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/bootmem.h>
23 #include <linux/compiler.h>
24 #include <linux/kernel.h>
25 #include <linux/module.h>
26 #include <linux/suspend.h>
27 #include <linux/pagevec.h>
28 #include <linux/blkdev.h>
29 #include <linux/slab.h>
30 #include <linux/notifier.h>
31 #include <linux/topology.h>
32 #include <linux/sysctl.h>
33 #include <linux/cpu.h>
34 #include <linux/cpuset.h>
35 #include <linux/memory_hotplug.h>
36 #include <linux/nodemask.h>
37 #include <linux/vmalloc.h>
38 #include <linux/mempolicy.h>
39 #include <linux/stop_machine.h>
40 #include <linux/sort.h>
41 #include <linux/pfn.h>
42 #include <linux/backing-dev.h>
43 #include <linux/fault-inject.h>
45 #include <asm/tlbflush.h>
46 #include <asm/div64.h>
50 * Array of node states.
52 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
53 [N_POSSIBLE
] = NODE_MASK_ALL
,
54 [N_ONLINE
] = { { [0] = 1UL } },
56 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
58 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
60 [N_CPU
] = { { [0] = 1UL } },
63 EXPORT_SYMBOL(node_states
);
65 unsigned long totalram_pages __read_mostly
;
66 unsigned long totalreserve_pages __read_mostly
;
68 int percpu_pagelist_fraction
;
70 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
71 int pageblock_order __read_mostly
;
74 static void __free_pages_ok(struct page
*page
, unsigned int order
);
77 * results with 256, 32 in the lowmem_reserve sysctl:
78 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
79 * 1G machine -> (16M dma, 784M normal, 224M high)
80 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
81 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
82 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
84 * TBD: should special case ZONE_DMA32 machines here - in those we normally
85 * don't need any ZONE_NORMAL reservation
87 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
88 #ifdef CONFIG_ZONE_DMA
91 #ifdef CONFIG_ZONE_DMA32
100 EXPORT_SYMBOL(totalram_pages
);
102 static char * const zone_names
[MAX_NR_ZONES
] = {
103 #ifdef CONFIG_ZONE_DMA
106 #ifdef CONFIG_ZONE_DMA32
110 #ifdef CONFIG_HIGHMEM
116 int min_free_kbytes
= 1024;
118 unsigned long __meminitdata nr_kernel_pages
;
119 unsigned long __meminitdata nr_all_pages
;
120 static unsigned long __meminitdata dma_reserve
;
122 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
124 * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
125 * ranges of memory (RAM) that may be registered with add_active_range().
126 * Ranges passed to add_active_range() will be merged if possible
127 * so the number of times add_active_range() can be called is
128 * related to the number of nodes and the number of holes
130 #ifdef CONFIG_MAX_ACTIVE_REGIONS
131 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
132 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
134 #if MAX_NUMNODES >= 32
135 /* If there can be many nodes, allow up to 50 holes per node */
136 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
138 /* By default, allow up to 256 distinct regions */
139 #define MAX_ACTIVE_REGIONS 256
143 static struct node_active_region __meminitdata early_node_map
[MAX_ACTIVE_REGIONS
];
144 static int __meminitdata nr_nodemap_entries
;
145 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
146 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
147 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
148 static unsigned long __meminitdata node_boundary_start_pfn
[MAX_NUMNODES
];
149 static unsigned long __meminitdata node_boundary_end_pfn
[MAX_NUMNODES
];
150 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
151 unsigned long __initdata required_kernelcore
;
152 static unsigned long __initdata required_movablecore
;
153 unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
155 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
157 EXPORT_SYMBOL(movable_zone
);
158 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
161 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
162 EXPORT_SYMBOL(nr_node_ids
);
165 int page_group_by_mobility_disabled __read_mostly
;
167 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
169 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
170 PB_migrate
, PB_migrate_end
);
173 #ifdef CONFIG_DEBUG_VM
174 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
178 unsigned long pfn
= page_to_pfn(page
);
181 seq
= zone_span_seqbegin(zone
);
182 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
184 else if (pfn
< zone
->zone_start_pfn
)
186 } while (zone_span_seqretry(zone
, seq
));
191 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
193 if (!pfn_valid_within(page_to_pfn(page
)))
195 if (zone
!= page_zone(page
))
201 * Temporary debugging check for pages not lying within a given zone.
203 static int bad_range(struct zone
*zone
, struct page
*page
)
205 if (page_outside_zone_boundaries(zone
, page
))
207 if (!page_is_consistent(zone
, page
))
213 static inline int bad_range(struct zone
*zone
, struct page
*page
)
219 static void bad_page(struct page
*page
)
221 printk(KERN_EMERG
"Bad page state in process '%s'\n"
222 KERN_EMERG
"page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
223 KERN_EMERG
"Trying to fix it up, but a reboot is needed\n"
224 KERN_EMERG
"Backtrace:\n",
225 current
->comm
, page
, (int)(2*sizeof(unsigned long)),
226 (unsigned long)page
->flags
, page
->mapping
,
227 page_mapcount(page
), page_count(page
));
229 page
->flags
&= ~(1 << PG_lru
|
239 set_page_count(page
, 0);
240 reset_page_mapcount(page
);
241 page
->mapping
= NULL
;
242 add_taint(TAINT_BAD_PAGE
);
246 * Higher-order pages are called "compound pages". They are structured thusly:
248 * The first PAGE_SIZE page is called the "head page".
250 * The remaining PAGE_SIZE pages are called "tail pages".
