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/compiler.h>
25 #include <linux/kernel.h>
26 #include <linux/module.h>
27 #include <linux/suspend.h>
28 #include <linux/pagevec.h>
29 #include <linux/blkdev.h>
30 #include <linux/slab.h>
31 #include <linux/oom.h>
32 #include <linux/notifier.h>
33 #include <linux/topology.h>
34 #include <linux/sysctl.h>
35 #include <linux/cpu.h>
36 #include <linux/cpuset.h>
37 #include <linux/memory_hotplug.h>
38 #include <linux/nodemask.h>
39 #include <linux/vmalloc.h>
40 #include <linux/mempolicy.h>
41 #include <linux/stop_machine.h>
42 #include <linux/sort.h>
43 #include <linux/pfn.h>
44 #include <linux/backing-dev.h>
45 #include <linux/fault-inject.h>
46 #include <linux/page-isolation.h>
47 #include <linux/page_cgroup.h>
48 #include <linux/debugobjects.h>
49 #include <linux/kmemleak.h>
51 #include <asm/tlbflush.h>
52 #include <asm/div64.h>
56 * Array of node states.
58 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
59 [N_POSSIBLE
] = NODE_MASK_ALL
,
60 [N_ONLINE
] = { { [0] = 1UL } },
62 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
64 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
66 [N_CPU
] = { { [0] = 1UL } },
69 EXPORT_SYMBOL(node_states
);
71 unsigned long totalram_pages __read_mostly
;
72 unsigned long totalreserve_pages __read_mostly
;
73 unsigned long highest_memmap_pfn __read_mostly
;
74 int percpu_pagelist_fraction
;
76 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
77 int pageblock_order __read_mostly
;
80 static void __free_pages_ok(struct page
*page
, unsigned int order
);
83 * results with 256, 32 in the lowmem_reserve sysctl:
84 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
85 * 1G machine -> (16M dma, 784M normal, 224M high)
86 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
87 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
88 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
90 * TBD: should special case ZONE_DMA32 machines here - in those we normally
91 * don't need any ZONE_NORMAL reservation
93 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
94 #ifdef CONFIG_ZONE_DMA
97 #ifdef CONFIG_ZONE_DMA32
100 #ifdef CONFIG_HIGHMEM
106 EXPORT_SYMBOL(totalram_pages
);
108 static char * const zone_names
[MAX_NR_ZONES
] = {
109 #ifdef CONFIG_ZONE_DMA
112 #ifdef CONFIG_ZONE_DMA32
116 #ifdef CONFIG_HIGHMEM
122 int min_free_kbytes
= 1024;
124 unsigned long __meminitdata nr_kernel_pages
;
125 unsigned long __meminitdata nr_all_pages
;
126 static unsigned long __meminitdata dma_reserve
;
128 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
130 * MAX_ACTIVE_REGIONS determines the maximum number of distinct
131 * ranges of memory (RAM) that may be registered with add_active_range().
132 * Ranges passed to add_active_range() will be merged if possible
133 * so the number of times add_active_range() can be called is
134 * related to the number of nodes and the number of holes
136 #ifdef CONFIG_MAX_ACTIVE_REGIONS
137 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
138 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
140 #if MAX_NUMNODES >= 32
141 /* If there can be many nodes, allow up to 50 holes per node */
142 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
144 /* By default, allow up to 256 distinct regions */
145 #define MAX_ACTIVE_REGIONS 256
149 static struct node_active_region __meminitdata early_node_map
[MAX_ACTIVE_REGIONS
];
150 static int __meminitdata nr_nodemap_entries
;
151 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
152 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
153 static unsigned long __initdata required_kernelcore
;
154 static unsigned long __initdata required_movablecore
;
155 static unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
157 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
159 EXPORT_SYMBOL(movable_zone
);
160 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
163 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
164 EXPORT_SYMBOL(nr_node_ids
);
167 int page_group_by_mobility_disabled __read_mostly
;
169 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
171 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
172 PB_migrate
, PB_migrate_end
);
175 #ifdef CONFIG_DEBUG_VM
176 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
180 unsigned long pfn
= page_to_pfn(page
);
183 seq
= zone_span_seqbegin(zone
);
184 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
186 else if (pfn
< zone
->zone_start_pfn
)
188 } while (zone_span_seqretry(zone
, seq
));
193 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
195 if (!pfn_valid_within(page_to_pfn(page
)))
197 if (zone
!= page_zone(page
))
203 * Temporary debugging check for pages not lying within a given zone.
205 static int bad_range(struct zone
*zone
, struct page
*page
)
207 if (page_outside_zone_boundaries(zone
, page
))
209 if (!page_is_consistent(zone
, page
))
215 static inline int bad_range(struct zone
*zone
, struct page
*page
)
221 static void bad_page(struct page
*page
)
223 static unsigned long resume
;
224 static unsigned long nr_shown
;
225 static unsigned long nr_unshown
;
228 * Allow a burst of 60 reports, then keep quiet for that minute;
229 * or allow a steady drip of one report per second.
231 if (nr_shown
== 60) {
232 if (time_before(jiffies
, resume
)) {
238 "BUG: Bad page state: %lu messages suppressed\n",
245 resume
= jiffies
+ 60 * HZ
;
247 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
248 current
->comm
, page_to_pfn(page
));
250 "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
251 page
, (void *)page
->flags
, page_count(page
),
252 page_mapcount(page
), page
->mapping
, page
->index
);
256 /* Leave bad fields for debug, except PageBuddy could make trouble */
257 __ClearPageBuddy(page
);
258 add_taint(TAINT_BAD_PAGE
);
262 * Higher-order pages are called "compound pages". They are structured thusly:
264 * The first PAGE_SIZE page is called the "head page".
266 * The remaining PAGE_SIZE pages are called "tail pages".
268 * All pages have PG_compound set. All pages have their ->private pointing at
269 * the head page (even the head page has this).
271 * The first tail page's ->lru.next holds the address of the compound page's
272 * put_page() function. Its ->lru.prev holds the order of allocation.
273 * This usage means that zero-order pages may not be compound.
276 static void free_compound_page(struct page
*page
)
278 __free_pages_ok(page
, compound_order(page
));
281 void prep_compound_page(struct page
*page
, unsigned long order
)
284 int nr_pages
= 1 << order
;
286 set_compound_page_dtor(page
, free_compound_page
);
287 set_compound_order(page
, order
);
289 for (i
= 1; i
< nr_pages
; i
++) {
290 struct page
*p
= page
+ i
;
293 p
->first_page
= page
;
297 #ifdef CONFIG_HUGETLBFS
298 void prep_compound_gigantic_page(struct page
*page
, unsigned long order
)
301 int nr_pages
= 1 << order
;
302 struct page
*p
= page
+ 1;
304 set_compound_page_dtor(page
, free_compound_page
);
305 set_compound_order(page
, order
);
307 for (i
= 1; i
< nr_pages
; i
++, p
= mem_map_next(p
, page
, i
)) {
309 p
->first_page
= page
;
314 static int destroy_compound_page(struct page
*page
, unsigned long order
)
317 int nr_pages
= 1 << order
;
320 if (unlikely(compound_order(page
) != order
) ||
321 unlikely(!PageHead(page
))) {
326 __ClearPageHead(page
);
328 for (i
= 1; i
< nr_pages
; i
++) {
329 struct page
*p
= page
+ i
;
331 if (unlikely(!PageTail(p
) || (p
->first_page
!= page
))) {
341 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
346 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
347 * and __GFP_HIGHMEM from hard or soft interrupt context.
349 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
350 for (i
= 0; i
< (1 << order
); i
++)
351 clear_highpage(page
+ i
);
354 static inline void set_page_order(struct page
*page
, int order
)
356 set_page_private(page
, order
);
357 __SetPageBuddy(page
);
360 static inline void rmv_page_order(struct page
*page
)
362 __ClearPageBuddy(page
);
363 set_page_private(page
, 0);
367 * Locate the struct page for both the matching buddy in our
368 * pair (buddy1) and the combined O(n+1) page they form (page).
370 * 1) Any buddy B1 will have an order O twin B2 which satisfies
371 * the following equation:
373 * For example, if the starting buddy (buddy2) is #8 its order
375 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
377 * 2) Any buddy B will have an order O+1 parent P which
378 * satisfies the following equation:
381 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
383 static inline struct page
*
384 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
386 unsigned long buddy_idx
= page_idx
^ (1 << order
);
388 return page
+ (buddy_idx
- page_idx
);
391 static inline unsigned long
392 __find_combined_index(unsigned long page_idx
, unsigned int order
)
394 return (page_idx
& ~(1 << order
));
398 * This function checks whether a page is free && is the buddy
399 * we can do coalesce a page and its buddy if
400 * (a) the buddy is not in a hole &&
401 * (b) the buddy is in the buddy system &&
402 * (c) a page and its buddy have the same order &&
403 * (d) a page and its buddy are in the same zone.
405 * For recording whether a page is in the buddy system, we use PG_buddy.
406 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
408 * For recording page's order, we use page_private(page).
410 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
413 if (!pfn_valid_within(page_to_pfn(buddy
)))
416 if (page_zone_id(page
) != page_zone_id(buddy
))
419 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
420 BUG_ON(page_count(buddy
) != 0);
427 * Freeing function for a buddy system allocator.
429 * The concept of a buddy system is to maintain direct-mapped table
430 * (containing bit values) for memory blocks of various "orders".
431 * The bottom level table contains the map for the smallest allocatable
432 * units of memory (here, pages), and each level above it describes
433 * pairs of units from the levels below, hence, "buddies".
434 * At a high level, all that happens here is marking the table entry
435 * at the bottom level available, and propagating the changes upward
436 * as necessary, plus some accounting needed to play nicely with other
437 * parts of the VM system.
438 * At each level, we keep a list of pages, which are heads of continuous
439 * free pages of length of (1 << order) and marked with PG_buddy. Page's
440 * order is recorded in page_private(page) field.
441 * So when we are allocating or freeing one, we can derive the state of the
442 * other. That is, if we allocate a small block, and both were
443 * free, the remainder of the region must be split into blocks.
444 * If a block is freed, and its buddy is also free, then this
445 * triggers coalescing into a block of larger size.
450 static inline void __free_one_page(struct page
*page
,
451 struct zone
*zone
, unsigned int order
)
453 unsigned long page_idx
;
454 int order_size
= 1 << order
;
455 int migratetype
= get_pageblock_migratetype(page
);
457 if (unlikely(PageCompound(page
)))
458 if (unlikely(destroy_compound_page(page
, order
)))
461 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
463 VM_BUG_ON(page_idx
& (order_size
- 1));
464 VM_BUG_ON(bad_range(zone
, page
));
466 __mod_zone_page_state(zone
, NR_FREE_PAGES
, order_size
);
467 while (order
< MAX_ORDER
-1) {
468 unsigned long combined_idx
;
471 buddy
= __page_find_buddy(page
, page_idx
, order
);
472 if (!page_is_buddy(page
, buddy
, order
))
475 /* Our buddy is free, merge with it and move up one order. */
476 list_del(&buddy
->lru
);
477 zone
->free_area
[order
].nr_free
--;
478 rmv_page_order(buddy
);
479 combined_idx
= __find_combined_index(page_idx
, order
);
480 page
= page
+ (combined_idx
- page_idx
);
481 page_idx
= combined_idx
;
484 set_page_order(page
, order
);
486 &zone
->free_area
[order
].free_list
[migratetype
]);
487 zone
->free_area
[order
].nr_free
++;
490 static inline int free_pages_check(struct page
*page
)
492 free_page_mlock(page
);
493 if (unlikely(page_mapcount(page
) |
494 (page
->mapping
!= NULL
) |
495 (page_count(page
) != 0) |
496 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
))) {
500 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
501 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
506 * Frees a list of pages.
507 * Assumes all pages on list are in same zone, and of same order.
508 * count is the number of pages to free.
510 * If the zone was previously in an "all pages pinned" state then look to
511 * see if this freeing clears that state.
513 * And clear the zone's pages_scanned counter, to hold off the "all pages are
514 * pinned" detection logic.
516 static void free_pages_bulk(struct zone
*zone
, int count
,
517 struct list_head
*list
, int order
)
519 spin_lock(&zone
->lock
);
520 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
521 zone
->pages_scanned
= 0;
525 VM_BUG_ON(list_empty(list
));
526 page
= list_entry(list
->prev
, struct page
, lru
);
527 /* have to delete it as __free_one_page list manipulates */
528 list_del(&page
->lru
);
529 __free_one_page(page
, zone
, order
);
531 spin_unlock(&zone
->lock
);
534 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
536 spin_lock(&zone
->lock
);
537 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
538 zone
->pages_scanned
= 0;
539 __free_one_page(page
, zone
, order
);
540 spin_unlock(&zone
->lock
);
543 static void __free_pages_ok(struct page
*page
, unsigned int order
)
549 for (i
= 0 ; i
< (1 << order
) ; ++i
)
550 bad
+= free_pages_check(page
+ i
);
554 if (!PageHighMem(page
)) {
555 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
556 debug_check_no_obj_freed(page_address(page
),
559 arch_free_page(page
, order
);
560 kernel_map_pages(page
, 1 << order
, 0);
562 local_irq_save(flags
);
563 __count_vm_events(PGFREE
, 1 << order
);
564 free_one_page(page_zone(page
), page
, order
);
565 local_irq_restore(flags
);
569 * permit the bootmem allocator to evade page validation on high-order frees
571 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
574 __ClearPageReserved(page
);
575 set_page_count(page
, 0);
576 set_page_refcounted(page
);
582 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
583 struct page
*p
= &page
[loop
];
585 if (loop
+ 1 < BITS_PER_LONG
)
587 __ClearPageReserved(p
);
588 set_page_count(p
, 0);
591 set_page_refcounted(page
);
592 __free_pages(page
, order
);
598 * The order of subdivision here is critical for the IO subsystem.
599 * Please do not alter this order without good reasons and regression
600 * testing. Specifically, as large blocks of memory are subdivided,
601 * the order in which smaller blocks are delivered depends on the order
602 * they're subdivided in this function. This is the primary factor
603 * influencing the order in which pages are delivered to the IO
604 * subsystem according to empirical testing, and this is also justified
605 * by considering the behavior of a buddy system containing a single
606 * large block of memory acted on by a series of small allocations.
607 * This behavior is a critical factor in sglist merging's success.
