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
)
172 if (unlikely(page_group_by_mobility_disabled
))
173 migratetype
= MIGRATE_UNMOVABLE
;
175 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
176 PB_migrate
, PB_migrate_end
);
179 #ifdef CONFIG_DEBUG_VM
180 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
184 unsigned long pfn
= page_to_pfn(page
);
187 seq
= zone_span_seqbegin(zone
);
188 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
190 else if (pfn
< zone
->zone_start_pfn
)
192 } while (zone_span_seqretry(zone
, seq
));
197 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
199 if (!pfn_valid_within(page_to_pfn(page
)))
201 if (zone
!= page_zone(page
))
207 * Temporary debugging check for pages not lying within a given zone.
209 static int bad_range(struct zone
*zone
, struct page
*page
)
211 if (page_outside_zone_boundaries(zone
, page
))
213 if (!page_is_consistent(zone
, page
))
219 static inline int bad_range(struct zone
*zone
, struct page
*page
)
225 static void bad_page(struct page
*page
)
227 static unsigned long resume
;
228 static unsigned long nr_shown
;
229 static unsigned long nr_unshown
;
232 * Allow a burst of 60 reports, then keep quiet for that minute;
233 * or allow a steady drip of one report per second.
235 if (nr_shown
== 60) {
236 if (time_before(jiffies
, resume
)) {
242 "BUG: Bad page state: %lu messages suppressed\n",
249 resume
= jiffies
+ 60 * HZ
;
251 printk(KERN_ALERT
"BUG: Bad page state in process %s pfn:%05lx\n",
252 current
->comm
, page_to_pfn(page
));
254 "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
255 page
, (void *)page
->flags
, page_count(page
),
256 page_mapcount(page
), page
->mapping
, page
->index
);
260 /* Leave bad fields for debug, except PageBuddy could make trouble */
261 __ClearPageBuddy(page
);
262 add_taint(TAINT_BAD_PAGE
);
266 * Higher-order pages are called "compound pages". They are structured thusly:
268 * The first PAGE_SIZE page is called the "head page".
270 * The remaining PAGE_SIZE pages are called "tail pages".
272 * All pages have PG_compound set. All pages have their ->private pointing at
273 * the head page (even the head page has this).
275 * The first tail page's ->lru.next holds the address of the compound page's
276 * put_page() function. Its ->lru.prev holds the order of allocation.
277 * This usage means that zero-order pages may not be compound.
280 static void free_compound_page(struct page
*page
)
282 __free_pages_ok(page
, compound_order(page
));
285 void prep_compound_page(struct page
*page
, unsigned long order
)
288 int nr_pages
= 1 << order
;
290 set_compound_page_dtor(page
, free_compound_page
);
291 set_compound_order(page
, order
);
293 for (i
= 1; i
< nr_pages
; i
++) {
294 struct page
*p
= page
+ i
;
297 p
->first_page
= page
;
301 #ifdef CONFIG_HUGETLBFS
302 void prep_compound_gigantic_page(struct page
*page
, unsigned long order
)
305 int nr_pages
= 1 << order
;
306 struct page
*p
= page
+ 1;
308 set_compound_page_dtor(page
, free_compound_page
);
309 set_compound_order(page
, order
);
311 for (i
= 1; i
< nr_pages
; i
++, p
= mem_map_next(p
, page
, i
)) {
313 p
->first_page
= page
;
318 static int destroy_compound_page(struct page
*page
, unsigned long order
)
321 int nr_pages
= 1 << order
;
324 if (unlikely(compound_order(page
) != order
) ||
325 unlikely(!PageHead(page
))) {
330 __ClearPageHead(page
);
332 for (i
= 1; i
< nr_pages
; i
++) {
333 struct page
*p
= page
+ i
;
335 if (unlikely(!PageTail(p
) || (p
->first_page
!= page
))) {
345 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
350 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
351 * and __GFP_HIGHMEM from hard or soft interrupt context.
353 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
354 for (i
= 0; i
< (1 << order
); i
++)
355 clear_highpage(page
+ i
);
358 static inline void set_page_order(struct page
*page
, int order
)
360 set_page_private(page
, order
);
361 __SetPageBuddy(page
);
364 static inline void rmv_page_order(struct page
*page
)
366 __ClearPageBuddy(page
);
367 set_page_private(page
, 0);
371 * Locate the struct page for both the matching buddy in our
372 * pair (buddy1) and the combined O(n+1) page they form (page).
374 * 1) Any buddy B1 will have an order O twin B2 which satisfies
375 * the following equation:
377 * For example, if the starting buddy (buddy2) is #8 its order
379 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
381 * 2) Any buddy B will have an order O+1 parent P which
382 * satisfies the following equation:
385 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
387 static inline struct page
*
388 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
390 unsigned long buddy_idx
= page_idx
^ (1 << order
);
392 return page
+ (buddy_idx
- page_idx
);
395 static inline unsigned long
396 __find_combined_index(unsigned long page_idx
, unsigned int order
)
398 return (page_idx
& ~(1 << order
));
402 * This function checks whether a page is free && is the buddy
403 * we can do coalesce a page and its buddy if
404 * (a) the buddy is not in a hole &&
405 * (b) the buddy is in the buddy system &&
406 * (c) a page and its buddy have the same order &&
407 * (d) a page and its buddy are in the same zone.
409 * For recording whether a page is in the buddy system, we use PG_buddy.
410 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
412 * For recording page's order, we use page_private(page).
414 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
417 if (!pfn_valid_within(page_to_pfn(buddy
)))
420 if (page_zone_id(page
) != page_zone_id(buddy
))
423 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
424 BUG_ON(page_count(buddy
) != 0);
431 * Freeing function for a buddy system allocator.
433 * The concept of a buddy system is to maintain direct-mapped table
434 * (containing bit values) for memory blocks of various "orders".
435 * The bottom level table contains the map for the smallest allocatable
436 * units of memory (here, pages), and each level above it describes
437 * pairs of units from the levels below, hence, "buddies".
438 * At a high level, all that happens here is marking the table entry
439 * at the bottom level available, and propagating the changes upward
440 * as necessary, plus some accounting needed to play nicely with other
441 * parts of the VM system.
442 * At each level, we keep a list of pages, which are heads of continuous
443 * free pages of length of (1 << order) and marked with PG_buddy. Page's
444 * order is recorded in page_private(page) field.
445 * So when we are allocating or freeing one, we can derive the state of the
446 * other. That is, if we allocate a small block, and both were
447 * free, the remainder of the region must be split into blocks.
448 * If a block is freed, and its buddy is also free, then this
449 * triggers coalescing into a block of larger size.
454 static inline void __free_one_page(struct page
*page
,
455 struct zone
*zone
, unsigned int order
)
457 unsigned long page_idx
;
458 int order_size
= 1 << order
;
459 int migratetype
= get_pageblock_migratetype(page
);
461 if (unlikely(PageCompound(page
)))
462 if (unlikely(destroy_compound_page(page
, order
)))
465 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
467 VM_BUG_ON(page_idx
& (order_size
- 1));
468 VM_BUG_ON(bad_range(zone
, page
));
470 __mod_zone_page_state(zone
, NR_FREE_PAGES
, order_size
);
471 while (order
< MAX_ORDER
-1) {
472 unsigned long combined_idx
;
475 buddy
= __page_find_buddy(page
, page_idx
, order
);
476 if (!page_is_buddy(page
, buddy
, order
))
479 /* Our buddy is free, merge with it and move up one order. */
480 list_del(&buddy
->lru
);
481 zone
->free_area
[order
].nr_free
--;
482 rmv_page_order(buddy
);
483 combined_idx
= __find_combined_index(page_idx
, order
);
484 page
= page
+ (combined_idx
- page_idx
);
485 page_idx
= combined_idx
;
488 set_page_order(page
, order
);
490 &zone
->free_area
[order
].free_list
[migratetype
]);
491 zone
->free_area
[order
].nr_free
++;
494 static inline int free_pages_check(struct page
*page
)
496 free_page_mlock(page
);
497 if (unlikely(page_mapcount(page
) |
498 (page
->mapping
!= NULL
) |
499 (page_count(page
) != 0) |
500 (page
->flags
& PAGE_FLAGS_CHECK_AT_FREE
))) {
504 if (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
)
505 page
->flags
&= ~PAGE_FLAGS_CHECK_AT_PREP
;
510 * Frees a list of pages.
511 * Assumes all pages on list are in same zone, and of same order.
512 * count is the number of pages to free.
514 * If the zone was previously in an "all pages pinned" state then look to
515 * see if this freeing clears that state.
517 * And clear the zone's pages_scanned counter, to hold off the "all pages are
518 * pinned" detection logic.
520 static void free_pages_bulk(struct zone
*zone
, int count
,
521 struct list_head
*list
, int order
)
523 spin_lock(&zone
->lock
);
524 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
525 zone
->pages_scanned
= 0;
529 VM_BUG_ON(list_empty(list
));
530 page
= list_entry(list
->prev
, struct page
, lru
);
531 /* have to delete it as __free_one_page list manipulates */
532 list_del(&page
->lru
);
533 __free_one_page(page
, zone
, order
);
535 spin_unlock(&zone
->lock
);
538 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
540 spin_lock(&zone
->lock
);
541 zone_clear_flag(zone
, ZONE_ALL_UNRECLAIMABLE
);
542 zone
->pages_scanned
= 0;
543 __free_one_page(page
, zone
, order
);
544 spin_unlock(&zone
->lock
);
547 static void __free_pages_ok(struct page
*page
, unsigned int order
)
553 for (i
= 0 ; i
< (1 << order
) ; ++i
)
554 bad
+= free_pages_check(page
+ i
);
558 if (!PageHighMem(page
)) {
559 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
560 debug_check_no_obj_freed(page_address(page
),
563 arch_free_page(page
, order
);
564 kernel_map_pages(page
, 1 << order
, 0);
566 local_irq_save(flags
);
567 __count_vm_events(PGFREE
, 1 << order
);
568 free_one_page(page_zone(page
), page
, order
);
569 local_irq_restore(flags
);
573 * permit the bootmem allocator to evade page validation on high-order frees
575 void __meminit
__free_pages_bootmem(struct page
*page
, unsigned int order
)
578 __ClearPageReserved(page
);
579 set_page_count(page
, 0);
580 set_page_refcounted(page
);
586 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
587 struct page
*p
= &page
[loop
];
589 if (loop
+ 1 < BITS_PER_LONG
)
591 __ClearPageReserved(p
);
592 set_page_count(p
, 0);
595 set_page_refcounted(page
);
596 __free_pages(page
, order
);
602 * The order of subdivision here is critical for the IO subsystem.
603 * Please do not alter this order without good reasons and regression
604 * testing. Specifically, as large blocks of memory are subdivided,
605 * the order in which smaller blocks are delivered depends on the order
606 * they're subdivided in this function. This is the primary factor
607 * influencing the order in which pages are delivered to the IO
608 * subsystem according to empirical testing, and this is also justified
609 * by considering the behavior of a buddy system containing a single
610 * large block of memory acted on by a series of small allocations.
611 * This behavior is a critical factor in sglist merging's success.
615 static inline void expand(struct zone
*zone
, struct page
*page
,
616 int low
, int high
, struct free_area
*area
,
619 unsigned long size
= 1 << high
;
625 VM_BUG_ON(bad_range(zone
, &page
[size
]));
626 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
628 set_page_order(&page
[size
], high
);
633 * This page is about to be returned from the page allocator
635 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
637 if (unlikely(page_mapcount(page
) |
638 (page
->mapping
!= NULL
) |
639 (page_count(page
) != 0) |
640 (page
->flags
& PAGE_FLAGS_CHECK_AT_PREP
))) {
645 set_page_private(page
, 0);
646 set_page_refcounted(page
);
648 arch_alloc_page(page
, order
);
649 kernel_map_pages(page
, 1 << order
, 1);
651 if (gfp_flags
& __GFP_ZERO
)
652 prep_zero_page(page
, order
, gfp_flags
);
654 if (order
&& (gfp_flags
& __GFP_COMP
))
655 prep_compound_page(page
, order
);
661 * Go through the free lists for the given migratetype and remove
662 * the smallest available page from the freelists
664 static struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
667 unsigned int current_order
;
668 struct free_area
* area
;
671 /* Find a page of the appropriate size in the preferred list */
672 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
673 area
= &(zone
->free_area
[current_order
]);
674 if (list_empty(&area
->free_list
[migratetype
]))
677 page
= list_entry(area
->free_list
[migratetype
].next
,
679 list_del(&page
->lru
);
680 rmv_page_order(page
);
682 __mod_zone_page_state(zone
, NR_FREE_PAGES
, - (1UL << order
));
683 expand(zone
, page
, order
, current_order
, area
, migratetype
);
692 * This array describes the order lists are fallen back to when
693 * the free lists for the desirable migrate type are depleted
695 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
696 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
697 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
698 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
699 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
703 * Move the free pages in a range to the free lists of the requested type.
704 * Note that start_page and end_pages are not aligned on a pageblock
705 * boundary. If alignment is required, use move_freepages_block()
707 static int move_freepages(struct zone
*zone
,
708 struct page
*start_page
, struct page
*end_page
,
715 #ifndef CONFIG_HOLES_IN_ZONE
717 * page_zone is not safe to call in this context when
718 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
719 * anyway as we check zone boundaries in move_freepages_block().