252 * All pages have PG_compound set. All pages have their ->private pointing at
253 * the head page (even the head page has this).
255 * The first tail page's ->lru.next holds the address of the compound page's
256 * put_page() function. Its ->lru.prev holds the order of allocation.
257 * This usage means that zero-order pages may not be compound.
260 static void free_compound_page(struct page
*page
)
262 __free_pages_ok(page
, compound_order(page
));
265 static void prep_compound_page(struct page
*page
, unsigned long order
)
268 int nr_pages
= 1 << order
;
270 set_compound_page_dtor(page
, free_compound_page
);
271 set_compound_order(page
, order
);
273 for (i
= 1; i
< nr_pages
; i
++) {
274 struct page
*p
= page
+ i
;
277 p
->first_page
= page
;
281 static void destroy_compound_page(struct page
*page
, unsigned long order
)
284 int nr_pages
= 1 << order
;
286 if (unlikely(compound_order(page
) != order
))
289 if (unlikely(!PageHead(page
)))
291 __ClearPageHead(page
);
292 for (i
= 1; i
< nr_pages
; i
++) {
293 struct page
*p
= page
+ i
;
295 if (unlikely(!PageTail(p
) |
296 (p
->first_page
!= page
)))
302 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
306 VM_BUG_ON((gfp_flags
& (__GFP_WAIT
| __GFP_HIGHMEM
)) == __GFP_HIGHMEM
);
308 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
309 * and __GFP_HIGHMEM from hard or soft interrupt context.
311 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
312 for (i
= 0; i
< (1 << order
); i
++)
313 clear_highpage(page
+ i
);
316 static inline void set_page_order(struct page
*page
, int order
)
318 set_page_private(page
, order
);
319 __SetPageBuddy(page
);
322 static inline void rmv_page_order(struct page
*page
)
324 __ClearPageBuddy(page
);
325 set_page_private(page
, 0);
329 * Locate the struct page for both the matching buddy in our
330 * pair (buddy1) and the combined O(n+1) page they form (page).
332 * 1) Any buddy B1 will have an order O twin B2 which satisfies
333 * the following equation:
335 * For example, if the starting buddy (buddy2) is #8 its order
337 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
339 * 2) Any buddy B will have an order O+1 parent P which
340 * satisfies the following equation:
343 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
345 static inline struct page
*
346 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
348 unsigned long buddy_idx
= page_idx
^ (1 << order
);
350 return page
+ (buddy_idx
- page_idx
);
353 static inline unsigned long
354 __find_combined_index(unsigned long page_idx
, unsigned int order
)
356 return (page_idx
& ~(1 << order
));
360 * This function checks whether a page is free && is the buddy
361 * we can do coalesce a page and its buddy if
362 * (a) the buddy is not in a hole &&
363 * (b) the buddy is in the buddy system &&
364 * (c) a page and its buddy have the same order &&
365 * (d) a page and its buddy are in the same zone.
367 * For recording whether a page is in the buddy system, we use PG_buddy.
368 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
370 * For recording page's order, we use page_private(page).
372 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
375 if (!pfn_valid_within(page_to_pfn(buddy
)))
378 if (page_zone_id(page
) != page_zone_id(buddy
))
381 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
382 BUG_ON(page_count(buddy
) != 0);
389 * Freeing function for a buddy system allocator.
391 * The concept of a buddy system is to maintain direct-mapped table
392 * (containing bit values) for memory blocks of various "orders".
393 * The bottom level table contains the map for the smallest allocatable
394 * units of memory (here, pages), and each level above it describes
395 * pairs of units from the levels below, hence, "buddies".
396 * At a high level, all that happens here is marking the table entry
397 * at the bottom level available, and propagating the changes upward
398 * as necessary, plus some accounting needed to play nicely with other
399 * parts of the VM system.
400 * At each level, we keep a list of pages, which are heads of continuous
401 * free pages of length of (1 << order) and marked with PG_buddy. Page's
402 * order is recorded in page_private(page) field.
403 * So when we are allocating or freeing one, we can derive the state of the
404 * other. That is, if we allocate a small block, and both were
405 * free, the remainder of the region must be split into blocks.
406 * If a block is freed, and its buddy is also free, then this
407 * triggers coalescing into a block of larger size.
412 static inline void __free_one_page(struct page
*page
,
413 struct zone
*zone
, unsigned int order
)
415 unsigned long page_idx
;
416 int order_size
= 1 << order
;
417 int migratetype
= get_pageblock_migratetype(page
);
419 if (unlikely(PageCompound(page
)))
420 destroy_compound_page(page
, order
);
422 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
424 VM_BUG_ON(page_idx
& (order_size
- 1));
425 VM_BUG_ON(bad_range(zone
, page
));
427 __mod_zone_page_state(zone
, NR_FREE_PAGES
, order_size
);
428 while (order
< MAX_ORDER
-1) {
429 unsigned long combined_idx
;
432 buddy
= __page_find_buddy(page
, page_idx
, order
);
433 if (!page_is_buddy(page
, buddy
, order
))
434 break; /* Move the buddy up one level. */
436 list_del(&buddy
->lru
);
437 zone
->free_area
[order
].nr_free
--;
438 rmv_page_order(buddy
);
439 combined_idx
= __find_combined_index(page_idx
, order
);
440 page
= page
+ (combined_idx
- page_idx
);
441 page_idx
= combined_idx
;
444 set_page_order(page
, order
);
446 &zone
->free_area
[order
].free_list
[migratetype
]);
447 zone
->free_area
[order
].nr_free
++;
450 static inline int free_pages_check(struct page
*page
)
452 if (unlikely(page_mapcount(page
) |
453 (page
->mapping
!= NULL
) |
454 (page_count(page
) != 0) |
467 __ClearPageDirty(page
);
469 * For now, we report if PG_reserved was found set, but do not
470 * clear it, and do not free the page. But we shall soon need
471 * to do more, for when the ZERO_PAGE count wraps negative.