611 static inline void expand(struct zone
*zone
, struct page
*page
,
612 int low
, int high
, struct free_area
*area
,
615 unsigned long size
= 1 << high
;
621 VM_BUG_ON(bad_range(zone
, &page
[size
]));
622 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
624 set_page_order(&page
[size
], high
);
629 * This page is about to be returned from the page allocator
631 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
633 if (unlikely(page_mapcount(page
) |
634 (page
->mapping
!= NULL
) |
635 (page_count(page
) != 0) |
636 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
))) {
641 set_page_private(page
, 0);
642 set_page_refcounted(page
);
644 arch_alloc_page(page
, order
);
645 kernel_map_pages(page
, 1 << order
, 1);
647 if (gfp_flags
& __GFP_ZERO
)
648 prep_zero_page(page
, order
, gfp_flags
);
650 if (order
&& (gfp_flags
& __GFP_COMP
))
651 prep_compound_page(page
, order
);
657 * Go through the free lists for the given migratetype and remove
658 * the smallest available page from the freelists
660 static struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
663 unsigned int current_order
;
664 struct free_area
* area
;
667 /* Find a page of the appropriate size in the preferred list */
668 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
669 area
= &(zone
->free_area
[current_order
]);
670 if (list_empty(&area
->free_list
[migratetype
]))
673 page
= list_entry(area
->free_list
[migratetype
].next
,
675 list_del(&page
->lru
);
676 rmv_page_order(page
);
678 __mod_zone_page_state(zone
, NR_FREE_PAGES
, - (1UL << order
));
679 expand(zone
, page
, order
, current_order
, area
, migratetype
);
688 * This array describes the order lists are fallen back to when
689 * the free lists for the desirable migrate type are depleted
691 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
692 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
693 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
694 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
695 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
699 * Move the free pages in a range to the free lists of the requested type.
700 * Note that start_page and end_pages are not aligned on a pageblock
701 * boundary. If alignment is required, use move_freepages_block()
703 static int move_freepages(struct zone
*zone
,
704 struct page
*start_page
, struct page
*end_page
,
711 #ifndef CONFIG_HOLES_IN_ZONE
713 * page_zone is not safe to call in this context when
714 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
715 * anyway as we check zone boundaries in move_freepages_block().
716 * Remove at a later date when no bug reports exist related to
717 * grouping pages by mobility
719 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
722 for (page
= start_page
; page
<= end_page
;) {
723 /* Make sure we are not inadvertently changing nodes */
724 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
726 if (!pfn_valid_within(page_to_pfn(page
))) {
731 if (!PageBuddy(page
)) {
736 order
= page_order(page
);
737 list_del(&page
->lru
);
739 &zone
->free_area
[order
].free_list
[migratetype
]);
741 pages_moved
+= 1 << order
;
747 static int move_freepages_block(struct zone
*zone
, struct page
*page
,
750 unsigned long start_pfn
, end_pfn
;
751 struct page
*start_page
, *end_page
;
753 start_pfn
= page_to_pfn(page
);
754 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
755 start_page
= pfn_to_page(start_pfn
);
756 end_page
= start_page
+ pageblock_nr_pages
- 1;
757 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
759 /* Do not cross zone boundaries */
760 if (start_pfn
< zone
->zone_start_pfn
)
762 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
765 return move_freepages(zone
, start_page
, end_page
, migratetype
);
768 /* Remove an element from the buddy allocator from the fallback list */
769 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
770 int start_migratetype
)
772 struct free_area
* area
;
777 /* Find the largest possible block of pages in the other list */
778 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
780 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
781 migratetype
= fallbacks
[start_migratetype
][i
];
783 /* MIGRATE_RESERVE handled later if necessary */
784 if (migratetype
== MIGRATE_RESERVE
)
787 area
= &(zone
->free_area
[current_order
]);
788 if (list_empty(&area
->free_list
[migratetype
]))
791 page
= list_entry(area
->free_list
[migratetype
].next
,
796 * If breaking a large block of pages, move all free
797 * pages to the preferred allocation list. If falling
798 * back for a reclaimable kernel allocation, be more
799 * agressive about taking ownership of free pages
801 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
802 start_migratetype
== MIGRATE_RECLAIMABLE
) {
804 pages
= move_freepages_block(zone
, page
,
807 /* Claim the whole block if over half of it is free */
808 if (pages
>= (1 << (pageblock_order
-1)))
809 set_pageblock_migratetype(page
,
812 migratetype
= start_migratetype
;
815 /* Remove the page from the freelists */
816 list_del(&page
->lru
);
817 rmv_page_order(page
);
818 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
821 if (current_order
== pageblock_order
)
822 set_pageblock_migratetype(page
,
825 expand(zone
, page
, order
, current_order
, area
, migratetype
);
830 /* Use MIGRATE_RESERVE rather than fail an allocation */
831 return __rmqueue_smallest(zone
, order
, MIGRATE_RESERVE
);
835 * Do the hard work of removing an element from the buddy allocator.
836 * Call me with the zone->lock already held.
838 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
843 page
= __rmqueue_smallest(zone
, order
, migratetype
);
846 page
= __rmqueue_fallback(zone
, order
, migratetype
);
852 * Obtain a specified number of elements from the buddy allocator, all under
853 * a single hold of the lock, for efficiency. Add them to the supplied list.
854 * Returns the number of new pages which were placed at *list.
856 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
857 unsigned long count
, struct list_head
*list
,
862 spin_lock(&zone
->lock
);
863 for (i
= 0; i
< count
; ++i
) {
864 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
865 if (unlikely(page
== NULL
))
869 * Split buddy pages returned by expand() are received here
870 * in physical page order. The page is added to the callers and
871 * list and the list head then moves forward. From the callers
872 * perspective, the linked list is ordered by page number in
873 * some conditions. This is useful for IO devices that can
874 * merge IO requests if the physical pages are ordered
877 list_add(&page
->lru
, list
);
878 set_page_private(page
, migratetype
);
881 spin_unlock(&zone
->lock
);
887 * Called from the vmstat counter updater to drain pagesets of this
888 * currently executing processor on remote nodes after they have
891 * Note that this function must be called with the thread pinned to
892 * a single processor.
894 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
899 local_irq_save(flags
);
900 if (pcp
->count
>= pcp
->batch
)
901 to_drain
= pcp
->batch
;
903 to_drain
= pcp
->count
;
904 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
905 pcp
->count
-= to_drain
;
906 local_irq_restore(flags
);
911 * Drain pages of the indicated processor.
913 * The processor must either be the current processor and the
914 * thread pinned to the current processor or a processor that
917 static void drain_pages(unsigned int cpu
)
922 for_each_populated_zone(zone
) {
923 struct per_cpu_pageset
*pset
;
924 struct per_cpu_pages
*pcp
;
926 pset
= zone_pcp(zone
, cpu
);
929 local_irq_save(flags
);
930 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
932 local_irq_restore(flags
);
937 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
939 void drain_local_pages(void *arg
)
941 drain_pages(smp_processor_id());
945 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
947 void drain_all_pages(void)
949 on_each_cpu(drain_local_pages
, NULL
, 1);
952 #ifdef CONFIG_HIBERNATION
954 void mark_free_pages(struct zone
*zone
)
956 unsigned long pfn
, max_zone_pfn
;
959 struct list_head
*curr
;
961 if (!zone
->spanned_pages
)
964 spin_lock_irqsave(&zone
->lock
, flags
);
966 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
967 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
968 if (pfn_valid(pfn
)) {
969 struct page
*page
= pfn_to_page(pfn
);
971 if (!swsusp_page_is_forbidden(page
))
972 swsusp_unset_page_free(page
);
975 for_each_migratetype_order(order
, t
) {
976 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
979 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
980 for (i
= 0; i
< (1UL << order
); i
++)
981 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
984 spin_unlock_irqrestore(&zone
->lock
, flags
);
986 #endif /* CONFIG_PM */
989 * Free a 0-order page
991 static void free_hot_cold_page(struct page
*page
, int cold
)
993 struct zone
*zone
= page_zone(page
);
994 struct per_cpu_pages
*pcp
;
998 page
->mapping
= NULL
;
999 if (free_pages_check(page
))
1002 if (!PageHighMem(page
)) {
1003 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
1004 debug_check_no_obj_freed(page_address(page
), PAGE_SIZE
);
1006 arch_free_page(page
, 0);
1007 kernel_map_pages(page
, 1, 0);
1009 pcp
= &zone_pcp(zone
, get_cpu())->pcp
;
1010 local_irq_save(flags
);
1011 __count_vm_event(PGFREE
);
1013 list_add_tail(&page
->lru
, &pcp
->list
);
1015 list_add(&page
->lru
, &pcp
->list
);
1016 set_page_private(page
, get_pageblock_migratetype(page
));
1018 if (pcp
->count
>= pcp
->high
) {
1019 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
1020 pcp
->count
-= pcp
->batch
;
1022 local_irq_restore(flags
);
1026 void free_hot_page(struct page
*page
)
1028 free_hot_cold_page(page
, 0);
1031 void free_cold_page(struct page
*page
)
1033 free_hot_cold_page(page
, 1);
1037 * split_page takes a non-compound higher-order page, and splits it into
1038 * n (1<<order) sub-pages: page[0..n]
1039 * Each sub-page must be freed individually.
1041 * Note: this is probably too low level an operation for use in drivers.
1042 * Please consult with lkml before using this in your driver.
1044 void split_page(struct page
*page
, unsigned int order
)
1048 VM_BUG_ON(PageCompound(page
));
1049 VM_BUG_ON(!page_count(page
));
1050 for (i
= 1; i
< (1 << order
); i
++)
1051 set_page_refcounted(page
+ i
);
1055 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1056 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1059 static struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1060 struct zone
*zone
, int order
, gfp_t gfp_flags
)
1062 unsigned long flags
;
1064 int cold
= !!(gfp_flags
& __GFP_COLD
);
1066 int migratetype
= allocflags_to_migratetype(gfp_flags
);
1070 if (likely(order
== 0)) {
1071 struct per_cpu_pages
*pcp
;
1073 pcp
= &zone_pcp(zone
, cpu
)->pcp
;
1074 local_irq_save(flags
);
1076 pcp
->count
= rmqueue_bulk(zone
, 0,
1077 pcp
->batch
, &pcp
->list
, migratetype
);
1078 if (unlikely(!pcp
->count
))
1082 /* Find a page of the appropriate migrate type */
1084 list_for_each_entry_reverse(page
, &pcp
->list
, lru
)
1085 if (page_private(page
) == migratetype
)
1088 list_for_each_entry(page
, &pcp
->list
, lru
)
1089 if (page_private(page
) == migratetype
)
1093 /* Allocate more to the pcp list if necessary */
1094 if (unlikely(&page
->lru
== &pcp
->list
)) {
1095 pcp
->count
+= rmqueue_bulk(zone
, 0,
1096 pcp
->batch
, &pcp
->list
, migratetype
);
1097 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1100 list_del(&page
->lru
);
1103 spin_lock_irqsave(&zone
->lock
, flags
);
1104 page
= __rmqueue(zone
, order
, migratetype
);
1105 spin_unlock(&zone
->lock
);
1110 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1111 zone_statistics(preferred_zone
, zone
);
1112 local_irq_restore(flags
);
1115 VM_BUG_ON(bad_range(zone
, page
));
1116 if (prep_new_page(page
, order
, gfp_flags
))
1121 local_irq_restore(flags
);
1126 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1127 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1128 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1129 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1130 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1131 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1132 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1134 #ifdef CONFIG_FAIL_PAGE_ALLOC
1136 static struct fail_page_alloc_attr
{
1137 struct fault_attr attr
;
1139 u32 ignore_gfp_highmem
;
1140 u32 ignore_gfp_wait
;
1143 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1145 struct dentry
*ignore_gfp_highmem_file
;
1146 struct dentry
*ignore_gfp_wait_file
;
1147 struct dentry
*min_order_file
;
1149 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1151 } fail_page_alloc
= {
1152 .attr
= FAULT_ATTR_INITIALIZER
,
1153 .ignore_gfp_wait
= 1,
1154 .ignore_gfp_highmem
= 1,
1158 static int __init
setup_fail_page_alloc(char *str
)
1160 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1162 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1164 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1166 if (order
< fail_page_alloc
.min_order
)
1168 if (gfp_mask
& __GFP_NOFAIL
)
1170 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1172 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1175 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1178 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1180 static int __init
fail_page_alloc_debugfs(void)
1182 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1186 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1190 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1192 fail_page_alloc
.ignore_gfp_wait_file
=
1193 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1194 &fail_page_alloc
.ignore_gfp_wait
);
1196 fail_page_alloc
.ignore_gfp_highmem_file
=
1197 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1198 &fail_page_alloc
.ignore_gfp_highmem
);
1199 fail_page_alloc
.min_order_file
=
1200 debugfs_create_u32("min-order", mode
, dir
,
1201 &fail_page_alloc
.min_order
);
1203 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1204 !fail_page_alloc
.ignore_gfp_highmem_file
||
1205 !fail_page_alloc
.min_order_file
) {
1207 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1208 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1209 debugfs_remove(fail_page_alloc
.min_order_file
);
1210 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1216 late_initcall(fail_page_alloc_debugfs
);
1218 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1220 #else /* CONFIG_FAIL_PAGE_ALLOC */
1222 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1227 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1230 * Return 1 if free pages are above 'mark'. This takes into account the order
1231 * of the allocation.
1233 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1234 int classzone_idx
, int alloc_flags
)
1236 /* free_pages my go negative - that's OK */
1238 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1241 if (alloc_flags
& ALLOC_HIGH
)
1243 if (alloc_flags
& ALLOC_HARDER
)
1246 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1248 for (o
= 0; o
< order
; o
++) {
1249 /* At the next order, this order's pages become unavailable */
1250 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1252 /* Require fewer higher order pages to be free */
1255 if (free_pages
<= min
)
1263 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1264 * skip over zones that are not allowed by the cpuset, or that have
1265 * been recently (in last second) found to be nearly full. See further
1266 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1267 * that have to skip over a lot of full or unallowed zones.
1269 * If the zonelist cache is present in the passed in zonelist, then
1270 * returns a pointer to the allowed node mask (either the current
1271 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1273 * If the zonelist cache is not available for this zonelist, does
1274 * nothing and returns NULL.
1276 * If the fullzones BITMAP in the zonelist cache is stale (more than
1277 * a second since last zap'd) then we zap it out (clear its bits.)