720 * Remove at a later date when no bug reports exist related to
721 * grouping pages by mobility
723 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
726 for (page
= start_page
; page
<= end_page
;) {
727 /* Make sure we are not inadvertently changing nodes */
728 VM_BUG_ON(page_to_nid(page
) != zone_to_nid(zone
));
730 if (!pfn_valid_within(page_to_pfn(page
))) {
735 if (!PageBuddy(page
)) {
740 order
= page_order(page
);
741 list_del(&page
->lru
);
743 &zone
->free_area
[order
].free_list
[migratetype
]);
745 pages_moved
+= 1 << order
;
751 static int move_freepages_block(struct zone
*zone
, struct page
*page
,
754 unsigned long start_pfn
, end_pfn
;
755 struct page
*start_page
, *end_page
;
757 start_pfn
= page_to_pfn(page
);
758 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
759 start_page
= pfn_to_page(start_pfn
);
760 end_page
= start_page
+ pageblock_nr_pages
- 1;
761 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
763 /* Do not cross zone boundaries */
764 if (start_pfn
< zone
->zone_start_pfn
)
766 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
769 return move_freepages(zone
, start_page
, end_page
, migratetype
);
772 /* Remove an element from the buddy allocator from the fallback list */
773 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
774 int start_migratetype
)
776 struct free_area
* area
;
781 /* Find the largest possible block of pages in the other list */
782 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
784 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
785 migratetype
= fallbacks
[start_migratetype
][i
];
787 /* MIGRATE_RESERVE handled later if necessary */
788 if (migratetype
== MIGRATE_RESERVE
)
791 area
= &(zone
->free_area
[current_order
]);
792 if (list_empty(&area
->free_list
[migratetype
]))
795 page
= list_entry(area
->free_list
[migratetype
].next
,
800 * If breaking a large block of pages, move all free
801 * pages to the preferred allocation list. If falling
802 * back for a reclaimable kernel allocation, be more
803 * agressive about taking ownership of free pages
805 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
806 start_migratetype
== MIGRATE_RECLAIMABLE
) {
808 pages
= move_freepages_block(zone
, page
,
811 /* Claim the whole block if over half of it is free */
812 if (pages
>= (1 << (pageblock_order
-1)))
813 set_pageblock_migratetype(page
,
816 migratetype
= start_migratetype
;
819 /* Remove the page from the freelists */
820 list_del(&page
->lru
);
821 rmv_page_order(page
);
822 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
825 if (current_order
== pageblock_order
)
826 set_pageblock_migratetype(page
,
829 expand(zone
, page
, order
, current_order
, area
, migratetype
);
834 /* Use MIGRATE_RESERVE rather than fail an allocation */
835 return __rmqueue_smallest(zone
, order
, MIGRATE_RESERVE
);
839 * Do the hard work of removing an element from the buddy allocator.
840 * Call me with the zone->lock already held.
842 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
847 page
= __rmqueue_smallest(zone
, order
, migratetype
);
850 page
= __rmqueue_fallback(zone
, order
, migratetype
);
856 * Obtain a specified number of elements from the buddy allocator, all under
857 * a single hold of the lock, for efficiency. Add them to the supplied list.
858 * Returns the number of new pages which were placed at *list.
860 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
861 unsigned long count
, struct list_head
*list
,
866 spin_lock(&zone
->lock
);
867 for (i
= 0; i
< count
; ++i
) {
868 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
869 if (unlikely(page
== NULL
))
873 * Split buddy pages returned by expand() are received here
874 * in physical page order. The page is added to the callers and
875 * list and the list head then moves forward. From the callers
876 * perspective, the linked list is ordered by page number in
877 * some conditions. This is useful for IO devices that can
878 * merge IO requests if the physical pages are ordered
881 list_add(&page
->lru
, list
);
882 set_page_private(page
, migratetype
);
885 spin_unlock(&zone
->lock
);
891 * Called from the vmstat counter updater to drain pagesets of this
892 * currently executing processor on remote nodes after they have
895 * Note that this function must be called with the thread pinned to
896 * a single processor.
898 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
903 local_irq_save(flags
);
904 if (pcp
->count
>= pcp
->batch
)
905 to_drain
= pcp
->batch
;
907 to_drain
= pcp
->count
;
908 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
909 pcp
->count
-= to_drain
;
910 local_irq_restore(flags
);
915 * Drain pages of the indicated processor.
917 * The processor must either be the current processor and the
918 * thread pinned to the current processor or a processor that
921 static void drain_pages(unsigned int cpu
)
926 for_each_populated_zone(zone
) {
927 struct per_cpu_pageset
*pset
;
928 struct per_cpu_pages
*pcp
;
930 pset
= zone_pcp(zone
, cpu
);
933 local_irq_save(flags
);
934 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
936 local_irq_restore(flags
);
941 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
943 void drain_local_pages(void *arg
)
945 drain_pages(smp_processor_id());
949 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
951 void drain_all_pages(void)
953 on_each_cpu(drain_local_pages
, NULL
, 1);
956 #ifdef CONFIG_HIBERNATION
958 void mark_free_pages(struct zone
*zone
)
960 unsigned long pfn
, max_zone_pfn
;
963 struct list_head
*curr
;
965 if (!zone
->spanned_pages
)
968 spin_lock_irqsave(&zone
->lock
, flags
);
970 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
971 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
972 if (pfn_valid(pfn
)) {
973 struct page
*page
= pfn_to_page(pfn
);
975 if (!swsusp_page_is_forbidden(page
))
976 swsusp_unset_page_free(page
);
979 for_each_migratetype_order(order
, t
) {
980 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
983 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
984 for (i
= 0; i
< (1UL << order
); i
++)
985 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
988 spin_unlock_irqrestore(&zone
->lock
, flags
);
990 #endif /* CONFIG_PM */
993 * Free a 0-order page
995 static void free_hot_cold_page(struct page
*page
, int cold
)
997 struct zone
*zone
= page_zone(page
);
998 struct per_cpu_pages
*pcp
;
1002 page
->mapping
= NULL
;
1003 if (free_pages_check(page
))
1006 if (!PageHighMem(page
)) {
1007 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
1008 debug_check_no_obj_freed(page_address(page
), PAGE_SIZE
);
1010 arch_free_page(page
, 0);
1011 kernel_map_pages(page
, 1, 0);
1013 pcp
= &zone_pcp(zone
, get_cpu())->pcp
;
1014 local_irq_save(flags
);
1015 __count_vm_event(PGFREE
);
1017 list_add_tail(&page
->lru
, &pcp
->list
);
1019 list_add(&page
->lru
, &pcp
->list
);
1020 set_page_private(page
, get_pageblock_migratetype(page
));
1022 if (pcp
->count
>= pcp
->high
) {
1023 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
1024 pcp
->count
-= pcp
->batch
;
1026 local_irq_restore(flags
);
1030 void free_hot_page(struct page
*page
)
1032 free_hot_cold_page(page
, 0);
1035 void free_cold_page(struct page
*page
)
1037 free_hot_cold_page(page
, 1);
1041 * split_page takes a non-compound higher-order page, and splits it into
1042 * n (1<<order) sub-pages: page[0..n]
1043 * Each sub-page must be freed individually.
1045 * Note: this is probably too low level an operation for use in drivers.
1046 * Please consult with lkml before using this in your driver.
1048 void split_page(struct page
*page
, unsigned int order
)
1052 VM_BUG_ON(PageCompound(page
));
1053 VM_BUG_ON(!page_count(page
));
1054 for (i
= 1; i
< (1 << order
); i
++)
1055 set_page_refcounted(page
+ i
);
1059 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1060 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1063 static struct page
*buffered_rmqueue(struct zone
*preferred_zone
,
1064 struct zone
*zone
, int order
, gfp_t gfp_flags
,
1067 unsigned long flags
;
1069 int cold
= !!(gfp_flags
& __GFP_COLD
);
1074 if (likely(order
== 0)) {
1075 struct per_cpu_pages
*pcp
;
1077 pcp
= &zone_pcp(zone
, cpu
)->pcp
;
1078 local_irq_save(flags
);
1080 pcp
->count
= rmqueue_bulk(zone
, 0,
1081 pcp
->batch
, &pcp
->list
, migratetype
);
1082 if (unlikely(!pcp
->count
))
1086 /* Find a page of the appropriate migrate type */
1088 list_for_each_entry_reverse(page
, &pcp
->list
, lru
)
1089 if (page_private(page
) == migratetype
)
1092 list_for_each_entry(page
, &pcp
->list
, lru
)
1093 if (page_private(page
) == migratetype
)
1097 /* Allocate more to the pcp list if necessary */
1098 if (unlikely(&page
->lru
== &pcp
->list
)) {
1099 pcp
->count
+= rmqueue_bulk(zone
, 0,
1100 pcp
->batch
, &pcp
->list
, migratetype
);
1101 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1104 list_del(&page
->lru
);
1107 spin_lock_irqsave(&zone
->lock
, flags
);
1108 page
= __rmqueue(zone
, order
, migratetype
);
1109 spin_unlock(&zone
->lock
);
1114 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1115 zone_statistics(preferred_zone
, zone
);
1116 local_irq_restore(flags
);
1119 VM_BUG_ON(bad_range(zone
, page
));
1120 if (prep_new_page(page
, order
, gfp_flags
))
1125 local_irq_restore(flags
);
1130 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1131 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1132 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1133 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1134 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1135 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1136 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1138 #ifdef CONFIG_FAIL_PAGE_ALLOC
1140 static struct fail_page_alloc_attr
{
1141 struct fault_attr attr
;
1143 u32 ignore_gfp_highmem
;
1144 u32 ignore_gfp_wait
;
1147 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1149 struct dentry
*ignore_gfp_highmem_file
;
1150 struct dentry
*ignore_gfp_wait_file
;
1151 struct dentry
*min_order_file
;
1153 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1155 } fail_page_alloc
= {
1156 .attr
= FAULT_ATTR_INITIALIZER
,
1157 .ignore_gfp_wait
= 1,
1158 .ignore_gfp_highmem
= 1,
1162 static int __init
setup_fail_page_alloc(char *str
)
1164 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1166 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1168 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1170 if (order
< fail_page_alloc
.min_order
)
1172 if (gfp_mask
& __GFP_NOFAIL
)
1174 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1176 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1179 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1182 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1184 static int __init
fail_page_alloc_debugfs(void)
1186 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1190 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1194 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1196 fail_page_alloc
.ignore_gfp_wait_file
=
1197 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1198 &fail_page_alloc
.ignore_gfp_wait
);
1200 fail_page_alloc
.ignore_gfp_highmem_file
=
1201 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1202 &fail_page_alloc
.ignore_gfp_highmem
);
1203 fail_page_alloc
.min_order_file
=
1204 debugfs_create_u32("min-order", mode
, dir
,
1205 &fail_page_alloc
.min_order
);
1207 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1208 !fail_page_alloc
.ignore_gfp_highmem_file
||
1209 !fail_page_alloc
.min_order_file
) {
1211 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1212 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1213 debugfs_remove(fail_page_alloc
.min_order_file
);
1214 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1220 late_initcall(fail_page_alloc_debugfs
);
1222 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1224 #else /* CONFIG_FAIL_PAGE_ALLOC */
1226 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1231 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1234 * Return 1 if free pages are above 'mark'. This takes into account the order
1235 * of the allocation.
1237 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1238 int classzone_idx
, int alloc_flags
)
1240 /* free_pages my go negative - that's OK */
1242 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1245 if (alloc_flags
& ALLOC_HIGH
)
1247 if (alloc_flags
& ALLOC_HARDER
)
1250 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1252 for (o
= 0; o
< order
; o
++) {
1253 /* At the next order, this order's pages become unavailable */
1254 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1256 /* Require fewer higher order pages to be free */
1259 if (free_pages
<= min
)
1267 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1268 * skip over zones that are not allowed by the cpuset, or that have
1269 * been recently (in last second) found to be nearly full. See further
1270 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1271 * that have to skip over a lot of full or unallowed zones.
1273 * If the zonelist cache is present in the passed in zonelist, then
1274 * returns a pointer to the allowed node mask (either the current
1275 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1277 * If the zonelist cache is not available for this zonelist, does
1278 * nothing and returns NULL.
1280 * If the fullzones BITMAP in the zonelist cache is stale (more than
1281 * a second since last zap'd) then we zap it out (clear its bits.)
1283 * We hold off even calling zlc_setup, until after we've checked the
1284 * first zone in the zonelist, on the theory that most allocations will
1285 * be satisfied from that first zone, so best to examine that zone as
1286 * quickly as we can.
1288 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1290 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1291 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1293 zlc
= zonelist
->zlcache_ptr
;
1297 if (time_after(jiffies
, zlc
->last_full_zap
+ HZ
)) {
1298 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1299 zlc
->last_full_zap
= jiffies
;
1302 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1303 &cpuset_current_mems_allowed
:
1304 &node_states
[N_HIGH_MEMORY
];
1305 return allowednodes
;
1309 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1310 * if it is worth looking at further for free memory:
1311 * 1) Check that the zone isn't thought to be full (doesn't have its
1312 * bit set in the zonelist_cache fullzones BITMAP).
1313 * 2) Check that the zones node (obtained from the zonelist_cache
1314 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1315 * Return true (non-zero) if zone is worth looking at further, or
1316 * else return false (zero) if it is not.
1318 * This check -ignores- the distinction between various watermarks,
1319 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1320 * found to be full for any variation of these watermarks, it will
1321 * be considered full for up to one second by all requests, unless
1322 * we are so low on memory on all allowed nodes that we are forced
1323 * into the second scan of the zonelist.
1325 * In the second scan we ignore this zonelist cache and exactly
1326 * apply the watermarks to all zones, even it is slower to do so.
1327 * We are low on memory in the second scan, and should leave no stone
1328 * unturned looking for a free page.
1330 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1331 nodemask_t
*allowednodes
)
1333 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1334 int i
; /* index of *z in zonelist zones */
1335 int n
; /* node that zone *z is on */
1337 zlc
= zonelist
->zlcache_ptr
;
1341 i
= z
- zonelist
->_zonerefs
;
1344 /* This zone is worth trying if it is allowed but not full */
1345 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1349 * Given 'z' scanning a zonelist, set the corresponding bit in
1350 * zlc->fullzones, so that subsequent attempts to allocate a page
1351 * from that zone don't waste time re-examining it.
1353 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1355 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1356 int i
; /* index of *z in zonelist zones */
1358 zlc
= zonelist
->zlcache_ptr
;
1362 i
= z
- zonelist
->_zonerefs
;
1364 set_bit(i
, zlc
->fullzones
);
1367 #else /* CONFIG_NUMA */
1369 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1374 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zoneref
*z
,
1375 nodemask_t
*allowednodes
)
1380 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zoneref
*z
)
1383 #endif /* CONFIG_NUMA */
1386 * get_page_from_freelist goes through the zonelist trying to allocate
1389 static struct page
*
1390 get_page_from_freelist(gfp_t gfp_mask
, nodemask_t
*nodemask
, unsigned int order
,
1391 struct zonelist
*zonelist
, int high_zoneidx
, int alloc_flags
,
1392 struct zone
*preferred_zone
, int migratetype
)
1395 struct page
*page
= NULL
;
1398 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1399 int zlc_active
= 0; /* set if using zonelist_cache */
1400 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1402 if (WARN_ON_ONCE(order
>= MAX_ORDER
))
1405 classzone_idx
= zone_idx(preferred_zone
);
1408 * Scan zonelist, looking for a zone with enough free.
