfn
, zid
, nid
);
1200 if (!free_base_page
) {
1201 free_base_page
= page
;
1202 free_base_pfn
= pfn
;
1207 /* Where possible, batch up pages for a single free */
1210 /* Free the current block of pages to allocator */
1211 nr_pages
+= nr_to_free
;
1212 deferred_free_range(free_base_page
, free_base_pfn
,
1214 free_base_page
= NULL
;
1215 free_base_pfn
= nr_to_free
= 0;
1218 first_init_pfn
= max(end_pfn
, first_init_pfn
);
1221 /* Sanity check that the next zone really is unpopulated */
1222 WARN_ON(++zid
< MAX_NR_ZONES
&& populated_zone(++zone
));
1224 pr_info("node %d initialised, %lu pages in %ums\n", nid
, nr_pages
,
1225 jiffies_to_msecs(jiffies
- start
));
1227 pgdat_init_report_one_done();
1231 void __init
page_alloc_init_late(void)
1235 /* There will be num_node_state(N_MEMORY) threads */
1236 atomic_set(&pgdat_init_n_undone
, num_node_state(N_MEMORY
));
1237 for_each_node_state(nid
, N_MEMORY
) {
1238 kthread_run(deferred_init_memmap
, NODE_DATA(nid
), "pgdatinit%d", nid
);
1241 /* Block until all are initialised */
1242 wait_for_completion(&pgdat_init_all_done_comp
);
1244 /* Reinit limits that are based on free pages after the kernel is up */
1245 files_maxfiles_init();
1247 #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
1250 /* Free whole pageblock and set its migration type to MIGRATE_CMA. */
1251 void __init
init_cma_reserved_pageblock(struct page
*page
)
1253 unsigned i
= pageblock_nr_pages
;
1254 struct page
*p
= page
;
1257 __ClearPageReserved(p
);
1258 set_page_count(p
, 0);
1261 set_pageblock_migratetype(page
, MIGRATE_CMA
);
1263 if (pageblock_order
>= MAX_ORDER
) {
1264 i
= pageblock_nr_pages
;
1267 set_page_refcounted(p
);
1268 __free_pages(p
, MAX_ORDER
- 1);
1269 p
+= MAX_ORDER_NR_PAGES
;
1270 } while (i
-= MAX_ORDER_NR_PAGES
);
1272 set_page_refcounted(page
);
1273 __free_pages(page
, pageblock_order
);
1276 adjust_managed_page_count(page
, pageblock_nr_pages
);
1281 * The order of subdivision here is critical for the IO subsystem.
1282 * Please do not alter this order without good reasons and regression
1283 * testing. Specifically, as large blocks of memory are subdivided,
1284 * the order in which smaller blocks are delivered depends on the order
1285 * they're subdivided in this function. This is the primary factor
1286 * influencing the order in which pages are delivered to the IO
1287 * subsystem according to empirical testing, and this is also justified
1288 * by considering the behavior of a buddy system containing a single
1289 * large block of memory acted on by a series of small allocations.
1290 * This behavior is a critical factor in sglist merging's success.
1294 static inline void expand(struct zone
*zone
, struct page
*page
,
1295 int low
, int high
, struct free_area
*area
,
1298 unsigned long size
= 1 << high
;
1300 while (high
> low
) {
1304 VM_BUG_ON_PAGE(bad_range(zone
, &page
[size
]), &page
[size
]);
1306 if (IS_ENABLED(CONFIG_DEBUG_PAGEALLOC
) &&
1307 debug_guardpage_enabled() &&
1308 high
< debug_guardpage_minorder()) {
1310 * Mark as guard pages (or page), that will allow to
1311 * merge back to allocator when buddy will be freed.
1312 * Corresponding page table entries will not be touched,
1313 * pages will stay not present in virtual address space
1315 set_page_guard(zone
, &page
[size
], high
, migratetype
);
1318 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
1320 set_page_order(&page
[size
], high
);
1325 * This page is about to be returned from the page allocator
1327 static inline int check_new_page(struct page
*page
)
1329 const char *bad_reason
= NULL
;
1330 unsigned long bad_flags
= 0;
1332 if (unlikely(page_mapcount(page
)))
1333 bad_reason
= "nonzero mapcount";
1334 if (unlikely(page
->mapping
!= NULL
))
1335 bad_reason
= "non-NULL mapping";
1336 if (unlikely(atomic_read(&page
->_count
) != 0))
1337 bad_reason
= "nonzero _count";
1338 if (unlikely(page
->flags
& __PG_HWPOISON
)) {
1339 bad_reason
= "HWPoisoned (hardware-corrupted)";
1340 bad_flags
= __PG_HWPOISON
;
1342 if (unlikely(page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)) {
1343 bad_reason
= "PAGE_FLAGS_CHECK_AT_PREP flag set";
1344 bad_flags
= PAGE_FLAGS_CHECK_AT_PREP
;
1347 if (unlikely(page
->mem_cgroup
))
1348 bad_reason
= "page still charged to cgroup";
1350 if (unlikely(bad_reason
)) {
1351 bad_page(page
, bad_reason
, bad_flags
);
1357 static int prep_new_page(struct page
*page
, unsigned int order
, gfp_t gfp_flags
,
1362 for (i
= 0; i
< (1 << order
); i
++) {
1363 struct page
*p
= page
+ i
;
1364 if (unlikely(check_new_page(p
)))
1368 set_page_private(page
, 0);
1369 set_page_refcounted(page
);
1371 arch_alloc_page(page
, order
);
1372 kernel_map_pages(page
, 1 << order
, 1);
1373 kasan_alloc_pages(page
, order
);
1375 if (gfp_flags
& __GFP_ZERO
)
1376 for (i
= 0; i
< (1 << order
); i
++)
1377 clear_highpage(page
+ i
);
1379 if (order
&& (gfp_flags
& __GFP_COMP
))
1380 prep_compound_page(page
, order
);
1382 set_page_owner(page
, order
, gfp_flags
);
1385 * page is set pfmemalloc when ALLOC_NO_WATERMARKS was necessary to
1386 * allocate the page. The expectation is that the caller is taking
1387 * steps that will free more memory. The caller should avoid the page
1388 * being used for !PFMEMALLOC purposes.
1390 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
1391 set_page_pfmemalloc(page
);
1393 clear_page_pfmemalloc(page
);
1399 * Go through the free lists for the given migratetype and remove
1400 * the smallest available page from the freelists
1403 struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
1406 unsigned int current_order
;
1407 struct free_area
*area
;
1410 /* Find a page of the appropriate size in the preferred list */
1411 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
1412 area
= &(zone
->free_area
[current_order
]);
1413 if (list_empty(&area
->free_list
[migratetype
]))
1416 page
= list_entry(area
->free_list
[migratetype
].next
,
1418 list_del(&page
->lru
);
1419 rmv_page_order(page
);
1421 expand(zone
, page
, order
, current_order
, area
, migratetype
);
1422 set_pcppage_migratetype(page
, migratetype
);
1431 * This array describes the order lists are fallen back to when
1432 * the free lists for the desirable migrate type are depleted
1434 static int fallbacks
[MIGRATE_TYPES
][4] = {
1435 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1436 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_TYPES
},
1437 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_TYPES
},
1439 [MIGRATE_CMA
] = { MIGRATE_TYPES
}, /* Never used */
1441 #ifdef CONFIG_MEMORY_ISOLATION
1442 [MIGRATE_ISOLATE
] = { MIGRATE_TYPES
}, /* Never used */
1447 static struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1450 return __rmqueue_smallest(zone
, order
, MIGRATE_CMA
);
1453 static inline struct page
*__rmqueue_cma_fallback(struct zone
*zone
,
1454 unsigned int order
) { return NULL
; }
1458 * Move the free pages in a range to the free lists of the requested type.
1459 * Note that start_page and end_pages are not aligned on a pageblock
1460 * boundary. If alignment is required, use move_freepages_block()
1462 int move_freepages(struct zone
*zone
,
1463 struct page
*start_page
, struct page
*end_page
,
1468 int pages_moved
= 0;
1470 #ifndef CONFIG_HOLES_IN_ZONE
1472 * page_zone is not safe to call in this context when
1473 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
1474 * anyway as we check zone boundaries in move_freepages_block().
1475 * Remove at a later date when no bug reports exist related to
1476 * grouping pages by mobility
1478 VM_BUG_ON(page_zone(start_page
) != page_zone(end_page
));
1481 for (page
= start_page
; page
<= end_page
;) {
1482 /* Make sure we are not inadvertently changing nodes */
1483 VM_BUG_ON_PAGE(page_to_nid(page
) != zone_to_nid(zone
), page
);
1485 if (!pfn_valid_within(page_to_pfn(page
))) {
1490 if (!PageBuddy(page
)) {
1495 order
= page_order(page
);
1496 list_move(&page
->lru
,
1497 &zone
->free_area
[order
].free_list
[migratetype
]);
1499 pages_moved
+= 1 << order
;
1505 int move_freepages_block(struct zone
*zone
, struct page
*page
,
1508 unsigned long start_pfn
, end_pfn
;
1509 struct page
*start_page
, *end_page
;
1511 start_pfn
= page_to_pfn(page
);
1512 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
1513 start_page
= pfn_to_page(start_pfn
);
1514 end_page
= start_page
+ pageblock_nr_pages
- 1;
1515 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
1517 /* Do not cross zone boundaries */
1518 if (!zone_spans_pfn(zone
, start_pfn
))
1520 if (!zone_spans_pfn(zone
, end_pfn
))
1523 return move_freepages(zone
, start_page
, end_page
, migratetype
);
1526 static void change_pageblock_range(struct page
*pageblock_page
,
1527 int start_order
, int migratetype
)
1529 int nr_pageblocks
= 1 << (start_order
- pageblock_order
);
1531 while (nr_pageblocks
--) {
1532 set_pageblock_migratetype(pageblock_page
, migratetype
);
1533 pageblock_page
+= pageblock_nr_pages
;
1538 * When we are falling back to another migratetype during allocation, try to
1539 * steal extra free pages from the same pageblocks to satisfy further
1540 * allocations, instead of polluting multiple pageblocks.
1542 * If we are stealing a relatively large buddy page, it is likely there will
1543 * be more free pages in the pageblock, so try to steal them all. For
1544 * reclaimable and unmovable allocations, we steal regardless of page size,
1545 * as fragmentation caused by those allocations polluting movable pageblocks
1546 * is worse than movable allocations stealing from unmovable and reclaimable
1549 static bool can_steal_fallback(unsigned int order
, int start_mt
)
1552 * Leaving this order check is intended, although there is
1553 * relaxed order check in next check. The reason is that
1554 * we can actually steal whole pageblock if this condition met,
1555 * but, below check doesn't guarantee it and that is just heuristic
1556 * so could be changed anytime.
1558 if (order
>= pageblock_order
)
1561 if (order
>= pageblock_order
/ 2 ||
1562 start_mt
== MIGRATE_RECLAIMABLE
||
1563 start_mt
== MIGRATE_UNMOVABLE
||
1564 page_group_by_mobility_disabled
)
1571 * This function implements actual steal behaviour. If order is large enough,
1572 * we can steal whole pageblock. If not, we first move freepages in this
1573 * pageblock and check whether half of pages are moved or not. If half of
1574 * pages are moved, we can change migratetype of pageblock and permanently
1575 * use it's pages as requested migratetype in the future.
1577 static void steal_suitable_fallback(struct zone
*zone
, struct page
*page
,
1580 unsigned int current_order
= page_order(page
);
1583 /* Take ownership for orders >= pageblock_order */
1584 if (current_order
>= pageblock_order
) {
1585 change_pageblock_range(page
, current_order
, start_type
);
1589 pages
= move_freepages_block(zone
, page
, start_type
);
1591 /* Claim the whole block if over half of it is free */
1592 if (pages
>= (1 << (pageblock_order
-1)) ||
1593 page_group_by_mobility_disabled
)
1594 set_pageblock_migratetype(page
, start_type
);
1598 * Check whether there is a suitable fallback freepage with requested order.
1599 * If only_stealable is true, this function returns fallback_mt only if
1600 * we can steal other freepages all together. This would help to reduce
1601 * fragmentation due to mixed migratetype pages in one pageblock.
1603 int find_suitable_fallback(struct free_area
*area
, unsigned int order
,
1604 int migratetype
, bool only_stealable
, bool *can_steal
)
1609 if (area
->nr_free
== 0)
1614 fallback_mt
= fallbacks
[migratetype
][i
];
1615 if (fallback_mt
== MIGRATE_TYPES
)
1618 if (list_empty(&area
->free_list
[fallback_mt
]))
1621 if (can_steal_fallback(order
, migratetype
))
1624 if (!only_stealable
)
1635 * Reserve a pageblock for exclusive use of high-order atomic allocations if
1636 * there are no empty page blocks that contain a page with a suitable order
1638 static void reserve_highatomic_pageblock(struct page
*page
, struct zone
*zone
,
1639 unsigned int alloc_order
)
1642 unsigned long max_managed
, flags
;
1645 * Limit the number reserved to 1 pageblock or roughly 1% of a zone.
1646 * Check is race-prone but harmless.
1648 max_managed
= (zone
->managed_pages
/ 100) + pageblock_nr_pages
;
1649 if (zone
->nr_reserved_highatomic
>= max_managed
)
1652 spin_lock_irqsave(&zone
->lock
, flags
);
1654 /* Recheck the nr_reserved_highatomic limit under the lock */
1655 if (zone
->nr_reserved_highatomic
>= max_managed
)
1659 mt
= get_pageblock_migratetype(page
);
1660 if (mt
!= MIGRATE_HIGHATOMIC
&&
1661 !is_migrate_isolate(mt
) && !is_migrate_cma(mt
)) {
1662 zone
->nr_reserved_highatomic
+= pageblock_nr_pages
;
1663 set_pageblock_migratetype(page
, MIGRATE_HIGHATOMIC
);
1664 move_freepages_block(zone
, page
, MIGRATE_HIGHATOMIC
);
1668 spin_unlock_irqrestore(&zone
->lock
, flags
);
1672 * Used when an allocation is about to fail under memory pressure. This
1673 * potentially hurts the reliability of high-order allocations when under
1674 * intense memory pressure but failed atomic allocations should be easier
1675 * to recover from than an OOM.
1677 static void unreserve_highatomic_pageblock(const struct alloc_context
*ac
)
1679 struct zonelist
*zonelist
= ac
->zonelist
;
1680 unsigned long flags
;
1686 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
, ac
->high_zoneidx
,
1688 /* Preserve at least one pageblock */
1689 if (zone
->nr_reserved_highatomic
<= pageblock_nr_pages
)
1692 spin_lock_irqsave(&zone
->lock
, flags
);
1693 for (order
= 0; order
< MAX_ORDER
; order
++) {
1694 struct free_area
*area
= &(zone
->free_area
[order
]);
1696 if (list_empty(&area
->free_list
[MIGRATE_HIGHATOMIC
]))
1699 page
= list_entry(area
->free_list
[MIGRATE_HIGHATOMIC
].next
,
1703 * It should never happen but changes to locking could
1704 * inadvertently allow a per-cpu drain to add pages
1705 * to MIGRATE_HIGHATOMIC while unreserving so be safe
1706 * and watch for underflows.
1708 zone
->nr_reserved_highatomic
-= min(pageblock_nr_pages
,
1709 zone
->nr_reserved_highatomic
);
1712 * Convert to ac->migratetype and avoid the normal
1713 * pageblock stealing heuristics. Minimally, the caller
1714 * is doing the work and needs the pages. More
1715 * importantly, if the block was always converted to
1716 * MIGRATE_UNMOVABLE or another type then the number
1717 * of pageblocks that cannot be completely freed
1720 set_pageblock_migratetype(page
, ac
->migratetype
);
1721 move_freepages_block(zone
, page
, ac
->migratetype
);
1722 spin_unlock_irqrestore(&zone
->lock
, flags
);
1725 spin_unlock_irqrestore(&zone
->lock
, flags
);
1729 /* Remove an element from the buddy allocator from the fallback list */
1730 static inline struct page
*
1731 __rmqueue_fallback(struct zone
*zone
, unsigned int order
, int start_migratetype
)
1733 struct free_area
*area
;
1734 unsigned int current_order
;
1739 /* Find the largest possible block of pages in the other list */
1740 for (current_order
= MAX_ORDER
-1;
1741 current_order
>= order
&& current_order
<= MAX_ORDER
-1;
1743 area
= &(zone
->free_area
[current_order
]);
1744 fallback_mt
= find_suitable_fallback(area
, current_order
,
1745 start_migratetype
, false, &can_steal
);
1746 if (fallback_mt
== -1)
1749 page
= list_entry(area
->free_list
[fallback_mt
].next
,
1752 steal_suitable_fallback(zone
, page
, start_migratetype
);
1754 /* Remove the page from the freelists */
1756 list_del(&page
->lru
);
1757 rmv_page_order(page
);
1759 expand(zone
, page
, order
, current_order
, area
,
1762 * The pcppage_migratetype may differ from pageblock's
1763 * migratetype depending on the decisions in
1764 * find_suitable_fallback(). This is OK as long as it does not
1765 * differ for MIGRATE_CMA pageblocks. Those can be used as
1766 * fallback only via special __rmqueue_cma_fallback() function
1768 set_pcppage_migratetype(page
, start_migratetype
);
1770 trace_mm_page_alloc_extfrag(page
, order
, current_order
,
1771 start_migratetype
, fallback_mt
);
1780 * Do the hard work of removing an element from the buddy allocator.
1781 * Call me with the zone->lock already held.
1783 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
1784 int migratetype
, gfp_t gfp_flags
)
1788 page
= __rmqueue_smallest(zone
, order
, migratetype
);
1789 if (unlikely(!page
)) {
1790 if (migratetype
== MIGRATE_MOVABLE
)
1791 page
= __rmqueue_cma_fallback(zone
, order
);
1794 page
= __rmqueue_fallback(zone
, order
, migratetype
);
1797 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
1802 * Obtain a specified number of elements from the buddy allocator, all under
1803 * a single hold of the lock, for efficiency. Add them to the supplied list.
1804 * Returns the number of new pages which were placed at *list.
1806 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
1807 unsigned long count
, struct list_head
*list
,
1808 int migratetype
, bool cold
)
1812 spin_lock(&zone
->lock
);
1813 for (i
= 0; i
< count
; ++i
) {
1814 struct page
*page
= __rmqueue(zone
, order
, migratetype
, 0);
1815 if (unlikely(page
== NULL
))
1819 * Split buddy pages returned by expand() are received here
1820 * in physical page order. The page is added to the callers and
1821 * list and the list head then moves forward. From the callers
1822 * perspective, the linked list is ordered by page number in
1823 * some conditions. This is useful for IO devices that can
1824 * merge IO requests if the physical pages are ordered
1828 list_add(&page
->lru
, list
);
1830 list_add_tail(&page
->lru
, list
);
1832 if (is_migrate_cma(get_pcppage_migratetype(page
)))
1833 __mod_zone_page_state(zone
, NR_FREE_CMA_PAGES
,
1836 __mod_zone_page_state(zone
, NR_FREE_PAGES
, -(i
<< order
));
1837 spin_unlock(&zone
->lock
);
1843 * Called from the vmstat counter updater to drain pagesets of this
1844 * currently executing processor on remote nodes after they have
1847 * Note that this function must be called with the thread pinned to
1848 * a single processor.
1850 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
1852 unsigned long flags
;
1853 int to_drain
, batch
;
1855 local_irq_save(flags
);
1856 batch
= READ_ONCE(pcp
->batch
);
1857 to_drain
= min(pcp
->count
, batch
);
1859 free_pcppages_bulk(zone
, to_drain
, pcp
);
1860 pcp
->count
-= to_drain
;
1862 local_irq_restore(flags
);
1867 * Drain pcplists of the indicated processor and zone.
1869 * The processor must either be the current processor and the
1870 * thread pinned to the current processor or a processor that
1873 static void drain_pages_zone(unsigned int cpu
, struct zone
*zone
)
1875 unsigned long flags
;
1876 struct per_cpu_pageset
*pset
;
1877 struct per_cpu_pages
*pcp
;
1879 local_irq_save(flags
);
1880 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
1884 free_pcppages_bulk(zone
, pcp
->count
, pcp
);
1887 local_irq_restore(flags
);
1891 * Drain pcplists of all zones on the indicated processor.
1893 * The processor must either be the current processor and the
1894 * thread pinned to the current processor or a processor that
1897 static void drain_pages(unsigned int cpu
)
1901 for_each_populated_zone(zone
) {
1902 drain_pages_zone(cpu
, zone
);
1907 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
1909 * The CPU has to be pinned. When zone parameter is non-NULL, spill just
1910 * the single zone's pages.
1912 void drain_local_pages(struct zone
*zone
)
1914 int cpu
= smp_processor_id();
1917 drain_pages_zone(cpu
, zone
);
1923 * Spill all the per-cpu pages from all CPUs back into the buddy allocator.
1925 * When zone parameter is non-NULL, spill just the single zone's pages.