473 return PageReserved(page
);
477 * Frees a list of pages.
478 * Assumes all pages on list are in same zone, and of same order.
479 * count is the number of pages to free.
481 * If the zone was previously in an "all pages pinned" state then look to
482 * see if this freeing clears that state.
484 * And clear the zone's pages_scanned counter, to hold off the "all pages are
485 * pinned" detection logic.
487 static void free_pages_bulk(struct zone
*zone
, int count
,
488 struct list_head
*list
, int order
)
490 spin_lock(&zone
->lock
);
491 zone
->all_unreclaimable
= 0;
492 zone
->pages_scanned
= 0;
496 VM_BUG_ON(list_empty(list
));
497 page
= list_entry(list
->prev
, struct page
, lru
);
498 /* have to delete it as __free_one_page list manipulates */
499 list_del(&page
->lru
);
500 __free_one_page(page
, zone
, order
);
502 spin_unlock(&zone
->lock
);
505 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
507 spin_lock(&zone
->lock
);
508 zone
->all_unreclaimable
= 0;
509 zone
->pages_scanned
= 0;
510 __free_one_page(page
, zone
, order
);
511 spin_unlock(&zone
->lock
);
514 static void __free_pages_ok(struct page
*page
, unsigned int order
)
520 for (i
= 0 ; i
< (1 << order
) ; ++i
)
521 reserved
+= free_pages_check(page
+ i
);
525 if (!PageHighMem(page
))
526 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
527 arch_free_page(page
, order
);
528 kernel_map_pages(page
, 1 << order
, 0);
530 local_irq_save(flags
);
531 __count_vm_events(PGFREE
, 1 << order
);
532 free_one_page(page_zone(page
), page
, order
);
533 local_irq_restore(flags
);
537 * permit the bootmem allocator to evade page validation on high-order frees
539 void fastcall __init
__free_pages_bootmem(struct page
*page
, unsigned int order
)
542 __ClearPageReserved(page
);
543 set_page_count(page
, 0);
544 set_page_refcounted(page
);
550 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
551 struct page
*p
= &page
[loop
];
553 if (loop
+ 1 < BITS_PER_LONG
)
555 __ClearPageReserved(p
);
556 set_page_count(p
, 0);
559 set_page_refcounted(page
);
560 __free_pages(page
, order
);
566 * The order of subdivision here is critical for the IO subsystem.
567 * Please do not alter this order without good reasons and regression
568 * testing. Specifically, as large blocks of memory are subdivided,
569 * the order in which smaller blocks are delivered depends on the order
570 * they're subdivided in this function. This is the primary factor
571 * influencing the order in which pages are delivered to the IO
572 * subsystem according to empirical testing, and this is also justified
573 * by considering the behavior of a buddy system containing a single
574 * large block of memory acted on by a series of small allocations.
575 * This behavior is a critical factor in sglist merging's success.
579 static inline void expand(struct zone
*zone
, struct page
*page
,
580 int low
, int high
, struct free_area
*area
,
583 unsigned long size
= 1 << high
;
589 VM_BUG_ON(bad_range(zone
, &page
[size
]));
590 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
592 set_page_order(&page
[size
], high
);
597 * This page is about to be returned from the page allocator
599 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
601 if (unlikely(page_mapcount(page
) |
602 (page
->mapping
!= NULL
) |
603 (page_count(page
) != 0) |
618 * For now, we report if PG_reserved was found set, but do not
619 * clear it, and do not allocate the page: as a safety net.
621 if (PageReserved(page
))
624 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
| 1 << PG_readahead
|
625 1 << PG_referenced
| 1 << PG_arch_1
|
626 1 << PG_owner_priv_1
| 1 << PG_mappedtodisk
);
627 set_page_private(page
, 0);
628 set_page_refcounted(page
);
630 arch_alloc_page(page
, order
);
631 kernel_map_pages(page
, 1 << order
, 1);
633 if (gfp_flags
& __GFP_ZERO
)
634 prep_zero_page(page
, order
, gfp_flags
);
636 if (order
&& (gfp_flags
& __GFP_COMP
))
637 prep_compound_page(page
, order
);
643 * Go through the free lists for the given migratetype and remove
644 * the smallest available page from the freelists
646 static struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
649 unsigned int current_order
;
650 struct free_area
* area
;
653 /* Find a page of the appropriate size in the preferred list */
654 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
655 area
= &(zone
->free_area
[current_order
]);
656 if (list_empty(&area
->free_list
[migratetype
]))
659 page
= list_entry(area
->free_list
[migratetype
].next
,
661 list_del(&page
->lru
);
662 rmv_page_order(page
);
664 __mod_zone_page_state(zone
, NR_FREE_PAGES
, - (1UL << order
));
665 expand(zone
, page
, order
, current_order
, area
, migratetype
);
674 * This array describes the order lists are fallen back to when
675 * the free lists for the desirable migrate type are depleted
677 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
678 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
679 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
680 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
681 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
685 * Move the free pages in a range to the free lists of the requested type.