1409 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1411 for_each_zone_zonelist_nodemask(zone
, z
, zonelist
,
1412 high_zoneidx
, nodemask
) {
1413 if (NUMA_BUILD
&& zlc_active
&&
1414 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1416 if ((alloc_flags
& ALLOC_CPUSET
) &&
1417 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1420 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1422 if (alloc_flags
& ALLOC_WMARK_MIN
)
1423 mark
= zone
->pages_min
;
1424 else if (alloc_flags
& ALLOC_WMARK_LOW
)
1425 mark
= zone
->pages_low
;
1427 mark
= zone
->pages_high
;
1428 if (!zone_watermark_ok(zone
, order
, mark
,
1429 classzone_idx
, alloc_flags
)) {
1430 if (!zone_reclaim_mode
||
1431 !zone_reclaim(zone
, gfp_mask
, order
))
1432 goto this_zone_full
;
1436 page
= buffered_rmqueue(preferred_zone
, zone
, order
,
1437 gfp_mask
, migratetype
);
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
, struct zone
*preferred_zone
,
1504 /* Acquire the OOM killer lock for the zones in zonelist */
1505 if (!try_set_zone_oom(zonelist
, gfp_mask
)) {
1506 schedule_timeout_uninterruptible(1);
1511 * Go through the zonelist yet one more time, keep very high watermark
1512 * here, this is only to catch a parallel oom killing, we must fail if
1513 * we're still under heavy pressure.
1515 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
,
1516 order
, zonelist
, high_zoneidx
,
1517 ALLOC_WMARK_HIGH
|ALLOC_CPUSET
,
1518 preferred_zone
, migratetype
);
1522 /* The OOM killer will not help higher order allocs */
1523 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
1526 /* Exhausted what can be done so it's blamo time */
1527 out_of_memory(zonelist
, gfp_mask
, order
);
1530 clear_zonelist_oom(zonelist
, gfp_mask
);
1534 /* The really slow allocator path where we enter direct reclaim */
1535 static inline struct page
*
1536 __alloc_pages_direct_reclaim(gfp_t gfp_mask
, unsigned int order
,
1537 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1538 nodemask_t
*nodemask
, int alloc_flags
, struct zone
*preferred_zone
,
1539 int migratetype
, unsigned long *did_some_progress
)
1541 struct page
*page
= NULL
;
1542 struct reclaim_state reclaim_state
;
1543 struct task_struct
*p
= current
;
1547 /* We now go into synchronous reclaim */
1548 cpuset_memory_pressure_bump();
1551 * The task's cpuset might have expanded its set of allowable nodes
1553 p
->flags
|= PF_MEMALLOC
;
1554 lockdep_set_current_reclaim_state(gfp_mask
);
1555 reclaim_state
.reclaimed_slab
= 0;
1556 p
->reclaim_state
= &reclaim_state
;
1558 *did_some_progress
= try_to_free_pages(zonelist
, order
, gfp_mask
, nodemask
);
1560 p
->reclaim_state
= NULL
;
1561 lockdep_clear_current_reclaim_state();
1562 p
->flags
&= ~PF_MEMALLOC
;
1569 if (likely(*did_some_progress
))
1570 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1571 zonelist
, high_zoneidx
,
1572 alloc_flags
, preferred_zone
,
1578 * This is called in the allocator slow-path if the allocation request is of
1579 * sufficient urgency to ignore watermarks and take other desperate measures
1581 static inline struct page
*
1582 __alloc_pages_high_priority(gfp_t gfp_mask
, unsigned int order
,
1583 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1584 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1590 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
,
1591 zonelist
, high_zoneidx
, ALLOC_NO_WATERMARKS
,
1592 preferred_zone
, migratetype
);
1594 if (!page
&& gfp_mask
& __GFP_NOFAIL
)
1595 congestion_wait(WRITE
, HZ
/50);
1596 } while (!page
&& (gfp_mask
& __GFP_NOFAIL
));
1602 void wake_all_kswapd(unsigned int order
, struct zonelist
*zonelist
,
1603 enum zone_type high_zoneidx
)
1608 for_each_zone_zonelist(zone
, z
, zonelist
, high_zoneidx
)
1609 wakeup_kswapd(zone
, order
);
1613 gfp_to_alloc_flags(gfp_t gfp_mask
)
1615 struct task_struct
*p
= current
;
1616 int alloc_flags
= ALLOC_WMARK_MIN
| ALLOC_CPUSET
;
1617 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1619 /* __GFP_HIGH is assumed to be the same as ALLOC_HIGH to save a branch. */
1620 BUILD_BUG_ON(__GFP_HIGH
!= ALLOC_HIGH
);
1623 * The caller may dip into page reserves a bit more if the caller
1624 * cannot run direct reclaim, or if the caller has realtime scheduling
1625 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1626 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1628 alloc_flags
|= (gfp_mask
& __GFP_HIGH
);
1631 alloc_flags
|= ALLOC_HARDER
;
1633 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1634 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1636 alloc_flags
&= ~ALLOC_CPUSET
;
1637 } else if (unlikely(rt_task(p
)))
1638 alloc_flags
|= ALLOC_HARDER
;
1640 if (likely(!(gfp_mask
& __GFP_NOMEMALLOC
))) {
1641 if (!in_interrupt() &&
1642 ((p
->flags
& PF_MEMALLOC
) ||
1643 unlikely(test_thread_flag(TIF_MEMDIE
))))
1644 alloc_flags
|= ALLOC_NO_WATERMARKS
;
1650 static inline struct page
*
1651 __alloc_pages_slowpath(gfp_t gfp_mask
, unsigned int order
,
1652 struct zonelist
*zonelist
, enum zone_type high_zoneidx
,
1653 nodemask_t
*nodemask
, struct zone
*preferred_zone
,
1656 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1657 struct page
*page
= NULL
;
1659 unsigned long pages_reclaimed
= 0;
1660 unsigned long did_some_progress
;
1661 struct task_struct
*p
= current
;
1664 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1665 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1666 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1667 * using a larger set of nodes after it has established that the
1668 * allowed per node queues are empty and that nodes are
1671 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1674 wake_all_kswapd(order
, zonelist
, high_zoneidx
);
1677 * OK, we're below the kswapd watermark and have kicked background
1678 * reclaim. Now things get more complex, so set up alloc_flags according
1679 * to how we want to proceed.
1681 alloc_flags
= gfp_to_alloc_flags(gfp_mask
);
1684 /* This is the last chance, in general, before the goto nopage. */
1685 page
= get_page_from_freelist(gfp_mask
, nodemask
, order
, zonelist
,
1686 high_zoneidx
, alloc_flags
& ~ALLOC_NO_WATERMARKS
,
1687 preferred_zone
, migratetype
);
1692 /* Allocate without watermarks if the context allows */
1693 if (alloc_flags
& ALLOC_NO_WATERMARKS
) {
1694 page
= __alloc_pages_high_priority(gfp_mask
, order
,
1695 zonelist
, high_zoneidx
, nodemask
,
1696 preferred_zone
, migratetype
);
1701 /* Atomic allocations - we can't balance anything */
1705 /* Avoid recursion of direct reclaim */
1706 if (p
->flags
& PF_MEMALLOC
)
1709 /* Try direct reclaim and then allocating */
1710 page
= __alloc_pages_direct_reclaim(gfp_mask
, order
,
1711 zonelist
, high_zoneidx
,
1713 alloc_flags
, preferred_zone
,
1714 migratetype
, &did_some_progress
);
1719 * If we failed to make any progress reclaiming, then we are
1720 * running out of options and have to consider going OOM
1722 if (!did_some_progress
) {
1723 if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1724 page
= __alloc_pages_may_oom(gfp_mask
, order
,
1725 zonelist
, high_zoneidx
,
1726 nodemask
, preferred_zone
,
1732 * The OOM killer does not trigger for high-order allocations
1733 * but if no progress is being made, there are no other
1734 * options and retrying is unlikely to help
1736 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
1743 /* Check if we should retry the allocation */
1744 pages_reclaimed
+= did_some_progress
;
1745 if (should_alloc_retry(gfp_mask
, order
, pages_reclaimed
)) {
1746 /* Wait for some write requests to complete then retry */
1747 congestion_wait(WRITE
, HZ
/50);
1752 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1753 printk(KERN_WARNING
"%s: page allocation failure."
1754 " order:%d, mode:0x%x\n",
1755 p
->comm
, order
, gfp_mask
);
1765 * This is the 'heart' of the zoned buddy allocator.
1768 __alloc_pages_nodemask(gfp_t gfp_mask
, unsigned int order
,
1769 struct zonelist
*zonelist
, nodemask_t
*nodemask
)
1771 enum zone_type high_zoneidx
= gfp_zone(gfp_mask
);
1772 struct zone
*preferred_zone
;
1774 int migratetype
= allocflags_to_migratetype(gfp_mask
);
1776 lockdep_trace_alloc(gfp_mask
);
1778 might_sleep_if(gfp_mask
& __GFP_WAIT
);
1780 if (should_fail_alloc_page(gfp_mask
, order
))
1784 * Check the zones suitable for the gfp_mask contain at least one
1785 * valid zone. It's possible to have an empty zonelist as a result
1786 * of GFP_THISNODE and a memoryless node
1788 if (unlikely(!zonelist
->_zonerefs
->zone
))
1791 /* The preferred zone is used for statistics later */
1792 first_zones_zonelist(zonelist
, high_zoneidx
, nodemask
, &preferred_zone
);
1793 if (!preferred_zone
)
1796 /* First allocation attempt */
1797 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, nodemask
, order
,
1798 zonelist
, high_zoneidx
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
,
1799 preferred_zone
, migratetype
);
1800 if (unlikely(!page
))
1801 page
= __alloc_pages_slowpath(gfp_mask
, order
,
1802 zonelist
, high_zoneidx
, nodemask
,
1803 preferred_zone
, migratetype
);
1807 EXPORT_SYMBOL(__alloc_pages_nodemask
);
1810 * Common helper functions.
1812 unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1815 page
= alloc_pages(gfp_mask
, order
);
1818 return (unsigned long) page_address(page
);
1821 EXPORT_SYMBOL(__get_free_pages
);
1823 unsigned long get_zeroed_page(gfp_t gfp_mask
)
1828 * get_zeroed_page() returns a 32-bit address, which cannot represent
1831 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1833 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1835 return (unsigned long) page_address(page
);
1839 EXPORT_SYMBOL(get_zeroed_page
);
1841 void __pagevec_free(struct pagevec
*pvec
)
1843 int i
= pagevec_count(pvec
);
1846 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1849 void __free_pages(struct page
*page
, unsigned int order
)
1851 if (put_page_testzero(page
)) {
1853 free_hot_page(page
);
1855 __free_pages_ok(page
, order
);
1859 EXPORT_SYMBOL(__free_pages
);
1861 void free_pages(unsigned long addr
, unsigned int order
)
1864 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1865 __free_pages(virt_to_page((void *)addr
), order
);
1869 EXPORT_SYMBOL(free_pages
);
1872 * alloc_pages_exact - allocate an exact number physically-contiguous pages.
1873 * @size: the number of bytes to allocate
1874 * @gfp_mask: GFP flags for the allocation
1876 * This function is similar to alloc_pages(), except that it allocates the
1877 * minimum number of pages to satisfy the request. alloc_pages() can only
1878 * allocate memory in power-of-two pages.
1880 * This function is also limited by MAX_ORDER.
1882 * Memory allocated by this function must be released by free_pages_exact().
1884 void *alloc_pages_exact(size_t size
, gfp_t gfp_mask
)
1886 unsigned int order
= get_order(size
);
1889 addr
= __get_free_pages(gfp_mask
, order
);
1891 unsigned long alloc_end
= addr
+ (PAGE_SIZE
<< order
);
1892 unsigned long used
= addr
+ PAGE_ALIGN(size
);
1894 split_page(virt_to_page(addr
), order
);
1895 while (used
< alloc_end
) {
1901 return (void *)addr
;
1903 EXPORT_SYMBOL(alloc_pages_exact
);
1906 * free_pages_exact - release memory allocated via alloc_pages_exact()
1907 * @virt: the value returned by alloc_pages_exact.
1908 * @size: size of allocation, same value as passed to alloc_pages_exact().
1910 * Release the memory allocated by a previous call to alloc_pages_exact.
1912 void free_pages_exact(void *virt
, size_t size
)
1914 unsigned long addr
= (unsigned long)virt
;
1915 unsigned long end
= addr
+ PAGE_ALIGN(size
);
1917 while (addr
< end
) {
1922 EXPORT_SYMBOL(free_pages_exact
);
1924 static unsigned int nr_free_zone_pages(int offset
)
1929 /* Just pick one node, since fallback list is circular */
1930 unsigned int sum
= 0;
1932 struct zonelist
*zonelist
= node_zonelist(numa_node_id(), GFP_KERNEL
);
1934 for_each_zone_zonelist(zone
, z
, zonelist
, offset
) {
1935 unsigned long size
= zone
->present_pages
;
1936 unsigned long high
= zone
->pages_high
;
1945 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1947 unsigned int nr_free_buffer_pages(void)
1949 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1951 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
1954 * Amount of free RAM allocatable within all zones
1956 unsigned int nr_free_pagecache_pages(void)
1958 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
1961 static inline void show_node(struct zone
*zone
)
1964 printk("Node %d ", zone_to_nid(zone
));
1967 void si_meminfo(struct sysinfo
*val
)
1969 val
->totalram
= totalram_pages
;
1971 val
->freeram
= global_page_state(NR_FREE_PAGES
);
1972 val
->bufferram
= nr_blockdev_pages();
1973 val
->totalhigh
= totalhigh_pages
;
1974 val
->freehigh
= nr_free_highpages();
1975 val
->mem_unit
= PAGE_SIZE
;
1978 EXPORT_SYMBOL(si_meminfo
);
1981 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1983 pg_data_t
*pgdat
= NODE_DATA(nid
);
1985 val
->totalram
= pgdat
->node_present_pages
;
1986 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
1987 #ifdef CONFIG_HIGHMEM
1988 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1989 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
1995 val
->mem_unit
= PAGE_SIZE
;
1999 #define K(x) ((x) << (PAGE_SHIFT-10))
2002 * Show free area list (used inside shift_scroll-lock stuff)
2003 * We also calculate the percentage fragmentation. We do this by counting the
2004 * memory on each free list with the exception of the first item on the list.