1927 * Note that this code is protected against sending an IPI to an offline
1928 * CPU but does not guarantee sending an IPI to newly hotplugged CPUs:
1929 * on_each_cpu_mask() blocks hotplug and won't talk to offlined CPUs but
1930 * nothing keeps CPUs from showing up after we populated the cpumask and
1931 * before the call to on_each_cpu_mask().
1933 void drain_all_pages(struct zone
*zone
)
1938 * Allocate in the BSS so we wont require allocation in
1939 * direct reclaim path for CONFIG_CPUMASK_OFFSTACK=y
1941 static cpumask_t cpus_with_pcps
;
1944 * We don't care about racing with CPU hotplug event
1945 * as offline notification will cause the notified
1946 * cpu to drain that CPU pcps and on_each_cpu_mask
1947 * disables preemption as part of its processing
1949 for_each_online_cpu(cpu
) {
1950 struct per_cpu_pageset
*pcp
;
1952 bool has_pcps
= false;
1955 pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
1959 for_each_populated_zone(z
) {
1960 pcp
= per_cpu_ptr(z
->pageset
, cpu
);
1961 if (pcp
->pcp
.count
) {
1969 cpumask_set_cpu(cpu
, &cpus_with_pcps
);
1971 cpumask_clear_cpu(cpu
, &cpus_with_pcps
);
1973 on_each_cpu_mask(&cpus_with_pcps
, (smp_call_func_t
) drain_local_pages
,
1977 #ifdef CONFIG_HIBERNATION
1979 void mark_free_pages(struct zone
*zone
)
1981 unsigned long pfn
, max_zone_pfn
;
1982 unsigned long flags
;
1983 unsigned int order
, t
;
1984 struct list_head
*curr
;
1986 if (zone_is_empty(zone
))
1989 spin_lock_irqsave(&zone
->lock
, flags
);
1991 max_zone_pfn
= zone_end_pfn(zone
);
1992 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
1993 if (pfn_valid(pfn
)) {
1994 struct page
*page
= pfn_to_page(pfn
);
1996 if (!swsusp_page_is_forbidden(page
))
1997 swsusp_unset_page_free(page
);
2000 for_each_migratetype_order(order
, t
) {
2001 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
2004 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
2005 for (i
= 0; i
< (1UL << order
); i
++)
2006 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
2009 spin_unlock_irqrestore(&zone
->lock
, flags
);
2011 #endif /* CONFIG_PM */
2014 * Free a 0-order page
2015 * cold == true ? free a cold page : free a hot page
2017 void free_hot_cold_page(struct page
*page
, bool cold
)
2019 struct zone
*zone
= page_zone(page
);
2020 struct per_cpu_pages
*pcp
;
2021 unsigned long flags
;
2022 unsigned long pfn
= page_to_pfn(page
);
2025 if (!free_pages_prepare(page
, 0))
2028 migratetype
= get_pfnblock_migratetype(page
, pfn
);
2029 set_pcppage_migratetype(page
, migratetype
);
2030 local_irq_save(flags
);
2031 __count_vm_event(PGFREE
);
2034 * We only track unmovable, reclaimable and movable on pcp lists.
2035 * Free ISOLATE pages back to the allocator because they are being
2036 * offlined but treat RESERVE as movable pages so we can get those
2037 * areas back if necessary. Otherwise, we may have to free
2038 * excessively into the page allocator
2040 if (migratetype
>= MIGRATE_PCPTYPES
) {
2041 if (unlikely(is_migrate_isolate(migratetype
))) {
2042 free_one_page(zone
, page
, pfn
, 0, migratetype
);
2045 migratetype
= MIGRATE_MOVABLE
;
2048 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2050 list_add(&page
->lru
, &pcp
->lists
[migratetype
]);
2052 list_add_tail(&page
->lru
, &pcp
->lists
[migratetype
]);
2054 if (pcp
->count
>= pcp
->high
) {
2055 unsigned long batch
= READ_ONCE(pcp
->batch
);
2056 free_pcppages_bulk(zone
, batch
, pcp
);
2057 pcp
->count
-= batch
;
2061 local_irq_restore(flags
);
2065 * Free a list of 0-order pages
2067 void free_hot_cold_page_list(struct list_head
*list
, bool cold
)
2069 struct page
*page
, *next
;
2071 list_for_each_entry_safe(page
, next
, list
, lru
) {
2072 trace_mm_page_free_batched(page
, cold
);
2073 free_hot_cold_page(page
, cold
);
2078 * split_page takes a non-compound higher-order page, and splits it into
2079 * n (1<<order) sub-pages: page[0..n]
2080 * Each sub-page must be freed individually.
2082 * Note: this is probably too low level an operation for use in drivers.
2083 * Please consult with lkml before using this in your driver.
2085 void split_page(struct page
*page
, unsigned int order
)
2090 VM_BUG_ON_PAGE(PageCompound(page
), page
);
2091 VM_BUG_ON_PAGE(!page_count(page
), page
);
2093 #ifdef CONFIG_KMEMCHECK
2095 * Split shadow pages too, because free(page[0]) would
2096 * otherwise free the whole shadow.
2098 if (kmemcheck_page_is_tracked(page
))
2099 split_page(virt_to_page(page
[0].shadow
), order
);
2102 gfp_mask
= get_page_owner_gfp(page
);
2103 set_page_owner(page
, 0, gfp_mask
);
2104 for (i
= 1; i
< (1 << order
); i
++) {
2105 set_page_refcounted(page
+ i
);
2106 set_page_owner(page
+ i
, 0, gfp_mask
);
2109 EXPORT_SYMBOL_GPL(split_page
);
2111 int __isolate_free_page(struct page
*page
, unsigned int order
)
2113 unsigned long watermark
;
2117 BUG_ON(!PageBuddy(page
));
2119 zone
= page_zone(page
);
2120 mt
= get_pageblock_migratetype(page
);
2122 if (!is_migrate_isolate(mt
)) {
2123 /* Obey watermarks as if the page was being allocated */
2124 watermark
= low_wmark_pages(zone
) + (1 << order
);
2125 if (!zone_watermark_ok(zone
, 0, watermark
, 0, 0))
2128 __mod_zone_freepage_state(zone
, -(1UL << order
), mt
);
2131 /* Remove page from free list */
2132 list_del(&page
->lru
);
2133 zone
->free_area
[order
].nr_free
--;
2134 rmv_page_order(page
);
2136 set_page_owner(page
, order
, __GFP_MOVABLE
);
2138 /* Set the pageblock if the isolated page is at least a pageblock */
2139 if (order
>= pageblock_order
- 1) {
2140 struct page
*endpage
= page
+ (1 << order
) - 1;
2141 for (; page
< endpage
; page
+= pageblock_nr_pages
) {
2142 int mt
= get_pageblock_migratetype(page
);
2143 if (!is_migrate_isolate(mt
) && !is_migrate_cma(mt
))
2144 set_pageblock_migratetype(page
,
2150 return 1UL << order
;
2154 * Similar to split_page except the page is already free. As this is only
2155 * being used for migration, the migratetype of the block also changes.
2156 * As this is called with interrupts disabled, the caller is responsible
2157 * for calling arch_alloc_page() and kernel_map_page() after interrupts
2160 * Note: this is probably too low level an operation for use in drivers.
2161 * Please consult with lkml before using this in your driver.
2163 int split_free_page(struct page
*page
)
2168 order
= page_order(page
);
2170 nr_pages
= __isolate_free_page(page
, order
);
2174 /* Split into individual pages */
2175 set_page_refcounted(page
);
2176 split_page(page
, order
);
2181 * Allocate a page from the given zone. Use pcplists for order-0 allocations.
2184 struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
2185 struct zone
*zone
, unsigned int order
,
2186 gfp_t gfp_flags
, int alloc_flags
, int migratetype
)
2188 unsigned long flags
;
2190 bool cold
= ((gfp_flags
& __GFP_COLD
) != 0);
2192 if (likely(order
== 0)) {
2193 struct per_cpu_pages
*pcp
;
2194 struct list_head
*list
;
2196 local_irq_save(flags
);
2197 pcp
= &this_cpu_ptr(zone
->pageset
)->pcp
;
2198 list
= &pcp
->lists
[migratetype
];
2199 if (list_empty(list
)) {
2200 pcp
->count
+= rmqueue_bulk(zone
, 0,
2203 if (unlikely(list_empty(list
)))
2208 page
= list_entry(list
->prev
, struct page
, lru
);
2210 page
= list_entry(list
->next
, struct page
, lru
);
2212 list_del(&page
->lru
);
2215 if (unlikely(gfp_flags
& __GFP_NOFAIL
)) {
2217 * __GFP_NOFAIL is not to be used in new code.
2219 * All __GFP_NOFAIL callers should be fixed so that they
2220 * properly detect and handle allocation failures.
2222 * We most definitely don't want callers attempting to
2223 * allocate greater than order-1 page units with
2226 WARN_ON_ONCE(order
> 1);
2228 spin_lock_irqsave(&zone
->lock
, flags
);
2231 if (alloc_flags
& ALLOC_HARDER
) {
2232 page
= __rmqueue_smallest(zone
, order
, MIGRATE_HIGHATOMIC
);
2234 trace_mm_page_alloc_zone_locked(page
, order
, migratetype
);
2237 page
= __rmqueue(zone
, order
, migratetype
, gfp_flags
);
2238 spin_unlock(&zone
->lock
);
2241 __mod_zone_freepage_state(zone
, -(1 << order
),
2242 get_pcppage_migratetype(page
));
2245 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
, -(1 << order
));
2246 if (atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]) <= 0 &&
2247 !test_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
))
2248 set_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
);
2250 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
2251 zone_statistics(preferred_zone
, zone
, gfp_flags
);
2252 local_irq_restore(flags
);
2254 VM_BUG_ON_PAGE(bad_range(zone
, page
), page
);
2258 local_irq_restore(flags
);
2262 #ifdef CONFIG_FAIL_PAGE_ALLOC
2265 struct fault_attr attr
;
2267 bool ignore_gfp_highmem
;
2268 bool ignore_gfp_reclaim
;
2270 } fail_page_alloc
= {
2271 .attr
= FAULT_ATTR_INITIALIZER
,
2272 .ignore_gfp_reclaim
= true,
2273 .ignore_gfp_highmem
= true,
2277 static int __init
setup_fail_page_alloc(char *str
)
2279 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
2281 __setup("fail_page_alloc=", setup_fail_page_alloc
);
2283 static bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2285 if (order
< fail_page_alloc
.min_order
)
2287 if (gfp_mask
& __GFP_NOFAIL
)
2289 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
2291 if (fail_page_alloc
.ignore_gfp_reclaim
&&
2292 (gfp_mask
& __GFP_DIRECT_RECLAIM
))
2295 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
2298 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
2300 static int __init
fail_page_alloc_debugfs(void)
2302 umode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
2305 dir
= fault_create_debugfs_attr("fail_page_alloc", NULL
,
2306 &fail_page_alloc
.attr
);
2308 return PTR_ERR(dir
);
2310 if (!debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
2311 &fail_page_alloc
.ignore_gfp_reclaim
))
2313 if (!debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
2314 &fail_page_alloc
.ignore_gfp_highmem
))
2316 if (!debugfs_create_u32("min-order", mode
, dir
,
2317 &fail_page_alloc
.min_order
))
2322 debugfs_remove_recursive(dir
);
2327 late_initcall(fail_page_alloc_debugfs
);
2329 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
2331 #else /* CONFIG_FAIL_PAGE_ALLOC */
2333 static inline bool should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
2338 #endif /* CONFIG_FAIL_PAGE_ALLOC */
2341 * Return true if free base pages are above 'mark'. For high-order checks it
2342 * will return true of the order-0 watermark is reached and there is at least
2343 * one free page of a suitable size. Checking now avoids taking the zone lock
2344 * to check in the allocation paths if no pages are free.
2346 static bool __zone_watermark_ok(struct zone
*z
, unsigned int order
,
2347 unsigned long mark
, int classzone_idx
, int alloc_flags
,
2352 const int alloc_harder
= (alloc_flags
& ALLOC_HARDER
);
2354 /* free_pages may go negative - that's OK */
2355 free_pages
-= (1 << order
) - 1;
2357 if (alloc_flags
& ALLOC_HIGH
)
2361 * If the caller does not have rights to ALLOC_HARDER then subtract
2362 * the high-atomic reserves. This will over-estimate the size of the
2363 * atomic reserve but it avoids a search.
2365 if (likely(!alloc_harder
))
2366 free_pages
-= z
->nr_reserved_highatomic
;
2371 /* If allocation can't use CMA areas don't use free CMA pages */
2372 if (!(alloc_flags
& ALLOC_CMA
))
2373 free_pages
-= zone_page_state(z
, NR_FREE_CMA_PAGES
);
2377 * Check watermarks for an order-0 allocation request. If these
2378 * are not met, then a high-order request also cannot go ahead
2379 * even if a suitable page happened to be free.
2381 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
2384 /* If this is an order-0 request then the watermark is fine */
2388 /* For a high-order request, check at least one suitable page is free */
2389 for (o
= order
; o
< MAX_ORDER
; o
++) {
2390 struct free_area
*area
= &z
->free_area
[o
];
2399 for (mt
= 0; mt
< MIGRATE_PCPTYPES
; mt
++) {
2400 if (!list_empty(&area
->free_list
[mt
]))
2405 if ((alloc_flags
& ALLOC_CMA
) &&
2406 !list_empty(&area
->free_list
[MIGRATE_CMA
])) {
2414 bool zone_watermark_ok(struct zone
*z
, unsigned int order
, unsigned long mark
,
2415 int classzone_idx
, int alloc_flags
)
2417 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, alloc_flags
,
2418 zone_page_state(z
, NR_FREE_PAGES
));
2421 bool zone_watermark_ok_safe(struct zone
*z
, unsigned int order
,
2422 unsigned long mark
, int classzone_idx
)
2424 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
);
2426 if (z
->percpu_drift_mark
&& free_pages
< z
->percpu_drift_mark
)
2427 free_pages
= zone_page_state_snapshot(z
, NR_FREE_PAGES
);
2429 return __zone_watermark_ok(z
, order
, mark
, classzone_idx
, 0,
2434 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
2436 return local_zone
->node
== zone
->node
;
2439 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2441 return node_distance(zone_to_nid(local_zone
), zone_to_nid(zone
)) <
2444 #else /* CONFIG_NUMA */
2445 static bool zone_local(struct zone
*local_zone
, struct zone
*zone
)
2450 static bool zone_allows_reclaim(struct zone
*local_zone
, struct zone
*zone
)
2454 #endif /* CONFIG_NUMA */
2456 static void reset_alloc_batches(struct zone
*preferred_zone
)
2458 struct zone
*zone
= preferred_zone
->zone_pgdat
->node_zones
;
2461 mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
2462 high_wmark_pages(zone
) - low_wmark_pages(zone
) -
2463 atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]));
2464 clear_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
);
2465 } while (zone
++ != preferred_zone
);
2469 * get_page_from_freelist goes through the zonelist trying to allocate
2472 static struct page
*
2473 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
, int alloc_flags
,
2474 const struct alloc_context
*ac
)
2476 struct zonelist
*zonelist
= ac
->zonelist
;
2478 struct page
*page
= NULL
;
2480 int nr_fair_skipped
= 0;
2481 bool zonelist_rescan
;
2484 zonelist_rescan
= false;
2487 * Scan zonelist, looking for a zone with enough free.
2488 * See also __cpuset_node_allowed() comment in kernel/cpuset.c.
2490 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
, ac
->high_zoneidx
,
2494 if (cpusets_enabled() &&
2495 (alloc_flags
& ALLOC_CPUSET
) &&
2496 !cpuset_zone_allowed(zone
, gfp_mask
))
2499 * Distribute pages in proportion to the individual
2500 * zone size to ensure fair page aging. The zone a
2501 * page was allocated in should have no effect on the
2502 * time the page has in memory before being reclaimed.
2504 if (alloc_flags
& ALLOC_FAIR
) {
2505 if (!zone_local(ac
->preferred_zone
, zone
))
2507 if (test_bit(ZONE_FAIR_DEPLETED
, &zone
->flags
)) {
2513 * When allocating a page cache page for writing, we
2514 * want to get it from a zone that is within its dirty
2515 * limit, such that no single zone holds more than its
2516 * proportional share of globally allowed dirty pages.
2517 * The dirty limits take into account the zone's
2518 * lowmem reserves and high watermark so that kswapd
2519 * should be able to balance it without having to
2520 * write pages from its LRU list.
2522 * This may look like it could increase pressure on
2523 * lower zones by failing allocations in higher zones
2524 * before they are full. But the pages that do spill
2525 * over are limited as the lower zones are protected
2526 * by this very same mechanism. It should not become
2527 * a practical burden to them.
2529 * XXX: For now, allow allocations to potentially
2530 * exceed the per-zone dirty limit in the slowpath
2531 * (spread_dirty_pages unset) before going into reclaim,
2532 * which is important when on a NUMA setup the allowed
2533 * zones are together not big enough to reach the
2534 * global limit. The proper fix for these situations
2535 * will require awareness of zones in the
2536 * dirty-throttling and the flusher threads.
2538 if (ac
->spread_dirty_pages
&& !zone_dirty_ok(zone
))
2541 mark
= zone
->watermark
[alloc_flags
& ALLOC_WMARK_MASK
];
2542 if (!zone_watermark_ok(zone
, order
, mark
,
2543 ac
->classzone_idx
, alloc_flags
)) {
2546 /* Checked here to keep the fast path fast */
2547 BUILD_BUG_ON(ALLOC_NO_WATERMARKS
< NR_WMARK
);
2548 if (alloc_flags
& ALLOC_NO_WATERMARKS
)
2551 if (zone_reclaim_mode
== 0 ||
2552 !zone_allows_reclaim(ac
->preferred_zone
, zone
))
2555 ret
= zone_reclaim(zone
, gfp_mask
, order
);
2557 case ZONE_RECLAIM_NOSCAN
:
2560 case ZONE_RECLAIM_FULL
:
2561 /* scanned but unreclaimable */
2564 /* did we reclaim enough */
2565 if (zone_watermark_ok(zone
, order
, mark
,
2566 ac
->classzone_idx
, alloc_flags
))
2574 page
= buffered_rmqueue(ac
->preferred_zone
, zone
, order
,
2575 gfp_mask
, alloc_flags
, ac
->migratetype
);
2577 if (prep_new_page(page
, order
, gfp_mask
, alloc_flags
))
2581 * If this is a high-order atomic allocation then check
2582 * if the pageblock should be reserved for the future
2584 if (unlikely(order
&& (alloc_flags
& ALLOC_HARDER
)))
2585 reserve_highatomic_pageblock(page
, zone
, order
);
2592 * The first pass makes sure allocations are spread fairly within the
2593 * local node. However, the local node might have free pages left
2594 * after the fairness batches are exhausted, and remote zones haven't
2595 * even been considered yet. Try once more without fairness, and
2596 * include remote zones now, before entering the slowpath and waking
2597 * kswapd: prefer spilling to a remote zone over swapping locally.
2599 if (alloc_flags
& ALLOC_FAIR
) {
2600 alloc_flags
&= ~ALLOC_FAIR
;
2601 if (nr_fair_skipped
) {
2602 zonelist_rescan
= true;
2603 reset_alloc_batches(ac
->preferred_zone
);
2605 if (nr_online_nodes
> 1)
2606 zonelist_rescan
= true;
2609 if (zonelist_rescan
)
2616 * Large machines with many possible nodes should not always dump per-node
2617 * meminfo in irq context.
2619 static inline bool should_suppress_show_mem(void)
2624 ret
= in_interrupt();
2629 static DEFINE_RATELIMIT_STATE(nopage_rs
,
2630 DEFAULT_RATELIMIT_INTERVAL
,
2631 DEFAULT_RATELIMIT_BURST
);
2633 void warn_alloc_failed(gfp_t gfp_mask
, unsigned int order
, const char *fmt
, ...)
2635 unsigned int filter
= SHOW_MEM_FILTER_NODES
;
2637 if ((gfp_mask
& __GFP_NOWARN
) || !__ratelimit(&nopage_rs
) ||
2638 debug_guardpage_minorder() > 0)
2642 * This documents exceptions given to allocations in certain
2643 * contexts that are allowed to allocate outside current's set
2646 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2647 if (test_thread_flag(TIF_MEMDIE
) ||
2648 (current
->flags
& (PF_MEMALLOC
| PF_EXITING
)))
2649 filter
&= ~SHOW_MEM_FILTER_NODES
;
2650 if (in_interrupt() || !(gfp_mask
& __GFP_DIRECT_RECLAIM
))
2651 filter
&= ~SHOW_MEM_FILTER_NODES
;
2654 struct va_format vaf
;
2657 va_start(args
, fmt
);
2662 pr_warn("%pV", &vaf
);
2667 pr_warn("%s: page allocation failure: order:%u, mode:0x%x\n",
2668 current
->comm
, order
, gfp_mask
);
2671 if (!should_suppress_show_mem())
2675 static inline struct page
*
2676 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
2677 const struct alloc_context
*ac
, unsigned long *did_some_progress
)
2679 struct oom_control oc
= {
2680 .zonelist
= ac
->zonelist
,
2681 .nodemask
= ac
->nodemask
,
2682 .gfp_mask
= gfp_mask
,
2687 *did_some_progress
= 0;
2690 * Acquire the oom lock. If that fails, somebody else is
2691 * making progress for us.