686 * Note that start_page and end_pages are not aligned on a pageblock
687 * boundary. If alignment is required, use move_freepages_block()
689 int move_freepages(struct zone
*zone
,
690 struct page
*start_page
, struct page
*end_page
,
697 #ifndef CONFIG_HOLES_IN_ZONE
699 * page_zone is not safe to call in this context when
700 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
701 * anyway as we check zone boundaries in move_freepages_block().
702 * Remove at a later date when no bug reports exist related to
703 * grouping pages by mobility
705 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
708 for (page
= start_page
; page
<= end_page
;) {
709 if (!pfn_valid_within(page_to_pfn(page
))) {
714 if (!PageBuddy(page
)) {
719 order
= page_order(page
);
720 list_del(&page
->lru
);
722 &zone
->free_area
[order
].free_list
[migratetype
]);
724 pages_moved
+= 1 << order
;
730 int move_freepages_block(struct zone
*zone
, struct page
*page
, int migratetype
)
732 unsigned long start_pfn
, end_pfn
;
733 struct page
*start_page
, *end_page
;
735 start_pfn
= page_to_pfn(page
);
736 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
737 start_page
= pfn_to_page(start_pfn
);
738 end_page
= start_page
+ pageblock_nr_pages
- 1;
739 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
741 /* Do not cross zone boundaries */
742 if (start_pfn
< zone
->zone_start_pfn
)
744 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
747 return move_freepages(zone
, start_page
, end_page
, migratetype
);
750 /* Return the page with the lowest PFN in the list */
751 static struct page
*min_page(struct list_head
*list
)
753 unsigned long min_pfn
= -1UL;
754 struct page
*min_page
= NULL
, *page
;;
756 list_for_each_entry(page
, list
, lru
) {
757 unsigned long pfn
= page_to_pfn(page
);
767 /* Remove an element from the buddy allocator from the fallback list */
768 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
769 int start_migratetype
)
771 struct free_area
* area
;
776 /* Find the largest possible block of pages in the other list */
777 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
779 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
780 migratetype
= fallbacks
[start_migratetype
][i
];
782 /* MIGRATE_RESERVE handled later if necessary */
783 if (migratetype
== MIGRATE_RESERVE
)
786 area
= &(zone
->free_area
[current_order
]);
787 if (list_empty(&area
->free_list
[migratetype
]))
790 /* Bias kernel allocations towards low pfns */
791 page
= list_entry(area
->free_list
[migratetype
].next
,
793 if (unlikely(start_migratetype
!= MIGRATE_MOVABLE
))
794 page
= min_page(&area
->free_list
[migratetype
]);
798 * If breaking a large block of pages, move all free
799 * pages to the preferred allocation list. If falling
800 * back for a reclaimable kernel allocation, be more
801 * agressive about taking ownership of free pages
803 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
804 start_migratetype
== MIGRATE_RECLAIMABLE
) {
806 pages
= move_freepages_block(zone
, page
,
809 /* Claim the whole block if over half of it is free */
810 if (pages
>= (1 << (pageblock_order
-1)))
811 set_pageblock_migratetype(page
,
814 migratetype
= start_migratetype
;
817 /* Remove the page from the freelists */
818 list_del(&page
->lru
);
819 rmv_page_order(page
);
820 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
823 if (current_order
== pageblock_order
)
824 set_pageblock_migratetype(page
,
827 expand(zone
, page
, order
, current_order
, area
, migratetype
);
832 /* Use MIGRATE_RESERVE rather than fail an allocation */
833 return __rmqueue_smallest(zone
, order
, MIGRATE_RESERVE
);
837 * Do the hard work of removing an element from the buddy allocator.
838 * Call me with the zone->lock already held.
840 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
845 page
= __rmqueue_smallest(zone
, order
, migratetype
);
848 page
= __rmqueue_fallback(zone
, order
, migratetype
);
854 * Obtain a specified number of elements from the buddy allocator, all under
855 * a single hold of the lock, for efficiency. Add them to the supplied list.
856 * Returns the number of new pages which were placed at *list.
858 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
859 unsigned long count
, struct list_head
*list
,
864 spin_lock(&zone
->lock
);
865 for (i
= 0; i
< count
; ++i
) {
866 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
867 if (unlikely(page
== NULL
))
869 list_add(&page
->lru
, list
);
870 set_page_private(page
, migratetype
);
872 spin_unlock(&zone
->lock
);
878 * Called from the vmstat counter updater to drain pagesets of this
879 * currently executing processor on remote nodes after they have
882 * Note that this function must be called with the thread pinned to
883 * a single processor.