2006 void show_free_areas(void)
2011 for_each_populated_zone(zone
) {
2013 printk("%s per-cpu:\n", zone
->name
);
2015 for_each_online_cpu(cpu
) {
2016 struct per_cpu_pageset
*pageset
;
2018 pageset
= zone_pcp(zone
, cpu
);
2020 printk("CPU %4d: hi:%5d, btch:%4d usd:%4d\n",
2021 cpu
, pageset
->pcp
.high
,
2022 pageset
->pcp
.batch
, pageset
->pcp
.count
);
2026 printk("Active_anon:%lu active_file:%lu inactive_anon:%lu\n"
2027 " inactive_file:%lu"
2028 //TODO: check/adjust line lengths
2029 #ifdef CONFIG_UNEVICTABLE_LRU
2032 " dirty:%lu writeback:%lu unstable:%lu\n"
2033 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
2034 global_page_state(NR_ACTIVE_ANON
),
2035 global_page_state(NR_ACTIVE_FILE
),
2036 global_page_state(NR_INACTIVE_ANON
),
2037 global_page_state(NR_INACTIVE_FILE
),
2038 #ifdef CONFIG_UNEVICTABLE_LRU
2039 global_page_state(NR_UNEVICTABLE
),
2041 global_page_state(NR_FILE_DIRTY
),
2042 global_page_state(NR_WRITEBACK
),
2043 global_page_state(NR_UNSTABLE_NFS
),
2044 global_page_state(NR_FREE_PAGES
),
2045 global_page_state(NR_SLAB_RECLAIMABLE
) +
2046 global_page_state(NR_SLAB_UNRECLAIMABLE
),
2047 global_page_state(NR_FILE_MAPPED
),
2048 global_page_state(NR_PAGETABLE
),
2049 global_page_state(NR_BOUNCE
));
2051 for_each_populated_zone(zone
) {
2060 " active_anon:%lukB"
2061 " inactive_anon:%lukB"
2062 " active_file:%lukB"
2063 " inactive_file:%lukB"
2064 #ifdef CONFIG_UNEVICTABLE_LRU
2065 " unevictable:%lukB"
2068 " pages_scanned:%lu"
2069 " all_unreclaimable? %s"
2072 K(zone_page_state(zone
, NR_FREE_PAGES
)),
2075 K(zone
->pages_high
),
2076 K(zone_page_state(zone
, NR_ACTIVE_ANON
)),
2077 K(zone_page_state(zone
, NR_INACTIVE_ANON
)),
2078 K(zone_page_state(zone
, NR_ACTIVE_FILE
)),
2079 K(zone_page_state(zone
, NR_INACTIVE_FILE
)),
2080 #ifdef CONFIG_UNEVICTABLE_LRU
2081 K(zone_page_state(zone
, NR_UNEVICTABLE
)),
2083 K(zone
->present_pages
),
2084 zone
->pages_scanned
,
2085 (zone_is_all_unreclaimable(zone
) ? "yes" : "no")
2087 printk("lowmem_reserve[]:");
2088 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2089 printk(" %lu", zone
->lowmem_reserve
[i
]);
2093 for_each_populated_zone(zone
) {
2094 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
2097 printk("%s: ", zone
->name
);
2099 spin_lock_irqsave(&zone
->lock
, flags
);
2100 for (order
= 0; order
< MAX_ORDER
; order
++) {
2101 nr
[order
] = zone
->free_area
[order
].nr_free
;
2102 total
+= nr
[order
] << order
;
2104 spin_unlock_irqrestore(&zone
->lock
, flags
);
2105 for (order
= 0; order
< MAX_ORDER
; order
++)
2106 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
2107 printk("= %lukB\n", K(total
));
2110 printk("%ld total pagecache pages\n", global_page_state(NR_FILE_PAGES
));
2112 show_swap_cache_info();
2115 static void zoneref_set_zone(struct zone
*zone
, struct zoneref
*zoneref
)
2117 zoneref
->zone
= zone
;
2118 zoneref
->zone_idx
= zone_idx(zone
);
2122 * Builds allocation fallback zone lists.
2124 * Add all populated zones of a node to the zonelist.
2126 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
2127 int nr_zones
, enum zone_type zone_type
)
2131 BUG_ON(zone_type
>= MAX_NR_ZONES
);
2136 zone
= pgdat
->node_zones
+ zone_type
;
2137 if (populated_zone(zone
)) {
2138 zoneref_set_zone(zone
,
2139 &zonelist
->_zonerefs
[nr_zones
++]);
2140 check_highest_zone(zone_type
);
2143 } while (zone_type
);
2150 * 0 = automatic detection of better ordering.
2151 * 1 = order by ([node] distance, -zonetype)
2152 * 2 = order by (-zonetype, [node] distance)
2154 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
2155 * the same zonelist. So only NUMA can configure this param.
2157 #define ZONELIST_ORDER_DEFAULT 0
2158 #define ZONELIST_ORDER_NODE 1
2159 #define ZONELIST_ORDER_ZONE 2
2161 /* zonelist order in the kernel.
2162 * set_zonelist_order() will set this to NODE or ZONE.
2164 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2165 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
2169 /* The value user specified ....changed by config */
2170 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2171 /* string for sysctl */
2172 #define NUMA_ZONELIST_ORDER_LEN 16
2173 char numa_zonelist_order
[16] = "default";
2176 * interface for configure zonelist ordering.
2177 * command line option "numa_zonelist_order"
2178 * = "[dD]efault - default, automatic configuration.
2179 * = "[nN]ode - order by node locality, then by zone within node
2180 * = "[zZ]one - order by zone, then by locality within zone
2183 static int __parse_numa_zonelist_order(char *s
)
2185 if (*s
== 'd' || *s
== 'D') {
2186 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
2187 } else if (*s
== 'n' || *s
== 'N') {
2188 user_zonelist_order
= ZONELIST_ORDER_NODE
;
2189 } else if (*s
== 'z' || *s
== 'Z') {
2190 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
2193 "Ignoring invalid numa_zonelist_order value: "
2200 static __init
int setup_numa_zonelist_order(char *s
)
2203 return __parse_numa_zonelist_order(s
);
2206 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
2209 * sysctl handler for numa_zonelist_order
2211 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
2212 struct file
*file
, void __user
*buffer
, size_t *length
,
2215 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
2219 strncpy(saved_string
, (char*)table
->data
,
2220 NUMA_ZONELIST_ORDER_LEN
);
2221 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
2225 int oldval
= user_zonelist_order
;
2226 if (__parse_numa_zonelist_order((char*)table
->data
)) {
2228 * bogus value. restore saved string
2230 strncpy((char*)table
->data
, saved_string
,
2231 NUMA_ZONELIST_ORDER_LEN
);
2232 user_zonelist_order
= oldval
;
2233 } else if (oldval
!= user_zonelist_order
)
2234 build_all_zonelists();
2240 #define MAX_NODE_LOAD (num_online_nodes())
2241 static int node_load
[MAX_NUMNODES
];
2244 * find_next_best_node - find the next node that should appear in a given node's fallback list
2245 * @node: node whose fallback list we're appending
2246 * @used_node_mask: nodemask_t of already used nodes
2248 * We use a number of factors to determine which is the next node that should
2249 * appear on a given node's fallback list. The node should not have appeared
2250 * already in @node's fallback list, and it should be the next closest node
2251 * according to the distance array (which contains arbitrary distance values
2252 * from each node to each node in the system), and should also prefer nodes
2253 * with no CPUs, since presumably they'll have very little allocation pressure
2254 * on them otherwise.
2255 * It returns -1 if no node is found.
2257 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2260 int min_val
= INT_MAX
;
2262 const struct cpumask
*tmp
= cpumask_of_node(0);
2264 /* Use the local node if we haven't already */
2265 if (!node_isset(node
, *used_node_mask
)) {
2266 node_set(node
, *used_node_mask
);
2270 for_each_node_state(n
, N_HIGH_MEMORY
) {
2272 /* Don't want a node to appear more than once */
2273 if (node_isset(n
, *used_node_mask
))
2276 /* Use the distance array to find the distance */
2277 val
= node_distance(node
, n
);
2279 /* Penalize nodes under us ("prefer the next node") */
2282 /* Give preference to headless and unused nodes */
2283 tmp
= cpumask_of_node(n
);
2284 if (!cpumask_empty(tmp
))
2285 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2287 /* Slight preference for less loaded node */
2288 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2289 val
+= node_load
[n
];
2291 if (val
< min_val
) {
2298 node_set(best_node
, *used_node_mask
);
2305 * Build zonelists ordered by node and zones within node.
2306 * This results in maximum locality--normal zone overflows into local
2307 * DMA zone, if any--but risks exhausting DMA zone.
2309 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2312 struct zonelist
*zonelist
;
2314 zonelist
= &pgdat
->node_zonelists
[0];
2315 for (j
= 0; zonelist
->_zonerefs
[j
].zone
!= NULL
; j
++)
2317 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2319 zonelist
->_zonerefs
[j
].zone
= NULL
;
2320 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2324 * Build gfp_thisnode zonelists
2326 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2329 struct zonelist
*zonelist
;
2331 zonelist
= &pgdat
->node_zonelists
[1];
2332 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2333 zonelist
->_zonerefs
[j
].zone
= NULL
;
2334 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2338 * Build zonelists ordered by zone and nodes within zones.
2339 * This results in conserving DMA zone[s] until all Normal memory is
2340 * exhausted, but results in overflowing to remote node while memory
2341 * may still exist in local DMA zone.
2343 static int node_order
[MAX_NUMNODES
];
2345 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2348 int zone_type
; /* needs to be signed */
2350 struct zonelist
*zonelist
;
2352 zonelist
= &pgdat
->node_zonelists
[0];
2354 for (zone_type
= MAX_NR_ZONES
- 1; zone_type
>= 0; zone_type
--) {
2355 for (j
= 0; j
< nr_nodes
; j
++) {
2356 node
= node_order
[j
];
2357 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2358 if (populated_zone(z
)) {
2360 &zonelist
->_zonerefs
[pos
++]);
2361 check_highest_zone(zone_type
);
2365 zonelist
->_zonerefs
[pos
].zone
= NULL
;
2366 zonelist
->_zonerefs
[pos
].zone_idx
= 0;
2369 static int default_zonelist_order(void)
2372 unsigned long low_kmem_size
,total_size
;
2376 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2377 * If they are really small and used heavily, the system can fall
2378 * into OOM very easily.
2379 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2381 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2384 for_each_online_node(nid
) {
2385 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2386 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2387 if (populated_zone(z
)) {
2388 if (zone_type
< ZONE_NORMAL
)
2389 low_kmem_size
+= z
->present_pages
;
2390 total_size
+= z
->present_pages
;
2394 if (!low_kmem_size
|| /* there are no DMA area. */
2395 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2396 return ZONELIST_ORDER_NODE
;
2398 * look into each node's config.
2399 * If there is a node whose DMA/DMA32 memory is very big area on
2400 * local memory, NODE_ORDER may be suitable.
2402 average_size
= total_size
/
2403 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2404 for_each_online_node(nid
) {
2407 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2408 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2409 if (populated_zone(z
)) {
2410 if (zone_type
< ZONE_NORMAL
)
2411 low_kmem_size
+= z
->present_pages
;
2412 total_size
+= z
->present_pages
;
2415 if (low_kmem_size
&&
2416 total_size
> average_size
&& /* ignore small node */
2417 low_kmem_size
> total_size
* 70/100)
2418 return ZONELIST_ORDER_NODE
;
2420 return ZONELIST_ORDER_ZONE
;
2423 static void set_zonelist_order(void)
2425 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2426 current_zonelist_order
= default_zonelist_order();
2428 current_zonelist_order
= user_zonelist_order
;
2431 static void build_zonelists(pg_data_t
*pgdat
)
2435 nodemask_t used_mask
;
2436 int local_node
, prev_node
;
2437 struct zonelist
*zonelist
;
2438 int order
= current_zonelist_order
;
2440 /* initialize zonelists */
2441 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2442 zonelist
= pgdat
->node_zonelists
+ i
;
2443 zonelist
->_zonerefs
[0].zone
= NULL
;
2444 zonelist
->_zonerefs
[0].zone_idx
= 0;
2447 /* NUMA-aware ordering of nodes */
2448 local_node
= pgdat
->node_id
;
2449 load
= num_online_nodes();
2450 prev_node
= local_node
;
2451 nodes_clear(used_mask
);
2453 memset(node_load
, 0, sizeof(node_load
));
2454 memset(node_order
, 0, sizeof(node_order
));
2457 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2458 int distance
= node_distance(local_node
, node
);
2461 * If another node is sufficiently far away then it is better
2462 * to reclaim pages in a zone before going off node.
2464 if (distance
> RECLAIM_DISTANCE
)
2465 zone_reclaim_mode
= 1;
2468 * We don't want to pressure a particular node.
2469 * So adding penalty to the first node in same
2470 * distance group to make it round-robin.
2472 if (distance
!= node_distance(local_node
, prev_node
))
2473 node_load
[node
] = load
;
2477 if (order
== ZONELIST_ORDER_NODE
)
2478 build_zonelists_in_node_order(pgdat
, node
);
2480 node_order
[j
++] = node
; /* remember order */
2483 if (order
== ZONELIST_ORDER_ZONE
) {
2484 /* calculate node order -- i.e., DMA last! */
2485 build_zonelists_in_zone_order(pgdat
, j
);
2488 build_thisnode_zonelists(pgdat
);
2491 /* Construct the zonelist performance cache - see further mmzone.h */
2492 static void build_zonelist_cache(pg_data_t
*pgdat
)
2494 struct zonelist
*zonelist
;
2495 struct zonelist_cache
*zlc
;
2498 zonelist
= &pgdat
->node_zonelists
[0];
2499 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2500 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2501 for (z
= zonelist
->_zonerefs
; z
->zone
; z
++)
2502 zlc
->z_to_n
[z
- zonelist
->_zonerefs
] = zonelist_node_idx(z
);
2506 #else /* CONFIG_NUMA */
2508 static void set_zonelist_order(void)
2510 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2513 static void build_zonelists(pg_data_t
*pgdat
)
2515 int node
, local_node
;
2517 struct zonelist
*zonelist
;
2519 local_node
= pgdat
->node_id
;
2521 zonelist
= &pgdat
->node_zonelists
[0];
2522 j
= build_zonelists_node(pgdat
, zonelist
, 0, MAX_NR_ZONES
- 1);
2525 * Now we build the zonelist so that it contains the zones
2526 * of all the other nodes.