2693 if (!mutex_trylock(&oom_lock
)) {
2694 *did_some_progress
= 1;
2695 schedule_timeout_uninterruptible(1);
2700 * Go through the zonelist yet one more time, keep very high watermark
2701 * here, this is only to catch a parallel oom killing, we must fail if
2702 * we're still under heavy pressure.
2704 page
= get_page_from_freelist(gfp_mask
| __GFP_HARDWALL
, order
,
2705 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
, ac
);
2709 if (!(gfp_mask
& __GFP_NOFAIL
)) {
2710 /* Coredumps can quickly deplete all memory reserves */
2711 if (current
->flags
& PF_DUMPCORE
)
2713 /* The OOM killer will not help higher order allocs */
2714 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
2716 /* The OOM killer does not needlessly kill tasks for lowmem */
2717 if (ac
->high_zoneidx
< ZONE_NORMAL
)
2719 /* The OOM killer does not compensate for IO-less reclaim */
2720 if (!(gfp_mask
& __GFP_FS
)) {
2722 * XXX: Page reclaim didn't yield anything,
2723 * and the OOM killer can't be invoked, but
2724 * keep looping as per tradition.
2726 *did_some_progress
= 1;
2729 if (pm_suspended_storage())
2731 /* The OOM killer may not free memory on a specific node */
2732 if (gfp_mask
& __GFP_THISNODE
)
2735 /* Exhausted what can be done so it's blamo time */
2736 if (out_of_memory(&oc
) || WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
))
2737 *did_some_progress
= 1;
2739 mutex_unlock(&oom_lock
);
2743 #ifdef CONFIG_COMPACTION
2744 /* Try memory compaction for high-order allocations before reclaim */
2745 static struct page
*
2746 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2747 int alloc_flags
, const struct alloc_context
*ac
,
2748 enum migrate_mode mode
, int *contended_compaction
,
2749 bool *deferred_compaction
)
2751 unsigned long compact_result
;
2757 current
->flags
|= PF_MEMALLOC
;
2758 compact_result
= try_to_compact_pages(gfp_mask
, order
, alloc_flags
, ac
,
2759 mode
, contended_compaction
);
2760 current
->flags
&= ~PF_MEMALLOC
;
2762 switch (compact_result
) {
2763 case COMPACT_DEFERRED
:
2764 *deferred_compaction
= true;
2766 case COMPACT_SKIPPED
:
2773 * At least in one zone compaction wasn't deferred or skipped, so let's
2774 * count a compaction stall
2776 count_vm_event(COMPACTSTALL
);
2778 page
= get_page_from_freelist(gfp_mask
, order
,
2779 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
2782 struct zone
*zone
= page_zone(page
);
2784 zone
->compact_blockskip_flush
= false;
2785 compaction_defer_reset(zone
, order
, true);
2786 count_vm_event(COMPACTSUCCESS
);
2791 * It's bad if compaction run occurs and fails. The most likely reason
2792 * is that pages exist, but not enough to satisfy watermarks.
2794 count_vm_event(COMPACTFAIL
);
2801 static inline struct page
*
2802 __alloc_pages_direct_compact(gfp_t gfp_mask
, unsigned int order
,
2803 int alloc_flags
, const struct alloc_context
*ac
,
2804 enum migrate_mode mode
, int *contended_compaction
,
2805 bool *deferred_compaction
)
2809 #endif /* CONFIG_COMPACTION */
2811 /* Perform direct synchronous page reclaim */
2813 __perform_reclaim(gfp_t gfp_mask
, unsigned int order
,
2814 const struct alloc_context
*ac
)
2816 struct reclaim_state reclaim_state
;
2821 /* We now go into synchronous reclaim */
2822 cpuset_memory_pressure_bump();
2823 current
->flags
|= PF_MEMALLOC
;
2824 lockdep_set_current_reclaim_state(gfp_mask
);
2825 reclaim_state
.reclaimed_slab
= 0;
2826 current
->reclaim_state
= &reclaim_state
;
2828 progress
= try_to_free_pages(ac
->zonelist
, order
, gfp_mask
,
2831 current
->reclaim_state
= NULL
;
2832 lockdep_clear_current_reclaim_state();
2833 current
->flags
&= ~PF_MEMALLOC
;
2840 /* The really slow allocator path where we enter direct reclaim */
2841 static inline struct page
*
2842 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
2843 int alloc_flags
, const struct alloc_context
*ac
,
2844 unsigned long *did_some_progress
)
2846 struct page
*page
= NULL
;
2847 bool drained
= false;
2849 *did_some_progress
= __perform_reclaim(gfp_mask
, order
, ac
);
2850 if (unlikely(!(*did_some_progress
)))
2854 page
= get_page_from_freelist(gfp_mask
, order
,
2855 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
2858 * If an allocation failed after direct reclaim, it could be because
2859 * pages are pinned on the per-cpu lists or in high alloc reserves.
2860 * Shrink them them and try again
2862 if (!page
&& !drained
) {
2863 unreserve_highatomic_pageblock(ac
);
2864 drain_all_pages(NULL
);
2872 static void wake_all_kswapds(unsigned int order
, const struct alloc_context
*ac
)
2877 for_each_zone_zonelist_nodemask(zone
, z
, ac
->zonelist
,
2878 ac
->high_zoneidx
, ac
->nodemask
)
2879 wakeup_kswapd(zone
, order
, zone_idx(ac
->preferred_zone
));
2883 gfp_to_alloc_flags(gfp_t gfp_mask
)
2885 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
2887 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
2888 BUILD_BUG_ON(__GFP_HIGH
!= (__force gfp_t
) ALLOC_HIGH
);
2891 * The caller may dip into page reserves a bit more if the caller
2892 * cannot run direct reclaim, or if the caller has realtime scheduling
2893 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
2894 * set both ALLOC_HARDER (__GFP_ATOMIC) and ALLOC_HIGH (__GFP_HIGH).
2896 alloc_flags
|= (__force
int) (gfp_mask
& __GFP_HIGH
);
2898 if (gfp_mask
& __GFP_ATOMIC
) {
2900 * Not worth trying to allocate harder for __GFP_NOMEMALLOC even
2901 * if it can't schedule.
2903 if (!(gfp_mask
& __GFP_NOMEMALLOC
))
2904 alloc_flags
|= ALLOC_HARDER
;
2906 * Ignore cpuset mems for GFP_ATOMIC rather than fail, see the
2907 * comment for __cpuset_node_allowed().
2909 alloc_flags
&= ~ALLOC_CPUSET
;
2910 } else if (unlikely(rt_task(current
)) && !in_interrupt())
2911 alloc_flags
|= ALLOC_HARDER
;
2913 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
2914 if (gfp_mask
& __GFP_MEMALLOC
)
2915 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2916 else if (in_serving_softirq() && (current
->flags
& PF_MEMALLOC
))
2917 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2918 else if (!in_interrupt() &&
2919 ((current
->flags
& PF_MEMALLOC
) ||
2920 unlikely(test_thread_flag(TIF_MEMDIE
))))
2921 alloc_flags
|= ALLOC_NO_WATERMARKS
;
2924 if (gfpflags_to_migratetype(gfp_mask
) == MIGRATE_MOVABLE
)
2925 alloc_flags
|= ALLOC_CMA
;
2930 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask
)
2932 return !!(gfp_to_alloc_flags(gfp_mask
) & ALLOC_NO_WATERMARKS
);
2935 static inline bool is_thp_gfp_mask(gfp_t gfp_mask
)
2937 return (gfp_mask
& (GFP_TRANSHUGE
| __GFP_KSWAPD_RECLAIM
)) == GFP_TRANSHUGE
;
2940 static inline struct page
*
2941 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
2942 struct alloc_context
*ac
)
2944 bool can_direct_reclaim
= gfp_mask
& __GFP_DIRECT_RECLAIM
;
2945 struct page
*page
= NULL
;
2947 unsigned long pages_reclaimed
= 0;
2948 unsigned long did_some_progress
;
2949 enum migrate_mode migration_mode
= MIGRATE_ASYNC
;
2950 bool deferred_compaction
= false;
2951 int contended_compaction
= COMPACT_CONTENDED_NONE
;
2954 * In the slowpath, we sanity check order to avoid ever trying to
2955 * reclaim >= MAX_ORDER areas which will never succeed. Callers may
2956 * be using allocators in order of preference for an area that is
2959 if (order
>= MAX_ORDER
) {
2960 WARN_ON_ONCE(!(gfp_mask
& __GFP_NOWARN
));
2965 * We also sanity check to catch abuse of atomic reserves being used by
2966 * callers that are not in atomic context.
2968 if (WARN_ON_ONCE((gfp_mask
& (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)) ==
2969 (__GFP_ATOMIC
|__GFP_DIRECT_RECLAIM
)))
2970 gfp_mask
&= ~__GFP_ATOMIC
;
2973 * If this allocation cannot block and it is for a specific node, then
2974 * fail early. There's no need to wakeup kswapd or retry for a
2975 * speculative node-specific allocation.
2977 if (IS_ENABLED(CONFIG_NUMA
) && (gfp_mask
& __GFP_THISNODE
) && !can_direct_reclaim
)
2981 if (gfp_mask
& __GFP_KSWAPD_RECLAIM
)
2982 wake_all_kswapds(order
, ac
);
2985 * OK, we're below the kswapd watermark and have kicked background
2986 * reclaim. Now things get more complex, so set up alloc_flags according
2987 * to how we want to proceed.
2989 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
2992 * Find the true preferred zone if the allocation is unconstrained by
2995 if (!(alloc_flags
& ALLOC_CPUSET
) && !ac
->nodemask
) {
2996 struct zoneref
*preferred_zoneref
;
2997 preferred_zoneref
= first_zones_zonelist(ac
->zonelist
,
2998 ac
->high_zoneidx
, NULL
, &ac
->preferred_zone
);
2999 ac
->classzone_idx
= zonelist_zone_idx(preferred_zoneref
);
3002 /* This is the last chance, in general, before the goto nopage. */
3003 page
= get_page_from_freelist(gfp_mask
, order
,
3004 alloc_flags
& ~ALLOC_NO_WATERMARKS
, ac
);
3008 /* Allocate without watermarks if the context allows */
3009 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
3011 * Ignore mempolicies if ALLOC_NO_WATERMARKS on the grounds
3012 * the allocation is high priority and these type of
3013 * allocations are system rather than user orientated
3015 ac
->zonelist
= node_zonelist(numa_node_id(), gfp_mask
);
3016 page
= get_page_from_freelist(gfp_mask
, order
,
3017 ALLOC_NO_WATERMARKS
, ac
);
3022 /* Caller is not willing to reclaim, we can't balance anything */
3023 if (!can_direct_reclaim
) {
3025 * All existing users of the __GFP_NOFAIL are blockable, so warn
3026 * of any new users that actually allow this type of allocation
3029 WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
);
3033 /* Avoid recursion of direct reclaim */
3034 if (current
->flags
& PF_MEMALLOC
) {
3036 * __GFP_NOFAIL request from this context is rather bizarre
3037 * because we cannot reclaim anything and only can loop waiting
3038 * for somebody to do a work for us.
3040 if (WARN_ON_ONCE(gfp_mask
& __GFP_NOFAIL
)) {
3047 /* Avoid allocations with no watermarks from looping endlessly */
3048 if (test_thread_flag(TIF_MEMDIE
) && !(gfp_mask
& __GFP_NOFAIL
))
3052 * Try direct compaction. The first pass is asynchronous. Subsequent
3053 * attempts after direct reclaim are synchronous
3055 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
, ac
,
3057 &contended_compaction
,
3058 &deferred_compaction
);
3062 /* Checks for THP-specific high-order allocations */
3063 if (is_thp_gfp_mask(gfp_mask
)) {
3065 * If compaction is deferred for high-order allocations, it is
3066 * because sync compaction recently failed. If this is the case
3067 * and the caller requested a THP allocation, we do not want
3068 * to heavily disrupt the system, so we fail the allocation
3069 * instead of entering direct reclaim.
3071 if (deferred_compaction
)
3075 * In all zones where compaction was attempted (and not
3076 * deferred or skipped), lock contention has been detected.
3077 * For THP allocation we do not want to disrupt the others
3078 * so we fallback to base pages instead.
3080 if (contended_compaction
== COMPACT_CONTENDED_LOCK
)
3084 * If compaction was aborted due to need_resched(), we do not
3085 * want to further increase allocation latency, unless it is
3086 * khugepaged trying to collapse.
3088 if (contended_compaction
== COMPACT_CONTENDED_SCHED
3089 && !(current
->flags
& PF_KTHREAD
))
3094 * It can become very expensive to allocate transparent hugepages at
3095 * fault, so use asynchronous memory compaction for THP unless it is
3096 * khugepaged trying to collapse.
3098 if (!is_thp_gfp_mask(gfp_mask
) || (current
->flags
& PF_KTHREAD
))
3099 migration_mode
= MIGRATE_SYNC_LIGHT
;
3101 /* Try direct reclaim and then allocating */
3102 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
, alloc_flags
, ac
,
3103 &did_some_progress
);
3107 /* Do not loop if specifically requested */
3108 if (gfp_mask
& __GFP_NORETRY
)
3111 /* Keep reclaiming pages as long as there is reasonable progress */
3112 pages_reclaimed
+= did_some_progress
;
3113 if ((did_some_progress
&& order
<= PAGE_ALLOC_COSTLY_ORDER
) ||
3114 ((gfp_mask
& __GFP_REPEAT
) && pages_reclaimed
< (1 << order
))) {
3115 /* Wait for some write requests to complete then retry */
3116 wait_iff_congested(ac
->preferred_zone
, BLK_RW_ASYNC
, HZ
/50);
3120 /* Reclaim has failed us, start killing things */
3121 page
= __alloc_pages_may_oom(gfp_mask
, order
, ac
, &did_some_progress
);
3125 /* Retry as long as the OOM killer is making progress */
3126 if (did_some_progress
)
3131 * High-order allocations do not necessarily loop after
3132 * direct reclaim and reclaim/compaction depends on compaction
3133 * being called after reclaim so call directly if necessary
3135 page
= __alloc_pages_direct_compact(gfp_mask
, order
, alloc_flags
,
3137 &contended_compaction
,
3138 &deferred_compaction
);
3142 warn_alloc_failed(gfp_mask
, order
, NULL
);
3148 * This is the 'heart' of the zoned buddy allocator.
3151 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
3152 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
3154 struct zoneref
*preferred_zoneref
;
3155 struct page
*page
= NULL
;
3156 unsigned int cpuset_mems_cookie
;
3157 int alloc_flags
= ALLOC_WMARK_LOW
|ALLOC_CPUSET
|ALLOC_FAIR
;
3158 gfp_t alloc_mask
; /* The gfp_t that was actually used for allocation */
3159 struct alloc_context ac
= {
3160 .high_zoneidx
= gfp_zone(gfp_mask
),
3161 .nodemask
= nodemask
,
3162 .migratetype
= gfpflags_to_migratetype(gfp_mask
),
3165 gfp_mask
&= gfp_allowed_mask
;
3167 lockdep_trace_alloc(gfp_mask
);
3169 might_sleep_if(gfp_mask
& __GFP_DIRECT_RECLAIM
);
3171 if (should_fail_alloc_page(gfp_mask
, order
))
3175 * Check the zones suitable for the gfp_mask contain at least one
3176 * valid zone. It's possible to have an empty zonelist as a result
3177 * of __GFP_THISNODE and a memoryless node
3179 if (unlikely(!zonelist
->_zonerefs
->zone
))
3182 if (IS_ENABLED(CONFIG_CMA
) && ac
.migratetype
== MIGRATE_MOVABLE
)
3183 alloc_flags
|= ALLOC_CMA
;
3186 cpuset_mems_cookie
= read_mems_allowed_begin();
3188 /* We set it here, as __alloc_pages_slowpath might have changed it */
3189 ac
.zonelist
= zonelist
;
3191 /* Dirty zone balancing only done in the fast path */
3192 ac
.spread_dirty_pages
= (gfp_mask
& __GFP_WRITE
);
3194 /* The preferred zone is used for statistics later */
3195 preferred_zoneref
= first_zones_zonelist(ac
.zonelist
, ac
.high_zoneidx
,
3196 ac
.nodemask
? : &cpuset_current_mems_allowed
,
3197 &ac
.preferred_zone
);
3198 if (!ac
.preferred_zone
)
3200 ac
.classzone_idx
= zonelist_zone_idx(preferred_zoneref
);
3202 /* First allocation attempt */
3203 alloc_mask
= gfp_mask
|__GFP_HARDWALL
;
3204 page
= get_page_from_freelist(alloc_mask
, order
, alloc_flags
, &ac
);
3205 if (unlikely(!page
)) {
3207 * Runtime PM, block IO and its error handling path
3208 * can deadlock because I/O on the device might not
3211 alloc_mask
= memalloc_noio_flags(gfp_mask
);
3212 ac
.spread_dirty_pages
= false;
3214 page
= __alloc_pages_slowpath(alloc_mask
, order
, &ac
);
3217 if (kmemcheck_enabled
&& page
)
3218 kmemcheck_pagealloc_alloc(page
, order
, gfp_mask
);
3220 trace_mm_page_alloc(page
, order
, alloc_mask
, ac
.migratetype
);
3224 * When updating a task's mems_allowed, it is possible to race with
3225 * parallel threads in such a way that an allocation can fail while
3226 * the mask is being updated. If a page allocation is about to fail,
3227 * check if the cpuset changed during allocation and if so, retry.
3229 if (unlikely(!page
&& read_mems_allowed_retry(cpuset_mems_cookie
)))
3234 EXPORT_SYMBOL(__alloc_pages_nodemask
);
3237 * Common helper functions.
3239 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
3244 * __get_free_pages() returns a 32-bit address, which cannot represent
3247 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
3249 page
= alloc_pages(gfp_mask
, order
);
3252 return (unsigned long) page_address(page
);
3254 EXPORT_SYMBOL(__get_free_pages
);
3256 unsigned long get_zeroed_page(gfp_t gfp_mask
)
3258 return __get_free_pages(gfp_mask
| __GFP_ZERO
, 0);
3260 EXPORT_SYMBOL(get_zeroed_page
);
3262 void __free_pages(struct page
*page
, unsigned int order
)
3264 if (put_page_testzero(page
)) {
3266 free_hot_cold_page(page
, false);
3268 __free_pages_ok(page
, order
);
3272 EXPORT_SYMBOL(__free_pages
);
3274 void free_pages(unsigned long addr
, unsigned int order
)
3277 VM_BUG_ON(!virt_addr_valid((void *)addr
));
3278 __free_pages(virt_to_page((void *)addr
), order
);
3282 EXPORT_SYMBOL(free_pages
);
3286 * An arbitrary-length arbitrary-offset area of memory which resides
3287 * within a 0 or higher order page. Multiple fragments within that page
3288 * are individually refcounted, in the page's reference counter.
3290 * The page_frag functions below provide a simple allocation framework for
3291 * page fragments. This is used by the network stack and network device
3292 * drivers to provide a backing region of memory for use as either an
3293 * sk_buff->head, or to be used in the "frags" portion of skb_shared_info.
3295 static struct page
*__page_frag_refill(struct page_frag_cache
*nc
,
3298 struct page
*page
= NULL
;
3299 gfp_t gfp
= gfp_mask
;
3301 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3302 gfp_mask
|= __GFP_COMP
| __GFP_NOWARN
| __GFP_NORETRY
|
3304 page
= alloc_pages_node(NUMA_NO_NODE
, gfp_mask
,
3305 PAGE_FRAG_CACHE_MAX_ORDER
);
3306 nc
->size
= page
? PAGE_FRAG_CACHE_MAX_SIZE
: PAGE_SIZE
;
3308 if (unlikely(!page
))
3309 page
= alloc_pages_node(NUMA_NO_NODE
, gfp
, 0);
3311 nc
->va
= page
? page_address(page
) : NULL
;
3316 void *__alloc_page_frag(struct page_frag_cache
*nc
,
3317 unsigned int fragsz
, gfp_t gfp_mask
)
3319 unsigned int size
= PAGE_SIZE
;
3323 if (unlikely(!nc
->va
)) {
3325 page
= __page_frag_refill(nc
, gfp_mask
);
3329 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3330 /* if size can vary use size else just use PAGE_SIZE */
3333 /* Even if we own the page, we do not use atomic_set().
3334 * This would break get_page_unless_zero() users.