885 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
890 local_irq_save(flags
);
891 if (pcp
->count
>= pcp
->batch
)
892 to_drain
= pcp
->batch
;
894 to_drain
= pcp
->count
;
895 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
896 pcp
->count
-= to_drain
;
897 local_irq_restore(flags
);
901 static void __drain_pages(unsigned int cpu
)
907 for_each_zone(zone
) {
908 struct per_cpu_pageset
*pset
;
910 if (!populated_zone(zone
))
913 pset
= zone_pcp(zone
, cpu
);
914 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
915 struct per_cpu_pages
*pcp
;
918 local_irq_save(flags
);
919 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
921 local_irq_restore(flags
);
926 #ifdef CONFIG_HIBERNATION
928 void mark_free_pages(struct zone
*zone
)
930 unsigned long pfn
, max_zone_pfn
;
933 struct list_head
*curr
;
935 if (!zone
->spanned_pages
)
938 spin_lock_irqsave(&zone
->lock
, flags
);
940 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
941 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
942 if (pfn_valid(pfn
)) {
943 struct page
*page
= pfn_to_page(pfn
);
945 if (!swsusp_page_is_forbidden(page
))
946 swsusp_unset_page_free(page
);
949 for_each_migratetype_order(order
, t
) {
950 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
953 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
954 for (i
= 0; i
< (1UL << order
); i
++)
955 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
958 spin_unlock_irqrestore(&zone
->lock
, flags
);
960 #endif /* CONFIG_PM */
963 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
965 void drain_local_pages(void)
969 local_irq_save(flags
);
970 __drain_pages(smp_processor_id());
971 local_irq_restore(flags
);
974 void smp_drain_local_pages(void *arg
)
980 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
982 void drain_all_local_pages(void)
986 local_irq_save(flags
);
987 __drain_pages(smp_processor_id());
988 local_irq_restore(flags
);
990 smp_call_function(smp_drain_local_pages
, NULL
, 0, 1);
994 * Free a 0-order page
996 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
)
998 struct zone
*zone
= page_zone(page
);
999 struct per_cpu_pages
*pcp
;
1000 unsigned long flags
;
1003 page
->mapping
= NULL
;
1004 if (free_pages_check(page
))
1007 if (!PageHighMem(page
))
1008 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
1009 arch_free_page(page
, 0);
1010 kernel_map_pages(page
, 1, 0);
1012 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
1013 local_irq_save(flags
);
1014 __count_vm_event(PGFREE
);
1015 list_add(&page
->lru
, &pcp
->list
);
1016 set_page_private(page
, get_pageblock_migratetype(page
));
1018 if (pcp
->count
>= pcp
->high
) {
1019 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
1020 pcp
->count
-= pcp
->batch
;
1022 local_irq_restore(flags
);
1026 void fastcall
free_hot_page(struct page
*page
)
1028 free_hot_cold_page(page
, 0);
1031 void fastcall
free_cold_page(struct page
*page
)
1033 free_hot_cold_page(page
, 1);
1037 * split_page takes a non-compound higher-order page, and splits it into
1038 * n (1<<order) sub-pages: page[0..n]
1039 * Each sub-page must be freed individually.
1041 * Note: this is probably too low level an operation for use in drivers.
1042 * Please consult with lkml before using this in your driver.
1044 void split_page(struct page
*page
, unsigned int order
)
1048 VM_BUG_ON(PageCompound(page
));
1049 VM_BUG_ON(!page_count(page
));
1050 for (i
= 1; i
< (1 << order
); i
++)
1051 set_page_refcounted(page
+ i
);
1055 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1056 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1059 static struct page
*buffered_rmqueue(struct zonelist
*zonelist
,
1060 struct zone
*zone
, int order
, gfp_t gfp_flags
)
1062 unsigned long flags
;
1064 int cold
= !!(gfp_flags
& __GFP_COLD
);
1066 int migratetype
= allocflags_to_migratetype(gfp_flags
);
1070 if (likely(order
== 0)) {
1071 struct per_cpu_pages
*pcp
;
1073 pcp
= &zone_pcp(zone
, cpu
)->pcp
[cold
];
1074 local_irq_save(flags
);
1076 pcp
->count
= rmqueue_bulk(zone
, 0,
1077 pcp
->batch
, &pcp
->list
, migratetype
);
1078 if (unlikely(!pcp
->count
))
1082 /* Find a page of the appropriate migrate type */
1083 list_for_each_entry(page
, &pcp
->list
, lru
)
1084 if (page_private(page
) == migratetype
)
1087 /* Allocate more to the pcp list if necessary */
1088 if (unlikely(&page
->lru
== &pcp
->list
)) {
1089 pcp
->count
+= rmqueue_bulk(zone
, 0,
1090 pcp
->batch
, &pcp
->list
, migratetype
);
1091 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1094 list_del(&page
->lru
);
1097 spin_lock_irqsave(&zone
->lock
, flags
);
1098 page
= __rmqueue(zone
, order
, migratetype
);
1099 spin_unlock(&zone
->lock
);
1104 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1105 zone_statistics(zonelist
, zone
);
1106 local_irq_restore(flags
);
1109 VM_BUG_ON(bad_range(zone
, page
));
1110 if (prep_new_page(page
, order
, gfp_flags
))
1115 local_irq_restore(flags
);
1120 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1121 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1122 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1123 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1124 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1125 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1126 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1128 #ifdef CONFIG_FAIL_PAGE_ALLOC
1130 static struct fail_page_alloc_attr
{
1131 struct fault_attr attr
;
1133 u32 ignore_gfp_highmem
;
1134 u32 ignore_gfp_wait
;
1137 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1139 struct dentry
*ignore_gfp_highmem_file
;
1140 struct dentry
*ignore_gfp_wait_file
;
1141 struct dentry
*min_order_file
;
1143 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1145 } fail_page_alloc
= {
1146 .attr
= FAULT_ATTR_INITIALIZER
,
1147 .ignore_gfp_wait
= 1,
1148 .ignore_gfp_highmem
= 1,
1152 static int __init
setup_fail_page_alloc(char *str
)
1154 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1156 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1158 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1160 if (order
< fail_page_alloc
.min_order
)
1162 if (gfp_mask
& __GFP_NOFAIL
)
1164 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1166 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1169 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1172 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1174 static int __init
fail_page_alloc_debugfs(void)
1176 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1180 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1184 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1186 fail_page_alloc
.ignore_gfp_wait_file
=
1187 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1188 &fail_page_alloc
.ignore_gfp_wait
);
1190 fail_page_alloc
.ignore_gfp_highmem_file
=
1191 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1192 &fail_page_alloc
.ignore_gfp_highmem
);
1193 fail_page_alloc
.min_order_file
=
1194 debugfs_create_u32("min-order", mode
, dir
,
1195 &fail_page_alloc
.min_order
);
1197 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1198 !fail_page_alloc
.ignore_gfp_highmem_file
||
1199 !fail_page_alloc
.min_order_file
) {
1201 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1202 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1203 debugfs_remove(fail_page_alloc
.min_order_file
);
1204 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1210 late_initcall(fail_page_alloc_debugfs
);
1212 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1214 #else /* CONFIG_FAIL_PAGE_ALLOC */
1216 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1221 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1224 * Return 1 if free pages are above 'mark'. This takes into account the order
1225 * of the allocation.