2527 * We don't want to pressure a particular node, so when
2528 * building the zones for node N, we make sure that the
2529 * zones coming right after the local ones are those from
2530 * node N+1 (modulo N)
2532 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2533 if (!node_online(node
))
2535 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2538 for (node
= 0; node
< local_node
; node
++) {
2539 if (!node_online(node
))
2541 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
,
2545 zonelist
->_zonerefs
[j
].zone
= NULL
;
2546 zonelist
->_zonerefs
[j
].zone_idx
= 0;
2549 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2550 static void build_zonelist_cache(pg_data_t
*pgdat
)
2552 pgdat
->node_zonelists
[0].zlcache_ptr
= NULL
;
2555 #endif /* CONFIG_NUMA */
2557 /* return values int ....just for stop_machine() */
2558 static int __build_all_zonelists(void *dummy
)
2562 for_each_online_node(nid
) {
2563 pg_data_t
*pgdat
= NODE_DATA(nid
);
2565 build_zonelists(pgdat
);
2566 build_zonelist_cache(pgdat
);
2571 void build_all_zonelists(void)
2573 set_zonelist_order();
2575 if (system_state
== SYSTEM_BOOTING
) {
2576 __build_all_zonelists(NULL
);
2577 mminit_verify_zonelist();
2578 cpuset_init_current_mems_allowed();
2580 /* we have to stop all cpus to guarantee there is no user
2582 stop_machine(__build_all_zonelists
, NULL
, NULL
);
2583 /* cpuset refresh routine should be here */
2585 vm_total_pages
= nr_free_pagecache_pages();
2587 * Disable grouping by mobility if the number of pages in the
2588 * system is too low to allow the mechanism to work. It would be
2589 * more accurate, but expensive to check per-zone. This check is
2590 * made on memory-hotadd so a system can start with mobility
2591 * disabled and enable it later
2593 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
2594 page_group_by_mobility_disabled
= 1;
2596 page_group_by_mobility_disabled
= 0;
2598 printk("Built %i zonelists in %s order, mobility grouping %s. "
2599 "Total pages: %ld\n",
2601 zonelist_order_name
[current_zonelist_order
],
2602 page_group_by_mobility_disabled
? "off" : "on",
2605 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2610 * Helper functions to size the waitqueue hash table.
2611 * Essentially these want to choose hash table sizes sufficiently
2612 * large so that collisions trying to wait on pages are rare.
2613 * But in fact, the number of active page waitqueues on typical
2614 * systems is ridiculously low, less than 200. So this is even
2615 * conservative, even though it seems large.
2617 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2618 * waitqueues, i.e. the size of the waitq table given the number of pages.
2620 #define PAGES_PER_WAITQUEUE 256
2622 #ifndef CONFIG_MEMORY_HOTPLUG
2623 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2625 unsigned long size
= 1;
2627 pages
/= PAGES_PER_WAITQUEUE
;
2629 while (size
< pages
)
2633 * Once we have dozens or even hundreds of threads sleeping
2634 * on IO we've got bigger problems than wait queue collision.
2635 * Limit the size of the wait table to a reasonable size.
2637 size
= min(size
, 4096UL);
2639 return max(size
, 4UL);
2643 * A zone's size might be changed by hot-add, so it is not possible to determine
2644 * a suitable size for its wait_table. So we use the maximum size now.
2646 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2648 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2649 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2650 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2652 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2653 * or more by the traditional way. (See above). It equals:
2655 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2656 * ia64(16K page size) : = ( 8G + 4M)byte.
2657 * powerpc (64K page size) : = (32G +16M)byte.
2659 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2666 * This is an integer logarithm so that shifts can be used later
2667 * to extract the more random high bits from the multiplicative
2668 * hash function before the remainder is taken.
2670 static inline unsigned long wait_table_bits(unsigned long size
)
2675 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2678 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2679 * of blocks reserved is based on zone->pages_min. The memory within the
2680 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2681 * higher will lead to a bigger reserve which will get freed as contiguous
2682 * blocks as reclaim kicks in
2684 static void setup_zone_migrate_reserve(struct zone
*zone
)
2686 unsigned long start_pfn
, pfn
, end_pfn
;
2688 unsigned long reserve
, block_migratetype
;
2690 /* Get the start pfn, end pfn and the number of blocks to reserve */
2691 start_pfn
= zone
->zone_start_pfn
;
2692 end_pfn
= start_pfn
+ zone
->spanned_pages
;
2693 reserve
= roundup(zone
->pages_min
, pageblock_nr_pages
) >>
2696 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
2697 if (!pfn_valid(pfn
))
2699 page
= pfn_to_page(pfn
);
2701 /* Watch out for overlapping nodes */
2702 if (page_to_nid(page
) != zone_to_nid(zone
))
2705 /* Blocks with reserved pages will never free, skip them. */
2706 if (PageReserved(page
))
2709 block_migratetype
= get_pageblock_migratetype(page
);
2711 /* If this block is reserved, account for it */
2712 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
2717 /* Suitable for reserving if this block is movable */
2718 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
2719 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
2720 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
2726 * If the reserve is met and this is a previous reserved block,
2729 if (block_migratetype
== MIGRATE_RESERVE
) {
2730 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2731 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
2737 * Initially all pages are reserved - free ones are freed
2738 * up by free_all_bootmem() once the early boot process is
2739 * done. Non-atomic initialization, single-pass.
2741 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2742 unsigned long start_pfn
, enum memmap_context context
)
2745 unsigned long end_pfn
= start_pfn
+ size
;
2749 if (highest_memmap_pfn
< end_pfn
- 1)
2750 highest_memmap_pfn
= end_pfn
- 1;
2752 z
= &NODE_DATA(nid
)->node_zones
[zone
];
2753 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2755 * There can be holes in boot-time mem_map[]s
2756 * handed to this function. They do not
2757 * exist on hotplugged memory.
2759 if (context
== MEMMAP_EARLY
) {
2760 if (!early_pfn_valid(pfn
))
2762 if (!early_pfn_in_nid(pfn
, nid
))
2765 page
= pfn_to_page(pfn
);
2766 set_page_links(page
, zone
, nid
, pfn
);
2767 mminit_verify_page_links(page
, zone
, nid
, pfn
);
2768 init_page_count(page
);
2769 reset_page_mapcount(page
);
2770 SetPageReserved(page
);
2772 * Mark the block movable so that blocks are reserved for
2773 * movable at startup. This will force kernel allocations
2774 * to reserve their blocks rather than leaking throughout
2775 * the address space during boot when many long-lived
2776 * kernel allocations are made. Later some blocks near
2777 * the start are marked MIGRATE_RESERVE by
2778 * setup_zone_migrate_reserve()
2780 * bitmap is created for zone's valid pfn range. but memmap
2781 * can be created for invalid pages (for alignment)
2782 * check here not to call set_pageblock_migratetype() against
2785 if ((z
->zone_start_pfn
<= pfn
)
2786 && (pfn
< z
->zone_start_pfn
+ z
->spanned_pages
)
2787 && !(pfn
& (pageblock_nr_pages
- 1)))
2788 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2790 INIT_LIST_HEAD(&page
->lru
);
2791 #ifdef WANT_PAGE_VIRTUAL
2792 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2793 if (!is_highmem_idx(zone
))
2794 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2799 static void __meminit
zone_init_free_lists(struct zone
*zone
)
2802 for_each_migratetype_order(order
, t
) {
2803 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
2804 zone
->free_area
[order
].nr_free
= 0;
2808 #ifndef __HAVE_ARCH_MEMMAP_INIT
2809 #define memmap_init(size, nid, zone, start_pfn) \
2810 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2813 static int zone_batchsize(struct zone
*zone
)
2819 * The per-cpu-pages pools are set to around 1000th of the
2820 * size of the zone. But no more than 1/2 of a meg.
2822 * OK, so we don't know how big the cache is. So guess.
2824 batch
= zone
->present_pages
/ 1024;
2825 if (batch
* PAGE_SIZE
> 512 * 1024)
2826 batch
= (512 * 1024) / PAGE_SIZE
;
2827 batch
/= 4; /* We effectively *= 4 below */
2832 * Clamp the batch to a 2^n - 1 value. Having a power
2833 * of 2 value was found to be more likely to have
2834 * suboptimal cache aliasing properties in some cases.
2836 * For example if 2 tasks are alternately allocating
2837 * batches of pages, one task can end up with a lot
2838 * of pages of one half of the possible page colors
2839 * and the other with pages of the other colors.
2841 batch
= rounddown_pow_of_two(batch
+ batch
/2) - 1;
2846 /* The deferral and batching of frees should be suppressed under NOMMU
2849 * The problem is that NOMMU needs to be able to allocate large chunks
2850 * of contiguous memory as there's no hardware page translation to
2851 * assemble apparent contiguous memory from discontiguous pages.
2853 * Queueing large contiguous runs of pages for batching, however,
2854 * causes the pages to actually be freed in smaller chunks. As there
2855 * can be a significant delay between the individual batches being
2856 * recycled, this leads to the once large chunks of space being
2857 * fragmented and becoming unavailable for high-order allocations.
2863 static void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2865 struct per_cpu_pages
*pcp
;
2867 memset(p
, 0, sizeof(*p
));
2871 pcp
->high
= 6 * batch
;
2872 pcp
->batch
= max(1UL, 1 * batch
);
2873 INIT_LIST_HEAD(&pcp
->list
);
2877 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2878 * to the value high for the pageset p.
2881 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2884 struct per_cpu_pages
*pcp
;
2888 pcp
->batch
= max(1UL, high
/4);
2889 if ((high
/4) > (PAGE_SHIFT
* 8))
2890 pcp
->batch
= PAGE_SHIFT
* 8;
2896 * Boot pageset table. One per cpu which is going to be used for all
2897 * zones and all nodes. The parameters will be set in such a way
2898 * that an item put on a list will immediately be handed over to
2899 * the buddy list. This is safe since pageset manipulation is done
2900 * with interrupts disabled.
2902 * Some NUMA counter updates may also be caught by the boot pagesets.
2904 * The boot_pagesets must be kept even after bootup is complete for
2905 * unused processors and/or zones. They do play a role for bootstrapping
2906 * hotplugged processors.
2908 * zoneinfo_show() and maybe other functions do
2909 * not check if the processor is online before following the pageset pointer.
2910 * Other parts of the kernel may not check if the zone is available.
2912 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
2915 * Dynamically allocate memory for the
2916 * per cpu pageset array in struct zone.
2918 static int __cpuinit
process_zones(int cpu
)
2920 struct zone
*zone
, *dzone
;
2921 int node
= cpu_to_node(cpu
);
2923 node_set_state(node
, N_CPU
); /* this node has a cpu */
2925 for_each_populated_zone(zone
) {
2926 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
2928 if (!zone_pcp(zone
, cpu
))
2931 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
2933 if (percpu_pagelist_fraction
)
2934 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
2935 (zone
->present_pages
/ percpu_pagelist_fraction
));
2940 for_each_zone(dzone
) {
2941 if (!populated_zone(dzone
))
2945 kfree(zone_pcp(dzone
, cpu
));
2946 zone_pcp(dzone
, cpu
) = NULL
;
2951 static inline void free_zone_pagesets(int cpu
)
2955 for_each_zone(zone
) {
2956 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
2958 /* Free per_cpu_pageset if it is slab allocated */
2959 if (pset
!= &boot_pageset
[cpu
])
2961 zone_pcp(zone
, cpu
) = NULL
;
2965 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
2966 unsigned long action
,
2969 int cpu
= (long)hcpu
;
2970 int ret
= NOTIFY_OK
;
2973 case CPU_UP_PREPARE
:
2974 case CPU_UP_PREPARE_FROZEN
:
2975 if (process_zones(cpu
))
2978 case CPU_UP_CANCELED
:
2979 case CPU_UP_CANCELED_FROZEN
:
2981 case CPU_DEAD_FROZEN
:
2982 free_zone_pagesets(cpu
);
2990 static struct notifier_block __cpuinitdata pageset_notifier
=
2991 { &pageset_cpuup_callback
, NULL
, 0 };
2993 void __init
setup_per_cpu_pageset(void)
2997 /* Initialize per_cpu_pageset for cpu 0.
2998 * A cpuup callback will do this for every cpu
2999 * as it comes online
3001 err
= process_zones(smp_processor_id());
3003 register_cpu_notifier(&pageset_notifier
);
3008 static noinline __init_refok
3009 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
3012 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3016 * The per-page waitqueue mechanism uses hashed waitqueues
3019 zone
->wait_table_hash_nr_entries
=
3020 wait_table_hash_nr_entries(zone_size_pages
);
3021 zone
->wait_table_bits
=
3022 wait_table_bits(zone
->wait_table_hash_nr_entries
);
3023 alloc_size
= zone
->wait_table_hash_nr_entries
3024 * sizeof(wait_queue_head_t
);
3026 if (!slab_is_available()) {
3027 zone
->wait_table
= (wait_queue_head_t
*)
3028 alloc_bootmem_node(pgdat
, alloc_size
);
3031 * This case means that a zone whose size was 0 gets new memory
3032 * via memory hot-add.
3033 * But it may be the case that a new node was hot-added. In
3034 * this case vmalloc() will not be able to use this new node's
3035 * memory - this wait_table must be initialized to use this new
3036 * node itself as well.
3037 * To use this new node's memory, further consideration will be
3040 zone
->wait_table
= vmalloc(alloc_size
);
3042 if (!zone
->wait_table
)
3045 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
3046 init_waitqueue_head(zone
->wait_table
+ i
);
3051 static __meminit
void zone_pcp_init(struct zone
*zone
)
3054 unsigned long batch
= zone_batchsize(zone
);
3056 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
3058 /* Early boot. Slab allocator not functional yet */
3059 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
3060 setup_pageset(&boot_pageset
[cpu
],0);
3062 setup_pageset(zone_pcp(zone
,cpu
), batch
);
3065 if (zone
->present_pages
)
3066 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
3067 zone
->name
, zone
->present_pages
, batch
);
3070 __meminit
int init_currently_empty_zone(struct zone
*zone
,
3071 unsigned long zone_start_pfn
,
3073 enum memmap_context context
)
3075 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
3077 ret
= zone_wait_table_init(zone
, size
);
3080 pgdat
->nr_zones
= zone_idx(zone
) + 1;
3082 zone
->zone_start_pfn
= zone_start_pfn
;
3084 mminit_dprintk(MMINIT_TRACE
, "memmap_init",
3085 "Initialising map node %d zone %lu pfns %lu -> %lu\n",
3087 (unsigned long)zone_idx(zone
),
3088 zone_start_pfn
, (zone_start_pfn
+ size
));
3090 zone_init_free_lists(zone
);
3095 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3097 * Basic iterator support. Return the first range of PFNs for a node
3098 * Note: nid == MAX_NUMNODES returns first region regardless of node
3100 static int __meminit
first_active_region_index_in_nid(int nid
)
3104 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3105 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
3112 * Basic iterator support. Return the next active range of PFNs for a node
3113 * Note: nid == MAX_NUMNODES returns next region regardless of node
3115 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
3117 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
3118 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
3124 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
3126 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
3127 * Architectures may implement their own version but if add_active_range()
3128 * was used and there are no special requirements, this is a convenient
3131 int __meminit
__early_pfn_to_nid(unsigned long pfn
)
3135 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3136 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
3137 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3139 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
3140 return early_node_map
[i
].nid
;
3142 /* This is a memory hole */
3145 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
3147 int __meminit
early_pfn_to_nid(unsigned long pfn
)
3151 nid
= __early_pfn_to_nid(pfn
);
3154 /* just returns 0 */
3158 #ifdef CONFIG_NODES_SPAN_OTHER_NODES
3159 bool __meminit
early_pfn_in_nid(unsigned long pfn
, int node
)
3163 nid
= __early_pfn_to_nid(pfn
);
3164 if (nid
>= 0 && nid
!= node
)
3170 /* Basic iterator support to walk early_node_map[] */
3171 #define for_each_active_range_index_in_nid(i, nid) \
3172 for (i = first_active_region_index_in_nid(nid); i != -1; \
3173 i = next_active_region_index_in_nid(i, nid))
3176 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
3177 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
3178 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
3180 * If an architecture guarantees that all ranges registered with
3181 * add_active_ranges() contain no holes and may be freed, this
3182 * this function may be used instead of calling free_bootmem() manually.