1279 * We hold off even calling zlc_setup, until after we've checked the
1280 * first zone in the zonelist, on the theory that most allocations will
1281 * be satisfied from that first zone, so best to examine that zone as
1282 * quickly as we can.
1284 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1286 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1287 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1289 zlc
= zonelist
->zlcache_ptr
;
1293 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1294 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1295 zlc
->last_full_zap
= jiffies
;
1298 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1299 &cpuset_current_mems_allowed
:
1300 &node_states
[N_HIGH_MEMORY
];
1301 return allowednodes
;
1305 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1306 * if it is worth looking at further for free memory:
1307 * 1) Check that the zone isn't thought to be full (doesn't have its
1308 * bit set in the zonelist_cache fullzones BITMAP).
1309 * 2) Check that the zones node (obtained from the zonelist_cache
1310 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1311 * Return true (non-zero) if zone is worth looking at further, or
1312 * else return false (zero) if it is not.
1314 * This check -ignores- the distinction between various watermarks,
1315 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1316 * found to be full for any variation of these watermarks, it will
1317 * be considered full for up to one second by all requests, unless
1318 * we are so low on memory on all allowed nodes that we are forced
1319 * into the second scan of the zonelist.
1321 * In the second scan we ignore this zonelist cache and exactly
1322 * apply the watermarks to all zones, even it is slower to do so.
1323 * We are low on memory in the second scan, and should leave no stone
1324 * unturned looking for a free page.
1326 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1327 nodemask_t
*allowednodes
)
1329 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1330 int i
; /* index of *z in zonelist zones */
1331 int n
; /* node that zone *z is on */
1333 zlc
= zonelist
->zlcache_ptr
;
1337 i
= z
- zonelist
->_zonerefs
;
1340 /* This zone is worth trying if it is allowed but not full */
1341 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1345 * Given 'z' scanning a zonelist, set the corresponding bit in
1346 * zlc->fullzones, so that subsequent attempts to allocate a page
1347 * from that zone don't waste time re-examining it.
1349 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1351 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1352 int i
; /* index of *z in zonelist zones */
1354 zlc
= zonelist
->zlcache_ptr
;
1358 i
= z
- zonelist
->_zonerefs
;
1360 set_bit(i
, zlc
->fullzones
);
1363 #else /* CONFIG_NUMA */
1365 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1370 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1371 nodemask_t
*allowednodes
)
1376 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1379 #endif /* CONFIG_NUMA */
1382 * get_page_from_freelist goes through the zonelist trying to allocate
1385 static struct page
*
1386 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1387 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
)
1390 struct page
*page
= NULL
;
1392 struct zone
*zone
, *preferred_zone
;
1393 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1394 int zlc_active
= 0; /* set if using zonelist_cache */
1395 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1397 (void)first_zones_zonelist(zonelist
, high_zoneidx
, nodemask
,
1399 if (!preferred_zone
)
1402 classzone_idx
= zone_idx(preferred_zone
);
1404 if (WARN_ON_ONCE(order
>= MAX_ORDER
))
1409 * Scan zonelist, looking for a zone with enough free.
1410 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1412 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1413 high_zoneidx
, nodemask
) {
1414 if (NUMA_BUILD
&& zlc_active
&&
1415 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1417 if ((alloc_flags
& ALLOC_CPUSET
) &&
1418 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1421 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1423 if (alloc_flags
& ALLOC_WMARK_MIN
)
1424 mark
= zone
->pages_min
;
1425 else if (alloc_flags
& ALLOC_WMARK_LOW
)
1426 mark
= zone
->pages_low
;
1428 mark
= zone
->pages_high
;
1429 if (!zone_watermark_ok(zone
, order
, mark
,
1430 classzone_idx
, alloc_flags
)) {
1431 if (!zone_reclaim_mode
||
1432 !zone_reclaim(zone
, gfp_mask
, order
))
1433 goto this_zone_full
;
1437 page
= buffered_rmqueue(preferred_zone
, zone
, order
, gfp_mask
);
1442 zlc_mark_zone_full(zonelist
, z
);
1444 if (NUMA_BUILD
&& !did_zlc_setup
) {
1445 /* we do zlc_setup after the first zone is tried */
1446 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1452 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1453 /* Disable zlc cache for second zonelist scan */
1461 should_alloc_retry(gfp_t gfp_mask
, unsigned int order
,
1462 unsigned long pages_reclaimed
)
1464 /* Do not loop if specifically requested */
1465 if (gfp_mask
& __GFP_NORETRY
)
1469 * In this implementation, order <= PAGE_ALLOC_COSTLY_ORDER
1470 * means __GFP_NOFAIL, but that may not be true in other
1473 if (order
<= PAGE_ALLOC_COSTLY_ORDER
)
1477 * For order > PAGE_ALLOC_COSTLY_ORDER, if __GFP_REPEAT is
1478 * specified, then we retry until we no longer reclaim any pages
1479 * (above), or we've reclaimed an order of pages at least as
1480 * large as the allocation's order. In both cases, if the
1481 * allocation still fails, we stop retrying.
1483 if (gfp_mask
& __GFP_REPEAT
&& pages_reclaimed
< (1 << order
))
1487 * Don't let big-order allocations loop unless the caller
1488 * explicitly requests that.
1490 if (gfp_mask
& __GFP_NOFAIL
)
1496 static inline struct page
*
1497 __alloc_pages_may_oom(gfp_t gfp_mask
, unsigned int order
,
1498 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1499 nodemask_t
*nodemask
)
1503 /* Acquire the OOM killer lock for the zones in zonelist */
1504 if (!try_set_zone_oom(zonelist
, gfp_mask
)) {
1505 schedule_timeout_uninterruptible(1);
1510 * Go through the zonelist yet one more time, keep very high watermark
1511 * here, this is only to catch a parallel oom killing, we must fail if
1512 * we're still under heavy pressure.
1514 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
1515 order
, zonelist
, high_zoneidx
,
1516 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1520 /* The OOM killer will not help higher order allocs */
1521 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
1524 /* Exhausted what can be done so it's blamo time */
1525 out_of_memory(zonelist
, gfp_mask
, order
);
1528 clear_zonelist_oom(zonelist
, gfp_mask
);
1532 /* The really slow allocator path where we enter direct reclaim */
1533 static inline struct page
*
1534 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
1535 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1536 nodemask_t
*nodemask
, int alloc_flags
, unsigned long *did_some_progress
)
1538 struct page
*page
= NULL
;
1539 struct reclaim_state reclaim_state
;
1540 struct task_struct
*p
= current
;
1544 /* We now go into synchronous reclaim */
1545 cpuset_memory_pressure_bump();
1548 * The task's cpuset might have expanded its set of allowable nodes
1550 p
->flags
|= PF_MEMALLOC
;
1551 lockdep_set_current_reclaim_state(gfp_mask
);
1552 reclaim_state
.reclaimed_slab
= 0;
1553 p
->reclaim_state
= &reclaim_state
;
1555 *did_some_progress
= try_to_free_pages(zonelist
, order
, gfp_mask
, nodemask
);
1557 p
->reclaim_state
= NULL
;
1558 lockdep_clear_current_reclaim_state();
1559 p
->flags
&= ~PF_MEMALLOC
;
1566 if (likely(*did_some_progress
))
1567 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1568 zonelist
, high_zoneidx
, alloc_flags
);
1573 is_allocation_high_priority(struct task_struct
*p
, gfp_t gfp_mask
)
1575 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1582 * This is called in the allocator slow-path if the allocation request is of
1583 * sufficient urgency to ignore watermarks and take other desperate measures
1585 static inline struct page
*
1586 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
1587 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1588 nodemask_t
*nodemask
)
1593 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1594 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
);
1596 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
1597 congestion_wait(WRITE
, HZ
/50);
1598 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
1604 void wake_all_kswapd(unsigned int order
, struct zonelist
*zonelist
,
1605 enum zone_type high_zoneidx
)
1610 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
1611 wakeup_kswapd(zone
, order
);
1614 static inline struct page
*
1615 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
1616 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1617 nodemask_t
*nodemask
)
1619 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1620 struct page
*page
= NULL
;
1622 unsigned long pages_reclaimed
= 0;
1623 unsigned long did_some_progress
;
1624 struct task_struct
*p
= current
;
1627 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1628 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1629 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1630 * using a larger set of nodes after it has established that the
1631 * allowed per node queues are empty and that nodes are
1634 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1637 wake_all_kswapd(order
, zonelist
, high_zoneidx
);
1640 * OK, we're below the kswapd watermark and have kicked background
1641 * reclaim. Now things get more complex, so set up alloc_flags according
1642 * to how we want to proceed.
1644 * The caller may dip into page reserves a bit more if the caller
1645 * cannot run direct reclaim, or if the caller has realtime scheduling
1646 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1647 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1649 alloc_flags
= ALLOC_WMARK_MIN
;
1650 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1651 alloc_flags
|= ALLOC_HARDER
;
1652 if (gfp_mask
& __GFP_HIGH
)
1653 alloc_flags
|= ALLOC_HIGH
;
1655 alloc_flags
|= ALLOC_CPUSET
;
1659 * Go through the zonelist again. Let __GFP_HIGH and allocations
1660 * coming from realtime tasks go deeper into reserves.
1662 * This is the last chance, in general, before the goto nopage.
1663 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1664 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1666 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
1667 high_zoneidx
, alloc_flags
);
1672 /* Allocate without watermarks if the context allows */
1673 if (is_allocation_high_priority(p
, gfp_mask
)) {
1674 /* Do not dip into emergency reserves if specified */
1675 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1676 page
= __alloc_pages_high_priority(gfp_mask
, order
,
1677 zonelist
, high_zoneidx
, nodemask
);
1682 /* Ensure no recursion into the allocator */
1686 /* Atomic allocations - we can't balance anything */
1690 /* Try direct reclaim and then allocating */
1691 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
,
1692 zonelist
, high_zoneidx
,
1694 alloc_flags
, &did_some_progress
);
1699 * If we failed to make any progress reclaiming, then we are
1700 * running out of options and have to consider going OOM
1702 if (!did_some_progress
) {
1703 if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1704 page
= __alloc_pages_may_oom(gfp_mask
, order
,
1705 zonelist
, high_zoneidx
,
1711 * The OOM killer does not trigger for high-order allocations
1712 * but if no progress is being made, there are no other
1713 * options and retrying is unlikely to help
1715 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
1722 /* Check if we should retry the allocation */
1723 pages_reclaimed
+= did_some_progress
;
1724 if (should_alloc_retry(gfp_mask
, order
, pages_reclaimed
)) {
1725 /* Wait for some write requests to complete then retry */
1726 congestion_wait(WRITE
, HZ
/50);
1731 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1732 printk(KERN_WARNING
"%s: page allocation failure."
1733 " order:%d, mode:0x%x\n",
1734 p
->comm
, order
, gfp_mask
);
1744 * This is the 'heart' of the zoned buddy allocator.
1747 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
1748 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
1750 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
1753 lockdep_trace_alloc(gfp_mask
);
1755 might_sleep_if(gfp_mask
& __GFP_WAIT
);
1757 if (should_fail_alloc_page(gfp_mask
, order
))
1761 * Check the zones suitable for the gfp_mask contain at least one
1762 * valid zone. It's possible to have an empty zonelist as a result
1763 * of GFP_THISNODE and a memoryless node
1765 if (unlikely(!zonelist
->_zonerefs
->zone
))
1768 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
1769 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1770 if (unlikely(!page
))
1771 page
= __alloc_pages_slowpath(gfp_mask
, order
,
1772 zonelist
, high_zoneidx
, nodemask
);
1776 EXPORT_SYMBOL(__alloc_pages_nodemask
);
1779 * Common helper functions.
1781 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1784 page
= alloc_pages(gfp_mask
, order
);
1787 return (unsigned long) page_address(page
);
1790 EXPORT_SYMBOL(__get_free_pages
);
1792 unsigned long get_zeroed_page(gfp_t gfp_mask
)
1797 * get_zeroed_page() returns a 32-bit address, which cannot represent
1800 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1802 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1804 return (unsigned long) page_address(page
);
1808 EXPORT_SYMBOL(get_zeroed_page
);
1810 void __pagevec_free(struct pagevec
*pvec
)
1812 int i
= pagevec_count(pvec
);
1815 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1818 void __free_pages(struct page
*page
, unsigned int order
)
1820 if (put_page_testzero(page
)) {
1822 free_hot_page(page
);
1824 __free_pages_ok(page
, order
);
1828 EXPORT_SYMBOL(__free_pages
);
1830 void free_pages(unsigned long addr
, unsigned int order
)
1833 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1834 __free_pages(virt_to_page((void *)addr
), order
);
1838 EXPORT_SYMBOL(free_pages
);
1841 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1842 * @size: the number of bytes to allocate
1843 * @gfp_mask: GFP flags for the allocation
1845 * This function is similar to alloc_pages(), except that it allocates the
1846 * minimum number of pages to satisfy the request. alloc_pages() can only
1847 * allocate memory in power-of-two pages.
1849 * This function is also limited by MAX_ORDER.
1851 * Memory allocated by this function must be released by free_pages_exact().
1853 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
1855 unsigned int order
= get_order(size
);
1858 addr
= __get_free_pages(gfp_mask
, order
);
1860 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
1861 unsigned long used
= addr
+ PAGE_ALIGN(size
);
1863 split_page(virt_to_page(addr
), order
);
1864 while (used
< alloc_end
) {
1870 return (void *)addr
;
1872 EXPORT_SYMBOL(alloc_pages_exact
);
1875 * free_pages_exact - release memory allocated via alloc_pages_exact()
1876 * @virt: the value returned by alloc_pages_exact.
1877 * @size: size of allocation, same value as passed to alloc_pages_exact().
1879 * Release the memory allocated by a previous call to alloc_pages_exact.