1227 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1228 int classzone_idx
, int alloc_flags
)
1230 /* free_pages my go negative - that's OK */
1232 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1235 if (alloc_flags
& ALLOC_HIGH
)
1237 if (alloc_flags
& ALLOC_HARDER
)
1240 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1242 for (o
= 0; o
< order
; o
++) {
1243 /* At the next order, this order's pages become unavailable */
1244 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1246 /* Require fewer higher order pages to be free */
1249 if (free_pages
<= min
)
1257 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1258 * skip over zones that are not allowed by the cpuset, or that have
1259 * been recently (in last second) found to be nearly full. See further
1260 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1261 * that have to skip over alot of full or unallowed zones.
1263 * If the zonelist cache is present in the passed in zonelist, then
1264 * returns a pointer to the allowed node mask (either the current
1265 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1267 * If the zonelist cache is not available for this zonelist, does
1268 * nothing and returns NULL.
1270 * If the fullzones BITMAP in the zonelist cache is stale (more than
1271 * a second since last zap'd) then we zap it out (clear its bits.)
1273 * We hold off even calling zlc_setup, until after we've checked the
1274 * first zone in the zonelist, on the theory that most allocations will
1275 * be satisfied from that first zone, so best to examine that zone as
1276 * quickly as we can.
1278 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1280 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1281 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1283 zlc
= zonelist
->zlcache_ptr
;
1287 if (jiffies
- zlc
->last_full_zap
> 1 * HZ
) {
1288 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1289 zlc
->last_full_zap
= jiffies
;
1292 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1293 &cpuset_current_mems_allowed
:
1294 &node_states
[N_HIGH_MEMORY
];
1295 return allowednodes
;
1299 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1300 * if it is worth looking at further for free memory:
1301 * 1) Check that the zone isn't thought to be full (doesn't have its
1302 * bit set in the zonelist_cache fullzones BITMAP).
1303 * 2) Check that the zones node (obtained from the zonelist_cache
1304 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1305 * Return true (non-zero) if zone is worth looking at further, or
1306 * else return false (zero) if it is not.
1308 * This check -ignores- the distinction between various watermarks,
1309 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1310 * found to be full for any variation of these watermarks, it will
1311 * be considered full for up to one second by all requests, unless
1312 * we are so low on memory on all allowed nodes that we are forced
1313 * into the second scan of the zonelist.
1315 * In the second scan we ignore this zonelist cache and exactly
1316 * apply the watermarks to all zones, even it is slower to do so.
1317 * We are low on memory in the second scan, and should leave no stone
1318 * unturned looking for a free page.
1320 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zone
**z
,
1321 nodemask_t
*allowednodes
)
1323 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1324 int i
; /* index of *z in zonelist zones */
1325 int n
; /* node that zone *z is on */
1327 zlc
= zonelist
->zlcache_ptr
;
1331 i
= z
- zonelist
->zones
;
1334 /* This zone is worth trying if it is allowed but not full */
1335 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1339 * Given 'z' scanning a zonelist, set the corresponding bit in
1340 * zlc->fullzones, so that subsequent attempts to allocate a page
1341 * from that zone don't waste time re-examining it.
1343 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zone
**z
)
1345 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1346 int i
; /* index of *z in zonelist zones */
1348 zlc
= zonelist
->zlcache_ptr
;
1352 i
= z
- zonelist
->zones
;
1354 set_bit(i
, zlc
->fullzones
);
1357 #else /* CONFIG_NUMA */
1359 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1364 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zone
**z
,
1365 nodemask_t
*allowednodes
)
1370 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zone
**z
)
1373 #endif /* CONFIG_NUMA */
1376 * get_page_from_freelist goes through the zonelist trying to allocate
1379 static struct page
*
1380 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
,
1381 struct zonelist
*zonelist
, int alloc_flags
)
1384 struct page
*page
= NULL
;
1385 int classzone_idx
= zone_idx(zonelist
->zones
[0]);
1387 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1388 int zlc_active
= 0; /* set if using zonelist_cache */
1389 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1390 enum zone_type highest_zoneidx
= -1; /* Gets set for policy zonelists */
1394 * Scan zonelist, looking for a zone with enough free.
1395 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1397 z
= zonelist
->zones
;
1401 * In NUMA, this could be a policy zonelist which contains
1402 * zones that may not be allowed by the current gfp_mask.
1403 * Check the zone is allowed by the current flags
1405 if (unlikely(alloc_should_filter_zonelist(zonelist
))) {
1406 if (highest_zoneidx
== -1)
1407 highest_zoneidx
= gfp_zone(gfp_mask
);
1408 if (zone_idx(*z
) > highest_zoneidx
)
1412 if (NUMA_BUILD
&& zlc_active
&&
1413 !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(zonelist
, zone
, order
, gfp_mask
);
1441 zlc_mark_zone_full(zonelist
, z
);
1443 if (NUMA_BUILD
&& !did_zlc_setup
) {
1444 /* we do zlc_setup after the first zone is tried */
1445 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1449 } while (*(++z
) != NULL
);
1451 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1452 /* Disable zlc cache for second zonelist scan */
1460 * This is the 'heart' of the zoned buddy allocator.