3184 void __init
free_bootmem_with_active_regions(int nid
,
3185 unsigned long max_low_pfn
)
3189 for_each_active_range_index_in_nid(i
, nid
) {
3190 unsigned long size_pages
= 0;
3191 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
3193 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
3196 if (end_pfn
> max_low_pfn
)
3197 end_pfn
= max_low_pfn
;
3199 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
3200 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
3201 PFN_PHYS(early_node_map
[i
].start_pfn
),
3202 size_pages
<< PAGE_SHIFT
);
3206 void __init
work_with_active_regions(int nid
, work_fn_t work_fn
, void *data
)
3211 for_each_active_range_index_in_nid(i
, nid
) {
3212 ret
= work_fn(early_node_map
[i
].start_pfn
,
3213 early_node_map
[i
].end_pfn
, data
);
3219 * sparse_memory_present_with_active_regions - Call memory_present for each active range
3220 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
3222 * If an architecture guarantees that all ranges registered with
3223 * add_active_ranges() contain no holes and may be freed, this
3224 * function may be used instead of calling memory_present() manually.
3226 void __init
sparse_memory_present_with_active_regions(int nid
)
3230 for_each_active_range_index_in_nid(i
, nid
)
3231 memory_present(early_node_map
[i
].nid
,
3232 early_node_map
[i
].start_pfn
,
3233 early_node_map
[i
].end_pfn
);
3237 * get_pfn_range_for_nid - Return the start and end page frames for a node
3238 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3239 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3240 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3242 * It returns the start and end page frame of a node based on information
3243 * provided by an arch calling add_active_range(). If called for a node
3244 * with no available memory, a warning is printed and the start and end
3247 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3248 unsigned long *start_pfn
, unsigned long *end_pfn
)
3254 for_each_active_range_index_in_nid(i
, nid
) {
3255 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3256 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3259 if (*start_pfn
== -1UL)
3264 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3265 * assumption is made that zones within a node are ordered in monotonic
3266 * increasing memory addresses so that the "highest" populated zone is used
3268 static void __init
find_usable_zone_for_movable(void)
3271 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3272 if (zone_index
== ZONE_MOVABLE
)
3275 if (arch_zone_highest_possible_pfn
[zone_index
] >
3276 arch_zone_lowest_possible_pfn
[zone_index
])
3280 VM_BUG_ON(zone_index
== -1);
3281 movable_zone
= zone_index
;
3285 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3286 * because it is sized independant of architecture. Unlike the other zones,
3287 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3288 * in each node depending on the size of each node and how evenly kernelcore
3289 * is distributed. This helper function adjusts the zone ranges
3290 * provided by the architecture for a given node by using the end of the
3291 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3292 * zones within a node are in order of monotonic increases memory addresses
3294 static void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3295 unsigned long zone_type
,
3296 unsigned long node_start_pfn
,
3297 unsigned long node_end_pfn
,
3298 unsigned long *zone_start_pfn
,
3299 unsigned long *zone_end_pfn
)
3301 /* Only adjust if ZONE_MOVABLE is on this node */
3302 if (zone_movable_pfn
[nid
]) {
3303 /* Size ZONE_MOVABLE */
3304 if (zone_type
== ZONE_MOVABLE
) {
3305 *zone_start_pfn
= zone_movable_pfn
[nid
];
3306 *zone_end_pfn
= min(node_end_pfn
,
3307 arch_zone_highest_possible_pfn
[movable_zone
]);
3309 /* Adjust for ZONE_MOVABLE starting within this range */
3310 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3311 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3312 *zone_end_pfn
= zone_movable_pfn
[nid
];
3314 /* Check if this whole range is within ZONE_MOVABLE */
3315 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3316 *zone_start_pfn
= *zone_end_pfn
;
3321 * Return the number of pages a zone spans in a node, including holes
3322 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3324 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3325 unsigned long zone_type
,
3326 unsigned long *ignored
)
3328 unsigned long node_start_pfn
, node_end_pfn
;
3329 unsigned long zone_start_pfn
, zone_end_pfn
;
3331 /* Get the start and end of the node and zone */
3332 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3333 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3334 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3335 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3336 node_start_pfn
, node_end_pfn
,
3337 &zone_start_pfn
, &zone_end_pfn
);
3339 /* Check that this node has pages within the zone's required range */
3340 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3343 /* Move the zone boundaries inside the node if necessary */
3344 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3345 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3347 /* Return the spanned pages */
3348 return zone_end_pfn
- zone_start_pfn
;
3352 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3353 * then all holes in the requested range will be accounted for.
3355 static unsigned long __meminit
__absent_pages_in_range(int nid
,
3356 unsigned long range_start_pfn
,
3357 unsigned long range_end_pfn
)
3360 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3361 unsigned long start_pfn
;
3363 /* Find the end_pfn of the first active range of pfns in the node */
3364 i
= first_active_region_index_in_nid(nid
);
3368 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3370 /* Account for ranges before physical memory on this node */
3371 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3372 hole_pages
= prev_end_pfn
- range_start_pfn
;
3374 /* Find all holes for the zone within the node */
3375 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3377 /* No need to continue if prev_end_pfn is outside the zone */
3378 if (prev_end_pfn
>= range_end_pfn
)
3381 /* Make sure the end of the zone is not within the hole */
3382 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3383 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3385 /* Update the hole size cound and move on */
3386 if (start_pfn
> range_start_pfn
) {
3387 BUG_ON(prev_end_pfn
> start_pfn
);
3388 hole_pages
+= start_pfn
- prev_end_pfn
;
3390 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3393 /* Account for ranges past physical memory on this node */
3394 if (range_end_pfn
> prev_end_pfn
)
3395 hole_pages
+= range_end_pfn
-
3396 max(range_start_pfn
, prev_end_pfn
);
3402 * absent_pages_in_range - Return number of page frames in holes within a range
3403 * @start_pfn: The start PFN to start searching for holes
3404 * @end_pfn: The end PFN to stop searching for holes
3406 * It returns the number of pages frames in memory holes within a range.
3408 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3409 unsigned long end_pfn
)
3411 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3414 /* Return the number of page frames in holes in a zone on a node */
3415 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3416 unsigned long zone_type
,
3417 unsigned long *ignored
)
3419 unsigned long node_start_pfn
, node_end_pfn
;
3420 unsigned long zone_start_pfn
, zone_end_pfn
;
3422 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3423 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3425 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3428 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3429 node_start_pfn
, node_end_pfn
,
3430 &zone_start_pfn
, &zone_end_pfn
);
3431 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3435 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3436 unsigned long zone_type
,
3437 unsigned long *zones_size
)
3439 return zones_size
[zone_type
];
3442 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3443 unsigned long zone_type
,
3444 unsigned long *zholes_size
)
3449 return zholes_size
[zone_type
];
3454 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3455 unsigned long *zones_size
, unsigned long *zholes_size
)
3457 unsigned long realtotalpages
, totalpages
= 0;
3460 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3461 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3463 pgdat
->node_spanned_pages
= totalpages
;
3465 realtotalpages
= totalpages
;
3466 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3468 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3470 pgdat
->node_present_pages
= realtotalpages
;
3471 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3475 #ifndef CONFIG_SPARSEMEM
3477 * Calculate the size of the zone->blockflags rounded to an unsigned long
3478 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3479 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3480 * round what is now in bits to nearest long in bits, then return it in
3483 static unsigned long __init
usemap_size(unsigned long zonesize
)
3485 unsigned long usemapsize
;
3487 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
3488 usemapsize
= usemapsize
>> pageblock_order
;
3489 usemapsize
*= NR_PAGEBLOCK_BITS
;
3490 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3492 return usemapsize
/ 8;
3495 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3496 struct zone
*zone
, unsigned long zonesize
)
3498 unsigned long usemapsize
= usemap_size(zonesize
);
3499 zone
->pageblock_flags
= NULL
;
3501 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3504 static void inline setup_usemap(struct pglist_data
*pgdat
,
3505 struct zone
*zone
, unsigned long zonesize
) {}
3506 #endif /* CONFIG_SPARSEMEM */
3508 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3510 /* Return a sensible default order for the pageblock size. */
3511 static inline int pageblock_default_order(void)
3513 if (HPAGE_SHIFT
> PAGE_SHIFT
)
3514 return HUGETLB_PAGE_ORDER
;
3519 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3520 static inline void __init
set_pageblock_order(unsigned int order
)
3522 /* Check that pageblock_nr_pages has not already been setup */
3523 if (pageblock_order
)
3527 * Assume the largest contiguous order of interest is a huge page.
3528 * This value may be variable depending on boot parameters on IA64
3530 pageblock_order
= order
;
3532 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3535 * When CONFIG_HUGETLB_PAGE_SIZE_VARIABLE is not set, set_pageblock_order()
3536 * and pageblock_default_order() are unused as pageblock_order is set
3537 * at compile-time. See include/linux/pageblock-flags.h for the values of
3538 * pageblock_order based on the kernel config
3540 static inline int pageblock_default_order(unsigned int order
)
3544 #define set_pageblock_order(x) do {} while (0)
3546 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3549 * Set up the zone data structures:
3550 * - mark all pages reserved
3551 * - mark all memory queues empty
3552 * - clear the memory bitmaps
3554 static void __paginginit
free_area_init_core(struct pglist_data
*pgdat
,
3555 unsigned long *zones_size
, unsigned long *zholes_size
)
3558 int nid
= pgdat
->node_id
;
3559 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
3562 pgdat_resize_init(pgdat
);
3563 pgdat
->nr_zones
= 0;
3564 init_waitqueue_head(&pgdat
->kswapd_wait
);
3565 pgdat
->kswapd_max_order
= 0;
3566 pgdat_page_cgroup_init(pgdat
);
3568 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3569 struct zone
*zone
= pgdat
->node_zones
+ j
;
3570 unsigned long size
, realsize
, memmap_pages
;
3573 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
3574 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
3578 * Adjust realsize so that it accounts for how much memory
3579 * is used by this zone for memmap. This affects the watermark
3580 * and per-cpu initialisations
3583 PAGE_ALIGN(size
* sizeof(struct page
)) >> PAGE_SHIFT
;
3584 if (realsize
>= memmap_pages
) {
3585 realsize
-= memmap_pages
;
3588 " %s zone: %lu pages used for memmap\n",
3589 zone_names
[j
], memmap_pages
);
3592 " %s zone: %lu pages exceeds realsize %lu\n",
3593 zone_names
[j
], memmap_pages
, realsize
);
3595 /* Account for reserved pages */
3596 if (j
== 0 && realsize
> dma_reserve
) {
3597 realsize
-= dma_reserve
;
3598 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
3599 zone_names
[0], dma_reserve
);
3602 if (!is_highmem_idx(j
))
3603 nr_kernel_pages
+= realsize
;
3604 nr_all_pages
+= realsize
;
3606 zone
->spanned_pages
= size
;
3607 zone
->present_pages
= realsize
;
3610 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3612 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3614 zone
->name
= zone_names
[j
];
3615 spin_lock_init(&zone
->lock
);
3616 spin_lock_init(&zone
->lru_lock
);
3617 zone_seqlock_init(zone
);
3618 zone
->zone_pgdat
= pgdat
;
3620 zone
->prev_priority
= DEF_PRIORITY
;
3622 zone_pcp_init(zone
);
3624 INIT_LIST_HEAD(&zone
->lru
[l
].list
);
3625 zone
->lru
[l
].nr_scan
= 0;
3627 zone
->reclaim_stat
.recent_rotated
[0] = 0;
3628 zone
->reclaim_stat
.recent_rotated
[1] = 0;
3629 zone
->reclaim_stat
.recent_scanned
[0] = 0;
3630 zone
->reclaim_stat
.recent_scanned
[1] = 0;
3631 zap_zone_vm_stats(zone
);
3636 set_pageblock_order(pageblock_default_order());
3637 setup_usemap(pgdat
, zone
, size
);
3638 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3639 size
, MEMMAP_EARLY
);
3641 memmap_init(size
, nid
, j
, zone_start_pfn
);
3642 zone_start_pfn
+= size
;
3646 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3648 /* Skip empty nodes */
3649 if (!pgdat
->node_spanned_pages
)
3652 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3653 /* ia64 gets its own node_mem_map, before this, without bootmem */
3654 if (!pgdat
->node_mem_map
) {
3655 unsigned long size
, start
, end
;
3659 * The zone's endpoints aren't required to be MAX_ORDER
3660 * aligned but the node_mem_map endpoints must be in order
3661 * for the buddy allocator to function correctly.
3663 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3664 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3665 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3666 size
= (end
- start
) * sizeof(struct page
);
3667 map
= alloc_remap(pgdat
->node_id
, size
);
3669 map
= alloc_bootmem_node(pgdat
, size
);
3670 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3672 #ifndef CONFIG_NEED_MULTIPLE_NODES
3674 * With no DISCONTIG, the global mem_map is just set as node 0's
3676 if (pgdat
== NODE_DATA(0)) {
3677 mem_map
= NODE_DATA(0)->node_mem_map
;
3678 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3679 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3680 mem_map
-= (pgdat
->node_start_pfn
- ARCH_PFN_OFFSET
);
3681 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3684 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3687 void __paginginit
free_area_init_node(int nid
, unsigned long *zones_size
,
3688 unsigned long node_start_pfn
, unsigned long *zholes_size
)
3690 pg_data_t
*pgdat
= NODE_DATA(nid
);
3692 pgdat
->node_id
= nid
;
3693 pgdat
->node_start_pfn
= node_start_pfn
;
3694 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3696 alloc_node_mem_map(pgdat
);
3697 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3698 printk(KERN_DEBUG
"free_area_init_node: node %d, pgdat %08lx, node_mem_map %08lx\n",
3699 nid
, (unsigned long)pgdat
,
3700 (unsigned long)pgdat
->node_mem_map
);
3703 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3706 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3708 #if MAX_NUMNODES > 1
3710 * Figure out the number of possible node ids.