3336 atomic_add(size
- 1, &page
->_count
);
3338 /* reset page count bias and offset to start of new frag */
3339 nc
->pfmemalloc
= page_is_pfmemalloc(page
);
3340 nc
->pagecnt_bias
= size
;
3344 offset
= nc
->offset
- fragsz
;
3345 if (unlikely(offset
< 0)) {
3346 page
= virt_to_page(nc
->va
);
3348 if (!atomic_sub_and_test(nc
->pagecnt_bias
, &page
->_count
))
3351 #if (PAGE_SIZE < PAGE_FRAG_CACHE_MAX_SIZE)
3352 /* if size can vary use size else just use PAGE_SIZE */
3355 /* OK, page count is 0, we can safely set it */
3356 atomic_set(&page
->_count
, size
);
3358 /* reset page count bias and offset to start of new frag */
3359 nc
->pagecnt_bias
= size
;
3360 offset
= size
- fragsz
;
3364 nc
->offset
= offset
;
3366 return nc
->va
+ offset
;
3368 EXPORT_SYMBOL(__alloc_page_frag
);
3371 * Frees a page fragment allocated out of either a compound or order 0 page.
3373 void __free_page_frag(void *addr
)
3375 struct page
*page
= virt_to_head_page(addr
);
3377 if (unlikely(put_page_testzero(page
)))
3378 __free_pages_ok(page
, compound_order(page
));
3380 EXPORT_SYMBOL(__free_page_frag
);
3383 * alloc_kmem_pages charges newly allocated pages to the kmem resource counter
3384 * of the current memory cgroup if __GFP_ACCOUNT is set, other than that it is
3385 * equivalent to alloc_pages.
3387 * It should be used when the caller would like to use kmalloc, but since the
3388 * allocation is large, it has to fall back to the page allocator.
3390 struct page
*alloc_kmem_pages(gfp_t gfp_mask
, unsigned int order
)
3394 page
= alloc_pages(gfp_mask
, order
);
3395 if (page
&& memcg_kmem_charge(page
, gfp_mask
, order
) != 0) {
3396 __free_pages(page
, order
);
3402 struct page
*alloc_kmem_pages_node(int nid
, gfp_t gfp_mask
, unsigned int order
)
3406 page
= alloc_pages_node(nid
, gfp_mask
, order
);
3407 if (page
&& memcg_kmem_charge(page
, gfp_mask
, order
) != 0) {
3408 __free_pages(page
, order
);
3415 * __free_kmem_pages and free_kmem_pages will free pages allocated with
3418 void __free_kmem_pages(struct page
*page
, unsigned int order
)
3420 memcg_kmem_uncharge(page
, order
);
3421 __free_pages(page
, order
);
3424 void free_kmem_pages(unsigned long addr
, unsigned int order
)
3427 VM_BUG_ON(!virt_addr_valid((void *)addr
));
3428 __free_kmem_pages(virt_to_page((void *)addr
), order
);
3432 static void *make_alloc_exact(unsigned long addr
, unsigned int order
,
3436 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
3437 unsigned long used
= addr
+ PAGE_ALIGN(size
);
3439 split_page(virt_to_page((void *)addr
), order
);
3440 while (used
< alloc_end
) {
3445 return (void *)addr
;
3449 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
3450 * @size: the number of bytes to allocate
3451 * @gfp_mask: GFP flags for the allocation
3453 * This function is similar to alloc_pages(), except that it allocates the
3454 * minimum number of pages to satisfy the request. alloc_pages() can only
3455 * allocate memory in power-of-two pages.
3457 * This function is also limited by MAX_ORDER.
3459 * Memory allocated by this function must be released by free_pages_exact().
3461 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
3463 unsigned int order
= get_order(size
);
3466 addr
= __get_free_pages(gfp_mask
, order
);
3467 return make_alloc_exact(addr
, order
, size
);
3469 EXPORT_SYMBOL(alloc_pages_exact
);
3472 * alloc_pages_exact_nid - allocate an exact number of physically-contiguous
3474 * @nid: the preferred node ID where memory should be allocated
3475 * @size: the number of bytes to allocate
3476 * @gfp_mask: GFP flags for the allocation
3478 * Like alloc_pages_exact(), but try to allocate on node nid first before falling
3481 void * __meminit
alloc_pages_exact_nid(int nid
, size_t size
, gfp_t gfp_mask
)
3483 unsigned int order
= get_order(size
);
3484 struct page
*p
= alloc_pages_node(nid
, gfp_mask
, order
);
3487 return make_alloc_exact((unsigned long)page_address(p
), order
, size
);
3491 * free_pages_exact - release memory allocated via alloc_pages_exact()
3492 * @virt: the value returned by alloc_pages_exact.
3493 * @size: size of allocation, same value as passed to alloc_pages_exact().
3495 * Release the memory allocated by a previous call to alloc_pages_exact.
3497 void free_pages_exact(void *virt
, size_t size
)
3499 unsigned long addr
= (unsigned long)virt
;
3500 unsigned long end
= addr
+ PAGE_ALIGN(size
);
3502 while (addr
< end
) {
3507 EXPORT_SYMBOL(free_pages_exact
);
3510 * nr_free_zone_pages - count number of pages beyond high watermark
3511 * @offset: The zone index of the highest zone
3513 * nr_free_zone_pages() counts the number of counts pages which are beyond the
3514 * high watermark within all zones at or below a given zone index. For each
3515 * zone, the number of pages is calculated as:
3516 * managed_pages - high_pages
3518 static unsigned long nr_free_zone_pages(int offset
)
3523 /* Just pick one node, since fallback list is circular */
3524 unsigned long sum
= 0;
3526 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
3528 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
3529 unsigned long size
= zone
->managed_pages
;
3530 unsigned long high
= high_wmark_pages(zone
);
3539 * nr_free_buffer_pages - count number of pages beyond high watermark
3541 * nr_free_buffer_pages() counts the number of pages which are beyond the high
3542 * watermark within ZONE_DMA and ZONE_NORMAL.
3544 unsigned long nr_free_buffer_pages(void)
3546 return nr_free_zone_pages(gfp_zone(GFP_USER
));
3548 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
3551 * nr_free_pagecache_pages - count number of pages beyond high watermark
3553 * nr_free_pagecache_pages() counts the number of pages which are beyond the
3554 * high watermark within all zones.
3556 unsigned long nr_free_pagecache_pages(void)
3558 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
3561 static inline void show_node(struct zone
*zone
)
3563 if (IS_ENABLED(CONFIG_NUMA
))
3564 printk("Node %d ", zone_to_nid(zone
));
3567 void si_meminfo(struct sysinfo
*val
)
3569 val
->totalram
= totalram_pages
;
3570 val
->sharedram
= global_page_state(NR_SHMEM
);
3571 val
->freeram
= global_page_state(NR_FREE_PAGES
);
3572 val
->bufferram
= nr_blockdev_pages();
3573 val
->totalhigh
= totalhigh_pages
;
3574 val
->freehigh
= nr_free_highpages();
3575 val
->mem_unit
= PAGE_SIZE
;
3578 EXPORT_SYMBOL(si_meminfo
);
3581 void si_meminfo_node(struct sysinfo
*val
, int nid
)
3583 int zone_type
; /* needs to be signed */
3584 unsigned long managed_pages
= 0;
3585 pg_data_t
*pgdat
= NODE_DATA(nid
);
3587 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++)
3588 managed_pages
+= pgdat
->node_zones
[zone_type
].managed_pages
;
3589 val
->totalram
= managed_pages
;
3590 val
->sharedram
= node_page_state(nid
, NR_SHMEM
);
3591 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
3592 #ifdef CONFIG_HIGHMEM
3593 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].managed_pages
;
3594 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
3600 val
->mem_unit
= PAGE_SIZE
;
3605 * Determine whether the node should be displayed or not, depending on whether
3606 * SHOW_MEM_FILTER_NODES was passed to show_free_areas().
3608 bool skip_free_areas_node(unsigned int flags
, int nid
)
3611 unsigned int cpuset_mems_cookie
;
3613 if (!(flags
& SHOW_MEM_FILTER_NODES
))
3617 cpuset_mems_cookie
= read_mems_allowed_begin();
3618 ret
= !node_isset(nid
, cpuset_current_mems_allowed
);
3619 } while (read_mems_allowed_retry(cpuset_mems_cookie
));
3624 #define K(x) ((x) << (PAGE_SHIFT-10))
3626 static void show_migration_types(unsigned char type
)
3628 static const char types
[MIGRATE_TYPES
] = {
3629 [MIGRATE_UNMOVABLE
] = 'U',
3630 [MIGRATE_MOVABLE
] = 'M',
3631 [MIGRATE_RECLAIMABLE
] = 'E',
3632 [MIGRATE_HIGHATOMIC
] = 'H',
3634 [MIGRATE_CMA
] = 'C',
3636 #ifdef CONFIG_MEMORY_ISOLATION
3637 [MIGRATE_ISOLATE
] = 'I',
3640 char tmp
[MIGRATE_TYPES
+ 1];
3644 for (i
= 0; i
< MIGRATE_TYPES
; i
++) {
3645 if (type
& (1 << i
))
3650 printk("(%s) ", tmp
);
3654 * Show free area list (used inside shift_scroll-lock stuff)
3655 * We also calculate the percentage fragmentation. We do this by counting the
3656 * memory on each free list with the exception of the first item on the list.
3659 * SHOW_MEM_FILTER_NODES: suppress nodes that are not allowed by current's
3662 void show_free_areas(unsigned int filter
)
3664 unsigned long free_pcp
= 0;
3668 for_each_populated_zone(zone
) {
3669 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3672 for_each_online_cpu(cpu
)
3673 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
3676 printk("active_anon:%lu inactive_anon:%lu isolated_anon:%lu\n"
3677 " active_file:%lu inactive_file:%lu isolated_file:%lu\n"
3678 " unevictable:%lu dirty:%lu writeback:%lu unstable:%lu\n"
3679 " slab_reclaimable:%lu slab_unreclaimable:%lu\n"
3680 " mapped:%lu shmem:%lu pagetables:%lu bounce:%lu\n"
3681 " free:%lu free_pcp:%lu free_cma:%lu\n",
3682 global_page_state(NR_ACTIVE_ANON
),
3683 global_page_state(NR_INACTIVE_ANON
),
3684 global_page_state(NR_ISOLATED_ANON
),
3685 global_page_state(NR_ACTIVE_FILE
),
3686 global_page_state(NR_INACTIVE_FILE
),
3687 global_page_state(NR_ISOLATED_FILE
),
3688 global_page_state(NR_UNEVICTABLE
),
3689 global_page_state(NR_FILE_DIRTY
),
3690 global_page_state(NR_WRITEBACK
),
3691 global_page_state(NR_UNSTABLE_NFS
),
3692 global_page_state(NR_SLAB_RECLAIMABLE
),
3693 global_page_state(NR_SLAB_UNRECLAIMABLE
),
3694 global_page_state(NR_FILE_MAPPED
),
3695 global_page_state(NR_SHMEM
),
3696 global_page_state(NR_PAGETABLE
),
3697 global_page_state(NR_BOUNCE
),
3698 global_page_state(NR_FREE_PAGES
),
3700 global_page_state(NR_FREE_CMA_PAGES
));
3702 for_each_populated_zone(zone
) {
3705 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3709 for_each_online_cpu(cpu
)
3710 free_pcp
+= per_cpu_ptr(zone
->pageset
, cpu
)->pcp
.count
;
3718 " active_anon:%lukB"
3719 " inactive_anon:%lukB"
3720 " active_file:%lukB"
3721 " inactive_file:%lukB"
3722 " unevictable:%lukB"
3723 " isolated(anon):%lukB"
3724 " isolated(file):%lukB"
3732 " slab_reclaimable:%lukB"
3733 " slab_unreclaimable:%lukB"
3734 " kernel_stack:%lukB"
3741 " writeback_tmp:%lukB"
3742 " pages_scanned:%lu"
3743 " all_unreclaimable? %s"
3746 K(zone_page_state(zone
, NR_FREE_PAGES
)),
3747 K(min_wmark_pages(zone
)),
3748 K(low_wmark_pages(zone
)),
3749 K(high_wmark_pages(zone
)),
3750 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
3751 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
3752 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
3753 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
3754 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
3755 K(zone_page_state(zone
, NR_ISOLATED_ANON
)),
3756 K(zone_page_state(zone
, NR_ISOLATED_FILE
)),
3757 K(zone
->present_pages
),
3758 K(zone
->managed_pages
),
3759 K(zone_page_state(zone
, NR_MLOCK
)),
3760 K(zone_page_state(zone
, NR_FILE_DIRTY
)),
3761 K(zone_page_state(zone
, NR_WRITEBACK
)),
3762 K(zone_page_state(zone
, NR_FILE_MAPPED
)),
3763 K(zone_page_state(zone
, NR_SHMEM
)),
3764 K(zone_page_state(zone
, NR_SLAB_RECLAIMABLE
)),
3765 K(zone_page_state(zone
, NR_SLAB_UNRECLAIMABLE
)),
3766 zone_page_state(zone
, NR_KERNEL_STACK
) *
3768 K(zone_page_state(zone
, NR_PAGETABLE
)),
3769 K(zone_page_state(zone
, NR_UNSTABLE_NFS
)),
3770 K(zone_page_state(zone
, NR_BOUNCE
)),
3772 K(this_cpu_read(zone
->pageset
->pcp
.count
)),
3773 K(zone_page_state(zone
, NR_FREE_CMA_PAGES
)),
3774 K(zone_page_state(zone
, NR_WRITEBACK_TEMP
)),
3775 K(zone_page_state(zone
, NR_PAGES_SCANNED
)),
3776 (!zone_reclaimable(zone
) ? "yes" : "no")
3778 printk("lowmem_reserve[]:");
3779 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3780 printk(" %ld", zone
->lowmem_reserve
[i
]);
3784 for_each_populated_zone(zone
) {
3786 unsigned long nr
[MAX_ORDER
], flags
, total
= 0;
3787 unsigned char types
[MAX_ORDER
];
3789 if (skip_free_areas_node(filter
, zone_to_nid(zone
)))
3792 printk("%s: ", zone
->name
);
3794 spin_lock_irqsave(&zone
->lock
, flags
);
3795 for (order
= 0; order
< MAX_ORDER
; order
++) {
3796 struct free_area
*area
= &zone
->free_area
[order
];
3799 nr
[order
] = area
->nr_free
;
3800 total
+= nr
[order
] << order
;
3803 for (type
= 0; type
< MIGRATE_TYPES
; type
++) {
3804 if (!list_empty(&area
->free_list
[type
]))
3805 types
[order
] |= 1 << type
;
3808 spin_unlock_irqrestore(&zone
->lock
, flags
);
3809 for (order
= 0; order
< MAX_ORDER
; order
++) {
3810 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
3812 show_migration_types(types
[order
]);
3814 printk("= %lukB\n", K(total
));
3817 hugetlb_show_meminfo();
3819 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
3821 show_swap_cache_info();
3824 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
3826 zoneref
->zone
= zone
;
3827 zoneref
->zone_idx
= zone_idx(zone
);
3831 * Builds allocation fallback zone lists.
3833 * Add all populated zones of a node to the zonelist.
3835 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
3839 enum zone_type zone_type
= MAX_NR_ZONES
;
3843 zone
= pgdat
->node_zones
+ zone_type
;
3844 if (populated_zone(zone
)) {
3845 zoneref_set_zone(zone
,
3846 &zonelist
->_zonerefs
[nr_zones
++]);
3847 check_highest_zone(zone_type
);
3849 } while (zone_type
);
3857 * 0 = automatic detection of better ordering.
3858 * 1 = order by ([node] distance, -zonetype)
3859 * 2 = order by (-zonetype, [node] distance)
3861 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
3862 * the same zonelist. So only NUMA can configure this param.
3864 #define ZONELIST_ORDER_DEFAULT 0
3865 #define ZONELIST_ORDER_NODE 1
3866 #define ZONELIST_ORDER_ZONE 2
3868 /* zonelist order in the kernel.
3869 * set_zonelist_order() will set this to NODE or ZONE.
3871 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3872 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
3876 /* The value user specified ....changed by config */
3877 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3878 /* string for sysctl */
3879 #define NUMA_ZONELIST_ORDER_LEN 16
3880 char numa_zonelist_order
[16] = "default";
3883 * interface for configure zonelist ordering.
3884 * command line option "numa_zonelist_order"
3885 * = "[dD]efault - default, automatic configuration.
3886 * = "[nN]ode - order by node locality, then by zone within node
3887 * = "[zZ]one - order by zone, then by locality within zone
3890 static int __parse_numa_zonelist_order(char *s
)
3892 if (*s
== 'd' || *s
== 'D') {
3893 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
3894 } else if (*s
== 'n' || *s
== 'N') {
3895 user_zonelist_order
= ZONELIST_ORDER_NODE
;
3896 } else if (*s
== 'z' || *s
== 'Z') {
3897 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
3900 "Ignoring invalid numa_zonelist_order value: "
3907 static __init
int setup_numa_zonelist_order(char *s
)
3914 ret
= __parse_numa_zonelist_order(s
);
3916 strlcpy(numa_zonelist_order
, s
, NUMA_ZONELIST_ORDER_LEN
);
3920 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
3923 * sysctl handler for numa_zonelist_order
3925 int numa_zonelist_order_handler(struct ctl_table
*table
, int write
,
3926 void __user
*buffer
, size_t *length
,
3929 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
3931 static DEFINE_MUTEX(zl_order_mutex
);
3933 mutex_lock(&zl_order_mutex
);
3935 if (strlen((char *)table
->data
) >= NUMA_ZONELIST_ORDER_LEN
) {
3939 strcpy(saved_string
, (char *)table
->data
);
3941 ret
= proc_dostring(table
, write
, buffer
, length
, ppos
);
3945 int oldval
= user_zonelist_order
;
3947 ret
= __parse_numa_zonelist_order((char *)table
->data
);
3950 * bogus value. restore saved string
3952 strncpy((char *)table
->data
, saved_string
,
3953 NUMA_ZONELIST_ORDER_LEN
);
3954 user_zonelist_order
= oldval
;
3955 } else if (oldval
!= user_zonelist_order
) {
3956 mutex_lock(&zonelists_mutex
);
3957 build_all_zonelists(NULL
, NULL
);
3958 mutex_unlock(&zonelists_mutex
);
3962 mutex_unlock(&zl_order_mutex
);
3967 #define MAX_NODE_LOAD (nr_online_nodes)
3968 static int node_load
[MAX_NUMNODES
];
3971 * find_next_best_node - find the next node that should appear in a given node's fallback list
3972 * @node: node whose fallback list we're appending
3973 * @used_node_mask: nodemask_t of already used nodes
3975 * We use a number of factors to determine which is the next node that should
3976 * appear on a given node's fallback list. The node should not have appeared
3977 * already in @node's fallback list, and it should be the next closest node
3978 * according to the distance array (which contains arbitrary distance values
3979 * from each node to each node in the system), and should also prefer nodes
3980 * with no CPUs, since presumably they'll have very little allocation pressure
3981 * on them otherwise.
3982 * It returns -1 if no node is found.
3984 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
3987 int min_val
= INT_MAX
;
3988 int best_node
= NUMA_NO_NODE
;
3989 const struct cpumask
*tmp
= cpumask_of_node(0);
3991 /* Use the local node if we haven't already */
3992 if (!node_isset(node
, *used_node_mask
)) {
3993 node_set(node
, *used_node_mask
);
3997 for_each_node_state(n
, N_MEMORY
) {
3999 /* Don't want a node to appear more than once */
4000 if (node_isset(n
, *used_node_mask
))
4003 /* Use the distance array to find the distance */
4004 val
= node_distance(node
, n
);
4006 /* Penalize nodes under us ("prefer the next node") */
4009 /* Give preference to headless and unused nodes */
4010 tmp
= cpumask_of_node(n
);
4011 if (!cpumask_empty(tmp
))
4012 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
4014 /* Slight preference for less loaded node */
4015 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
4016 val
+= node_load
[n
];
4018 if (val
< min_val
) {
4025 node_set(best_node
, *used_node_mask
);
4032 * Build zonelists ordered by node and zones within node.
4033 * This results in maximum locality--normal zone overflows into local
4034 * DMA zone, if any--but risks exhausting DMA zone.
4036 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
4039 struct zonelist
*zonelist
;
4041 zonelist
= &pgdat
->node_zonelists
[0];
4042 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
4044 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4045 zonelist
->_zonerefs
[j
].zone
= NULL
;
4046 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4050 * Build gfp_thisnode zonelists
4052 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
4055 struct zonelist
*zonelist
;
4057 zonelist
= &pgdat
->node_zonelists
[1];
4058 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4059 zonelist
->_zonerefs
[j
].zone
= NULL
;
4060 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4064 * Build zonelists ordered by zone and nodes within zones.
4065 * This results in conserving DMA zone[s] until all Normal memory is
4066 * exhausted, but results in overflowing to remote node while memory
4067 * may still exist in local DMA zone.