1462 struct page
* fastcall
1463 __alloc_pages(gfp_t gfp_mask
, unsigned int order
,
1464 struct zonelist
*zonelist
)
1466 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1469 struct reclaim_state reclaim_state
;
1470 struct task_struct
*p
= current
;
1473 int did_some_progress
;
1475 might_sleep_if(wait
);
1477 if (should_fail_alloc_page(gfp_mask
, order
))
1481 z
= zonelist
->zones
; /* the list of zones suitable for gfp_mask */
1483 if (unlikely(*z
== NULL
)) {
1485 * Happens if we have an empty zonelist as a result of
1486 * GFP_THISNODE being used on a memoryless node
1491 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1492 zonelist
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1497 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1498 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1499 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1500 * using a larger set of nodes after it has established that the
1501 * allowed per node queues are empty and that nodes are
1504 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1507 for (z
= zonelist
->zones
; *z
; z
++)
1508 wakeup_kswapd(*z
, order
);
1511 * OK, we're below the kswapd watermark and have kicked background
1512 * reclaim. Now things get more complex, so set up alloc_flags according
1513 * to how we want to proceed.
1515 * The caller may dip into page reserves a bit more if the caller
1516 * cannot run direct reclaim, or if the caller has realtime scheduling
1517 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1518 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1520 alloc_flags
= ALLOC_WMARK_MIN
;
1521 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1522 alloc_flags
|= ALLOC_HARDER
;
1523 if (gfp_mask
& __GFP_HIGH
)
1524 alloc_flags
|= ALLOC_HIGH
;
1526 alloc_flags
|= ALLOC_CPUSET
;
1529 * Go through the zonelist again. Let __GFP_HIGH and allocations
1530 * coming from realtime tasks go deeper into reserves.
1532 * This is the last chance, in general, before the goto nopage.
1533 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1534 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1536 page
= get_page_from_freelist(gfp_mask
, order
, zonelist
, alloc_flags
);
1540 /* This allocation should allow future memory freeing. */
1543 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1544 && !in_interrupt()) {
1545 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1547 /* go through the zonelist yet again, ignoring mins */
1548 page
= get_page_from_freelist(gfp_mask
, order
,
1549 zonelist
, ALLOC_NO_WATERMARKS
);
1552 if (gfp_mask
& __GFP_NOFAIL
) {
1553 congestion_wait(WRITE
, HZ
/50);
1560 /* Atomic allocations - we can't balance anything */
1566 /* We now go into synchronous reclaim */
1567 cpuset_memory_pressure_bump();
1568 p
->flags
|= PF_MEMALLOC
;
1569 reclaim_state
.reclaimed_slab
= 0;
1570 p
->reclaim_state
= &reclaim_state
;
1572 did_some_progress
= try_to_free_pages(zonelist
->zones
, order
, gfp_mask
);
1574 p
->reclaim_state
= NULL
;
1575 p
->flags
&= ~PF_MEMALLOC
;
1580 drain_all_local_pages();
1582 if (likely(did_some_progress
)) {
1583 page
= get_page_from_freelist(gfp_mask
, order
,
1584 zonelist
, alloc_flags
);
1587 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1589 * Go through the zonelist yet one more time, keep
1590 * very high watermark here, this is only to catch
1591 * a parallel oom killing, we must fail if we're still
1592 * under heavy pressure.
1594 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1595 zonelist
, ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1599 /* The OOM killer will not help higher order allocs so fail */
1600 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
1603 out_of_memory(zonelist
, gfp_mask
, order
);
1608 * Don't let big-order allocations loop unless the caller explicitly
1609 * requests that. Wait for some write requests to complete then retry.
1611 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1612 * <= 3, but that may not be true in other implementations.
1615 if (!(gfp_mask
& __GFP_NORETRY
)) {
1616 if ((order
<= PAGE_ALLOC_COSTLY_ORDER
) ||
1617 (gfp_mask
& __GFP_REPEAT
))
1619 if (gfp_mask
& __GFP_NOFAIL
)
1623 congestion_wait(WRITE
, HZ
/50);
1628 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1629 printk(KERN_WARNING
"%s: page allocation failure."
1630 " order:%d, mode:0x%x\n",
1631 p
->comm
, order
, gfp_mask
);
1639 EXPORT_SYMBOL(__alloc_pages
);
1642 * Common helper functions.