1881 void free_pages_exact(void *virt
, size_t size
)
1883 unsigned long addr
= (unsigned long)virt
;
1884 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1886 while (addr
< end
) {
1891 EXPORT_SYMBOL(free_pages_exact
);
1893 static unsigned int nr_free_zone_pages(int offset
)
1898 /* Just pick one node, since fallback list is circular */
1899 unsigned int sum
= 0;
1901 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
1903 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
1904 unsigned long size
= zone
->present_pages
;
1905 unsigned long high
= zone
->pages_high
;
1914 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1916 unsigned int nr_free_buffer_pages(void)
1918 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1920 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
1923 * Amount of free RAM allocatable within all zones
1925 unsigned int nr_free_pagecache_pages(void)
1927 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
1930 static inline void show_node(struct zone
*zone
)
1933 printk("Node %d ", zone_to_nid(zone
));
1936 void si_meminfo(struct sysinfo
*val
)
1938 val
->totalram
= totalram_pages
;
1940 val
->freeram
= global_page_state(NR_FREE_PAGES
);
1941 val
->bufferram
= nr_blockdev_pages();
1942 val
->totalhigh
= totalhigh_pages
;
1943 val
->freehigh
= nr_free_highpages();
1944 val
->mem_unit
= PAGE_SIZE
;
1947 EXPORT_SYMBOL(si_meminfo
);
1950 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1952 pg_data_t
*pgdat
= NODE_DATA(nid
);
1954 val
->totalram
= pgdat
->node_present_pages
;
1955 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
1956 #ifdef CONFIG_HIGHMEM
1957 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1958 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
1964 val
->mem_unit
= PAGE_SIZE
;
1968 #define K(x) ((x) << (PAGE_SHIFT-10))
1971 * Show free area list (used inside shift_scroll-lock stuff)
1972 * We also calculate the percentage fragmentation. We do this by counting the
1973 * memory on each free list with the exception of the first item on the list.
1975 void show_free_areas(void)
1980 for_each_populated_zone(zone
) {
1982 printk("%s per-cpu:\n", zone
->name
);
1984 for_each_online_cpu(cpu
) {
1985 struct per_cpu_pageset
*pageset
;
1987 pageset
= zone_pcp(zone
, cpu
);
1989 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
1990 cpu
, pageset
->pcp
.high
,
1991 pageset
->pcp
.batch
, pageset
->pcp
.count
);
1995 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
1996 " inactive_file:%lu"
1997 //TODO: check/adjust line lengths
1998 #ifdef CONFIG_UNEVICTABLE_LRU
2001 " dirty:%lu writeback:%lu unstable:%lu\n"
2002 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
2003 global_page_state(NR_ACTIVE_ANON
),
2004 global_page_state(NR_ACTIVE_FILE
),
2005 global_page_state(NR_INACTIVE_ANON
),
2006 global_page_state(NR_INACTIVE_FILE
),
2007 #ifdef CONFIG_UNEVICTABLE_LRU
2008 global_page_state(NR_UNEVICTABLE
),
2010 global_page_state(NR_FILE_DIRTY
),
2011 global_page_state(NR_WRITEBACK
),
2012 global_page_state(NR_UNSTABLE_NFS
),
2013 global_page_state(NR_FREE_PAGES
),
2014 global_page_state(NR_SLAB_RECLAIMABLE
) +
2015 global_page_state(NR_SLAB_UNRECLAIMABLE
),
2016 global_page_state(NR_FILE_MAPPED
),
2017 global_page_state(NR_PAGETABLE
),
2018 global_page_state(NR_BOUNCE
));
2020 for_each_populated_zone(zone
) {
2029 " active_anon:%lukB"
2030 " inactive_anon:%lukB"
2031 " active_file:%lukB"
2032 " inactive_file:%lukB"
2033 #ifdef CONFIG_UNEVICTABLE_LRU
2034 " unevictable:%lukB"
2037 " pages_scanned:%lu"
2038 " all_unreclaimable? %s"
2041 K(zone_page_state(zone
, NR_FREE_PAGES
)),
2044 K(zone
->pages_high
),
2045 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
2046 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
2047 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
2048 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
2049 #ifdef CONFIG_UNEVICTABLE_LRU
2050 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
2052 K(zone
->present_pages
),
2053 zone
->pages_scanned
,
2054 (zone_is_all_unreclaimable(zone
) ? "yes" : "no")
2056 printk("lowmem_reserve[]:");
2057 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2058 printk(" %lu", zone
->lowmem_reserve
[i
]);
2062 for_each_populated_zone(zone
) {
2063 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
2066 printk("%s: ", zone
->name
);
2068 spin_lock_irqsave(&zone
->lock
, flags
);
2069 for (order
= 0; order
< MAX_ORDER
; order
++) {
2070 nr
[order
] = zone
->free_area
[order
].nr_free
;
2071 total
+= nr
[order
] << order
;
2073 spin_unlock_irqrestore(&zone
->lock
, flags
);
2074 for (order
= 0; order
< MAX_ORDER
; order
++)
2075 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
2076 printk("= %lukB\n", K(total
));
2079 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
2081 show_swap_cache_info();
2084 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
2086 zoneref
->zone
= zone
;
2087 zoneref
->zone_idx
= zone_idx(zone
);
2091 * Builds allocation fallback zone lists.
2093 * Add all populated zones of a node to the zonelist.
2095 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
2096 int nr_zones
, enum zone_type zone_type
)
2100 BUG_ON(zone_type
>= MAX_NR_ZONES
);
2105 zone
= pgdat
->node_zones
+ zone_type
;
2106 if (populated_zone(zone
)) {
2107 zoneref_set_zone(zone
,
2108 &zonelist
->_zonerefs
[nr_zones
++]);
2109 check_highest_zone(zone_type
);
2112 } while (zone_type
);
2119 * 0 = automatic detection of better ordering.
2120 * 1 = order by ([node] distance, -zonetype)
2121 * 2 = order by (-zonetype, [node] distance)
2123 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2124 * the same zonelist. So only NUMA can configure this param.
2126 #define ZONELIST_ORDER_DEFAULT 0
2127 #define ZONELIST_ORDER_NODE 1
2128 #define ZONELIST_ORDER_ZONE 2
2130 /* zonelist order in the kernel.
2131 * set_zonelist_order() will set this to NODE or ZONE.
2133 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2134 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
2138 /* The value user specified ....changed by config */
2139 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2140 /* string for sysctl */
2141 #define NUMA_ZONELIST_ORDER_LEN 16
2142 char numa_zonelist_order
[16] = "default";
2145 * interface for configure zonelist ordering.
2146 * command line option "numa_zonelist_order"
2147 * = "[dD]efault - default, automatic configuration.
2148 * = "[nN]ode - order by node locality, then by zone within node
2149 * = "[zZ]one - order by zone, then by locality within zone
2152 static int __parse_numa_zonelist_order(char *s
)
2154 if (*s
== 'd' || *s
== 'D') {
2155 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2156 } else if (*s
== 'n' || *s
== 'N') {
2157 user_zonelist_order
= ZONELIST_ORDER_NODE
;
2158 } else if (*s
== 'z' || *s
== 'Z') {
2159 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
2162 "Ignoring invalid numa_zonelist_order value: "
2169 static __init
int setup_numa_zonelist_order(char *s
)
2172 return __parse_numa_zonelist_order(s
);
2175 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
2178 * sysctl handler for numa_zonelist_order
2180 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
2181 struct file
*file
, void __user
*buffer
, size_t *length
,
2184 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2188 strncpy(saved_string
, (char*)table
->data
,
2189 NUMA_ZONELIST_ORDER_LEN
);
2190 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
2194 int oldval
= user_zonelist_order
;
2195 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2197 * bogus value. restore saved string
2199 strncpy((char*)table
->data
, saved_string
,
2200 NUMA_ZONELIST_ORDER_LEN
);
2201 user_zonelist_order
= oldval
;
2202 } else if (oldval
!= user_zonelist_order
)
2203 build_all_zonelists();
2209 #define MAX_NODE_LOAD (num_online_nodes())
2210 static int node_load
[MAX_NUMNODES
];
2213 * find_next_best_node - find the next node that should appear in a given node's fallback list
2214 * @node: node whose fallback list we're appending
2215 * @used_node_mask: nodemask_t of already used nodes
2217 * We use a number of factors to determine which is the next node that should
2218 * appear on a given node's fallback list. The node should not have appeared
2219 * already in @node's fallback list, and it should be the next closest node
2220 * according to the distance array (which contains arbitrary distance values
2221 * from each node to each node in the system), and should also prefer nodes
2222 * with no CPUs, since presumably they'll have very little allocation pressure
2223 * on them otherwise.
2224 * It returns -1 if no node is found.
2226 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2229 int min_val
= INT_MAX
;
2231 const struct cpumask
*tmp
= cpumask_of_node(0);
2233 /* Use the local node if we haven't already */
2234 if (!node_isset(node
, *used_node_mask
)) {
2235 node_set(node
, *used_node_mask
);
2239 for_each_node_state(n
, N_HIGH_MEMORY
) {
2241 /* Don't want a node to appear more than once */
2242 if (node_isset(n
, *used_node_mask
))
2245 /* Use the distance array to find the distance */
2246 val
= node_distance(node
, n
);
2248 /* Penalize nodes under us ("prefer the next node") */
2251 /* Give preference to headless and unused nodes */
2252 tmp
= cpumask_of_node(n
);
2253 if (!cpumask_empty(tmp
))
2254 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2256 /* Slight preference for less loaded node */
2257 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2258 val
+= node_load
[n
];
2260 if (val
< min_val
) {
2267 node_set(best_node
, *used_node_mask
);
2274 * Build zonelists ordered by node and zones within node.
2275 * This results in maximum locality--normal zone overflows into local
2276 * DMA zone, if any--but risks exhausting DMA zone.
2278 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2281 struct zonelist
*zonelist
;
2283 zonelist
= &pgdat
->node_zonelists
[0];
2284 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
2286 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2288 zonelist
->_zonerefs
[j
].zone
= NULL
;
2289 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2293 * Build gfp_thisnode zonelists
2295 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2298 struct zonelist
*zonelist
;
2300 zonelist
= &pgdat
->node_zonelists
[1];
2301 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2302 zonelist
->_zonerefs
[j
].zone
= NULL
;
2303 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2307 * Build zonelists ordered by zone and nodes within zones.
2308 * This results in conserving DMA zone[s] until all Normal memory is
2309 * exhausted, but results in overflowing to remote node while memory
2310 * may still exist in local DMA zone.
2312 static int node_order
[MAX_NUMNODES
];
2314 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2317 int zone_type
; /* needs to be signed */
2319 struct zonelist
*zonelist
;
2321 zonelist
= &pgdat
->node_zonelists
[0];
2323 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
2324 for (j
= 0; j
< nr_nodes
; j
++) {
2325 node
= node_order
[j
];
2326 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2327 if (populated_zone(z
)) {
2329 &zonelist
->_zonerefs
[pos
++]);
2330 check_highest_zone(zone_type
);
2334 zonelist
->_zonerefs
[pos
].zone
= NULL
;
2335 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
2338 static int default_zonelist_order(void)
2341 unsigned long low_kmem_size
,total_size
;
2345 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2346 * If they are really small and used heavily, the system can fall
2347 * into OOM very easily.
2348 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2350 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2353 for_each_online_node(nid
) {
2354 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2355 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2356 if (populated_zone(z
)) {
2357 if (zone_type
< ZONE_NORMAL
)
2358 low_kmem_size
+= z
->present_pages
;
2359 total_size
+= z
->present_pages
;
2363 if (!low_kmem_size
|| /* there are no DMA area. */
2364 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2365 return ZONELIST_ORDER_NODE
;
2367 * look into each node's config.
2368 * If there is a node whose DMA/DMA32 memory is very big area on
2369 * local memory, NODE_ORDER may be suitable.
2371 average_size
= total_size
/
2372 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2373 for_each_online_node(nid
) {
2376 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2377 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2378 if (populated_zone(z
)) {
2379 if (zone_type
< ZONE_NORMAL
)
2380 low_kmem_size
+= z
->present_pages
;
2381 total_size
+= z
->present_pages
;
2384 if (low_kmem_size
&&
2385 total_size
> average_size
&& /* ignore small node */
2386 low_kmem_size
> total_size
* 70/100)
2387 return ZONELIST_ORDER_NODE
;
2389 return ZONELIST_ORDER_ZONE
;
2392 static void set_zonelist_order(void)
2394 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2395 current_zonelist_order
= default_zonelist_order();
2397 current_zonelist_order
= user_zonelist_order
;
2400 static void build_zonelists(pg_data_t
*pgdat
)
2404 nodemask_t used_mask
;
2405 int local_node
, prev_node
;
2406 struct zonelist
*zonelist
;
2407 int order
= current_zonelist_order
;
2409 /* initialize zonelists */
2410 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2411 zonelist
= pgdat
->node_zonelists
+ i
;
2412 zonelist
->_zonerefs
[0].zone
= NULL
;
2413 zonelist
->_zonerefs
[0].zone_idx
= 0;
2416 /* NUMA-aware ordering of nodes */
2417 local_node
= pgdat
->node_id
;
2418 load
= num_online_nodes();
2419 prev_node
= local_node
;
2420 nodes_clear(used_mask
);
2422 memset(node_load
, 0, sizeof(node_load
));
2423 memset(node_order
, 0, sizeof(node_order
));
2426 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2427 int distance
= node_distance(local_node
, node
);
2430 * If another node is sufficiently far away then it is better
2431 * to reclaim pages in a zone before going off node.
2433 if (distance
> RECLAIM_DISTANCE
)
2434 zone_reclaim_mode
= 1;
2437 * We don't want to pressure a particular node.
2438 * So adding penalty to the first node in same
2439 * distance group to make it round-robin.
2441 if (distance
!= node_distance(local_node
, prev_node
))
2442 node_load
[node
] = load
;
2446 if (order
== ZONELIST_ORDER_NODE
)
2447 build_zonelists_in_node_order(pgdat
, node
);
2449 node_order
[j
++] = node
; /* remember order */
2452 if (order
== ZONELIST_ORDER_ZONE
) {
2453 /* calculate node order -- i.e., DMA last! */
2454 build_zonelists_in_zone_order(pgdat
, j
);
2457 build_thisnode_zonelists(pgdat
);
2460 /* Construct the zonelist performance cache - see further mmzone.h */
2461 static void build_zonelist_cache(pg_data_t
*pgdat
)
2463 struct zonelist
*zonelist
;
2464 struct zonelist_cache
*zlc
;
2467 zonelist
= &pgdat
->node_zonelists
[0];
2468 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2469 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2470 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
2471 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
2475 #else /* CONFIG_NUMA */
2477 static void set_zonelist_order(void)
2479 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2482 static void build_zonelists(pg_data_t
*pgdat
)
2484 int node
, local_node
;
2486 struct zonelist
*zonelist
;
2488 local_node
= pgdat
->node_id
;
2490 zonelist
= &pgdat
->node_zonelists
[0];
2491 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2494 * Now we build the zonelist so that it contains the zones
2495 * of all the other nodes.