4069 static int node_order
[MAX_NUMNODES
];
4071 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
4074 int zone_type
; /* needs to be signed */
4076 struct zonelist
*zonelist
;
4078 zonelist
= &pgdat
->node_zonelists
[0];
4080 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
4081 for (j
= 0; j
< nr_nodes
; j
++) {
4082 node
= node_order
[j
];
4083 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
4084 if (populated_zone(z
)) {
4086 &zonelist
->_zonerefs
[pos
++]);
4087 check_highest_zone(zone_type
);
4091 zonelist
->_zonerefs
[pos
].zone
= NULL
;
4092 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
4095 #if defined(CONFIG_64BIT)
4097 * Devices that require DMA32/DMA are relatively rare and do not justify a
4098 * penalty to every machine in case the specialised case applies. Default
4099 * to Node-ordering on 64-bit NUMA machines
4101 static int default_zonelist_order(void)
4103 return ZONELIST_ORDER_NODE
;
4107 * On 32-bit, the Normal zone needs to be preserved for allocations accessible
4108 * by the kernel. If processes running on node 0 deplete the low memory zone
4109 * then reclaim will occur more frequency increasing stalls and potentially
4110 * be easier to OOM if a large percentage of the zone is under writeback or
4111 * dirty. The problem is significantly worse if CONFIG_HIGHPTE is not set.
4112 * Hence, default to zone ordering on 32-bit.
4114 static int default_zonelist_order(void)
4116 return ZONELIST_ORDER_ZONE
;
4118 #endif /* CONFIG_64BIT */
4120 static void set_zonelist_order(void)
4122 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
4123 current_zonelist_order
= default_zonelist_order();
4125 current_zonelist_order
= user_zonelist_order
;
4128 static void build_zonelists(pg_data_t
*pgdat
)
4131 nodemask_t used_mask
;
4132 int local_node
, prev_node
;
4133 struct zonelist
*zonelist
;
4134 unsigned int order
= current_zonelist_order
;
4136 /* initialize zonelists */
4137 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
4138 zonelist
= pgdat
->node_zonelists
+ i
;
4139 zonelist
->_zonerefs
[0].zone
= NULL
;
4140 zonelist
->_zonerefs
[0].zone_idx
= 0;
4143 /* NUMA-aware ordering of nodes */
4144 local_node
= pgdat
->node_id
;
4145 load
= nr_online_nodes
;
4146 prev_node
= local_node
;
4147 nodes_clear(used_mask
);
4149 memset(node_order
, 0, sizeof(node_order
));
4152 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
4154 * We don't want to pressure a particular node.
4155 * So adding penalty to the first node in same
4156 * distance group to make it round-robin.
4158 if (node_distance(local_node
, node
) !=
4159 node_distance(local_node
, prev_node
))
4160 node_load
[node
] = load
;
4164 if (order
== ZONELIST_ORDER_NODE
)
4165 build_zonelists_in_node_order(pgdat
, node
);
4167 node_order
[i
++] = node
; /* remember order */
4170 if (order
== ZONELIST_ORDER_ZONE
) {
4171 /* calculate node order -- i.e., DMA last! */
4172 build_zonelists_in_zone_order(pgdat
, i
);
4175 build_thisnode_zonelists(pgdat
);
4178 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4180 * Return node id of node used for "local" allocations.
4181 * I.e., first node id of first zone in arg node's generic zonelist.
4182 * Used for initializing percpu 'numa_mem', which is used primarily
4183 * for kernel allocations, so use GFP_KERNEL flags to locate zonelist.
4185 int local_memory_node(int node
)
4189 (void)first_zones_zonelist(node_zonelist(node
, GFP_KERNEL
),
4190 gfp_zone(GFP_KERNEL
),
4197 #else /* CONFIG_NUMA */
4199 static void set_zonelist_order(void)
4201 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
4204 static void build_zonelists(pg_data_t
*pgdat
)
4206 int node
, local_node
;
4208 struct zonelist
*zonelist
;
4210 local_node
= pgdat
->node_id
;
4212 zonelist
= &pgdat
->node_zonelists
[0];
4213 j
= build_zonelists_node(pgdat
, zonelist
, 0);
4216 * Now we build the zonelist so that it contains the zones
4217 * of all the other nodes.
4218 * We don't want to pressure a particular node, so when
4219 * building the zones for node N, we make sure that the
4220 * zones coming right after the local ones are those from
4221 * node N+1 (modulo N)
4223 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
4224 if (!node_online(node
))
4226 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4228 for (node
= 0; node
< local_node
; node
++) {
4229 if (!node_online(node
))
4231 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
);
4234 zonelist
->_zonerefs
[j
].zone
= NULL
;
4235 zonelist
->_zonerefs
[j
].zone_idx
= 0;
4238 #endif /* CONFIG_NUMA */
4241 * Boot pageset table. One per cpu which is going to be used for all
4242 * zones and all nodes. The parameters will be set in such a way
4243 * that an item put on a list will immediately be handed over to
4244 * the buddy list. This is safe since pageset manipulation is done
4245 * with interrupts disabled.
4247 * The boot_pagesets must be kept even after bootup is complete for
4248 * unused processors and/or zones. They do play a role for bootstrapping
4249 * hotplugged processors.
4251 * zoneinfo_show() and maybe other functions do
4252 * not check if the processor is online before following the pageset pointer.
4253 * Other parts of the kernel may not check if the zone is available.
4255 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
);
4256 static DEFINE_PER_CPU(struct per_cpu_pageset
, boot_pageset
);
4257 static void setup_zone_pageset(struct zone
*zone
);
4260 * Global mutex to protect against size modification of zonelists
4261 * as well as to serialize pageset setup for the new populated zone.
4263 DEFINE_MUTEX(zonelists_mutex
);
4265 /* return values int ....just for stop_machine() */
4266 static int __build_all_zonelists(void *data
)
4270 pg_data_t
*self
= data
;
4273 memset(node_load
, 0, sizeof(node_load
));
4276 if (self
&& !node_online(self
->node_id
)) {
4277 build_zonelists(self
);
4280 for_each_online_node(nid
) {
4281 pg_data_t
*pgdat
= NODE_DATA(nid
);
4283 build_zonelists(pgdat
);
4287 * Initialize the boot_pagesets that are going to be used
4288 * for bootstrapping processors. The real pagesets for
4289 * each zone will be allocated later when the per cpu
4290 * allocator is available.
4292 * boot_pagesets are used also for bootstrapping offline
4293 * cpus if the system is already booted because the pagesets
4294 * are needed to initialize allocators on a specific cpu too.
4295 * F.e. the percpu allocator needs the page allocator which
4296 * needs the percpu allocator in order to allocate its pagesets
4297 * (a chicken-egg dilemma).
4299 for_each_possible_cpu(cpu
) {
4300 setup_pageset(&per_cpu(boot_pageset
, cpu
), 0);
4302 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
4304 * We now know the "local memory node" for each node--
4305 * i.e., the node of the first zone in the generic zonelist.
4306 * Set up numa_mem percpu variable for on-line cpus. During
4307 * boot, only the boot cpu should be on-line; we'll init the
4308 * secondary cpus' numa_mem as they come on-line. During
4309 * node/memory hotplug, we'll fixup all on-line cpus.
4311 if (cpu_online(cpu
))
4312 set_cpu_numa_mem(cpu
, local_memory_node(cpu_to_node(cpu
)));
4319 static noinline
void __init
4320 build_all_zonelists_init(void)
4322 __build_all_zonelists(NULL
);
4323 mminit_verify_zonelist();
4324 cpuset_init_current_mems_allowed();
4328 * Called with zonelists_mutex held always
4329 * unless system_state == SYSTEM_BOOTING.
4331 * __ref due to (1) call of __meminit annotated setup_zone_pageset
4332 * [we're only called with non-NULL zone through __meminit paths] and
4333 * (2) call of __init annotated helper build_all_zonelists_init
4334 * [protected by SYSTEM_BOOTING].
4336 void __ref
build_all_zonelists(pg_data_t
*pgdat
, struct zone
*zone
)
4338 set_zonelist_order();
4340 if (system_state
== SYSTEM_BOOTING
) {
4341 build_all_zonelists_init();
4343 #ifdef CONFIG_MEMORY_HOTPLUG
4345 setup_zone_pageset(zone
);
4347 /* we have to stop all cpus to guarantee there is no user
4349 stop_machine(__build_all_zonelists
, pgdat
, NULL
);
4350 /* cpuset refresh routine should be here */
4352 vm_total_pages
= nr_free_pagecache_pages();
4354 * Disable grouping by mobility if the number of pages in the
4355 * system is too low to allow the mechanism to work. It would be
4356 * more accurate, but expensive to check per-zone. This check is
4357 * made on memory-hotadd so a system can start with mobility
4358 * disabled and enable it later
4360 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
4361 page_group_by_mobility_disabled
= 1;
4363 page_group_by_mobility_disabled
= 0;
4365 pr_info("Built %i zonelists in %s order, mobility grouping %s. "
4366 "Total pages: %ld\n",
4368 zonelist_order_name
[current_zonelist_order
],
4369 page_group_by_mobility_disabled
? "off" : "on",
4372 pr_info("Policy zone: %s\n", zone_names
[policy_zone
]);
4377 * Helper functions to size the waitqueue hash table.
4378 * Essentially these want to choose hash table sizes sufficiently
4379 * large so that collisions trying to wait on pages are rare.
4380 * But in fact, the number of active page waitqueues on typical
4381 * systems is ridiculously low, less than 200. So this is even
4382 * conservative, even though it seems large.
4384 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
4385 * waitqueues, i.e. the size of the waitq table given the number of pages.
4387 #define PAGES_PER_WAITQUEUE 256
4389 #ifndef CONFIG_MEMORY_HOTPLUG
4390 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
4392 unsigned long size
= 1;
4394 pages
/= PAGES_PER_WAITQUEUE
;
4396 while (size
< pages
)
4400 * Once we have dozens or even hundreds of threads sleeping
4401 * on IO we've got bigger problems than wait queue collision.
4402 * Limit the size of the wait table to a reasonable size.
4404 size
= min(size
, 4096UL);
4406 return max(size
, 4UL);
4410 * A zone's size might be changed by hot-add, so it is not possible to determine
4411 * a suitable size for its wait_table. So we use the maximum size now.
4413 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
4415 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
4416 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
4417 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
4419 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
4420 * or more by the traditional way. (See above). It equals:
4422 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
4423 * ia64(16K page size) : = ( 8G + 4M)byte.
4424 * powerpc (64K page size) : = (32G +16M)byte.
4426 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
4433 * This is an integer logarithm so that shifts can be used later
4434 * to extract the more random high bits from the multiplicative
4435 * hash function before the remainder is taken.
4437 static inline unsigned long wait_table_bits(unsigned long size
)
4443 * Initially all pages are reserved - free ones are freed
4444 * up by free_all_bootmem() once the early boot process is
4445 * done. Non-atomic initialization, single-pass.
4447 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
4448 unsigned long start_pfn
, enum memmap_context context
)
4450 pg_data_t
*pgdat
= NODE_DATA(nid
);
4451 unsigned long end_pfn
= start_pfn
+ size
;
4454 unsigned long nr_initialised
= 0;
4456 if (highest_memmap_pfn
< end_pfn
- 1)
4457 highest_memmap_pfn
= end_pfn
- 1;
4459 z
= &pgdat
->node_zones
[zone
];
4460 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
4462 * There can be holes in boot-time mem_map[]s
4463 * handed to this function. They do not
4464 * exist on hotplugged memory.
4466 if (context
== MEMMAP_EARLY
) {
4467 if (!early_pfn_valid(pfn
))
4469 if (!early_pfn_in_nid(pfn
, nid
))
4471 if (!update_defer_init(pgdat
, pfn
, end_pfn
,
4477 * Mark the block movable so that blocks are reserved for
4478 * movable at startup. This will force kernel allocations
4479 * to reserve their blocks rather than leaking throughout
4480 * the address space during boot when many long-lived
4481 * kernel allocations are made.
4483 * bitmap is created for zone's valid pfn range. but memmap
4484 * can be created for invalid pages (for alignment)
4485 * check here not to call set_pageblock_migratetype() against
4488 if (!(pfn
& (pageblock_nr_pages
- 1))) {
4489 struct page
*page
= pfn_to_page(pfn
);
4491 __init_single_page(page
, pfn
, zone
, nid
);
4492 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4494 __init_single_pfn(pfn
, zone
, nid
);
4499 static void __meminit
zone_init_free_lists(struct zone
*zone
)
4501 unsigned int order
, t
;
4502 for_each_migratetype_order(order
, t
) {
4503 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
4504 zone
->free_area
[order
].nr_free
= 0;
4508 #ifndef __HAVE_ARCH_MEMMAP_INIT
4509 #define memmap_init(size, nid, zone, start_pfn) \
4510 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
4513 static int zone_batchsize(struct zone
*zone
)
4519 * The per-cpu-pages pools are set to around 1000th of the
4520 * size of the zone. But no more than 1/2 of a meg.
4522 * OK, so we don't know how big the cache is. So guess.
4524 batch
= zone
->managed_pages
/ 1024;
4525 if (batch
* PAGE_SIZE
> 512 * 1024)
4526 batch
= (512 * 1024) / PAGE_SIZE
;
4527 batch
/= 4; /* We effectively *= 4 below */
4532 * Clamp the batch to a 2^n - 1 value. Having a power
4533 * of 2 value was found to be more likely to have
4534 * suboptimal cache aliasing properties in some cases.
4536 * For example if 2 tasks are alternately allocating
4537 * batches of pages, one task can end up with a lot
4538 * of pages of one half of the possible page colors
4539 * and the other with pages of the other colors.
4541 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
4546 /* The deferral and batching of frees should be suppressed under NOMMU
4549 * The problem is that NOMMU needs to be able to allocate large chunks
4550 * of contiguous memory as there's no hardware page translation to
4551 * assemble apparent contiguous memory from discontiguous pages.
4553 * Queueing large contiguous runs of pages for batching, however,
4554 * causes the pages to actually be freed in smaller chunks. As there
4555 * can be a significant delay between the individual batches being
4556 * recycled, this leads to the once large chunks of space being
4557 * fragmented and becoming unavailable for high-order allocations.
4564 * pcp->high and pcp->batch values are related and dependent on one another:
4565 * ->batch must never be higher then ->high.
4566 * The following function updates them in a safe manner without read side
4569 * Any new users of pcp->batch and pcp->high should ensure they can cope with
4570 * those fields changing asynchronously (acording the the above rule).
4572 * mutex_is_locked(&pcp_batch_high_lock) required when calling this function
4573 * outside of boot time (or some other assurance that no concurrent updaters
4576 static void pageset_update(struct per_cpu_pages
*pcp
, unsigned long high
,
4577 unsigned long batch
)
4579 /* start with a fail safe value for batch */
4583 /* Update high, then batch, in order */
4590 /* a companion to pageset_set_high() */
4591 static void pageset_set_batch(struct per_cpu_pageset
*p
, unsigned long batch
)
4593 pageset_update(&p
->pcp
, 6 * batch
, max(1UL, 1 * batch
));
4596 static void pageset_init(struct per_cpu_pageset
*p
)
4598 struct per_cpu_pages
*pcp
;
4601 memset(p
, 0, sizeof(*p
));
4605 for (migratetype
= 0; migratetype
< MIGRATE_PCPTYPES
; migratetype
++)
4606 INIT_LIST_HEAD(&pcp
->lists
[migratetype
]);
4609 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
4612 pageset_set_batch(p
, batch
);
4616 * pageset_set_high() sets the high water mark for hot per_cpu_pagelist
4617 * to the value high for the pageset p.
4619 static void pageset_set_high(struct per_cpu_pageset
*p
,
4622 unsigned long batch
= max(1UL, high
/ 4);
4623 if ((high
/ 4) > (PAGE_SHIFT
* 8))
4624 batch
= PAGE_SHIFT
* 8;
4626 pageset_update(&p
->pcp
, high
, batch
);
4629 static void pageset_set_high_and_batch(struct zone
*zone
,
4630 struct per_cpu_pageset
*pcp
)
4632 if (percpu_pagelist_fraction
)
4633 pageset_set_high(pcp
,
4634 (zone
->managed_pages
/
4635 percpu_pagelist_fraction
));
4637 pageset_set_batch(pcp
, zone_batchsize(zone
));
4640 static void __meminit
zone_pageset_init(struct zone
*zone
, int cpu
)
4642 struct per_cpu_pageset
*pcp
= per_cpu_ptr(zone
->pageset
, cpu
);
4645 pageset_set_high_and_batch(zone
, pcp
);
4648 static void __meminit
setup_zone_pageset(struct zone
*zone
)
4651 zone
->pageset
= alloc_percpu(struct per_cpu_pageset
);
4652 for_each_possible_cpu(cpu
)
4653 zone_pageset_init(zone
, cpu
);
4657 * Allocate per cpu pagesets and initialize them.
4658 * Before this call only boot pagesets were available.
4660 void __init
setup_per_cpu_pageset(void)
4664 for_each_populated_zone(zone
)
4665 setup_zone_pageset(zone
);
4668 static noinline __init_refok
4669 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
4675 * The per-page waitqueue mechanism uses hashed waitqueues
4678 zone
->wait_table_hash_nr_entries
=
4679 wait_table_hash_nr_entries(zone_size_pages
);
4680 zone
->wait_table_bits
=
4681 wait_table_bits(zone
->wait_table_hash_nr_entries
);
4682 alloc_size
= zone
->wait_table_hash_nr_entries
4683 * sizeof(wait_queue_head_t
);
4685 if (!slab_is_available()) {
4686 zone
->wait_table
= (wait_queue_head_t
*)
4687 memblock_virt_alloc_node_nopanic(
4688 alloc_size
, zone
->zone_pgdat
->node_id
);
4691 * This case means that a zone whose size was 0 gets new memory
4692 * via memory hot-add.
4693 * But it may be the case that a new node was hot-added. In
4694 * this case vmalloc() will not be able to use this new node's
4695 * memory - this wait_table must be initialized to use this new
4696 * node itself as well.
4697 * To use this new node's memory, further consideration will be
4700 zone
->wait_table
= vmalloc(alloc_size
);
4702 if (!zone
->wait_table
)
4705 for (i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
4706 init_waitqueue_head(zone
->wait_table
+ i
);
4711 static __meminit
void zone_pcp_init(struct zone
*zone
)
4714 * per cpu subsystem is not up at this point. The following code
4715 * relies on the ability of the linker to provide the
4716 * offset of a (static) per cpu variable into the per cpu area.
4718 zone
->pageset
= &boot_pageset
;
4720 if (populated_zone(zone
))
4721 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%u\n",
4722 zone
->name
, zone
->present_pages
,
4723 zone_batchsize(zone
));
4726 int __meminit
init_currently_empty_zone(struct zone
*zone
,
4727 unsigned long zone_start_pfn
,
4730 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
4732 ret
= zone_wait_table_init(zone
, size
);
4735 pgdat
->nr_zones
= zone_idx(zone
) + 1;
4737 zone
->zone_start_pfn
= zone_start_pfn
;
4739 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
4740 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
4742 (unsigned long)zone_idx(zone
),
4743 zone_start_pfn
, (zone_start_pfn
+ size
));
4745 zone_init_free_lists(zone
);
4750 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
4751 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
4754 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
4756 int __meminit
__early_pfn_to_nid(unsigned long pfn
,
4757 struct mminit_pfnnid_cache
*state
)
4759 unsigned long start_pfn
, end_pfn
;
4762 if (state
->last_start
<= pfn
&& pfn
< state
->last_end
)
4763 return state
->last_nid
;
4765 nid
= memblock_search_pfn_nid(pfn
, &start_pfn
, &end_pfn
);
4767 state
->last_start
= start_pfn
;
4768 state
->last_end
= end_pfn
;
4769 state
->last_nid
= nid
;
4774 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
4777 * free_bootmem_with_active_regions - Call memblock_free_early_nid for each active range
4778 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
4779 * @max_low_pfn: The highest PFN that will be passed to memblock_free_early_nid
4781 * If an architecture guarantees that all ranges registered contain no holes
4782 * and may be freed, this this function may be used instead of calling
4783 * memblock_free_early_nid() manually.
4785 void __init
free_bootmem_with_active_regions(int nid
, unsigned long max_low_pfn
)
4787 unsigned long start_pfn
, end_pfn
;
4790 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
) {
4791 start_pfn
= min(start_pfn
, max_low_pfn
);
4792 end_pfn
= min(end_pfn
, max_low_pfn
);
4794 if (start_pfn
< end_pfn
)
4795 memblock_free_early_nid(PFN_PHYS(start_pfn
),
4796 (end_pfn
- start_pfn
) << PAGE_SHIFT
,
4802 * sparse_memory_present_with_active_regions - Call memory_present for each active range
4803 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
4805 * If an architecture guarantees that all ranges registered contain no holes and may
4806 * be freed, this function may be used instead of calling memory_present() manually.