3712 static void __init
setup_nr_node_ids(void)
3715 unsigned int highest
= 0;
3717 for_each_node_mask(node
, node_possible_map
)
3719 nr_node_ids
= highest
+ 1;
3722 static inline void setup_nr_node_ids(void)
3728 * add_active_range - Register a range of PFNs backed by physical memory
3729 * @nid: The node ID the range resides on
3730 * @start_pfn: The start PFN of the available physical memory
3731 * @end_pfn: The end PFN of the available physical memory
3733 * These ranges are stored in an early_node_map[] and later used by
3734 * free_area_init_nodes() to calculate zone sizes and holes. If the
3735 * range spans a memory hole, it is up to the architecture to ensure
3736 * the memory is not freed by the bootmem allocator. If possible
3737 * the range being registered will be merged with existing ranges.
3739 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3740 unsigned long end_pfn
)
3744 mminit_dprintk(MMINIT_TRACE
, "memory_register",
3745 "Entering add_active_range(%d, %#lx, %#lx) "
3746 "%d entries of %d used\n",
3747 nid
, start_pfn
, end_pfn
,
3748 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3750 mminit_validate_memmodel_limits(&start_pfn
, &end_pfn
);
3752 /* Merge with existing active regions if possible */
3753 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3754 if (early_node_map
[i
].nid
!= nid
)
3757 /* Skip if an existing region covers this new one */
3758 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3759 end_pfn
<= early_node_map
[i
].end_pfn
)
3762 /* Merge forward if suitable */
3763 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3764 end_pfn
> early_node_map
[i
].end_pfn
) {
3765 early_node_map
[i
].end_pfn
= end_pfn
;
3769 /* Merge backward if suitable */
3770 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3771 end_pfn
>= early_node_map
[i
].start_pfn
) {
3772 early_node_map
[i
].start_pfn
= start_pfn
;
3777 /* Check that early_node_map is large enough */
3778 if (i
>= MAX_ACTIVE_REGIONS
) {
3779 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3780 MAX_ACTIVE_REGIONS
);
3784 early_node_map
[i
].nid
= nid
;
3785 early_node_map
[i
].start_pfn
= start_pfn
;
3786 early_node_map
[i
].end_pfn
= end_pfn
;
3787 nr_nodemap_entries
= i
+ 1;
3791 * remove_active_range - Shrink an existing registered range of PFNs
3792 * @nid: The node id the range is on that should be shrunk
3793 * @start_pfn: The new PFN of the range
3794 * @end_pfn: The new PFN of the range
3796 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3797 * The map is kept near the end physical page range that has already been
3798 * registered. This function allows an arch to shrink an existing registered
3801 void __init
remove_active_range(unsigned int nid
, unsigned long start_pfn
,
3802 unsigned long end_pfn
)
3807 printk(KERN_DEBUG
"remove_active_range (%d, %lu, %lu)\n",
3808 nid
, start_pfn
, end_pfn
);
3810 /* Find the old active region end and shrink */
3811 for_each_active_range_index_in_nid(i
, nid
) {
3812 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3813 early_node_map
[i
].end_pfn
<= end_pfn
) {
3815 early_node_map
[i
].start_pfn
= 0;
3816 early_node_map
[i
].end_pfn
= 0;
3820 if (early_node_map
[i
].start_pfn
< start_pfn
&&
3821 early_node_map
[i
].end_pfn
> start_pfn
) {
3822 unsigned long temp_end_pfn
= early_node_map
[i
].end_pfn
;
3823 early_node_map
[i
].end_pfn
= start_pfn
;
3824 if (temp_end_pfn
> end_pfn
)
3825 add_active_range(nid
, end_pfn
, temp_end_pfn
);
3828 if (early_node_map
[i
].start_pfn
>= start_pfn
&&
3829 early_node_map
[i
].end_pfn
> end_pfn
&&
3830 early_node_map
[i
].start_pfn
< end_pfn
) {
3831 early_node_map
[i
].start_pfn
= end_pfn
;
3839 /* remove the blank ones */
3840 for (i
= nr_nodemap_entries
- 1; i
> 0; i
--) {
3841 if (early_node_map
[i
].nid
!= nid
)
3843 if (early_node_map
[i
].end_pfn
)
3845 /* we found it, get rid of it */
3846 for (j
= i
; j
< nr_nodemap_entries
- 1; j
++)
3847 memcpy(&early_node_map
[j
], &early_node_map
[j
+1],
3848 sizeof(early_node_map
[j
]));
3849 j
= nr_nodemap_entries
- 1;
3850 memset(&early_node_map
[j
], 0, sizeof(early_node_map
[j
]));
3851 nr_nodemap_entries
--;
3856 * remove_all_active_ranges - Remove all currently registered regions
3858 * During discovery, it may be found that a table like SRAT is invalid
3859 * and an alternative discovery method must be used. This function removes
3860 * all currently registered regions.
3862 void __init
remove_all_active_ranges(void)
3864 memset(early_node_map
, 0, sizeof(early_node_map
));
3865 nr_nodemap_entries
= 0;
3868 /* Compare two active node_active_regions */
3869 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3871 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3872 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3874 /* Done this way to avoid overflows */
3875 if (arange
->start_pfn
> brange
->start_pfn
)
3877 if (arange
->start_pfn
< brange
->start_pfn
)
3883 /* sort the node_map by start_pfn */
3884 static void __init
sort_node_map(void)
3886 sort(early_node_map
, (size_t)nr_nodemap_entries
,
3887 sizeof(struct node_active_region
),
3888 cmp_node_active_region
, NULL
);
3891 /* Find the lowest pfn for a node */
3892 static unsigned long __init
find_min_pfn_for_node(int nid
)
3895 unsigned long min_pfn
= ULONG_MAX
;
3897 /* Assuming a sorted map, the first range found has the starting pfn */
3898 for_each_active_range_index_in_nid(i
, nid
)
3899 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
3901 if (min_pfn
== ULONG_MAX
) {
3903 "Could not find start_pfn for node %d\n", nid
);
3911 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3913 * It returns the minimum PFN based on information provided via
3914 * add_active_range().
3916 unsigned long __init
find_min_pfn_with_active_regions(void)
3918 return find_min_pfn_for_node(MAX_NUMNODES
);
3922 * early_calculate_totalpages()
3923 * Sum pages in active regions for movable zone.
3924 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3926 static unsigned long __init
early_calculate_totalpages(void)
3929 unsigned long totalpages
= 0;
3931 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3932 unsigned long pages
= early_node_map
[i
].end_pfn
-
3933 early_node_map
[i
].start_pfn
;
3934 totalpages
+= pages
;
3936 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
3942 * Find the PFN the Movable zone begins in each node. Kernel memory
3943 * is spread evenly between nodes as long as the nodes have enough
3944 * memory. When they don't, some nodes will have more kernelcore than
3947 static void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
3950 unsigned long usable_startpfn
;
3951 unsigned long kernelcore_node
, kernelcore_remaining
;
3952 unsigned long totalpages
= early_calculate_totalpages();
3953 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
3956 * If movablecore was specified, calculate what size of
3957 * kernelcore that corresponds so that memory usable for
3958 * any allocation type is evenly spread. If both kernelcore
3959 * and movablecore are specified, then the value of kernelcore
3960 * will be used for required_kernelcore if it's greater than
3961 * what movablecore would have allowed.
3963 if (required_movablecore
) {
3964 unsigned long corepages
;
3967 * Round-up so that ZONE_MOVABLE is at least as large as what
3968 * was requested by the user
3970 required_movablecore
=
3971 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
3972 corepages
= totalpages
- required_movablecore
;
3974 required_kernelcore
= max(required_kernelcore
, corepages
);
3977 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3978 if (!required_kernelcore
)
3981 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3982 find_usable_zone_for_movable();
3983 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
3986 /* Spread kernelcore memory as evenly as possible throughout nodes */
3987 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3988 for_each_node_state(nid
, N_HIGH_MEMORY
) {
3990 * Recalculate kernelcore_node if the division per node
3991 * now exceeds what is necessary to satisfy the requested
3992 * amount of memory for the kernel
3994 if (required_kernelcore
< kernelcore_node
)
3995 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3998 * As the map is walked, we track how much memory is usable
3999 * by the kernel using kernelcore_remaining. When it is
4000 * 0, the rest of the node is usable by ZONE_MOVABLE
4002 kernelcore_remaining
= kernelcore_node
;
4004 /* Go through each range of PFNs within this node */
4005 for_each_active_range_index_in_nid(i
, nid
) {
4006 unsigned long start_pfn
, end_pfn
;
4007 unsigned long size_pages
;
4009 start_pfn
= max(early_node_map
[i
].start_pfn
,
4010 zone_movable_pfn
[nid
]);
4011 end_pfn
= early_node_map
[i
].end_pfn
;
4012 if (start_pfn
>= end_pfn
)
4015 /* Account for what is only usable for kernelcore */
4016 if (start_pfn
< usable_startpfn
) {
4017 unsigned long kernel_pages
;
4018 kernel_pages
= min(end_pfn
, usable_startpfn
)
4021 kernelcore_remaining
-= min(kernel_pages
,
4022 kernelcore_remaining
);
4023 required_kernelcore
-= min(kernel_pages
,
4024 required_kernelcore
);
4026 /* Continue if range is now fully accounted */
4027 if (end_pfn
<= usable_startpfn
) {
4030 * Push zone_movable_pfn to the end so
4031 * that if we have to rebalance
4032 * kernelcore across nodes, we will
4033 * not double account here
4035 zone_movable_pfn
[nid
] = end_pfn
;
4038 start_pfn
= usable_startpfn
;
4042 * The usable PFN range for ZONE_MOVABLE is from
4043 * start_pfn->end_pfn. Calculate size_pages as the
4044 * number of pages used as kernelcore
4046 size_pages
= end_pfn
- start_pfn
;
4047 if (size_pages
> kernelcore_remaining
)
4048 size_pages
= kernelcore_remaining
;
4049 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
4052 * Some kernelcore has been met, update counts and
4053 * break if the kernelcore for this node has been
4056 required_kernelcore
-= min(required_kernelcore
,
4058 kernelcore_remaining
-= size_pages
;
4059 if (!kernelcore_remaining
)
4065 * If there is still required_kernelcore, we do another pass with one
4066 * less node in the count. This will push zone_movable_pfn[nid] further
4067 * along on the nodes that still have memory until kernelcore is
4071 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
4074 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
4075 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
4076 zone_movable_pfn
[nid
] =
4077 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
4080 /* Any regular memory on that node ? */
4081 static void check_for_regular_memory(pg_data_t
*pgdat
)
4083 #ifdef CONFIG_HIGHMEM
4084 enum zone_type zone_type
;
4086 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
4087 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
4088 if (zone
->present_pages
)
4089 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
4095 * free_area_init_nodes - Initialise all pg_data_t and zone data
4096 * @max_zone_pfn: an array of max PFNs for each zone
4098 * This will call free_area_init_node() for each active node in the system.
4099 * Using the page ranges provided by add_active_range(), the size of each
4100 * zone in each node and their holes is calculated. If the maximum PFN
4101 * between two adjacent zones match, it is assumed that the zone is empty.
4102 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
4103 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
4104 * starts where the previous one ended. For example, ZONE_DMA32 starts
4105 * at arch_max_dma_pfn.
4107 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
4112 /* Sort early_node_map as initialisation assumes it is sorted */
4115 /* Record where the zone boundaries are */
4116 memset(arch_zone_lowest_possible_pfn
, 0,
4117 sizeof(arch_zone_lowest_possible_pfn
));
4118 memset(arch_zone_highest_possible_pfn
, 0,
4119 sizeof(arch_zone_highest_possible_pfn
));
4120 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
4121 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
4122 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
4123 if (i
== ZONE_MOVABLE
)
4125 arch_zone_lowest_possible_pfn
[i
] =
4126 arch_zone_highest_possible_pfn
[i
-1];
4127 arch_zone_highest_possible_pfn
[i
] =
4128 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
4130 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
4131 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
4133 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
4134 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
4135 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
4137 /* Print out the zone ranges */
4138 printk("Zone PFN ranges:\n");
4139 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4140 if (i
== ZONE_MOVABLE
)
4142 printk(" %-8s %0#10lx -> %0#10lx\n",
4144 arch_zone_lowest_possible_pfn
[i
],
4145 arch_zone_highest_possible_pfn
[i
]);
4148 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
4149 printk("Movable zone start PFN for each node\n");
4150 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
4151 if (zone_movable_pfn
[i
])
4152 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
4155 /* Print out the early_node_map[] */
4156 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
4157 for (i
= 0; i
< nr_nodemap_entries
; i
++)
4158 printk(" %3d: %0#10lx -> %0#10lx\n", early_node_map
[i
].nid
,
4159 early_node_map
[i
].start_pfn
,
4160 early_node_map
[i
].end_pfn
);
4162 /* Initialise every node */
4163 mminit_verify_pageflags_layout();
4164 setup_nr_node_ids();
4165 for_each_online_node(nid
) {
4166 pg_data_t
*pgdat
= NODE_DATA(nid
);
4167 free_area_init_node(nid
, NULL
,
4168 find_min_pfn_for_node(nid
), NULL
);
4170 /* Any memory on that node */
4171 if (pgdat
->node_present_pages
)
4172 node_set_state(nid
, N_HIGH_MEMORY
);
4173 check_for_regular_memory(pgdat
);
4177 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
4179 unsigned long long coremem
;
4183 coremem
= memparse(p
, &p
);
4184 *core
= coremem
>> PAGE_SHIFT
;
4186 /* Paranoid check that UL is enough for the coremem value */
4187 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
4193 * kernelcore=size sets the amount of memory for use for allocations that
4194 * cannot be reclaimed or migrated.
4196 static int __init
cmdline_parse_kernelcore(char *p
)
4198 return cmdline_parse_core(p
, &required_kernelcore
);
4202 * movablecore=size sets the amount of memory for use for allocations that
4203 * can be reclaimed or migrated.
4205 static int __init
cmdline_parse_movablecore(char *p
)
4207 return cmdline_parse_core(p
, &required_movablecore
);
4210 early_param("kernelcore", cmdline_parse_kernelcore
);
4211 early_param("movablecore", cmdline_parse_movablecore
);
4213 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
4216 * set_dma_reserve - set the specified number of pages reserved in the first zone
4217 * @new_dma_reserve: The number of pages to mark reserved
4219 * The per-cpu batchsize and zone watermarks are determined by present_pages.