2496 * We don't want to pressure a particular node, so when
2497 * building the zones for node N, we make sure that the
2498 * zones coming right after the local ones are those from
2499 * node N+1 (modulo N)
2501 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2502 if (!node_online(node
))
2504 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2507 for (node
= 0; node
< local_node
; node
++) {
2508 if (!node_online(node
))
2510 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2514 zonelist
->_zonerefs
[j
].zone
= NULL
;
2515 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2518 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2519 static void build_zonelist_cache(pg_data_t
*pgdat
)
2521 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
2524 #endif /* CONFIG_NUMA */
2526 /* return values int ....just for stop_machine() */
2527 static int __build_all_zonelists(void *dummy
)
2531 for_each_online_node(nid
) {
2532 pg_data_t
*pgdat
= NODE_DATA(nid
);
2534 build_zonelists(pgdat
);
2535 build_zonelist_cache(pgdat
);
2540 void build_all_zonelists(void)
2542 set_zonelist_order();
2544 if (system_state
== SYSTEM_BOOTING
) {
2545 __build_all_zonelists(NULL
);
2546 mminit_verify_zonelist();
2547 cpuset_init_current_mems_allowed();
2549 /* we have to stop all cpus to guarantee there is no user
2551 stop_machine(__build_all_zonelists
, NULL
, NULL
);
2552 /* cpuset refresh routine should be here */
2554 vm_total_pages
= nr_free_pagecache_pages();
2556 * Disable grouping by mobility if the number of pages in the
2557 * system is too low to allow the mechanism to work. It would be
2558 * more accurate, but expensive to check per-zone. This check is
2559 * made on memory-hotadd so a system can start with mobility
2560 * disabled and enable it later
2562 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
2563 page_group_by_mobility_disabled
= 1;
2565 page_group_by_mobility_disabled
= 0;
2567 printk("Built %i zonelists in %s order, mobility grouping %s. "
2568 "Total pages: %ld\n",
2570 zonelist_order_name
[current_zonelist_order
],
2571 page_group_by_mobility_disabled
? "off" : "on",
2574 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2579 * Helper functions to size the waitqueue hash table.
2580 * Essentially these want to choose hash table sizes sufficiently
2581 * large so that collisions trying to wait on pages are rare.
2582 * But in fact, the number of active page waitqueues on typical
2583 * systems is ridiculously low, less than 200. So this is even
2584 * conservative, even though it seems large.
2586 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2587 * waitqueues, i.e. the size of the waitq table given the number of pages.
2589 #define PAGES_PER_WAITQUEUE 256
2591 #ifndef CONFIG_MEMORY_HOTPLUG
2592 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2594 unsigned long size
= 1;
2596 pages
/= PAGES_PER_WAITQUEUE
;
2598 while (size
< pages
)
2602 * Once we have dozens or even hundreds of threads sleeping
2603 * on IO we've got bigger problems than wait queue collision.
2604 * Limit the size of the wait table to a reasonable size.
2606 size
= min(size
, 4096UL);
2608 return max(size
, 4UL);
2612 * A zone's size might be changed by hot-add, so it is not possible to determine
2613 * a suitable size for its wait_table. So we use the maximum size now.
2615 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2617 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2618 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2619 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2621 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2622 * or more by the traditional way. (See above). It equals:
2624 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2625 * ia64(16K page size) : = ( 8G + 4M)byte.
2626 * powerpc (64K page size) : = (32G +16M)byte.
2628 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2635 * This is an integer logarithm so that shifts can be used later
2636 * to extract the more random high bits from the multiplicative
2637 * hash function before the remainder is taken.
2639 static inline unsigned long wait_table_bits(unsigned long size
)
2644 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2647 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2648 * of blocks reserved is based on zone->pages_min. The memory within the
2649 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2650 * higher will lead to a bigger reserve which will get freed as contiguous
2651 * blocks as reclaim kicks in
2653 static void setup_zone_migrate_reserve(struct zone
*zone
)
2655 unsigned long start_pfn
, pfn
, end_pfn
;
2657 unsigned long reserve
, block_migratetype
;
2659 /* Get the start pfn, end pfn and the number of blocks to reserve */
2660 start_pfn
= zone
->zone_start_pfn
;
2661 end_pfn
= start_pfn
+ zone
->spanned_pages
;
2662 reserve
= roundup(zone
->pages_min
, pageblock_nr_pages
) >>
2665 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
2666 if (!pfn_valid(pfn
))
2668 page
= pfn_to_page(pfn
);
2670 /* Watch out for overlapping nodes */
2671 if (page_to_nid(page
) != zone_to_nid(zone
))
2674 /* Blocks with reserved pages will never free, skip them. */
2675 if (PageReserved(page
))
2678 block_migratetype
= get_pageblock_migratetype(page
);
2680 /* If this block is reserved, account for it */
2681 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
2686 /* Suitable for reserving if this block is movable */
2687 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
2688 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
2689 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
2695 * If the reserve is met and this is a previous reserved block,
2698 if (block_migratetype
== MIGRATE_RESERVE
) {
2699 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2700 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
2706 * Initially all pages are reserved - free ones are freed
2707 * up by free_all_bootmem() once the early boot process is
2708 * done. Non-atomic initialization, single-pass.
2710 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2711 unsigned long start_pfn
, enum memmap_context context
)
2714 unsigned long end_pfn
= start_pfn
+ size
;
2718 if (highest_memmap_pfn
< end_pfn
- 1)
2719 highest_memmap_pfn
= end_pfn
- 1;
2721 z
= &NODE_DATA(nid
)->node_zones
[zone
];
2722 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2724 * There can be holes in boot-time mem_map[]s
2725 * handed to this function. They do not
2726 * exist on hotplugged memory.
2728 if (context
== MEMMAP_EARLY
) {
2729 if (!early_pfn_valid(pfn
))
2731 if (!early_pfn_in_nid(pfn
, nid
))
2734 page
= pfn_to_page(pfn
);
2735 set_page_links(page
, zone
, nid
, pfn
);
2736 mminit_verify_page_links(page
, zone
, nid
, pfn
);
2737 init_page_count(page
);
2738 reset_page_mapcount(page
);
2739 SetPageReserved(page
);
2741 * Mark the block movable so that blocks are reserved for
2742 * movable at startup. This will force kernel allocations
2743 * to reserve their blocks rather than leaking throughout
2744 * the address space during boot when many long-lived
2745 * kernel allocations are made. Later some blocks near
2746 * the start are marked MIGRATE_RESERVE by
2747 * setup_zone_migrate_reserve()
2749 * bitmap is created for zone's valid pfn range. but memmap
2750 * can be created for invalid pages (for alignment)
2751 * check here not to call set_pageblock_migratetype() against
2754 if ((z
->zone_start_pfn
<= pfn
)
2755 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
2756 && !(pfn
& (pageblock_nr_pages
- 1)))
2757 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2759 INIT_LIST_HEAD(&page
->lru
);
2760 #ifdef WANT_PAGE_VIRTUAL
2761 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2762 if (!is_highmem_idx(zone
))
2763 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2768 static void __meminit
zone_init_free_lists(struct zone
*zone
)
2771 for_each_migratetype_order(order
, t
) {
2772 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
2773 zone
->free_area
[order
].nr_free
= 0;
2777 #ifndef __HAVE_ARCH_MEMMAP_INIT
2778 #define memmap_init(size, nid, zone, start_pfn) \
2779 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2782 static int zone_batchsize(struct zone
*zone
)
2788 * The per-cpu-pages pools are set to around 1000th of the
2789 * size of the zone. But no more than 1/2 of a meg.
2791 * OK, so we don't know how big the cache is. So guess.
2793 batch
= zone
->present_pages
/ 1024;
2794 if (batch
* PAGE_SIZE
> 512 * 1024)
2795 batch
= (512 * 1024) / PAGE_SIZE
;
2796 batch
/= 4; /* We effectively *= 4 below */
2801 * Clamp the batch to a 2^n - 1 value. Having a power
2802 * of 2 value was found to be more likely to have
2803 * suboptimal cache aliasing properties in some cases.
2805 * For example if 2 tasks are alternately allocating
2806 * batches of pages, one task can end up with a lot
2807 * of pages of one half of the possible page colors
2808 * and the other with pages of the other colors.
2810 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
2815 /* The deferral and batching of frees should be suppressed under NOMMU
2818 * The problem is that NOMMU needs to be able to allocate large chunks
2819 * of contiguous memory as there's no hardware page translation to
2820 * assemble apparent contiguous memory from discontiguous pages.
2822 * Queueing large contiguous runs of pages for batching, however,
2823 * causes the pages to actually be freed in smaller chunks. As there
2824 * can be a significant delay between the individual batches being
2825 * recycled, this leads to the once large chunks of space being
2826 * fragmented and becoming unavailable for high-order allocations.
2832 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2834 struct per_cpu_pages
*pcp
;
2836 memset(p
, 0, sizeof(*p
));
2840 pcp
->high
= 6 * batch
;
2841 pcp
->batch
= max(1UL, 1 * batch
);
2842 INIT_LIST_HEAD(&pcp
->list
);
2846 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2847 * to the value high for the pageset p.
2850 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2853 struct per_cpu_pages
*pcp
;
2857 pcp
->batch
= max(1UL, high
/4);
2858 if ((high
/4) > (PAGE_SHIFT
* 8))
2859 pcp
->batch
= PAGE_SHIFT
* 8;
2865 * Boot pageset table. One per cpu which is going to be used for all
2866 * zones and all nodes. The parameters will be set in such a way
2867 * that an item put on a list will immediately be handed over to
2868 * the buddy list. This is safe since pageset manipulation is done
2869 * with interrupts disabled.
2871 * Some NUMA counter updates may also be caught by the boot pagesets.
2873 * The boot_pagesets must be kept even after bootup is complete for
2874 * unused processors and/or zones. They do play a role for bootstrapping
2875 * hotplugged processors.
2877 * zoneinfo_show() and maybe other functions do
2878 * not check if the processor is online before following the pageset pointer.
2879 * Other parts of the kernel may not check if the zone is available.
2881 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
2884 * Dynamically allocate memory for the
2885 * per cpu pageset array in struct zone.
2887 static int __cpuinit
process_zones(int cpu
)
2889 struct zone
*zone
, *dzone
;
2890 int node
= cpu_to_node(cpu
);
2892 node_set_state(node
, N_CPU
); /* this node has a cpu */
2894 for_each_populated_zone(zone
) {
2895 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
2897 if (!zone_pcp(zone
, cpu
))
2900 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
2902 if (percpu_pagelist_fraction
)
2903 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
2904 (zone
->present_pages
/ percpu_pagelist_fraction
));
2909 for_each_zone(dzone
) {
2910 if (!populated_zone(dzone
))
2914 kfree(zone_pcp(dzone
, cpu
));
2915 zone_pcp(dzone
, cpu
) = NULL
;
2920 static inline void free_zone_pagesets(int cpu
)
2924 for_each_zone(zone
) {
2925 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
2927 /* Free per_cpu_pageset if it is slab allocated */
2928 if (pset
!= &boot_pageset
[cpu
])
2930 zone_pcp(zone
, cpu
) = NULL
;
2934 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
2935 unsigned long action
,
2938 int cpu
= (long)hcpu
;
2939 int ret
= NOTIFY_OK
;
2942 case CPU_UP_PREPARE
:
2943 case CPU_UP_PREPARE_FROZEN
:
2944 if (process_zones(cpu
))
2947 case CPU_UP_CANCELED
:
2948 case CPU_UP_CANCELED_FROZEN
:
2950 case CPU_DEAD_FROZEN
:
2951 free_zone_pagesets(cpu
);
2959 static struct notifier_block __cpuinitdata pageset_notifier
=
2960 { &pageset_cpuup_callback
, NULL
, 0 };
2962 void __init
setup_per_cpu_pageset(void)
2966 /* Initialize per_cpu_pageset for cpu 0.
2967 * A cpuup callback will do this for every cpu
2968 * as it comes online
2970 err
= process_zones(smp_processor_id());
2972 register_cpu_notifier(&pageset_notifier
);
2977 static noinline __init_refok
2978 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
2981 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2985 * The per-page waitqueue mechanism uses hashed waitqueues
2988 zone
->wait_table_hash_nr_entries
=
2989 wait_table_hash_nr_entries(zone_size_pages
);
2990 zone
->wait_table_bits
=
2991 wait_table_bits(zone
->wait_table_hash_nr_entries
);
2992 alloc_size
= zone
->wait_table_hash_nr_entries
2993 * sizeof(wait_queue_head_t
);
2995 if (!slab_is_available()) {
2996 zone
->wait_table
= (wait_queue_head_t
*)
2997 alloc_bootmem_node(pgdat
, alloc_size
);
3000 * This case means that a zone whose size was 0 gets new memory
3001 * via memory hot-add.
3002 * But it may be the case that a new node was hot-added. In
3003 * this case vmalloc() will not be able to use this new node's
3004 * memory - this wait_table must be initialized to use this new
3005 * node itself as well.