4808 void __init
sparse_memory_present_with_active_regions(int nid
)
4810 unsigned long start_pfn
, end_pfn
;
4813 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, &this_nid
)
4814 memory_present(this_nid
, start_pfn
, end_pfn
);
4818 * get_pfn_range_for_nid - Return the start and end page frames for a node
4819 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
4820 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
4821 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
4823 * It returns the start and end page frame of a node based on information
4824 * provided by memblock_set_node(). If called for a node
4825 * with no available memory, a warning is printed and the start and end
4828 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
4829 unsigned long *start_pfn
, unsigned long *end_pfn
)
4831 unsigned long this_start_pfn
, this_end_pfn
;
4837 for_each_mem_pfn_range(i
, nid
, &this_start_pfn
, &this_end_pfn
, NULL
) {
4838 *start_pfn
= min(*start_pfn
, this_start_pfn
);
4839 *end_pfn
= max(*end_pfn
, this_end_pfn
);
4842 if (*start_pfn
== -1UL)
4847 * This finds a zone that can be used for ZONE_MOVABLE pages. The
4848 * assumption is made that zones within a node are ordered in monotonic
4849 * increasing memory addresses so that the "highest" populated zone is used
4851 static void __init
find_usable_zone_for_movable(void)
4854 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
4855 if (zone_index
== ZONE_MOVABLE
)
4858 if (arch_zone_highest_possible_pfn
[zone_index
] >
4859 arch_zone_lowest_possible_pfn
[zone_index
])
4863 VM_BUG_ON(zone_index
== -1);
4864 movable_zone
= zone_index
;
4868 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
4869 * because it is sized independent of architecture. Unlike the other zones,
4870 * the starting point for ZONE_MOVABLE is not fixed. It may be different
4871 * in each node depending on the size of each node and how evenly kernelcore
4872 * is distributed. This helper function adjusts the zone ranges
4873 * provided by the architecture for a given node by using the end of the
4874 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
4875 * zones within a node are in order of monotonic increases memory addresses
4877 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
4878 unsigned long zone_type
,
4879 unsigned long node_start_pfn
,
4880 unsigned long node_end_pfn
,
4881 unsigned long *zone_start_pfn
,
4882 unsigned long *zone_end_pfn
)
4884 /* Only adjust if ZONE_MOVABLE is on this node */
4885 if (zone_movable_pfn
[nid
]) {
4886 /* Size ZONE_MOVABLE */
4887 if (zone_type
== ZONE_MOVABLE
) {
4888 *zone_start_pfn
= zone_movable_pfn
[nid
];
4889 *zone_end_pfn
= min(node_end_pfn
,
4890 arch_zone_highest_possible_pfn
[movable_zone
]);
4892 /* Adjust for ZONE_MOVABLE starting within this range */
4893 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
4894 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
4895 *zone_end_pfn
= zone_movable_pfn
[nid
];
4897 /* Check if this whole range is within ZONE_MOVABLE */
4898 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
4899 *zone_start_pfn
= *zone_end_pfn
;
4904 * Return the number of pages a zone spans in a node, including holes
4905 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
4907 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
4908 unsigned long zone_type
,
4909 unsigned long node_start_pfn
,
4910 unsigned long node_end_pfn
,
4911 unsigned long *ignored
)
4913 unsigned long zone_start_pfn
, zone_end_pfn
;
4915 /* When hotadd a new node from cpu_up(), the node should be empty */
4916 if (!node_start_pfn
&& !node_end_pfn
)
4919 /* Get the start and end of the zone */
4920 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
4921 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
4922 adjust_zone_range_for_zone_movable(nid
, zone_type
,
4923 node_start_pfn
, node_end_pfn
,
4924 &zone_start_pfn
, &zone_end_pfn
);
4926 /* Check that this node has pages within the zone's required range */
4927 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
4930 /* Move the zone boundaries inside the node if necessary */
4931 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
4932 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
4934 /* Return the spanned pages */
4935 return zone_end_pfn
- zone_start_pfn
;
4939 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
4940 * then all holes in the requested range will be accounted for.
4942 unsigned long __meminit
__absent_pages_in_range(int nid
,
4943 unsigned long range_start_pfn
,
4944 unsigned long range_end_pfn
)
4946 unsigned long nr_absent
= range_end_pfn
- range_start_pfn
;
4947 unsigned long start_pfn
, end_pfn
;
4950 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
4951 start_pfn
= clamp(start_pfn
, range_start_pfn
, range_end_pfn
);
4952 end_pfn
= clamp(end_pfn
, range_start_pfn
, range_end_pfn
);
4953 nr_absent
-= end_pfn
- start_pfn
;
4959 * absent_pages_in_range - Return number of page frames in holes within a range
4960 * @start_pfn: The start PFN to start searching for holes
4961 * @end_pfn: The end PFN to stop searching for holes
4963 * It returns the number of pages frames in memory holes within a range.
4965 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
4966 unsigned long end_pfn
)
4968 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
4971 /* Return the number of page frames in holes in a zone on a node */
4972 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
4973 unsigned long zone_type
,
4974 unsigned long node_start_pfn
,
4975 unsigned long node_end_pfn
,
4976 unsigned long *ignored
)
4978 unsigned long zone_low
= arch_zone_lowest_possible_pfn
[zone_type
];
4979 unsigned long zone_high
= arch_zone_highest_possible_pfn
[zone_type
];
4980 unsigned long zone_start_pfn
, zone_end_pfn
;
4982 /* When hotadd a new node from cpu_up(), the node should be empty */
4983 if (!node_start_pfn
&& !node_end_pfn
)
4986 zone_start_pfn
= clamp(node_start_pfn
, zone_low
, zone_high
);
4987 zone_end_pfn
= clamp(node_end_pfn
, zone_low
, zone_high
);
4989 adjust_zone_range_for_zone_movable(nid
, zone_type
,
4990 node_start_pfn
, node_end_pfn
,
4991 &zone_start_pfn
, &zone_end_pfn
);
4992 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
4995 #else /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
4996 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
4997 unsigned long zone_type
,
4998 unsigned long node_start_pfn
,
4999 unsigned long node_end_pfn
,
5000 unsigned long *zones_size
)
5002 return zones_size
[zone_type
];
5005 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
5006 unsigned long zone_type
,
5007 unsigned long node_start_pfn
,
5008 unsigned long node_end_pfn
,
5009 unsigned long *zholes_size
)
5014 return zholes_size
[zone_type
];
5017 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5019 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
5020 unsigned long node_start_pfn
,
5021 unsigned long node_end_pfn
,
5022 unsigned long *zones_size
,
5023 unsigned long *zholes_size
)
5025 unsigned long realtotalpages
= 0, totalpages
= 0;
5028 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5029 struct zone
*zone
= pgdat
->node_zones
+ i
;
5030 unsigned long size
, real_size
;
5032 size
= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
5036 real_size
= size
- zone_absent_pages_in_node(pgdat
->node_id
, i
,
5037 node_start_pfn
, node_end_pfn
,
5039 zone
->spanned_pages
= size
;
5040 zone
->present_pages
= real_size
;
5043 realtotalpages
+= real_size
;
5046 pgdat
->node_spanned_pages
= totalpages
;
5047 pgdat
->node_present_pages
= realtotalpages
;
5048 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
5052 #ifndef CONFIG_SPARSEMEM
5054 * Calculate the size of the zone->blockflags rounded to an unsigned long
5055 * Start by making sure zonesize is a multiple of pageblock_order by rounding
5056 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
5057 * round what is now in bits to nearest long in bits, then return it in
5060 static unsigned long __init
usemap_size(unsigned long zone_start_pfn
, unsigned long zonesize
)
5062 unsigned long usemapsize
;
5064 zonesize
+= zone_start_pfn
& (pageblock_nr_pages
-1);
5065 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
5066 usemapsize
= usemapsize
>> pageblock_order
;
5067 usemapsize
*= NR_PAGEBLOCK_BITS
;
5068 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
5070 return usemapsize
/ 8;
5073 static void __init
setup_usemap(struct pglist_data
*pgdat
,
5075 unsigned long zone_start_pfn
,
5076 unsigned long zonesize
)
5078 unsigned long usemapsize
= usemap_size(zone_start_pfn
, zonesize
);
5079 zone
->pageblock_flags
= NULL
;
5081 zone
->pageblock_flags
=
5082 memblock_virt_alloc_node_nopanic(usemapsize
,
5086 static inline void setup_usemap(struct pglist_data
*pgdat
, struct zone
*zone
,
5087 unsigned long zone_start_pfn
, unsigned long zonesize
) {}
5088 #endif /* CONFIG_SPARSEMEM */
5090 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
5092 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
5093 void __paginginit
set_pageblock_order(void)
5097 /* Check that pageblock_nr_pages has not already been setup */
5098 if (pageblock_order
)
5101 if (HPAGE_SHIFT
> PAGE_SHIFT
)
5102 order
= HUGETLB_PAGE_ORDER
;
5104 order
= MAX_ORDER
- 1;
5107 * Assume the largest contiguous order of interest is a huge page.
5108 * This value may be variable depending on boot parameters on IA64 and
5111 pageblock_order
= order
;
5113 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5116 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
5117 * is unused as pageblock_order is set at compile-time. See
5118 * include/linux/pageblock-flags.h for the values of pageblock_order based on
5121 void __paginginit
set_pageblock_order(void)
5125 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
5127 static unsigned long __paginginit
calc_memmap_size(unsigned long spanned_pages
,
5128 unsigned long present_pages
)
5130 unsigned long pages
= spanned_pages
;
5133 * Provide a more accurate estimation if there are holes within
5134 * the zone and SPARSEMEM is in use. If there are holes within the
5135 * zone, each populated memory region may cost us one or two extra
5136 * memmap pages due to alignment because memmap pages for each
5137 * populated regions may not naturally algined on page boundary.
5138 * So the (present_pages >> 4) heuristic is a tradeoff for that.
5140 if (spanned_pages
> present_pages
+ (present_pages
>> 4) &&
5141 IS_ENABLED(CONFIG_SPARSEMEM
))
5142 pages
= present_pages
;
5144 return PAGE_ALIGN(pages
* sizeof(struct page
)) >> PAGE_SHIFT
;
5148 * Set up the zone data structures:
5149 * - mark all pages reserved
5150 * - mark all memory queues empty
5151 * - clear the memory bitmaps
5153 * NOTE: pgdat should get zeroed by caller.
5155 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
)
5158 int nid
= pgdat
->node_id
;
5159 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
5162 pgdat_resize_init(pgdat
);
5163 #ifdef CONFIG_NUMA_BALANCING
5164 spin_lock_init(&pgdat
->numabalancing_migrate_lock
);
5165 pgdat
->numabalancing_migrate_nr_pages
= 0;
5166 pgdat
->numabalancing_migrate_next_window
= jiffies
;
5168 init_waitqueue_head(&pgdat
->kswapd_wait
);
5169 init_waitqueue_head(&pgdat
->pfmemalloc_wait
);
5170 pgdat_page_ext_init(pgdat
);
5172 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5173 struct zone
*zone
= pgdat
->node_zones
+ j
;
5174 unsigned long size
, realsize
, freesize
, memmap_pages
;
5176 size
= zone
->spanned_pages
;
5177 realsize
= freesize
= zone
->present_pages
;
5180 * Adjust freesize so that it accounts for how much memory
5181 * is used by this zone for memmap. This affects the watermark
5182 * and per-cpu initialisations
5184 memmap_pages
= calc_memmap_size(size
, realsize
);
5185 if (!is_highmem_idx(j
)) {
5186 if (freesize
>= memmap_pages
) {
5187 freesize
-= memmap_pages
;
5190 " %s zone: %lu pages used for memmap\n",
5191 zone_names
[j
], memmap_pages
);
5194 " %s zone: %lu pages exceeds freesize %lu\n",
5195 zone_names
[j
], memmap_pages
, freesize
);
5198 /* Account for reserved pages */
5199 if (j
== 0 && freesize
> dma_reserve
) {
5200 freesize
-= dma_reserve
;
5201 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
5202 zone_names
[0], dma_reserve
);
5205 if (!is_highmem_idx(j
))
5206 nr_kernel_pages
+= freesize
;
5207 /* Charge for highmem memmap if there are enough kernel pages */
5208 else if (nr_kernel_pages
> memmap_pages
* 2)
5209 nr_kernel_pages
-= memmap_pages
;
5210 nr_all_pages
+= freesize
;
5213 * Set an approximate value for lowmem here, it will be adjusted
5214 * when the bootmem allocator frees pages into the buddy system.
5215 * And all highmem pages will be managed by the buddy system.
5217 zone
->managed_pages
= is_highmem_idx(j
) ? realsize
: freesize
;
5220 zone
->min_unmapped_pages
= (freesize
*sysctl_min_unmapped_ratio
)
5222 zone
->min_slab_pages
= (freesize
* sysctl_min_slab_ratio
) / 100;
5224 zone
->name
= zone_names
[j
];
5225 spin_lock_init(&zone
->lock
);
5226 spin_lock_init(&zone
->lru_lock
);
5227 zone_seqlock_init(zone
);
5228 zone
->zone_pgdat
= pgdat
;
5229 zone_pcp_init(zone
);
5231 /* For bootup, initialized properly in watermark setup */
5232 mod_zone_page_state(zone
, NR_ALLOC_BATCH
, zone
->managed_pages
);
5234 lruvec_init(&zone
->lruvec
);
5238 set_pageblock_order();
5239 setup_usemap(pgdat
, zone
, zone_start_pfn
, size
);
5240 ret
= init_currently_empty_zone(zone
, zone_start_pfn
, size
);
5242 memmap_init(size
, nid
, j
, zone_start_pfn
);
5243 zone_start_pfn
+= size
;
5247 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
5249 unsigned long __maybe_unused start
= 0;
5250 unsigned long __maybe_unused offset
= 0;
5252 /* Skip empty nodes */
5253 if (!pgdat
->node_spanned_pages
)
5256 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5257 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
5258 offset
= pgdat
->node_start_pfn
- start
;
5259 /* ia64 gets its own node_mem_map, before this, without bootmem */
5260 if (!pgdat
->node_mem_map
) {
5261 unsigned long size
, end
;
5265 * The zone's endpoints aren't required to be MAX_ORDER
5266 * aligned but the node_mem_map endpoints must be in order
5267 * for the buddy allocator to function correctly.
5269 end
= pgdat_end_pfn(pgdat
);
5270 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
5271 size
= (end
- start
) * sizeof(struct page
);
5272 map
= alloc_remap(pgdat
->node_id
, size
);
5274 map
= memblock_virt_alloc_node_nopanic(size
,
5276 pgdat
->node_mem_map
= map
+ offset
;
5278 #ifndef CONFIG_NEED_MULTIPLE_NODES
5280 * With no DISCONTIG, the global mem_map is just set as node 0's
5282 if (pgdat
== NODE_DATA(0)) {
5283 mem_map
= NODE_DATA(0)->node_mem_map
;
5284 #if defined(CONFIG_HAVE_MEMBLOCK_NODE_MAP) || defined(CONFIG_FLATMEM)
5285 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
5287 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5290 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
5293 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
5294 unsigned long node_start_pfn
, unsigned long *zholes_size
)
5296 pg_data_t
*pgdat
= NODE_DATA(nid
);
5297 unsigned long start_pfn
= 0;
5298 unsigned long end_pfn
= 0;
5300 /* pg_data_t should be reset to zero when it's allocated */
5301 WARN_ON(pgdat
->nr_zones
|| pgdat
->classzone_idx
);
5303 reset_deferred_meminit(pgdat
);
5304 pgdat
->node_id
= nid
;
5305 pgdat
->node_start_pfn
= node_start_pfn
;
5306 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5307 get_pfn_range_for_nid(nid
, &start_pfn
, &end_pfn
);
5308 pr_info("Initmem setup node %d [mem %#018Lx-%#018Lx]\n", nid
,
5309 (u64
)start_pfn
<< PAGE_SHIFT
,
5310 end_pfn
? ((u64
)end_pfn
<< PAGE_SHIFT
) - 1 : 0);
5312 calculate_node_totalpages(pgdat
, start_pfn
, end_pfn
,
5313 zones_size
, zholes_size
);
5315 alloc_node_mem_map(pgdat
);
5316 #ifdef CONFIG_FLAT_NODE_MEM_MAP
5317 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
5318 nid
, (unsigned long)pgdat
,
5319 (unsigned long)pgdat
->node_mem_map
);
5322 free_area_init_core(pgdat
);
5325 #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
5327 #if MAX_NUMNODES > 1
5329 * Figure out the number of possible node ids.
5331 void __init
setup_nr_node_ids(void)
5333 unsigned int highest
;
5335 highest
= find_last_bit(node_possible_map
.bits
, MAX_NUMNODES
);
5336 nr_node_ids
= highest
+ 1;
5341 * node_map_pfn_alignment - determine the maximum internode alignment
5343 * This function should be called after node map is populated and sorted.
5344 * It calculates the maximum power of two alignment which can distinguish
5347 * For example, if all nodes are 1GiB and aligned to 1GiB, the return value
5348 * would indicate 1GiB alignment with (1 << (30 - PAGE_SHIFT)). If the
5349 * nodes are shifted by 256MiB, 256MiB. Note that if only the last node is
5350 * shifted, 1GiB is enough and this function will indicate so.
5352 * This is used to test whether pfn -> nid mapping of the chosen memory
5353 * model has fine enough granularity to avoid incorrect mapping for the
5354 * populated node map.
5356 * Returns the determined alignment in pfn's. 0 if there is no alignment
5357 * requirement (single node).
5359 unsigned long __init
node_map_pfn_alignment(void)
5361 unsigned long accl_mask
= 0, last_end
= 0;
5362 unsigned long start
, end
, mask
;
5366 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start
, &end
, &nid
) {
5367 if (!start
|| last_nid
< 0 || last_nid
== nid
) {
5374 * Start with a mask granular enough to pin-point to the
5375 * start pfn and tick off bits one-by-one until it becomes
5376 * too coarse to separate the current node from the last.
5378 mask
= ~((1 << __ffs(start
)) - 1);
5379 while (mask
&& last_end
<= (start
& (mask
<< 1)))
5382 /* accumulate all internode masks */
5386 /* convert mask to number of pages */
5387 return ~accl_mask
+ 1;
5390 /* Find the lowest pfn for a node */
5391 static unsigned long __init
find_min_pfn_for_node(int nid
)
5393 unsigned long min_pfn
= ULONG_MAX
;
5394 unsigned long start_pfn
;
5397 for_each_mem_pfn_range(i
, nid
, &start_pfn
, NULL
, NULL
)
5398 min_pfn
= min(min_pfn
, start_pfn
);
5400 if (min_pfn
== ULONG_MAX
) {
5402 "Could not find start_pfn for node %d\n", nid
);
5410 * find_min_pfn_with_active_regions - Find the minimum PFN registered
5412 * It returns the minimum PFN based on information provided via
5413 * memblock_set_node().
5415 unsigned long __init
find_min_pfn_with_active_regions(void)
5417 return find_min_pfn_for_node(MAX_NUMNODES
);
5421 * early_calculate_totalpages()
5422 * Sum pages in active regions for movable zone.
5423 * Populate N_MEMORY for calculating usable_nodes.
5425 static unsigned long __init
early_calculate_totalpages(void)
5427 unsigned long totalpages
= 0;
5428 unsigned long start_pfn
, end_pfn
;
5431 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
) {
5432 unsigned long pages
= end_pfn
- start_pfn
;
5434 totalpages
+= pages
;
5436 node_set_state(nid
, N_MEMORY
);
5442 * Find the PFN the Movable zone begins in each node. Kernel memory
5443 * is spread evenly between nodes as long as the nodes have enough
5444 * memory. When they don't, some nodes will have more kernelcore than
5447 static void __init
find_zone_movable_pfns_for_nodes(void)
5450 unsigned long usable_startpfn
;
5451 unsigned long kernelcore_node
, kernelcore_remaining
;
5452 /* save the state before borrow the nodemask */
5453 nodemask_t saved_node_state
= node_states
[N_MEMORY
];
5454 unsigned long totalpages
= early_calculate_totalpages();
5455 int usable_nodes
= nodes_weight(node_states
[N_MEMORY
]);
5456 struct memblock_region
*r
;
5458 /* Need to find movable_zone earlier when movable_node is specified. */
5459 find_usable_zone_for_movable();
5462 * If movable_node is specified, ignore kernelcore and movablecore
5465 if (movable_node_is_enabled()) {
5466 for_each_memblock(memory
, r
) {
5467 if (!memblock_is_hotpluggable(r
))
5472 usable_startpfn
= PFN_DOWN(r
->base
);
5473 zone_movable_pfn
[nid
] = zone_movable_pfn
[nid
] ?
5474 min(usable_startpfn
, zone_movable_pfn
[nid
]) :
5482 * If movablecore=nn[KMG] was specified, calculate what size of
5483 * kernelcore that corresponds so that memory usable for
5484 * any allocation type is evenly spread. If both kernelcore
5485 * and movablecore are specified, then the value of kernelcore
5486 * will be used for required_kernelcore if it's greater than
5487 * what movablecore would have allowed.