4220 * In the DMA zone, a significant percentage may be consumed by kernel image
4221 * and other unfreeable allocations which can skew the watermarks badly. This
4222 * function may optionally be used to account for unfreeable pages in the
4223 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
4224 * smaller per-cpu batchsize.
4226 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
4228 dma_reserve
= new_dma_reserve
;
4231 #ifndef CONFIG_NEED_MULTIPLE_NODES
4232 struct pglist_data __refdata contig_page_data
= { .bdata
= &bootmem_node_data
[0] };
4233 EXPORT_SYMBOL(contig_page_data
);
4236 void __init
free_area_init(unsigned long *zones_size
)
4238 free_area_init_node(0, zones_size
,
4239 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
4242 static int page_alloc_cpu_notify(struct notifier_block
*self
,
4243 unsigned long action
, void *hcpu
)
4245 int cpu
= (unsigned long)hcpu
;
4247 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
4251 * Spill the event counters of the dead processor
4252 * into the current processors event counters.
4253 * This artificially elevates the count of the current
4256 vm_events_fold_cpu(cpu
);
4259 * Zero the differential counters of the dead processor
4260 * so that the vm statistics are consistent.
4262 * This is only okay since the processor is dead and cannot
4263 * race with what we are doing.
4265 refresh_cpu_vm_stats(cpu
);
4270 void __init
page_alloc_init(void)
4272 hotcpu_notifier(page_alloc_cpu_notify
, 0);
4276 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
4277 * or min_free_kbytes changes.
4279 static void calculate_totalreserve_pages(void)
4281 struct pglist_data
*pgdat
;
4282 unsigned long reserve_pages
= 0;
4283 enum zone_type i
, j
;
4285 for_each_online_pgdat(pgdat
) {
4286 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
4287 struct zone
*zone
= pgdat
->node_zones
+ i
;
4288 unsigned long max
= 0;
4290 /* Find valid and maximum lowmem_reserve in the zone */
4291 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
4292 if (zone
->lowmem_reserve
[j
] > max
)
4293 max
= zone
->lowmem_reserve
[j
];
4296 /* we treat pages_high as reserved pages. */
4297 max
+= zone
->pages_high
;
4299 if (max
> zone
->present_pages
)
4300 max
= zone
->present_pages
;
4301 reserve_pages
+= max
;
4304 totalreserve_pages
= reserve_pages
;
4308 * setup_per_zone_lowmem_reserve - called whenever
4309 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4310 * has a correct pages reserved value, so an adequate number of
4311 * pages are left in the zone after a successful __alloc_pages().
4313 static void setup_per_zone_lowmem_reserve(void)
4315 struct pglist_data
*pgdat
;
4316 enum zone_type j
, idx
;
4318 for_each_online_pgdat(pgdat
) {
4319 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4320 struct zone
*zone
= pgdat
->node_zones
+ j
;
4321 unsigned long present_pages
= zone
->present_pages
;
4323 zone
->lowmem_reserve
[j
] = 0;
4327 struct zone
*lower_zone
;
4331 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4332 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4334 lower_zone
= pgdat
->node_zones
+ idx
;
4335 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4336 sysctl_lowmem_reserve_ratio
[idx
];
4337 present_pages
+= lower_zone
->present_pages
;
4342 /* update totalreserve_pages */
4343 calculate_totalreserve_pages();
4347 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4349 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4350 * with respect to min_free_kbytes.
4352 void setup_per_zone_pages_min(void)
4354 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4355 unsigned long lowmem_pages
= 0;
4357 unsigned long flags
;
4359 /* Calculate total number of !ZONE_HIGHMEM pages */
4360 for_each_zone(zone
) {
4361 if (!is_highmem(zone
))
4362 lowmem_pages
+= zone
->present_pages
;
4365 for_each_zone(zone
) {
4368 spin_lock_irqsave(&zone
->lock
, flags
);
4369 tmp
= (u64
)pages_min
* zone
->present_pages
;
4370 do_div(tmp
, lowmem_pages
);
4371 if (is_highmem(zone
)) {
4373 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4374 * need highmem pages, so cap pages_min to a small
4377 * The (pages_high-pages_low) and (pages_low-pages_min)
4378 * deltas controls asynch page reclaim, and so should
4379 * not be capped for highmem.
4383 min_pages
= zone
->present_pages
/ 1024;
4384 if (min_pages
< SWAP_CLUSTER_MAX
)
4385 min_pages
= SWAP_CLUSTER_MAX
;
4386 if (min_pages
> 128)
4388 zone
->pages_min
= min_pages
;
4391 * If it's a lowmem zone, reserve a number of pages
4392 * proportionate to the zone's size.
4394 zone
->pages_min
= tmp
;
4397 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
4398 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
4399 setup_zone_migrate_reserve(zone
);
4400 spin_unlock_irqrestore(&zone
->lock
, flags
);
4403 /* update totalreserve_pages */
4404 calculate_totalreserve_pages();
4408 * setup_per_zone_inactive_ratio - called when min_free_kbytes changes.
4410 * The inactive anon list should be small enough that the VM never has to
4411 * do too much work, but large enough that each inactive page has a chance
4412 * to be referenced again before it is swapped out.
4414 * The inactive_anon ratio is the target ratio of ACTIVE_ANON to
4415 * INACTIVE_ANON pages on this zone's LRU, maintained by the
4416 * pageout code. A zone->inactive_ratio of 3 means 3:1 or 25% of
4417 * the anonymous pages are kept on the inactive list.
4420 * memory ratio inactive anon
4421 * -------------------------------------
4430 static void setup_per_zone_inactive_ratio(void)
4434 for_each_zone(zone
) {
4435 unsigned int gb
, ratio
;
4437 /* Zone size in gigabytes */
4438 gb
= zone
->present_pages
>> (30 - PAGE_SHIFT
);
4439 ratio
= int_sqrt(10 * gb
);
4443 zone
->inactive_ratio
= ratio
;
4448 * Initialise min_free_kbytes.
4450 * For small machines we want it small (128k min). For large machines
4451 * we want it large (64MB max). But it is not linear, because network
4452 * bandwidth does not increase linearly with machine size. We use
4454 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4455 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4471 static int __init
init_per_zone_pages_min(void)
4473 unsigned long lowmem_kbytes
;
4475 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4477 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4478 if (min_free_kbytes
< 128)
4479 min_free_kbytes
= 128;
4480 if (min_free_kbytes
> 65536)
4481 min_free_kbytes
= 65536;
4482 setup_per_zone_pages_min();
4483 setup_per_zone_lowmem_reserve();
4484 setup_per_zone_inactive_ratio();
4487 module_init(init_per_zone_pages_min
)
4490 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4491 * that we can call two helper functions whenever min_free_kbytes
4494 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4495 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4497 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
4499 setup_per_zone_pages_min();
4504 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4505 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4510 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4515 zone
->min_unmapped_pages
= (zone
->present_pages
*
4516 sysctl_min_unmapped_ratio
) / 100;
4520 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4521 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4526 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4531 zone
->min_slab_pages
= (zone
->present_pages
*
4532 sysctl_min_slab_ratio
) / 100;
4538 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4539 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4540 * whenever sysctl_lowmem_reserve_ratio changes.
4542 * The reserve ratio obviously has absolutely no relation with the
4543 * pages_min watermarks. The lowmem reserve ratio can only make sense
4544 * if in function of the boot time zone sizes.
4546 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4547 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4549 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4550 setup_per_zone_lowmem_reserve();
4555 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4556 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4557 * can have before it gets flushed back to buddy allocator.
4560 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4561 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4567 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4568 if (!write
|| (ret
== -EINVAL
))
4570 for_each_zone(zone
) {
4571 for_each_online_cpu(cpu
) {
4573 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4574 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
4580 int hashdist
= HASHDIST_DEFAULT
;
4583 static int __init
set_hashdist(char *str
)
4587 hashdist
= simple_strtoul(str
, &str
, 0);
4590 __setup("hashdist=", set_hashdist
);
4594 * allocate a large system hash table from bootmem
4595 * - it is assumed that the hash table must contain an exact power-of-2
4596 * quantity of entries
4597 * - limit is the number of hash buckets, not the total allocation size
4599 void *__init
alloc_large_system_hash(const char *tablename
,
4600 unsigned long bucketsize
,
4601 unsigned long numentries
,
4604 unsigned int *_hash_shift
,
4605 unsigned int *_hash_mask
,
4606 unsigned long limit
)
4608 unsigned long long max
= limit
;
4609 unsigned long log2qty
, size
;
4612 /* allow the kernel cmdline to have a say */
4614 /* round applicable memory size up to nearest megabyte */
4615 numentries
= nr_kernel_pages
;
4616 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4617 numentries
>>= 20 - PAGE_SHIFT
;
4618 numentries
<<= 20 - PAGE_SHIFT
;
4620 /* limit to 1 bucket per 2^scale bytes of low memory */
4621 if (scale
> PAGE_SHIFT
)
4622 numentries
>>= (scale
- PAGE_SHIFT
);
4624 numentries
<<= (PAGE_SHIFT
- scale
);
4626 /* Make sure we've got at least a 0-order allocation.. */
4627 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4628 numentries
= PAGE_SIZE
/ bucketsize
;
4630 numentries
= roundup_pow_of_two(numentries
);
4632 /* limit allocation size to 1/16 total memory by default */
4634 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4635 do_div(max
, bucketsize
);
4638 if (numentries
> max
)
4641 log2qty
= ilog2(numentries
);
4644 size
= bucketsize
<< log2qty
;
4645 if (flags
& HASH_EARLY
)
4646 table
= alloc_bootmem_nopanic(size
);
4648 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4650 unsigned long order
= get_order(size
);
4652 if (order
< MAX_ORDER
)
4653 table
= (void *)__get_free_pages(GFP_ATOMIC
,
4656 * If bucketsize is not a power-of-two, we may free
4657 * some pages at the end of hash table.
4660 unsigned long alloc_end
= (unsigned long)table
+
4661 (PAGE_SIZE
<< order
);
4662 unsigned long used
= (unsigned long)table
+
4664 split_page(virt_to_page(table
), order
);
4665 while (used
< alloc_end
) {
4671 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4674 panic("Failed to allocate %s hash table\n", tablename
);
4676 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4679 ilog2(size
) - PAGE_SHIFT
,
4683 *_hash_shift
= log2qty
;
4685 *_hash_mask
= (1 << log2qty
) - 1;
4688 * If hashdist is set, the table allocation is done with __vmalloc()
4689 * which invokes the kmemleak_alloc() callback. This function may also
4690 * be called before the slab and kmemleak are initialised when
4691 * kmemleak simply buffers the request to be executed later
4692 * (GFP_ATOMIC flag ignored in this case).
4695 kmemleak_alloc(table
, size
, 1, GFP_ATOMIC
);
4700 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4701 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4704 #ifdef CONFIG_SPARSEMEM
4705 return __pfn_to_section(pfn
)->pageblock_flags
;
4707 return zone
->pageblock_flags
;
4708 #endif /* CONFIG_SPARSEMEM */
4711 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4713 #ifdef CONFIG_SPARSEMEM
4714 pfn
&= (PAGES_PER_SECTION
-1);
4715 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4717 pfn
= pfn
- zone
->zone_start_pfn
;
4718 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4719 #endif /* CONFIG_SPARSEMEM */
4723 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4724 * @page: The page within the block of interest
4725 * @start_bitidx: The first bit of interest to retrieve
4726 * @end_bitidx: The last bit of interest
4727 * returns pageblock_bits flags
4729 unsigned long get_pageblock_flags_group(struct page
*page
,
4730 int start_bitidx
, int end_bitidx
)
4733 unsigned long *bitmap
;
4734 unsigned long pfn
, bitidx
;
4735 unsigned long flags
= 0;
4736 unsigned long value
= 1;
4738 zone
= page_zone(page
);
4739 pfn
= page_to_pfn(page
);
4740 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4741 bitidx
= pfn_to_bitidx(zone
, pfn
);
4743 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4744 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
4751 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4752 * @page: The page within the block of interest
4753 * @start_bitidx: The first bit of interest
4754 * @end_bitidx: The last bit of interest
4755 * @flags: The flags to set
4757 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
4758 int start_bitidx
, int end_bitidx
)
4761 unsigned long *bitmap
;
4762 unsigned long pfn
, bitidx
;
4763 unsigned long value
= 1;
4765 zone
= page_zone(page
);
4766 pfn
= page_to_pfn(page
);
4767 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4768 bitidx
= pfn_to_bitidx(zone
, pfn
);
4769 VM_BUG_ON(pfn
< zone
->zone_start_pfn
);
4770 VM_BUG_ON(pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
);
4772 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4774 __set_bit(bitidx
+ start_bitidx
, bitmap
);
4776 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
4780 * This is designed as sub function...plz see page_isolation.c also.
4781 * set/clear page block's type to be ISOLATE.
4782 * page allocater never alloc memory from ISOLATE block.
4785 int set_migratetype_isolate(struct page
*page
)
4788 unsigned long flags
;
4791 zone
= page_zone(page
);
4792 spin_lock_irqsave(&zone
->lock
, flags
);
4794 * In future, more migrate types will be able to be isolation target.
4796 if (get_pageblock_migratetype(page
) != MIGRATE_MOVABLE
)
4798 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
4799 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
4802 spin_unlock_irqrestore(&zone
->lock
, flags
);
4808 void unset_migratetype_isolate(struct page
*page
)
4811 unsigned long flags
;
4812 zone
= page_zone(page
);
4813 spin_lock_irqsave(&zone
->lock
, flags
);
4814 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
4816 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4817 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
4819 spin_unlock_irqrestore(&zone
->lock
, flags
);
4822 #ifdef CONFIG_MEMORY_HOTREMOVE
4824 * All pages in the range must be isolated before calling this.
4827 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
4833 unsigned long flags
;
4834 /* find the first valid pfn */
4835 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
4840 zone
= page_zone(pfn_to_page(pfn
));
4841 spin_lock_irqsave(&zone
->lock
, flags
);
4843 while (pfn
< end_pfn
) {
4844 if (!pfn_valid(pfn
)) {
4848 page
= pfn_to_page(pfn
);
4849 BUG_ON(page_count(page
));
4850 BUG_ON(!PageBuddy(page
));
4851 order
= page_order(page
);
4852 #ifdef CONFIG_DEBUG_VM
4853 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
4854 pfn
, 1 << order
, end_pfn
);
4856 list_del(&page
->lru
);
4857 rmv_page_order(page
);
4858 zone
->free_area
[order
].nr_free
--;
4859 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
4861 for (i
= 0; i
< (1 << order
); i
++)
4862 SetPageReserved((page
+i
));
4863 pfn
+= (1 << order
);
4865 spin_unlock_irqrestore(&zone
->lock
, flags
);