3006 * To use this new node's memory, further consideration will be
3009 zone
->wait_table
= vmalloc(alloc_size
);
3011 if (!zone
->wait_table
)
3014 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
3015 init_waitqueue_head(zone
->wait_table
+ i
);
3020 static __meminit
void zone_pcp_init(struct zone
*zone
)
3023 unsigned long batch
= zone_batchsize(zone
);
3025 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
3027 /* Early boot. Slab allocator not functional yet */
3028 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
3029 setup_pageset(&boot_pageset
[cpu
],0);
3031 setup_pageset(zone_pcp(zone
,cpu
), batch
);
3034 if (zone
->present_pages
)
3035 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
3036 zone
->name
, zone
->present_pages
, batch
);
3039 __meminit
int init_currently_empty_zone(struct zone
*zone
,
3040 unsigned long zone_start_pfn
,
3042 enum memmap_context context
)
3044 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3046 ret
= zone_wait_table_init(zone
, size
);
3049 pgdat
->nr_zones
= zone_idx(zone
) + 1;
3051 zone
->zone_start_pfn
= zone_start_pfn
;
3053 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
3054 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3056 (unsigned long)zone_idx(zone
),
3057 zone_start_pfn
, (zone_start_pfn
+ size
));
3059 zone_init_free_lists(zone
);
3064 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3066 * Basic iterator support. Return the first range of PFNs for a node
3067 * Note: nid == MAX_NUMNODES returns first region regardless of node
3069 static int __meminit
first_active_region_index_in_nid(int nid
)
3073 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3074 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
3081 * Basic iterator support. Return the next active range of PFNs for a node
3082 * Note: nid == MAX_NUMNODES returns next region regardless of node
3084 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
3086 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
3087 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
3093 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3095 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3096 * Architectures may implement their own version but if add_active_range()
3097 * was used and there are no special requirements, this is a convenient
3100 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
3104 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3105 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
3106 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3108 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
3109 return early_node_map
[i
].nid
;
3111 /* This is a memory hole */
3114 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3116 int __meminit
early_pfn_to_nid(unsigned long pfn
)
3120 nid
= __early_pfn_to_nid(pfn
);
3123 /* just returns 0 */
3127 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3128 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
3132 nid
= __early_pfn_to_nid(pfn
);
3133 if (nid
>= 0 && nid
!= node
)
3139 /* Basic iterator support to walk early_node_map[] */
3140 #define for_each_active_range_index_in_nid(i, nid) \
3141 for (i = first_active_region_index_in_nid(nid); i != -1; \
3142 i = next_active_region_index_in_nid(i, nid))
3145 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3146 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3147 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3149 * If an architecture guarantees that all ranges registered with
3150 * add_active_ranges() contain no holes and may be freed, this
3151 * this function may be used instead of calling free_bootmem() manually.
3153 void __init
free_bootmem_with_active_regions(int nid
,
3154 unsigned long max_low_pfn
)
3158 for_each_active_range_index_in_nid(i
, nid
) {
3159 unsigned long size_pages
= 0;
3160 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3162 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
3165 if (end_pfn
> max_low_pfn
)
3166 end_pfn
= max_low_pfn
;
3168 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
3169 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
3170 PFN_PHYS(early_node_map
[i
].start_pfn
),
3171 size_pages
<< PAGE_SHIFT
);
3175 void __init
work_with_active_regions(int nid
, work_fn_t work_fn
, void *data
)
3180 for_each_active_range_index_in_nid(i
, nid
) {
3181 ret
= work_fn(early_node_map
[i
].start_pfn
,
3182 early_node_map
[i
].end_pfn
, data
);
3188 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3189 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3191 * If an architecture guarantees that all ranges registered with
3192 * add_active_ranges() contain no holes and may be freed, this
3193 * function may be used instead of calling memory_present() manually.
3195 void __init
sparse_memory_present_with_active_regions(int nid
)
3199 for_each_active_range_index_in_nid(i
, nid
)
3200 memory_present(early_node_map
[i
].nid
,
3201 early_node_map
[i
].start_pfn
,
3202 early_node_map
[i
].end_pfn
);
3206 * get_pfn_range_for_nid - Return the start and end page frames for a node
3207 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3208 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3209 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3211 * It returns the start and end page frame of a node based on information
3212 * provided by an arch calling add_active_range(). If called for a node
3213 * with no available memory, a warning is printed and the start and end
3216 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3217 unsigned long *start_pfn
, unsigned long *end_pfn
)
3223 for_each_active_range_index_in_nid(i
, nid
) {
3224 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3225 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3228 if (*start_pfn
== -1UL)
3233 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3234 * assumption is made that zones within a node are ordered in monotonic
3235 * increasing memory addresses so that the "highest" populated zone is used
3237 static void __init
find_usable_zone_for_movable(void)
3240 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3241 if (zone_index
== ZONE_MOVABLE
)
3244 if (arch_zone_highest_possible_pfn
[zone_index
] >
3245 arch_zone_lowest_possible_pfn
[zone_index
])
3249 VM_BUG_ON(zone_index
== -1);
3250 movable_zone
= zone_index
;
3254 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3255 * because it is sized independant of architecture. Unlike the other zones,
3256 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3257 * in each node depending on the size of each node and how evenly kernelcore
3258 * is distributed. This helper function adjusts the zone ranges
3259 * provided by the architecture for a given node by using the end of the
3260 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3261 * zones within a node are in order of monotonic increases memory addresses
3263 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3264 unsigned long zone_type
,
3265 unsigned long node_start_pfn
,
3266 unsigned long node_end_pfn
,
3267 unsigned long *zone_start_pfn
,
3268 unsigned long *zone_end_pfn
)
3270 /* Only adjust if ZONE_MOVABLE is on this node */
3271 if (zone_movable_pfn
[nid
]) {
3272 /* Size ZONE_MOVABLE */
3273 if (zone_type
== ZONE_MOVABLE
) {
3274 *zone_start_pfn
= zone_movable_pfn
[nid
];
3275 *zone_end_pfn
= min(node_end_pfn
,
3276 arch_zone_highest_possible_pfn
[movable_zone
]);
3278 /* Adjust for ZONE_MOVABLE starting within this range */
3279 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3280 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3281 *zone_end_pfn
= zone_movable_pfn
[nid
];
3283 /* Check if this whole range is within ZONE_MOVABLE */
3284 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3285 *zone_start_pfn
= *zone_end_pfn
;
3290 * Return the number of pages a zone spans in a node, including holes
3291 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3293 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3294 unsigned long zone_type
,
3295 unsigned long *ignored
)
3297 unsigned long node_start_pfn
, node_end_pfn
;
3298 unsigned long zone_start_pfn
, zone_end_pfn
;
3300 /* Get the start and end of the node and zone */
3301 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3302 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3303 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3304 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3305 node_start_pfn
, node_end_pfn
,
3306 &zone_start_pfn
, &zone_end_pfn
);
3308 /* Check that this node has pages within the zone's required range */
3309 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3312 /* Move the zone boundaries inside the node if necessary */
3313 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3314 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3316 /* Return the spanned pages */
3317 return zone_end_pfn
- zone_start_pfn
;
3321 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3322 * then all holes in the requested range will be accounted for.
3324 static unsigned long __meminit
__absent_pages_in_range(int nid
,
3325 unsigned long range_start_pfn
,
3326 unsigned long range_end_pfn
)
3329 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3330 unsigned long start_pfn
;
3332 /* Find the end_pfn of the first active range of pfns in the node */
3333 i
= first_active_region_index_in_nid(nid
);
3337 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3339 /* Account for ranges before physical memory on this node */
3340 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3341 hole_pages
= prev_end_pfn
- range_start_pfn
;
3343 /* Find all holes for the zone within the node */
3344 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3346 /* No need to continue if prev_end_pfn is outside the zone */
3347 if (prev_end_pfn
>= range_end_pfn
)
3350 /* Make sure the end of the zone is not within the hole */
3351 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3352 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3354 /* Update the hole size cound and move on */
3355 if (start_pfn
> range_start_pfn
) {
3356 BUG_ON(prev_end_pfn
> start_pfn
);
3357 hole_pages
+= start_pfn
- prev_end_pfn
;
3359 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3362 /* Account for ranges past physical memory on this node */
3363 if (range_end_pfn
> prev_end_pfn
)
3364 hole_pages
+= range_end_pfn
-
3365 max(range_start_pfn
, prev_end_pfn
);
3371 * absent_pages_in_range - Return number of page frames in holes within a range
3372 * @start_pfn: The start PFN to start searching for holes
3373 * @end_pfn: The end PFN to stop searching for holes
3375 * It returns the number of pages frames in memory holes within a range.
3377 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3378 unsigned long end_pfn
)
3380 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3383 /* Return the number of page frames in holes in a zone on a node */
3384 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3385 unsigned long zone_type
,
3386 unsigned long *ignored
)
3388 unsigned long node_start_pfn
, node_end_pfn
;
3389 unsigned long zone_start_pfn
, zone_end_pfn
;
3391 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3392 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3394 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3397 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3398 node_start_pfn
, node_end_pfn
,
3399 &zone_start_pfn
, &zone_end_pfn
);
3400 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3404 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3405 unsigned long zone_type
,
3406 unsigned long *zones_size
)
3408 return zones_size
[zone_type
];
3411 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3412 unsigned long zone_type
,
3413 unsigned long *zholes_size
)
3418 return zholes_size
[zone_type
];
3423 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3424 unsigned long *zones_size
, unsigned long *zholes_size
)
3426 unsigned long realtotalpages
, totalpages
= 0;
3429 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3430 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3432 pgdat
->node_spanned_pages
= totalpages
;
3434 realtotalpages
= totalpages
;
3435 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3437 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3439 pgdat
->node_present_pages
= realtotalpages
;
3440 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3444 #ifndef CONFIG_SPARSEMEM
3446 * Calculate the size of the zone->blockflags rounded to an unsigned long
3447 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3448 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3449 * round what is now in bits to nearest long in bits, then return it in
3452 static unsigned long __init
usemap_size(unsigned long zonesize
)
3454 unsigned long usemapsize
;
3456 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
3457 usemapsize
= usemapsize
>> pageblock_order
;
3458 usemapsize
*= NR_PAGEBLOCK_BITS
;
3459 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3461 return usemapsize
/ 8;
3464 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3465 struct zone
*zone
, unsigned long zonesize
)
3467 unsigned long usemapsize
= usemap_size(zonesize
);
3468 zone
->pageblock_flags
= NULL
;
3470 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3473 static void inline setup_usemap(struct pglist_data
*pgdat
,
3474 struct zone
*zone
, unsigned long zonesize
) {}
3475 #endif /* CONFIG_SPARSEMEM */
3477 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3479 /* Return a sensible default order for the pageblock size. */
3480 static inline int pageblock_default_order(void)
3482 if (HPAGE_SHIFT
> PAGE_SHIFT
)
3483 return HUGETLB_PAGE_ORDER
;
3488 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3489 static inline void __init
set_pageblock_order(unsigned int order
)
3491 /* Check that pageblock_nr_pages has not already been setup */
3492 if (pageblock_order
)
3496 * Assume the largest contiguous order of interest is a huge page.
3497 * This value may be variable depending on boot parameters on IA64
3499 pageblock_order
= order
;
3501 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3504 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3505 * and pageblock_default_order() are unused as pageblock_order is set
3506 * at compile-time. See include/linux/pageblock-flags.h for the values of
3507 * pageblock_order based on the kernel config
3509 static inline int pageblock_default_order(unsigned int order
)
3513 #define set_pageblock_order(x) do {} while (0)
3515 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3518 * Set up the zone data structures:
3519 * - mark all pages reserved
3520 * - mark all memory queues empty
3521 * - clear the memory bitmaps
3523 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
3524 unsigned long *zones_size
, unsigned long *zholes_size
)
3527 int nid
= pgdat
->node_id
;
3528 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
3531 pgdat_resize_init(pgdat
);
3532 pgdat
->nr_zones
= 0;
3533 init_waitqueue_head(&pgdat
->kswapd_wait
);
3534 pgdat
->kswapd_max_order
= 0;
3535 pgdat_page_cgroup_init(pgdat
);
3537 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3538 struct zone
*zone
= pgdat
->node_zones
+ j
;
3539 unsigned long size
, realsize
, memmap_pages
;
3542 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
3543 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
3547 * Adjust realsize so that it accounts for how much memory
3548 * is used by this zone for memmap. This affects the watermark
3549 * and per-cpu initialisations
3552 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
3553 if (realsize
>= memmap_pages
) {
3554 realsize
-= memmap_pages
;
3557 " %s zone: %lu pages used for memmap\n",
3558 zone_names
[j
], memmap_pages
);
3561 " %s zone: %lu pages exceeds realsize %lu\n",
3562 zone_names
[j
], memmap_pages
, realsize
);
3564 /* Account for reserved pages */
3565 if (j
== 0 && realsize
> dma_reserve
) {
3566 realsize
-= dma_reserve
;
3567 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
3568 zone_names
[0], dma_reserve
);
3571 if (!is_highmem_idx(j
))
3572 nr_kernel_pages
+= realsize
;
3573 nr_all_pages
+= realsize
;
3575 zone
->spanned_pages
= size
;
3576 zone
->present_pages
= realsize
;
3579 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3581 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3583 zone
->name
= zone_names
[j
];
3584 spin_lock_init(&zone
->lock
);
3585 spin_lock_init(&zone
->lru_lock
);
3586 zone_seqlock_init(zone
);
3587 zone
->zone_pgdat
= pgdat
;
3589 zone
->prev_priority
= DEF_PRIORITY
;
3591 zone_pcp_init(zone
);
3593 INIT_LIST_HEAD(&zone
->lru
[l
].list
);
3594 zone
->lru
[l
].nr_scan
= 0;
3596 zone
->reclaim_stat
.recent_rotated
[0] = 0;
3597 zone
->reclaim_stat
.recent_rotated
[1] = 0;
3598 zone
->reclaim_stat
.recent_scanned
[0] = 0;
3599 zone
->reclaim_stat
.recent_scanned
[1] = 0;
3600 zap_zone_vm_stats(zone
);
3605 set_pageblock_order(pageblock_default_order());
3606 setup_usemap(pgdat
, zone
, size
);
3607 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3608 size
, MEMMAP_EARLY
);
3610 memmap_init(size
, nid
, j
, zone_start_pfn
);
3611 zone_start_pfn
+= size
;
3615 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3617 /* Skip empty nodes */
3618 if (!pgdat
->node_spanned_pages
)
3621 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3622 /* ia64 gets its own node_mem_map, before this, without bootmem */
3623 if (!pgdat
->node_mem_map
) {
3624 unsigned long size
, start
, end
;
3628 * The zone's endpoints aren't required to be MAX_ORDER
3629 * aligned but the node_mem_map endpoints must be in order
3630 * for the buddy allocator to function correctly.
3632 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3633 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3634 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3635 size
= (end
- start
) * sizeof(struct page
);
3636 map
= alloc_remap(pgdat
->node_id
, size
);
3638 map
= alloc_bootmem_node(pgdat
, size
);
3639 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3641 #ifndef CONFIG_NEED_MULTIPLE_NODES
3643 * With no DISCONTIG, the global mem_map is just set as node 0's
3645 if (pgdat
== NODE_DATA(0)) {
3646 mem_map
= NODE_DATA(0)->node_mem_map
;
3647 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3648 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3649 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
3650 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3653 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3656 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
3657 unsigned long node_start_pfn
, unsigned long *zholes_size
)
3659 pg_data_t
*pgdat
= NODE_DATA(nid
);
3661 pgdat
->node_id
= nid
;
3662 pgdat
->node_start_pfn
= node_start_pfn
;
3663 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3665 alloc_node_mem_map(pgdat
);
3666 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3667 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3668 nid
, (unsigned long)pgdat
,
3669 (unsigned long)pgdat
->node_mem_map
);
3672 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3675 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3677 #if MAX_NUMNODES > 1
3679 * Figure out the number of possible node ids.