5489 if (required_movablecore
) {
5490 unsigned long corepages
;
5493 * Round-up so that ZONE_MOVABLE is at least as large as what
5494 * was requested by the user
5496 required_movablecore
=
5497 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
5498 required_movablecore
= min(totalpages
, required_movablecore
);
5499 corepages
= totalpages
- required_movablecore
;
5501 required_kernelcore
= max(required_kernelcore
, corepages
);
5505 * If kernelcore was not specified or kernelcore size is larger
5506 * than totalpages, there is no ZONE_MOVABLE.
5508 if (!required_kernelcore
|| required_kernelcore
>= totalpages
)
5511 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
5512 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
5515 /* Spread kernelcore memory as evenly as possible throughout nodes */
5516 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5517 for_each_node_state(nid
, N_MEMORY
) {
5518 unsigned long start_pfn
, end_pfn
;
5521 * Recalculate kernelcore_node if the division per node
5522 * now exceeds what is necessary to satisfy the requested
5523 * amount of memory for the kernel
5525 if (required_kernelcore
< kernelcore_node
)
5526 kernelcore_node
= required_kernelcore
/ usable_nodes
;
5529 * As the map is walked, we track how much memory is usable
5530 * by the kernel using kernelcore_remaining. When it is
5531 * 0, the rest of the node is usable by ZONE_MOVABLE
5533 kernelcore_remaining
= kernelcore_node
;
5535 /* Go through each range of PFNs within this node */
5536 for_each_mem_pfn_range(i
, nid
, &start_pfn
, &end_pfn
, NULL
) {
5537 unsigned long size_pages
;
5539 start_pfn
= max(start_pfn
, zone_movable_pfn
[nid
]);
5540 if (start_pfn
>= end_pfn
)
5543 /* Account for what is only usable for kernelcore */
5544 if (start_pfn
< usable_startpfn
) {
5545 unsigned long kernel_pages
;
5546 kernel_pages
= min(end_pfn
, usable_startpfn
)
5549 kernelcore_remaining
-= min(kernel_pages
,
5550 kernelcore_remaining
);
5551 required_kernelcore
-= min(kernel_pages
,
5552 required_kernelcore
);
5554 /* Continue if range is now fully accounted */
5555 if (end_pfn
<= usable_startpfn
) {
5558 * Push zone_movable_pfn to the end so
5559 * that if we have to rebalance
5560 * kernelcore across nodes, we will
5561 * not double account here
5563 zone_movable_pfn
[nid
] = end_pfn
;
5566 start_pfn
= usable_startpfn
;
5570 * The usable PFN range for ZONE_MOVABLE is from
5571 * start_pfn->end_pfn. Calculate size_pages as the
5572 * number of pages used as kernelcore
5574 size_pages
= end_pfn
- start_pfn
;
5575 if (size_pages
> kernelcore_remaining
)
5576 size_pages
= kernelcore_remaining
;
5577 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
5580 * Some kernelcore has been met, update counts and
5581 * break if the kernelcore for this node has been
5584 required_kernelcore
-= min(required_kernelcore
,
5586 kernelcore_remaining
-= size_pages
;
5587 if (!kernelcore_remaining
)
5593 * If there is still required_kernelcore, we do another pass with one
5594 * less node in the count. This will push zone_movable_pfn[nid] further
5595 * along on the nodes that still have memory until kernelcore is
5599 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
5603 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
5604 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
5605 zone_movable_pfn
[nid
] =
5606 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
5609 /* restore the node_state */
5610 node_states
[N_MEMORY
] = saved_node_state
;
5613 /* Any regular or high memory on that node ? */
5614 static void check_for_memory(pg_data_t
*pgdat
, int nid
)
5616 enum zone_type zone_type
;
5618 if (N_MEMORY
== N_NORMAL_MEMORY
)
5621 for (zone_type
= 0; zone_type
<= ZONE_MOVABLE
- 1; zone_type
++) {
5622 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
5623 if (populated_zone(zone
)) {
5624 node_set_state(nid
, N_HIGH_MEMORY
);
5625 if (N_NORMAL_MEMORY
!= N_HIGH_MEMORY
&&
5626 zone_type
<= ZONE_NORMAL
)
5627 node_set_state(nid
, N_NORMAL_MEMORY
);
5634 * free_area_init_nodes - Initialise all pg_data_t and zone data
5635 * @max_zone_pfn: an array of max PFNs for each zone
5637 * This will call free_area_init_node() for each active node in the system.
5638 * Using the page ranges provided by memblock_set_node(), the size of each
5639 * zone in each node and their holes is calculated. If the maximum PFN
5640 * between two adjacent zones match, it is assumed that the zone is empty.
5641 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
5642 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
5643 * starts where the previous one ended. For example, ZONE_DMA32 starts
5644 * at arch_max_dma_pfn.
5646 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
5648 unsigned long start_pfn
, end_pfn
;
5651 /* Record where the zone boundaries are */
5652 memset(arch_zone_lowest_possible_pfn
, 0,
5653 sizeof(arch_zone_lowest_possible_pfn
));
5654 memset(arch_zone_highest_possible_pfn
, 0,
5655 sizeof(arch_zone_highest_possible_pfn
));
5656 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
5657 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
5658 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
5659 if (i
== ZONE_MOVABLE
)
5661 arch_zone_lowest_possible_pfn
[i
] =
5662 arch_zone_highest_possible_pfn
[i
-1];
5663 arch_zone_highest_possible_pfn
[i
] =
5664 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
5666 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
5667 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
5669 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
5670 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
5671 find_zone_movable_pfns_for_nodes();
5673 /* Print out the zone ranges */
5674 pr_info("Zone ranges:\n");
5675 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5676 if (i
== ZONE_MOVABLE
)
5678 pr_info(" %-8s ", zone_names
[i
]);
5679 if (arch_zone_lowest_possible_pfn
[i
] ==
5680 arch_zone_highest_possible_pfn
[i
])
5683 pr_cont("[mem %#018Lx-%#018Lx]\n",
5684 (u64
)arch_zone_lowest_possible_pfn
[i
]
5686 ((u64
)arch_zone_highest_possible_pfn
[i
]
5687 << PAGE_SHIFT
) - 1);
5690 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
5691 pr_info("Movable zone start for each node\n");
5692 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
5693 if (zone_movable_pfn
[i
])
5694 pr_info(" Node %d: %#018Lx\n", i
,
5695 (u64
)zone_movable_pfn
[i
] << PAGE_SHIFT
);
5698 /* Print out the early node map */
5699 pr_info("Early memory node ranges\n");
5700 for_each_mem_pfn_range(i
, MAX_NUMNODES
, &start_pfn
, &end_pfn
, &nid
)
5701 pr_info(" node %3d: [mem %#018Lx-%#018Lx]\n", nid
,
5702 (u64
)start_pfn
<< PAGE_SHIFT
,
5703 ((u64
)end_pfn
<< PAGE_SHIFT
) - 1);
5705 /* Initialise every node */
5706 mminit_verify_pageflags_layout();
5707 setup_nr_node_ids();
5708 for_each_online_node(nid
) {
5709 pg_data_t
*pgdat
= NODE_DATA(nid
);
5710 free_area_init_node(nid
, NULL
,
5711 find_min_pfn_for_node(nid
), NULL
);
5713 /* Any memory on that node */
5714 if (pgdat
->node_present_pages
)
5715 node_set_state(nid
, N_MEMORY
);
5716 check_for_memory(pgdat
, nid
);
5720 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
5722 unsigned long long coremem
;
5726 coremem
= memparse(p
, &p
);
5727 *core
= coremem
>> PAGE_SHIFT
;
5729 /* Paranoid check that UL is enough for the coremem value */
5730 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
5736 * kernelcore=size sets the amount of memory for use for allocations that
5737 * cannot be reclaimed or migrated.
5739 static int __init
cmdline_parse_kernelcore(char *p
)
5741 return cmdline_parse_core(p
, &required_kernelcore
);
5745 * movablecore=size sets the amount of memory for use for allocations that
5746 * can be reclaimed or migrated.
5748 static int __init
cmdline_parse_movablecore(char *p
)
5750 return cmdline_parse_core(p
, &required_movablecore
);
5753 early_param("kernelcore", cmdline_parse_kernelcore
);
5754 early_param("movablecore", cmdline_parse_movablecore
);
5756 #endif /* CONFIG_HAVE_MEMBLOCK_NODE_MAP */
5758 void adjust_managed_page_count(struct page
*page
, long count
)
5760 spin_lock(&managed_page_count_lock
);
5761 page_zone(page
)->managed_pages
+= count
;
5762 totalram_pages
+= count
;
5763 #ifdef CONFIG_HIGHMEM
5764 if (PageHighMem(page
))
5765 totalhigh_pages
+= count
;
5767 spin_unlock(&managed_page_count_lock
);
5769 EXPORT_SYMBOL(adjust_managed_page_count
);
5771 unsigned long free_reserved_area(void *start
, void *end
, int poison
, char *s
)
5774 unsigned long pages
= 0;
5776 start
= (void *)PAGE_ALIGN((unsigned long)start
);
5777 end
= (void *)((unsigned long)end
& PAGE_MASK
);
5778 for (pos
= start
; pos
< end
; pos
+= PAGE_SIZE
, pages
++) {
5779 if ((unsigned int)poison
<= 0xFF)
5780 memset(pos
, poison
, PAGE_SIZE
);
5781 free_reserved_page(virt_to_page(pos
));
5785 pr_info("Freeing %s memory: %ldK (%p - %p)\n",
5786 s
, pages
<< (PAGE_SHIFT
- 10), start
, end
);
5790 EXPORT_SYMBOL(free_reserved_area
);
5792 #ifdef CONFIG_HIGHMEM
5793 void free_highmem_page(struct page
*page
)
5795 __free_reserved_page(page
);
5797 page_zone(page
)->managed_pages
++;
5803 void __init
mem_init_print_info(const char *str
)
5805 unsigned long physpages
, codesize
, datasize
, rosize
, bss_size
;
5806 unsigned long init_code_size
, init_data_size
;
5808 physpages
= get_num_physpages();
5809 codesize
= _etext
- _stext
;
5810 datasize
= _edata
- _sdata
;
5811 rosize
= __end_rodata
- __start_rodata
;
5812 bss_size
= __bss_stop
- __bss_start
;
5813 init_data_size
= __init_end
- __init_begin
;
5814 init_code_size
= _einittext
- _sinittext
;
5817 * Detect special cases and adjust section sizes accordingly:
5818 * 1) .init.* may be embedded into .data sections
5819 * 2) .init.text.* may be out of [__init_begin, __init_end],
5820 * please refer to arch/tile/kernel/vmlinux.lds.S.
5821 * 3) .rodata.* may be embedded into .text or .data sections.
5823 #define adj_init_size(start, end, size, pos, adj) \
5825 if (start <= pos && pos < end && size > adj) \
5829 adj_init_size(__init_begin
, __init_end
, init_data_size
,
5830 _sinittext
, init_code_size
);
5831 adj_init_size(_stext
, _etext
, codesize
, _sinittext
, init_code_size
);
5832 adj_init_size(_sdata
, _edata
, datasize
, __init_begin
, init_data_size
);
5833 adj_init_size(_stext
, _etext
, codesize
, __start_rodata
, rosize
);
5834 adj_init_size(_sdata
, _edata
, datasize
, __start_rodata
, rosize
);
5836 #undef adj_init_size
5838 pr_info("Memory: %luK/%luK available "
5839 "(%luK kernel code, %luK rwdata, %luK rodata, "
5840 "%luK init, %luK bss, %luK reserved, %luK cma-reserved"
5841 #ifdef CONFIG_HIGHMEM
5845 nr_free_pages() << (PAGE_SHIFT
-10), physpages
<< (PAGE_SHIFT
-10),
5846 codesize
>> 10, datasize
>> 10, rosize
>> 10,
5847 (init_data_size
+ init_code_size
) >> 10, bss_size
>> 10,
5848 (physpages
- totalram_pages
- totalcma_pages
) << (PAGE_SHIFT
-10),
5849 totalcma_pages
<< (PAGE_SHIFT
-10),
5850 #ifdef CONFIG_HIGHMEM
5851 totalhigh_pages
<< (PAGE_SHIFT
-10),
5853 str
? ", " : "", str
? str
: "");
5857 * set_dma_reserve - set the specified number of pages reserved in the first zone
5858 * @new_dma_reserve: The number of pages to mark reserved
5860 * The per-cpu batchsize and zone watermarks are determined by managed_pages.
5861 * In the DMA zone, a significant percentage may be consumed by kernel image
5862 * and other unfreeable allocations which can skew the watermarks badly. This
5863 * function may optionally be used to account for unfreeable pages in the
5864 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
5865 * smaller per-cpu batchsize.
5867 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
5869 dma_reserve
= new_dma_reserve
;
5872 void __init
free_area_init(unsigned long *zones_size
)
5874 free_area_init_node(0, zones_size
,
5875 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
5878 static int page_alloc_cpu_notify(struct notifier_block
*self
,
5879 unsigned long action
, void *hcpu
)
5881 int cpu
= (unsigned long)hcpu
;
5883 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
5884 lru_add_drain_cpu(cpu
);
5888 * Spill the event counters of the dead processor
5889 * into the current processors event counters.
5890 * This artificially elevates the count of the current
5893 vm_events_fold_cpu(cpu
);
5896 * Zero the differential counters of the dead processor
5897 * so that the vm statistics are consistent.
5899 * This is only okay since the processor is dead and cannot
5900 * race with what we are doing.
5902 cpu_vm_stats_fold(cpu
);
5907 void __init
page_alloc_init(void)
5909 hotcpu_notifier(page_alloc_cpu_notify
, 0);
5913 * calculate_totalreserve_pages - called when sysctl_lowmem_reserve_ratio
5914 * or min_free_kbytes changes.
5916 static void calculate_totalreserve_pages(void)
5918 struct pglist_data
*pgdat
;
5919 unsigned long reserve_pages
= 0;
5920 enum zone_type i
, j
;
5922 for_each_online_pgdat(pgdat
) {
5923 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
5924 struct zone
*zone
= pgdat
->node_zones
+ i
;
5927 /* Find valid and maximum lowmem_reserve in the zone */
5928 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
5929 if (zone
->lowmem_reserve
[j
] > max
)
5930 max
= zone
->lowmem_reserve
[j
];
5933 /* we treat the high watermark as reserved pages. */
5934 max
+= high_wmark_pages(zone
);
5936 if (max
> zone
->managed_pages
)
5937 max
= zone
->managed_pages
;
5939 zone
->totalreserve_pages
= max
;
5941 reserve_pages
+= max
;
5944 totalreserve_pages
= reserve_pages
;
5948 * setup_per_zone_lowmem_reserve - called whenever
5949 * sysctl_lowmem_reserve_ratio changes. Ensures that each zone
5950 * has a correct pages reserved value, so an adequate number of
5951 * pages are left in the zone after a successful __alloc_pages().
5953 static void setup_per_zone_lowmem_reserve(void)
5955 struct pglist_data
*pgdat
;
5956 enum zone_type j
, idx
;
5958 for_each_online_pgdat(pgdat
) {
5959 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
5960 struct zone
*zone
= pgdat
->node_zones
+ j
;
5961 unsigned long managed_pages
= zone
->managed_pages
;
5963 zone
->lowmem_reserve
[j
] = 0;
5967 struct zone
*lower_zone
;
5971 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
5972 sysctl_lowmem_reserve_ratio
[idx
] = 1;
5974 lower_zone
= pgdat
->node_zones
+ idx
;
5975 lower_zone
->lowmem_reserve
[j
] = managed_pages
/
5976 sysctl_lowmem_reserve_ratio
[idx
];
5977 managed_pages
+= lower_zone
->managed_pages
;
5982 /* update totalreserve_pages */
5983 calculate_totalreserve_pages();
5986 static void __setup_per_zone_wmarks(void)
5988 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
5989 unsigned long lowmem_pages
= 0;
5991 unsigned long flags
;
5993 /* Calculate total number of !ZONE_HIGHMEM pages */
5994 for_each_zone(zone
) {
5995 if (!is_highmem(zone
))
5996 lowmem_pages
+= zone
->managed_pages
;
5999 for_each_zone(zone
) {
6002 spin_lock_irqsave(&zone
->lock
, flags
);
6003 tmp
= (u64
)pages_min
* zone
->managed_pages
;
6004 do_div(tmp
, lowmem_pages
);
6005 if (is_highmem(zone
)) {
6007 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
6008 * need highmem pages, so cap pages_min to a small
6011 * The WMARK_HIGH-WMARK_LOW and (WMARK_LOW-WMARK_MIN)
6012 * deltas control asynch page reclaim, and so should
6013 * not be capped for highmem.
6015 unsigned long min_pages
;
6017 min_pages
= zone
->managed_pages
/ 1024;
6018 min_pages
= clamp(min_pages
, SWAP_CLUSTER_MAX
, 128UL);
6019 zone
->watermark
[WMARK_MIN
] = min_pages
;
6022 * If it's a lowmem zone, reserve a number of pages
6023 * proportionate to the zone's size.
6025 zone
->watermark
[WMARK_MIN
] = tmp
;
6028 zone
->watermark
[WMARK_LOW
] = min_wmark_pages(zone
) + (tmp
>> 2);
6029 zone
->watermark
[WMARK_HIGH
] = min_wmark_pages(zone
) + (tmp
>> 1);
6031 __mod_zone_page_state(zone
, NR_ALLOC_BATCH
,
6032 high_wmark_pages(zone
) - low_wmark_pages(zone
) -
6033 atomic_long_read(&zone
->vm_stat
[NR_ALLOC_BATCH
]));
6035 spin_unlock_irqrestore(&zone
->lock
, flags
);
6038 /* update totalreserve_pages */
6039 calculate_totalreserve_pages();
6043 * setup_per_zone_wmarks - called when min_free_kbytes changes
6044 * or when memory is hot-{added|removed}
6046 * Ensures that the watermark[min,low,high] values for each zone are set
6047 * correctly with respect to min_free_kbytes.
6049 void setup_per_zone_wmarks(void)
6051 mutex_lock(&zonelists_mutex
);
6052 __setup_per_zone_wmarks();
6053 mutex_unlock(&zonelists_mutex
);
6057 * The inactive anon list should be small enough that the VM never has to
6058 * do too much work, but large enough that each inactive page has a chance
6059 * to be referenced again before it is swapped out.
6061 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
6062 * INACTIVE_ANON pages on this zone's LRU, maintained by the
6063 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
6064 * the anonymous pages are kept on the inactive list.
6067 * memory ratio inactive anon
6068 * -------------------------------------
6077 static void __meminit
calculate_zone_inactive_ratio(struct zone
*zone
)
6079 unsigned int gb
, ratio
;
6081 /* Zone size in gigabytes */
6082 gb
= zone
->managed_pages
>> (30 - PAGE_SHIFT
);
6084 ratio
= int_sqrt(10 * gb
);
6088 zone
->inactive_ratio
= ratio
;
6091 static void __meminit
setup_per_zone_inactive_ratio(void)
6096 calculate_zone_inactive_ratio(zone
);
6100 * Initialise min_free_kbytes.
6102 * For small machines we want it small (128k min). For large machines
6103 * we want it large (64MB max). But it is not linear, because network
6104 * bandwidth does not increase linearly with machine size. We use
6106 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
6107 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
6123 int __meminit
init_per_zone_wmark_min(void)
6125 unsigned long lowmem_kbytes
;
6126 int new_min_free_kbytes
;
6128 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
6129 new_min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
6131 if (new_min_free_kbytes
> user_min_free_kbytes
) {
6132 min_free_kbytes
= new_min_free_kbytes
;
6133 if (min_free_kbytes
< 128)
6134 min_free_kbytes
= 128;
6135 if (min_free_kbytes
> 65536)
6136 min_free_kbytes
= 65536;
6138 pr_warn("min_free_kbytes is not updated to %d because user defined value %d is preferred\n",
6139 new_min_free_kbytes
, user_min_free_kbytes
);
6141 setup_per_zone_wmarks();
6142 refresh_zone_stat_thresholds();
6143 setup_per_zone_lowmem_reserve();
6144 setup_per_zone_inactive_ratio();
6147 module_init(init_per_zone_wmark_min
)
6150 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
6151 * that we can call two helper functions whenever min_free_kbytes
6154 int min_free_kbytes_sysctl_handler(struct ctl_table
*table
, int write
,
6155 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6159 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6164 user_min_free_kbytes
= min_free_kbytes
;
6165 setup_per_zone_wmarks();
6171 int sysctl_min_unmapped_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6172 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6177 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6182 zone
->min_unmapped_pages
= (zone
->managed_pages
*
6183 sysctl_min_unmapped_ratio
) / 100;
6187 int sysctl_min_slab_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6188 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6193 rc
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6198 zone
->min_slab_pages
= (zone
->managed_pages
*
6199 sysctl_min_slab_ratio
) / 100;
6205 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
6206 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
6207 * whenever sysctl_lowmem_reserve_ratio changes.