3681 static void __init
setup_nr_node_ids(void)
3684 unsigned int highest
= 0;
3686 for_each_node_mask(node
, node_possible_map
)
3688 nr_node_ids
= highest
+ 1;
3691 static inline void setup_nr_node_ids(void)
3697 * add_active_range - Register a range of PFNs backed by physical memory
3698 * @nid: The node ID the range resides on
3699 * @start_pfn: The start PFN of the available physical memory
3700 * @end_pfn: The end PFN of the available physical memory
3702 * These ranges are stored in an early_node_map[] and later used by
3703 * free_area_init_nodes() to calculate zone sizes and holes. If the
3704 * range spans a memory hole, it is up to the architecture to ensure
3705 * the memory is not freed by the bootmem allocator. If possible
3706 * the range being registered will be merged with existing ranges.
3708 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3709 unsigned long end_pfn
)
3713 mminit_dprintk(MMINIT_TRACE
, "memory_register",
3714 "Entering add_active_range(%d, %#lx, %#lx) "
3715 "%d entries of %d used\n",
3716 nid
, start_pfn
, end_pfn
,
3717 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3719 mminit_validate_memmodel_limits(&start_pfn
, &end_pfn
);
3721 /* Merge with existing active regions if possible */
3722 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3723 if (early_node_map
[i
].nid
!= nid
)
3726 /* Skip if an existing region covers this new one */
3727 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3728 end_pfn
<= early_node_map
[i
].end_pfn
)
3731 /* Merge forward if suitable */
3732 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3733 end_pfn
> early_node_map
[i
].end_pfn
) {
3734 early_node_map
[i
].end_pfn
= end_pfn
;
3738 /* Merge backward if suitable */
3739 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3740 end_pfn
>= early_node_map
[i
].start_pfn
) {
3741 early_node_map
[i
].start_pfn
= start_pfn
;
3746 /* Check that early_node_map is large enough */
3747 if (i
>= MAX_ACTIVE_REGIONS
) {
3748 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3749 MAX_ACTIVE_REGIONS
);
3753 early_node_map
[i
].nid
= nid
;
3754 early_node_map
[i
].start_pfn
= start_pfn
;
3755 early_node_map
[i
].end_pfn
= end_pfn
;
3756 nr_nodemap_entries
= i
+ 1;
3760 * remove_active_range - Shrink an existing registered range of PFNs
3761 * @nid: The node id the range is on that should be shrunk
3762 * @start_pfn: The new PFN of the range
3763 * @end_pfn: The new PFN of the range
3765 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3766 * The map is kept near the end physical page range that has already been
3767 * registered. This function allows an arch to shrink an existing registered
3770 void __init
remove_active_range(unsigned int nid
, unsigned long start_pfn
,
3771 unsigned long end_pfn
)
3776 printk(KERN_DEBUG
"remove_active_range (%d, %lu, %lu)\n",
3777 nid
, start_pfn
, end_pfn
);
3779 /* Find the old active region end and shrink */
3780 for_each_active_range_index_in_nid(i
, nid
) {
3781 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3782 early_node_map
[i
].end_pfn
<= end_pfn
) {
3784 early_node_map
[i
].start_pfn
= 0;
3785 early_node_map
[i
].end_pfn
= 0;
3789 if (early_node_map
[i
].start_pfn
< start_pfn
&&
3790 early_node_map
[i
].end_pfn
> start_pfn
) {
3791 unsigned long temp_end_pfn
= early_node_map
[i
].end_pfn
;
3792 early_node_map
[i
].end_pfn
= start_pfn
;
3793 if (temp_end_pfn
> end_pfn
)
3794 add_active_range(nid
, end_pfn
, temp_end_pfn
);
3797 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3798 early_node_map
[i
].end_pfn
> end_pfn
&&
3799 early_node_map
[i
].start_pfn
< end_pfn
) {
3800 early_node_map
[i
].start_pfn
= end_pfn
;
3808 /* remove the blank ones */
3809 for (i
= nr_nodemap_entries
- 1; i
> 0; i
--) {
3810 if (early_node_map
[i
].nid
!= nid
)
3812 if (early_node_map
[i
].end_pfn
)
3814 /* we found it, get rid of it */
3815 for (j
= i
; j
< nr_nodemap_entries
- 1; j
++)
3816 memcpy(&early_node_map
[j
], &early_node_map
[j
+1],
3817 sizeof(early_node_map
[j
]));
3818 j
= nr_nodemap_entries
- 1;
3819 memset(&early_node_map
[j
], 0, sizeof(early_node_map
[j
]));
3820 nr_nodemap_entries
--;
3825 * remove_all_active_ranges - Remove all currently registered regions
3827 * During discovery, it may be found that a table like SRAT is invalid
3828 * and an alternative discovery method must be used. This function removes
3829 * all currently registered regions.
3831 void __init
remove_all_active_ranges(void)
3833 memset(early_node_map
, 0, sizeof(early_node_map
));
3834 nr_nodemap_entries
= 0;
3837 /* Compare two active node_active_regions */
3838 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3840 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3841 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3843 /* Done this way to avoid overflows */
3844 if (arange
->start_pfn
> brange
->start_pfn
)
3846 if (arange
->start_pfn
< brange
->start_pfn
)
3852 /* sort the node_map by start_pfn */
3853 static void __init
sort_node_map(void)
3855 sort(early_node_map
, (size_t)nr_nodemap_entries
,
3856 sizeof(struct node_active_region
),
3857 cmp_node_active_region
, NULL
);
3860 /* Find the lowest pfn for a node */
3861 static unsigned long __init
find_min_pfn_for_node(int nid
)
3864 unsigned long min_pfn
= ULONG_MAX
;
3866 /* Assuming a sorted map, the first range found has the starting pfn */
3867 for_each_active_range_index_in_nid(i
, nid
)
3868 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
3870 if (min_pfn
== ULONG_MAX
) {
3872 "Could not find start_pfn for node %d\n", nid
);
3880 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3882 * It returns the minimum PFN based on information provided via
3883 * add_active_range().
3885 unsigned long __init
find_min_pfn_with_active_regions(void)
3887 return find_min_pfn_for_node(MAX_NUMNODES
);
3891 * early_calculate_totalpages()
3892 * Sum pages in active regions for movable zone.
3893 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3895 static unsigned long __init
early_calculate_totalpages(void)
3898 unsigned long totalpages
= 0;
3900 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3901 unsigned long pages
= early_node_map
[i
].end_pfn
-
3902 early_node_map
[i
].start_pfn
;
3903 totalpages
+= pages
;
3905 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
3911 * Find the PFN the Movable zone begins in each node. Kernel memory
3912 * is spread evenly between nodes as long as the nodes have enough
3913 * memory. When they don't, some nodes will have more kernelcore than
3916 static void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
3919 unsigned long usable_startpfn
;
3920 unsigned long kernelcore_node
, kernelcore_remaining
;
3921 unsigned long totalpages
= early_calculate_totalpages();
3922 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
3925 * If movablecore was specified, calculate what size of
3926 * kernelcore that corresponds so that memory usable for
3927 * any allocation type is evenly spread. If both kernelcore
3928 * and movablecore are specified, then the value of kernelcore
3929 * will be used for required_kernelcore if it's greater than
3930 * what movablecore would have allowed.
3932 if (required_movablecore
) {
3933 unsigned long corepages
;
3936 * Round-up so that ZONE_MOVABLE is at least as large as what
3937 * was requested by the user
3939 required_movablecore
=
3940 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
3941 corepages
= totalpages
- required_movablecore
;
3943 required_kernelcore
= max(required_kernelcore
, corepages
);
3946 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3947 if (!required_kernelcore
)
3950 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3951 find_usable_zone_for_movable();
3952 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
3955 /* Spread kernelcore memory as evenly as possible throughout nodes */
3956 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3957 for_each_node_state(nid
, N_HIGH_MEMORY
) {
3959 * Recalculate kernelcore_node if the division per node
3960 * now exceeds what is necessary to satisfy the requested
3961 * amount of memory for the kernel
3963 if (required_kernelcore
< kernelcore_node
)
3964 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3967 * As the map is walked, we track how much memory is usable
3968 * by the kernel using kernelcore_remaining. When it is
3969 * 0, the rest of the node is usable by ZONE_MOVABLE
3971 kernelcore_remaining
= kernelcore_node
;
3973 /* Go through each range of PFNs within this node */
3974 for_each_active_range_index_in_nid(i
, nid
) {
3975 unsigned long start_pfn
, end_pfn
;
3976 unsigned long size_pages
;
3978 start_pfn
= max(early_node_map
[i
].start_pfn
,
3979 zone_movable_pfn
[nid
]);
3980 end_pfn
= early_node_map
[i
].end_pfn
;
3981 if (start_pfn
>= end_pfn
)
3984 /* Account for what is only usable for kernelcore */
3985 if (start_pfn
< usable_startpfn
) {
3986 unsigned long kernel_pages
;
3987 kernel_pages
= min(end_pfn
, usable_startpfn
)
3990 kernelcore_remaining
-= min(kernel_pages
,
3991 kernelcore_remaining
);
3992 required_kernelcore
-= min(kernel_pages
,
3993 required_kernelcore
);
3995 /* Continue if range is now fully accounted */
3996 if (end_pfn
<= usable_startpfn
) {
3999 * Push zone_movable_pfn to the end so
4000 * that if we have to rebalance
4001 * kernelcore across nodes, we will
4002 * not double account here
4004 zone_movable_pfn
[nid
] = end_pfn
;
4007 start_pfn
= usable_startpfn
;
4011 * The usable PFN range for ZONE_MOVABLE is from
4012 * start_pfn->end_pfn. Calculate size_pages as the
4013 * number of pages used as kernelcore
4015 size_pages
= end_pfn
- start_pfn
;
4016 if (size_pages
> kernelcore_remaining
)
4017 size_pages
= kernelcore_remaining
;
4018 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
4021 * Some kernelcore has been met, update counts and
4022 * break if the kernelcore for this node has been
4025 required_kernelcore
-= min(required_kernelcore
,
4027 kernelcore_remaining
-= size_pages
;
4028 if (!kernelcore_remaining
)
4034 * If there is still required_kernelcore, we do another pass with one
4035 * less node in the count. This will push zone_movable_pfn[nid] further
4036 * along on the nodes that still have memory until kernelcore is
4040 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
4043 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4044 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
4045 zone_movable_pfn
[nid
] =
4046 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
4049 /* Any regular memory on that node ? */
4050 static void check_for_regular_memory(pg_data_t
*pgdat
)
4052 #ifdef CONFIG_HIGHMEM
4053 enum zone_type zone_type
;
4055 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
4056 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4057 if (zone
->present_pages
)
4058 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
4064 * free_area_init_nodes - Initialise all pg_data_t and zone data
4065 * @max_zone_pfn: an array of max PFNs for each zone
4067 * This will call free_area_init_node() for each active node in the system.
4068 * Using the page ranges provided by add_active_range(), the size of each
4069 * zone in each node and their holes is calculated. If the maximum PFN
4070 * between two adjacent zones match, it is assumed that the zone is empty.
4071 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4072 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4073 * starts where the previous one ended. For example, ZONE_DMA32 starts
4074 * at arch_max_dma_pfn.
4076 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
4081 /* Sort early_node_map as initialisation assumes it is sorted */
4084 /* Record where the zone boundaries are */
4085 memset(arch_zone_lowest_possible_pfn
, 0,
4086 sizeof(arch_zone_lowest_possible_pfn
));
4087 memset(arch_zone_highest_possible_pfn
, 0,
4088 sizeof(arch_zone_highest_possible_pfn
));
4089 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
4090 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
4091 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
4092 if (i
== ZONE_MOVABLE
)
4094 arch_zone_lowest_possible_pfn
[i
] =
4095 arch_zone_highest_possible_pfn
[i
-1];
4096 arch_zone_highest_possible_pfn
[i
] =
4097 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
4099 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
4100 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
4102 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4103 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
4104 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
4106 /* Print out the zone ranges */
4107 printk("Zone PFN ranges:\n");
4108 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4109 if (i
== ZONE_MOVABLE
)
4111 printk(" %-8s %0#10lx -> %0#10lx\n",
4113 arch_zone_lowest_possible_pfn
[i
],
4114 arch_zone_highest_possible_pfn
[i
]);
4117 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4118 printk("Movable zone start PFN for each node\n");
4119 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4120 if (zone_movable_pfn
[i
])
4121 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
4124 /* Print out the early_node_map[] */
4125 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
4126 for (i
= 0; i
< nr_nodemap_entries
; i
++)
4127 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map
[i
].nid
,
4128 early_node_map
[i
].start_pfn
,
4129 early_node_map
[i
].end_pfn
);
4131 /* Initialise every node */
4132 mminit_verify_pageflags_layout();
4133 setup_nr_node_ids();
4134 for_each_online_node(nid
) {
4135 pg_data_t
*pgdat
= NODE_DATA(nid
);
4136 free_area_init_node(nid
, NULL
,
4137 find_min_pfn_for_node(nid
), NULL
);
4139 /* Any memory on that node */
4140 if (pgdat
->node_present_pages
)
4141 node_set_state(nid
, N_HIGH_MEMORY
);
4142 check_for_regular_memory(pgdat
);
4146 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4148 unsigned long long coremem
;
4152 coremem
= memparse(p
, &p
);
4153 *core
= coremem
>> PAGE_SHIFT
;
4155 /* Paranoid check that UL is enough for the coremem value */
4156 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
4162 * kernelcore=size sets the amount of memory for use for allocations that
4163 * cannot be reclaimed or migrated.
4165 static int __init
cmdline_parse_kernelcore(char *p
)
4167 return cmdline_parse_core(p
, &required_kernelcore
);
4171 * movablecore=size sets the amount of memory for use for allocations that
4172 * can be reclaimed or migrated.
4174 static int __init
cmdline_parse_movablecore(char *p
)
4176 return cmdline_parse_core(p
, &required_movablecore
);
4179 early_param("kernelcore", cmdline_parse_kernelcore
);
4180 early_param("movablecore", cmdline_parse_movablecore
);
4182 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4185 * set_dma_reserve - set the specified number of pages reserved in the first zone
4186 * @new_dma_reserve: The number of pages to mark reserved
4188 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4189 * In the DMA zone, a significant percentage may be consumed by kernel image
4190 * and other unfreeable allocations which can skew the watermarks badly. This
4191 * function may optionally be used to account for unfreeable pages in the
4192 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4193 * smaller per-cpu batchsize.
4195 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
4197 dma_reserve
= new_dma_reserve
;
4200 #ifndef CONFIG_NEED_MULTIPLE_NODES
4201 struct pglist_data __refdata contig_page_data
= { .bdata
= &bootmem_node_data
[0] };
4202 EXPORT_SYMBOL(contig_page_data
);
4205 void __init
free_area_init(unsigned long *zones_size
)
4207 free_area_init_node(0, zones_size
,
4208 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
4211 static int page_alloc_cpu_notify(struct notifier_block
*self
,
4212 unsigned long action
, void *hcpu
)
4214 int cpu
= (unsigned long)hcpu
;
4216 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
4220 * Spill the event counters of the dead processor
4221 * into the current processors event counters.
4222 * This artificially elevates the count of the current
4225 vm_events_fold_cpu(cpu
);
4228 * Zero the differential counters of the dead processor
4229 * so that the vm statistics are consistent.
4231 * This is only okay since the processor is dead and cannot
4232 * race with what we are doing.
4234 refresh_cpu_vm_stats(cpu
);
4239 void __init
page_alloc_init(void)
4241 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4245 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4246 * or min_free_kbytes changes.
4248 static void calculate_totalreserve_pages(void)
4250 struct pglist_data
*pgdat
;
4251 unsigned long reserve_pages
= 0;
4252 enum zone_type i
, j
;
4254 for_each_online_pgdat(pgdat
) {
4255 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4256 struct zone
*zone
= pgdat
->node_zones
+ i
;
4257 unsigned long max
= 0;
4259 /* Find valid and maximum lowmem_reserve in the zone */
4260 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4261 if (zone
->lowmem_reserve
[j
] > max
)
4262 max
= zone
->lowmem_reserve
[j
];
4265 /* we treat pages_high as reserved pages. */
4266 max
+= zone
->pages_high
;
4268 if (max
> zone
->present_pages
)
4269 max
= zone
->present_pages
;
4270 reserve_pages
+= max
;
4273 totalreserve_pages
= reserve_pages
;
4277 * setup_per_zone_lowmem_reserve - called whenever
4278 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4279 * has a correct pages reserved value, so an adequate number of
4280 * pages are left in the zone after a successful __alloc_pages().
4282 static void setup_per_zone_lowmem_reserve(void)
4284 struct pglist_data
*pgdat
;
4285 enum zone_type j
, idx
;
4287 for_each_online_pgdat(pgdat
) {
4288 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4289 struct zone
*zone
= pgdat
->node_zones
+ j
;
4290 unsigned long present_pages
= zone
->present_pages
;
4292 zone
->lowmem_reserve
[j
] = 0;
4296 struct zone
*lower_zone
;
4300 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4301 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4303 lower_zone
= pgdat
->node_zones
+ idx
;
4304 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4305 sysctl_lowmem_reserve_ratio
[idx
];
4306 present_pages
+= lower_zone
->present_pages
;
4311 /* update totalreserve_pages */
4312 calculate_totalreserve_pages();
4316 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4318 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4319 * with respect to min_free_kbytes.
4321 void setup_per_zone_pages_min(void)
4323 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4324 unsigned long lowmem_pages
= 0;
4326 unsigned long flags
;
4328 /* Calculate total number of !ZONE_HIGHMEM pages */
4329 for_each_zone(zone
) {
4330 if (!is_highmem(zone
))
4331 lowmem_pages
+= zone
->present_pages
;
4334 for_each_zone(zone
) {
4337 spin_lock_irqsave(&zone
->lock
, flags
);
4338 tmp
= (u64
)pages_min
* zone
->present_pages
;
4339 do_div(tmp
, lowmem_pages
);
4340 if (is_highmem(zone
)) {
4342 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4343 * need highmem pages, so cap pages_min to a small
4346 * The (pages_high-pages_low) and (pages_low-pages_min)
4347 * deltas controls asynch page reclaim, and so should
4348 * not be capped for highmem.
4352 min_pages
= zone
->present_pages
/ 1024;
4353 if (min_pages
< SWAP_CLUSTER_MAX
)
4354 min_pages
= SWAP_CLUSTER_MAX
;
4355 if (min_pages
> 128)
4357 zone
->pages_min
= min_pages
;
4360 * If it's a lowmem zone, reserve a number of pages
4361 * proportionate to the zone's size.
4363 zone
->pages_min
= tmp
;
4366 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
4367 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
4368 setup_zone_migrate_reserve(zone
);
4369 spin_unlock_irqrestore(&zone
->lock
, flags
);
4372 /* update totalreserve_pages */
4373 calculate_totalreserve_pages();
4377 * setup_per_zone_inactive_ratio - called when min_free_kbytes changes.
4379 * The inactive anon list should be small enough that the VM never has to
4380 * do too much work, but large enough that each inactive page has a chance
4381 * to be referenced again before it is swapped out.
4383 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4384 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4385 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4386 * the anonymous pages are kept on the inactive list.
4389 * memory ratio inactive anon
4390 * -------------------------------------
4399 static void setup_per_zone_inactive_ratio(void)
4403 for_each_zone(zone
) {
4404 unsigned int gb
, ratio
;
4406 /* Zone size in gigabytes */
4407 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
4408 ratio
= int_sqrt(10 * gb
);
4412 zone
->inactive_ratio
= ratio
;
4417 * Initialise min_free_kbytes.
4419 * For small machines we want it small (128k min). For large machines
4420 * we want it large (64MB max). But it is not linear, because network
4421 * bandwidth does not increase linearly with machine size. We use
4423 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4424 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4440 static int __init
init_per_zone_pages_min(void)
4442 unsigned long lowmem_kbytes
;
4444 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4446 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4447 if (min_free_kbytes
< 128)
4448 min_free_kbytes
= 128;
4449 if (min_free_kbytes
> 65536)
4450 min_free_kbytes
= 65536;
4451 setup_per_zone_pages_min();
4452 setup_per_zone_lowmem_reserve();
4453 setup_per_zone_inactive_ratio();
4456 module_init(init_per_zone_pages_min
)
4459 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4460 * that we can call two helper functions whenever min_free_kbytes
4463 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4464 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4466 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
4468 setup_per_zone_pages_min();
4473 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4474 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4479 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4484 zone
->min_unmapped_pages
= (zone
->present_pages
*
4485 sysctl_min_unmapped_ratio
) / 100;
4489 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4490 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4495 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4500 zone
->min_slab_pages
= (zone
->present_pages
*
4501 sysctl_min_slab_ratio
) / 100;
4507 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4508 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4509 * whenever sysctl_lowmem_reserve_ratio changes.
4511 * The reserve ratio obviously has absolutely no relation with the
4512 * pages_min watermarks. The lowmem reserve ratio can only make sense
4513 * if in function of the boot time zone sizes.
4515 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4516 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4518 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4519 setup_per_zone_lowmem_reserve();
4524 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4525 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4526 * can have before it gets flushed back to buddy allocator.
4529 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4530 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4536 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4537 if (!write
|| (ret
== -EINVAL
))
4539 for_each_zone(zone
) {
4540 for_each_online_cpu(cpu
) {
4542 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4543 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
4549 int hashdist
= HASHDIST_DEFAULT
;
4552 static int __init
set_hashdist(char *str
)
4556 hashdist
= simple_strtoul(str
, &str
, 0);
4559 __setup("hashdist=", set_hashdist
);
4563 * allocate a large system hash table from bootmem
4564 * - it is assumed that the hash table must contain an exact power-of-2
4565 * quantity of entries
4566 * - limit is the number of hash buckets, not the total allocation size
4568 void *__init
alloc_large_system_hash(const char *tablename
,
4569 unsigned long bucketsize
,
4570 unsigned long numentries
,
4573 unsigned int *_hash_shift
,
4574 unsigned int *_hash_mask
,
4575 unsigned long limit
)
4577 unsigned long long max
= limit
;
4578 unsigned long log2qty
, size
;
4581 /* allow the kernel cmdline to have a say */
4583 /* round applicable memory size up to nearest megabyte */
4584 numentries
= nr_kernel_pages
;
4585 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4586 numentries
>>= 20 - PAGE_SHIFT
;
4587 numentries
<<= 20 - PAGE_SHIFT
;
4589 /* limit to 1 bucket per 2^scale bytes of low memory */
4590 if (scale
> PAGE_SHIFT
)
4591 numentries
>>= (scale
- PAGE_SHIFT
);
4593 numentries
<<= (PAGE_SHIFT
- scale
);
4595 /* Make sure we've got at least a 0-order allocation.. */
4596 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4597 numentries
= PAGE_SIZE
/ bucketsize
;
4599 numentries
= roundup_pow_of_two(numentries
);
4601 /* limit allocation size to 1/16 total memory by default */
4603 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4604 do_div(max
, bucketsize
);
4607 if (numentries
> max
)
4610 log2qty
= ilog2(numentries
);
4613 size
= bucketsize
<< log2qty
;
4614 if (flags
& HASH_EARLY
)
4615 table
= alloc_bootmem_nopanic(size
);
4617 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4619 unsigned long order
= get_order(size
);
4621 if (order
< MAX_ORDER
)
4622 table
= (void *)__get_free_pages(GFP_ATOMIC
,
4625 * If bucketsize is not a power-of-two, we may free
4626 * some pages at the end of hash table.
4629 unsigned long alloc_end
= (unsigned long)table
+
4630 (PAGE_SIZE
<< order
);
4631 unsigned long used
= (unsigned long)table
+
4633 split_page(virt_to_page(table
), order
);
4634 while (used
< alloc_end
) {
4640 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4643 panic("Failed to allocate %s hash table\n", tablename
);
4645 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4648 ilog2(size
) - PAGE_SHIFT
,
4652 *_hash_shift
= log2qty
;
4654 *_hash_mask
= (1 << log2qty
) - 1;
4657 * If hashdist is set, the table allocation is done with __vmalloc()
4658 * which invokes the kmemleak_alloc() callback. This function may also
4659 * be called before the slab and kmemleak are initialised when
4660 * kmemleak simply buffers the request to be executed later
4661 * (GFP_ATOMIC flag ignored in this case).
4664 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
4669 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4670 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4673 #ifdef CONFIG_SPARSEMEM
4674 return __pfn_to_section(pfn
)->pageblock_flags
;
4676 return zone
->pageblock_flags
;
4677 #endif /* CONFIG_SPARSEMEM */
4680 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4682 #ifdef CONFIG_SPARSEMEM
4683 pfn
&= (PAGES_PER_SECTION
-1);
4684 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4686 pfn
= pfn
- zone
->zone_start_pfn
;
4687 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4688 #endif /* CONFIG_SPARSEMEM */
4692 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4693 * @page: The page within the block of interest
4694 * @start_bitidx: The first bit of interest to retrieve
4695 * @end_bitidx: The last bit of interest
4696 * returns pageblock_bits flags
4698 unsigned long get_pageblock_flags_group(struct page
*page
,
4699 int start_bitidx
, int end_bitidx
)
4702 unsigned long *bitmap
;
4703 unsigned long pfn
, bitidx
;
4704 unsigned long flags
= 0;
4705 unsigned long value
= 1;
4707 zone
= page_zone(page
);
4708 pfn
= page_to_pfn(page
);
4709 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4710 bitidx
= pfn_to_bitidx(zone
, pfn
);
4712 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4713 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
4720 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4721 * @page: The page within the block of interest
4722 * @start_bitidx: The first bit of interest
4723 * @end_bitidx: The last bit of interest
4724 * @flags: The flags to set
4726 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
4727 int start_bitidx
, int end_bitidx
)
4730 unsigned long *bitmap
;
4731 unsigned long pfn
, bitidx
;
4732 unsigned long value
= 1;
4734 zone
= page_zone(page
);
4735 pfn
= page_to_pfn(page
);
4736 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4737 bitidx
= pfn_to_bitidx(zone
, pfn
);
4738 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
4739 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
4741 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4743 __set_bit(bitidx
+ start_bitidx
, bitmap
);
4745 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
4749 * This is designed as sub function...plz see page_isolation.c also.
4750 * set/clear page block's type to be ISOLATE.
4751 * page allocater never alloc memory from ISOLATE block.
4754 int set_migratetype_isolate(struct page
*page
)
4757 unsigned long flags
;
4760 zone
= page_zone(page
);
4761 spin_lock_irqsave(&zone
->lock
, flags
);
4763 * In future, more migrate types will be able to be isolation target.
4765 if (get_pageblock_migratetype(page
) != MIGRATE_MOVABLE
)
4767 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
4768 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
4771 spin_unlock_irqrestore(&zone
->lock
, flags
);
4777 void unset_migratetype_isolate(struct page
*page
)
4780 unsigned long flags
;
4781 zone
= page_zone(page
);
4782 spin_lock_irqsave(&zone
->lock
, flags
);
4783 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
4785 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4786 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
4788 spin_unlock_irqrestore(&zone
->lock
, flags
);
4791 #ifdef CONFIG_MEMORY_HOTREMOVE
4793 * All pages in the range must be isolated before calling this.
4796 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
4802 unsigned long flags
;
4803 /* find the first valid pfn */
4804 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
4809 zone
= page_zone(pfn_to_page(pfn
));
4810 spin_lock_irqsave(&zone
->lock
, flags
);
4812 while (pfn
< end_pfn
) {
4813 if (!pfn_valid(pfn
)) {
4817 page
= pfn_to_page(pfn
);
4818 BUG_ON(page_count(page
));
4819 BUG_ON(!PageBuddy(page
));
4820 order
= page_order(page
);
4821 #ifdef CONFIG_DEBUG_VM
4822 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
4823 pfn
, 1 << order
, end_pfn
);
4825 list_del(&page
->lru
);
4826 rmv_page_order(page
);
4827 zone
->free_area
[order
].nr_free
--;
4828 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
4830 for (i
= 0; i
< (1 << order
); i
++)
4831 SetPageReserved((page
+i
));
4832 pfn
+= (1 << order
);
4834 spin_unlock_irqrestore(&zone
->lock
, flags
);