6209 * The reserve ratio obviously has absolutely no relation with the
6210 * minimum watermarks. The lowmem reserve ratio can only make sense
6211 * if in function of the boot time zone sizes.
6213 int lowmem_reserve_ratio_sysctl_handler(struct ctl_table
*table
, int write
,
6214 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6216 proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6217 setup_per_zone_lowmem_reserve();
6222 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
6223 * cpu. It is the fraction of total pages in each zone that a hot per cpu
6224 * pagelist can have before it gets flushed back to buddy allocator.
6226 int percpu_pagelist_fraction_sysctl_handler(struct ctl_table
*table
, int write
,
6227 void __user
*buffer
, size_t *length
, loff_t
*ppos
)
6230 int old_percpu_pagelist_fraction
;
6233 mutex_lock(&pcp_batch_high_lock
);
6234 old_percpu_pagelist_fraction
= percpu_pagelist_fraction
;
6236 ret
= proc_dointvec_minmax(table
, write
, buffer
, length
, ppos
);
6237 if (!write
|| ret
< 0)
6240 /* Sanity checking to avoid pcp imbalance */
6241 if (percpu_pagelist_fraction
&&
6242 percpu_pagelist_fraction
< MIN_PERCPU_PAGELIST_FRACTION
) {
6243 percpu_pagelist_fraction
= old_percpu_pagelist_fraction
;
6249 if (percpu_pagelist_fraction
== old_percpu_pagelist_fraction
)
6252 for_each_populated_zone(zone
) {
6255 for_each_possible_cpu(cpu
)
6256 pageset_set_high_and_batch(zone
,
6257 per_cpu_ptr(zone
->pageset
, cpu
));
6260 mutex_unlock(&pcp_batch_high_lock
);
6265 int hashdist
= HASHDIST_DEFAULT
;
6267 static int __init
set_hashdist(char *str
)
6271 hashdist
= simple_strtoul(str
, &str
, 0);
6274 __setup("hashdist=", set_hashdist
);
6278 * allocate a large system hash table from bootmem
6279 * - it is assumed that the hash table must contain an exact power-of-2
6280 * quantity of entries
6281 * - limit is the number of hash buckets, not the total allocation size
6283 void *__init
alloc_large_system_hash(const char *tablename
,
6284 unsigned long bucketsize
,
6285 unsigned long numentries
,
6288 unsigned int *_hash_shift
,
6289 unsigned int *_hash_mask
,
6290 unsigned long low_limit
,
6291 unsigned long high_limit
)
6293 unsigned long long max
= high_limit
;
6294 unsigned long log2qty
, size
;
6297 /* allow the kernel cmdline to have a say */
6299 /* round applicable memory size up to nearest megabyte */
6300 numentries
= nr_kernel_pages
;
6302 /* It isn't necessary when PAGE_SIZE >= 1MB */
6303 if (PAGE_SHIFT
< 20)
6304 numentries
= round_up(numentries
, (1<<20)/PAGE_SIZE
);
6306 /* limit to 1 bucket per 2^scale bytes of low memory */
6307 if (scale
> PAGE_SHIFT
)
6308 numentries
>>= (scale
- PAGE_SHIFT
);
6310 numentries
<<= (PAGE_SHIFT
- scale
);
6312 /* Make sure we've got at least a 0-order allocation.. */
6313 if (unlikely(flags
& HASH_SMALL
)) {
6314 /* Makes no sense without HASH_EARLY */
6315 WARN_ON(!(flags
& HASH_EARLY
));
6316 if (!(numentries
>> *_hash_shift
)) {
6317 numentries
= 1UL << *_hash_shift
;
6318 BUG_ON(!numentries
);
6320 } else if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
6321 numentries
= PAGE_SIZE
/ bucketsize
;
6323 numentries
= roundup_pow_of_two(numentries
);
6325 /* limit allocation size to 1/16 total memory by default */
6327 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
6328 do_div(max
, bucketsize
);
6330 max
= min(max
, 0x80000000ULL
);
6332 if (numentries
< low_limit
)
6333 numentries
= low_limit
;
6334 if (numentries
> max
)
6337 log2qty
= ilog2(numentries
);
6340 size
= bucketsize
<< log2qty
;
6341 if (flags
& HASH_EARLY
)
6342 table
= memblock_virt_alloc_nopanic(size
, 0);
6344 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
6347 * If bucketsize is not a power-of-two, we may free
6348 * some pages at the end of hash table which
6349 * alloc_pages_exact() automatically does
6351 if (get_order(size
) < MAX_ORDER
) {
6352 table
= alloc_pages_exact(size
, GFP_ATOMIC
);
6353 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
6356 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
6359 panic("Failed to allocate %s hash table\n", tablename
);
6361 printk(KERN_INFO
"%s hash table entries: %ld (order: %d, %lu bytes)\n",
6364 ilog2(size
) - PAGE_SHIFT
,
6368 *_hash_shift
= log2qty
;
6370 *_hash_mask
= (1 << log2qty
) - 1;
6375 /* Return a pointer to the bitmap storing bits affecting a block of pages */
6376 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
6379 #ifdef CONFIG_SPARSEMEM
6380 return __pfn_to_section(pfn
)->pageblock_flags
;
6382 return zone
->pageblock_flags
;
6383 #endif /* CONFIG_SPARSEMEM */
6386 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
6388 #ifdef CONFIG_SPARSEMEM
6389 pfn
&= (PAGES_PER_SECTION
-1);
6390 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
6392 pfn
= pfn
- round_down(zone
->zone_start_pfn
, pageblock_nr_pages
);
6393 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
6394 #endif /* CONFIG_SPARSEMEM */
6398 * get_pfnblock_flags_mask - Return the requested group of flags for the pageblock_nr_pages block of pages
6399 * @page: The page within the block of interest
6400 * @pfn: The target page frame number
6401 * @end_bitidx: The last bit of interest to retrieve
6402 * @mask: mask of bits that the caller is interested in
6404 * Return: pageblock_bits flags
6406 unsigned long get_pfnblock_flags_mask(struct page
*page
, unsigned long pfn
,
6407 unsigned long end_bitidx
,
6411 unsigned long *bitmap
;
6412 unsigned long bitidx
, word_bitidx
;
6415 zone
= page_zone(page
);
6416 bitmap
= get_pageblock_bitmap(zone
, pfn
);
6417 bitidx
= pfn_to_bitidx(zone
, pfn
);
6418 word_bitidx
= bitidx
/ BITS_PER_LONG
;
6419 bitidx
&= (BITS_PER_LONG
-1);
6421 word
= bitmap
[word_bitidx
];
6422 bitidx
+= end_bitidx
;
6423 return (word
>> (BITS_PER_LONG
- bitidx
- 1)) & mask
;
6427 * set_pfnblock_flags_mask - Set the requested group of flags for a pageblock_nr_pages block of pages
6428 * @page: The page within the block of interest
6429 * @flags: The flags to set
6430 * @pfn: The target page frame number
6431 * @end_bitidx: The last bit of interest
6432 * @mask: mask of bits that the caller is interested in
6434 void set_pfnblock_flags_mask(struct page
*page
, unsigned long flags
,
6436 unsigned long end_bitidx
,
6440 unsigned long *bitmap
;
6441 unsigned long bitidx
, word_bitidx
;
6442 unsigned long old_word
, word
;
6444 BUILD_BUG_ON(NR_PAGEBLOCK_BITS
!= 4);
6446 zone
= page_zone(page
);
6447 bitmap
= get_pageblock_bitmap(zone
, pfn
);
6448 bitidx
= pfn_to_bitidx(zone
, pfn
);
6449 word_bitidx
= bitidx
/ BITS_PER_LONG
;
6450 bitidx
&= (BITS_PER_LONG
-1);
6452 VM_BUG_ON_PAGE(!zone_spans_pfn(zone
, pfn
), page
);
6454 bitidx
+= end_bitidx
;
6455 mask
<<= (BITS_PER_LONG
- bitidx
- 1);
6456 flags
<<= (BITS_PER_LONG
- bitidx
- 1);
6458 word
= READ_ONCE(bitmap
[word_bitidx
]);
6460 old_word
= cmpxchg(&bitmap
[word_bitidx
], word
, (word
& ~mask
) | flags
);
6461 if (word
== old_word
)
6468 * This function checks whether pageblock includes unmovable pages or not.
6469 * If @count is not zero, it is okay to include less @count unmovable pages
6471 * PageLRU check without isolation or lru_lock could race so that
6472 * MIGRATE_MOVABLE block might include unmovable pages. It means you can't
6473 * expect this function should be exact.
6475 bool has_unmovable_pages(struct zone
*zone
, struct page
*page
, int count
,
6476 bool skip_hwpoisoned_pages
)
6478 unsigned long pfn
, iter
, found
;
6482 * For avoiding noise data, lru_add_drain_all() should be called
6483 * If ZONE_MOVABLE, the zone never contains unmovable pages
6485 if (zone_idx(zone
) == ZONE_MOVABLE
)
6487 mt
= get_pageblock_migratetype(page
);
6488 if (mt
== MIGRATE_MOVABLE
|| is_migrate_cma(mt
))
6491 pfn
= page_to_pfn(page
);
6492 for (found
= 0, iter
= 0; iter
< pageblock_nr_pages
; iter
++) {
6493 unsigned long check
= pfn
+ iter
;
6495 if (!pfn_valid_within(check
))
6498 page
= pfn_to_page(check
);
6501 * Hugepages are not in LRU lists, but they're movable.
6502 * We need not scan over tail pages bacause we don't
6503 * handle each tail page individually in migration.
6505 if (PageHuge(page
)) {
6506 iter
= round_up(iter
+ 1, 1<<compound_order(page
)) - 1;
6511 * We can't use page_count without pin a page
6512 * because another CPU can free compound page.
6513 * This check already skips compound tails of THP
6514 * because their page->_count is zero at all time.
6516 if (!atomic_read(&page
->_count
)) {
6517 if (PageBuddy(page
))
6518 iter
+= (1 << page_order(page
)) - 1;
6523 * The HWPoisoned page may be not in buddy system, and
6524 * page_count() is not 0.
6526 if (skip_hwpoisoned_pages
&& PageHWPoison(page
))
6532 * If there are RECLAIMABLE pages, we need to check
6533 * it. But now, memory offline itself doesn't call
6534 * shrink_node_slabs() and it still to be fixed.
6537 * If the page is not RAM, page_count()should be 0.
6538 * we don't need more check. This is an _used_ not-movable page.
6540 * The problematic thing here is PG_reserved pages. PG_reserved
6541 * is set to both of a memory hole page and a _used_ kernel
6550 bool is_pageblock_removable_nolock(struct page
*page
)
6556 * We have to be careful here because we are iterating over memory
6557 * sections which are not zone aware so we might end up outside of
6558 * the zone but still within the section.
6559 * We have to take care about the node as well. If the node is offline
6560 * its NODE_DATA will be NULL - see page_zone.
6562 if (!node_online(page_to_nid(page
)))
6565 zone
= page_zone(page
);
6566 pfn
= page_to_pfn(page
);
6567 if (!zone_spans_pfn(zone
, pfn
))
6570 return !has_unmovable_pages(zone
, page
, 0, true);
6575 static unsigned long pfn_max_align_down(unsigned long pfn
)
6577 return pfn
& ~(max_t(unsigned long, MAX_ORDER_NR_PAGES
,
6578 pageblock_nr_pages
) - 1);
6581 static unsigned long pfn_max_align_up(unsigned long pfn
)
6583 return ALIGN(pfn
, max_t(unsigned long, MAX_ORDER_NR_PAGES
,
6584 pageblock_nr_pages
));
6587 /* [start, end) must belong to a single zone. */
6588 static int __alloc_contig_migrate_range(struct compact_control
*cc
,
6589 unsigned long start
, unsigned long end
)
6591 /* This function is based on compact_zone() from compaction.c. */
6592 unsigned long nr_reclaimed
;
6593 unsigned long pfn
= start
;
6594 unsigned int tries
= 0;
6599 while (pfn
< end
|| !list_empty(&cc
->migratepages
)) {
6600 if (fatal_signal_pending(current
)) {
6605 if (list_empty(&cc
->migratepages
)) {
6606 cc
->nr_migratepages
= 0;
6607 pfn
= isolate_migratepages_range(cc
, pfn
, end
);
6613 } else if (++tries
== 5) {
6614 ret
= ret
< 0 ? ret
: -EBUSY
;
6618 nr_reclaimed
= reclaim_clean_pages_from_list(cc
->zone
,
6620 cc
->nr_migratepages
-= nr_reclaimed
;
6622 ret
= migrate_pages(&cc
->migratepages
, alloc_migrate_target
,
6623 NULL
, 0, cc
->mode
, MR_CMA
);
6626 putback_movable_pages(&cc
->migratepages
);
6633 * alloc_contig_range() -- tries to allocate given range of pages
6634 * @start: start PFN to allocate
6635 * @end: one-past-the-last PFN to allocate
6636 * @migratetype: migratetype of the underlaying pageblocks (either
6637 * #MIGRATE_MOVABLE or #MIGRATE_CMA). All pageblocks
6638 * in range must have the same migratetype and it must
6639 * be either of the two.
6641 * The PFN range does not have to be pageblock or MAX_ORDER_NR_PAGES
6642 * aligned, however it's the caller's responsibility to guarantee that
6643 * we are the only thread that changes migrate type of pageblocks the
6646 * The PFN range must belong to a single zone.
6648 * Returns zero on success or negative error code. On success all
6649 * pages which PFN is in [start, end) are allocated for the caller and
6650 * need to be freed with free_contig_range().
6652 int alloc_contig_range(unsigned long start
, unsigned long end
,
6653 unsigned migratetype
)
6655 unsigned long outer_start
, outer_end
;
6659 struct compact_control cc
= {
6660 .nr_migratepages
= 0,
6662 .zone
= page_zone(pfn_to_page(start
)),
6663 .mode
= MIGRATE_SYNC
,
6664 .ignore_skip_hint
= true,
6666 INIT_LIST_HEAD(&cc
.migratepages
);
6669 * What we do here is we mark all pageblocks in range as
6670 * MIGRATE_ISOLATE. Because pageblock and max order pages may
6671 * have different sizes, and due to the way page allocator
6672 * work, we align the range to biggest of the two pages so
6673 * that page allocator won't try to merge buddies from
6674 * different pageblocks and change MIGRATE_ISOLATE to some
6675 * other migration type.
6677 * Once the pageblocks are marked as MIGRATE_ISOLATE, we
6678 * migrate the pages from an unaligned range (ie. pages that
6679 * we are interested in). This will put all the pages in
6680 * range back to page allocator as MIGRATE_ISOLATE.
6682 * When this is done, we take the pages in range from page
6683 * allocator removing them from the buddy system. This way
6684 * page allocator will never consider using them.
6686 * This lets us mark the pageblocks back as
6687 * MIGRATE_CMA/MIGRATE_MOVABLE so that free pages in the
6688 * aligned range but not in the unaligned, original range are
6689 * put back to page allocator so that buddy can use them.
6692 ret
= start_isolate_page_range(pfn_max_align_down(start
),
6693 pfn_max_align_up(end
), migratetype
,
6699 * In case of -EBUSY, we'd like to know which page causes problem.
6700 * So, just fall through. We will check it in test_pages_isolated().
6702 ret
= __alloc_contig_migrate_range(&cc
, start
, end
);
6703 if (ret
&& ret
!= -EBUSY
)
6707 * Pages from [start, end) are within a MAX_ORDER_NR_PAGES
6708 * aligned blocks that are marked as MIGRATE_ISOLATE. What's
6709 * more, all pages in [start, end) are free in page allocator.
6710 * What we are going to do is to allocate all pages from
6711 * [start, end) (that is remove them from page allocator).
6713 * The only problem is that pages at the beginning and at the
6714 * end of interesting range may be not aligned with pages that
6715 * page allocator holds, ie. they can be part of higher order
6716 * pages. Because of this, we reserve the bigger range and
6717 * once this is done free the pages we are not interested in.
6719 * We don't have to hold zone->lock here because the pages are
6720 * isolated thus they won't get removed from buddy.
6723 lru_add_drain_all();
6724 drain_all_pages(cc
.zone
);
6727 outer_start
= start
;
6728 while (!PageBuddy(pfn_to_page(outer_start
))) {
6729 if (++order
>= MAX_ORDER
) {
6730 outer_start
= start
;
6733 outer_start
&= ~0UL << order
;
6736 if (outer_start
!= start
) {
6737 order
= page_order(pfn_to_page(outer_start
));
6740 * outer_start page could be small order buddy page and
6741 * it doesn't include start page. Adjust outer_start
6742 * in this case to report failed page properly
6743 * on tracepoint in test_pages_isolated()
6745 if (outer_start
+ (1UL << order
) <= start
)
6746 outer_start
= start
;
6749 /* Make sure the range is really isolated. */
6750 if (test_pages_isolated(outer_start
, end
, false)) {
6751 pr_info("%s: [%lx, %lx) PFNs busy\n",
6752 __func__
, outer_start
, end
);
6757 /* Grab isolated pages from freelists. */
6758 outer_end
= isolate_freepages_range(&cc
, outer_start
, end
);
6764 /* Free head and tail (if any) */
6765 if (start
!= outer_start
)
6766 free_contig_range(outer_start
, start
- outer_start
);
6767 if (end
!= outer_end
)
6768 free_contig_range(end
, outer_end
- end
);
6771 undo_isolate_page_range(pfn_max_align_down(start
),
6772 pfn_max_align_up(end
), migratetype
);
6776 void free_contig_range(unsigned long pfn
, unsigned nr_pages
)
6778 unsigned int count
= 0;
6780 for (; nr_pages
--; pfn
++) {
6781 struct page
*page
= pfn_to_page(pfn
);
6783 count
+= page_count(page
) != 1;
6786 WARN(count
!= 0, "%d pages are still in use!\n", count
);
6790 #ifdef CONFIG_MEMORY_HOTPLUG
6792 * The zone indicated has a new number of managed_pages; batch sizes and percpu
6793 * page high values need to be recalulated.
6795 void __meminit
zone_pcp_update(struct zone
*zone
)
6798 mutex_lock(&pcp_batch_high_lock
);
6799 for_each_possible_cpu(cpu
)
6800 pageset_set_high_and_batch(zone
,
6801 per_cpu_ptr(zone
->pageset
, cpu
));
6802 mutex_unlock(&pcp_batch_high_lock
);
6806 void zone_pcp_reset(struct zone
*zone
)
6808 unsigned long flags
;
6810 struct per_cpu_pageset
*pset
;
6812 /* avoid races with drain_pages() */
6813 local_irq_save(flags
);
6814 if (zone
->pageset
!= &boot_pageset
) {
6815 for_each_online_cpu(cpu
) {
6816 pset
= per_cpu_ptr(zone
->pageset
, cpu
);
6817 drain_zonestat(zone
, pset
);
6819 free_percpu(zone
->pageset
);
6820 zone
->pageset
= &boot_pageset
;
6822 local_irq_restore(flags
);
6825 #ifdef CONFIG_MEMORY_HOTREMOVE
6827 * All pages in the range must be isolated before calling this.
6830 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
6834 unsigned int order
, i
;
6836 unsigned long flags
;
6837 /* find the first valid pfn */
6838 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
6843 zone
= page_zone(pfn_to_page(pfn
));
6844 spin_lock_irqsave(&zone
->lock
, flags
);
6846 while (pfn
< end_pfn
) {
6847 if (!pfn_valid(pfn
)) {
6851 page
= pfn_to_page(pfn
);
6853 * The HWPoisoned page may be not in buddy system, and
6854 * page_count() is not 0.
6856 if (unlikely(!PageBuddy(page
) && PageHWPoison(page
))) {
6858 SetPageReserved(page
);
6862 BUG_ON(page_count(page
));
6863 BUG_ON(!PageBuddy(page
));
6864 order
= page_order(page
);
6865 #ifdef CONFIG_DEBUG_VM
6866 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
6867 pfn
, 1 << order
, end_pfn
);
6869 list_del(&page
->lru
);
6870 rmv_page_order(page
);
6871 zone
->free_area
[order
].nr_free
--;
6872 for (i
= 0; i
< (1 << order
); i
++)
6873 SetPageReserved((page
+i
));
6874 pfn
+= (1 << order
);
6876 spin_unlock_irqrestore(&zone
->lock
, flags
);
6880 #ifdef CONFIG_MEMORY_FAILURE
6881 bool is_free_buddy_page(struct page
*page
)
6883 struct zone
*zone
= page_zone(page
);
6884 unsigned long pfn
= page_to_pfn(page
);
6885 unsigned long flags
;
6888 spin_lock_irqsave(&zone
->lock
, flags
);
6889 for (order
= 0; order
< MAX_ORDER
; order
++) {
6890 struct page
*page_head
= page
- (pfn
& ((1 << order
) - 1));
6892 if (PageBuddy(page_head
) && page_order(page_head
) >= order
)
6895 spin_unlock_irqrestore(&zone
->lock
, flags
);
6897 return order
< MAX_ORDER
;