1644 fastcall
unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1647 page
= alloc_pages(gfp_mask
, order
);
1650 return (unsigned long) page_address(page
);
1653 EXPORT_SYMBOL(__get_free_pages
);
1655 fastcall
unsigned long get_zeroed_page(gfp_t gfp_mask
)
1660 * get_zeroed_page() returns a 32-bit address, which cannot represent
1663 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1665 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1667 return (unsigned long) page_address(page
);
1671 EXPORT_SYMBOL(get_zeroed_page
);
1673 void __pagevec_free(struct pagevec
*pvec
)
1675 int i
= pagevec_count(pvec
);
1678 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1681 fastcall
void __free_pages(struct page
*page
, unsigned int order
)
1683 if (put_page_testzero(page
)) {
1685 free_hot_page(page
);
1687 __free_pages_ok(page
, order
);
1691 EXPORT_SYMBOL(__free_pages
);
1693 fastcall
void free_pages(unsigned long addr
, unsigned int order
)
1696 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1697 __free_pages(virt_to_page((void *)addr
), order
);
1701 EXPORT_SYMBOL(free_pages
);
1703 static unsigned int nr_free_zone_pages(int offset
)
1705 /* Just pick one node, since fallback list is circular */
1706 pg_data_t
*pgdat
= NODE_DATA(numa_node_id());
1707 unsigned int sum
= 0;
1709 struct zonelist
*zonelist
= pgdat
->node_zonelists
+ offset
;
1710 struct zone
**zonep
= zonelist
->zones
;
1713 for (zone
= *zonep
++; zone
; zone
= *zonep
++) {
1714 unsigned long size
= zone
->present_pages
;
1715 unsigned long high
= zone
->pages_high
;
1724 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1726 unsigned int nr_free_buffer_pages(void)
1728 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1730 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
1733 * Amount of free RAM allocatable within all zones
1735 unsigned int nr_free_pagecache_pages(void)
1737 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
1740 static inline void show_node(struct zone
*zone
)
1743 printk("Node %d ", zone_to_nid(zone
));
1746 void si_meminfo(struct sysinfo
*val
)
1748 val
->totalram
= totalram_pages
;
1750 val
->freeram
= global_page_state(NR_FREE_PAGES
);
1751 val
->bufferram
= nr_blockdev_pages();
1752 val
->totalhigh
= totalhigh_pages
;
1753 val
->freehigh
= nr_free_highpages();
1754 val
->mem_unit
= PAGE_SIZE
;
1757 EXPORT_SYMBOL(si_meminfo
);
1760 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1762 pg_data_t
*pgdat
= NODE_DATA(nid
);
1764 val
->totalram
= pgdat
->node_present_pages
;
1765 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
1766 #ifdef CONFIG_HIGHMEM
1767 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1768 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
1774 val
->mem_unit
= PAGE_SIZE
;
1778 #define K(x) ((x) << (PAGE_SHIFT-10))
1781 * Show free area list (used inside shift_scroll-lock stuff)
1782 * We also calculate the percentage fragmentation. We do this by counting the
1783 * memory on each free list with the exception of the first item on the list.
1785 void show_free_areas(void)
1790 for_each_zone(zone
) {
1791 if (!populated_zone(zone
))
1795 printk("%s per-cpu:\n", zone
->name
);
1797 for_each_online_cpu(cpu
) {
1798 struct per_cpu_pageset
*pageset
;
1800 pageset
= zone_pcp(zone
, cpu
);
1802 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1803 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1804 cpu
, pageset
->pcp
[0].high
,
1805 pageset
->pcp
[0].batch
, pageset
->pcp
[0].count
,
1806 pageset
->pcp
[1].high
, pageset
->pcp
[1].batch
,
1807 pageset
->pcp
[1].count
);
1811 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
1812 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1813 global_page_state(NR_ACTIVE
),
1814 global_page_state(NR_INACTIVE
),
1815 global_page_state(NR_FILE_DIRTY
),
1816 global_page_state(NR_WRITEBACK
),
1817 global_page_state(NR_UNSTABLE_NFS
),
1818 global_page_state(NR_FREE_PAGES
),
1819 global_page_state(NR_SLAB_RECLAIMABLE
) +
1820 global_page_state(NR_SLAB_UNRECLAIMABLE
),
1821 global_page_state(NR_FILE_MAPPED
),
1822 global_page_state(NR_PAGETABLE
),
1823 global_page_state(NR_BOUNCE
));
1825 for_each_zone(zone
) {
1828 if (!populated_zone(zone
))
1840 " pages_scanned:%lu"
1841 " all_unreclaimable? %s"
1844 K(zone_page_state(zone
, NR_FREE_PAGES
)),
1847 K(zone
->pages_high
),
1848 K(zone_page_state(zone
, NR_ACTIVE
)),
1849 K(zone_page_state(zone
, NR_INACTIVE
)),
1850 K(zone
->present_pages
),
1851 zone
->pages_scanned
,
1852 (zone
->all_unreclaimable
? "yes" : "no")
1854 printk("lowmem_reserve[]:");
1855 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1856 printk(" %lu", zone
->lowmem_reserve
[i
]);
1860 for_each_zone(zone
) {
1861 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1863 if (!populated_zone(zone
))
1867 printk("%s: ", zone
->name
);
1869 spin_lock_irqsave(&zone
->lock
, flags
);
1870 for (order
= 0; order
< MAX_ORDER
; order
++) {
1871 nr
[order
] = zone
->free_area
[order
].nr_free
;
1872 total
+= nr
[order
] << order
;
1874 spin_unlock_irqrestore(&zone
->lock
, flags
);
1875 for (order
= 0; order
< MAX_ORDER
; order
++)
1876 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1877 printk("= %lukB\n", K(total
));
1880 show_swap_cache_info();
1884 * Builds allocation fallback zone lists.
1886 * Add all populated zones of a node to the zonelist.
1888 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
1889 int nr_zones
, enum zone_type zone_type
)
1893 BUG_ON(zone_type
>= MAX_NR_ZONES
);
1898 zone
= pgdat
->node_zones
+ zone_type
;
1899 if (populated_zone(zone
)) {
1900 zonelist
->zones
[nr_zones
++] = zone
;
1901 check_highest_zone(zone_type
);
1904 } while (zone_type
);
1911 * 0 = automatic detection of better ordering.
1912 * 1 = order by ([node] distance, -zonetype)
1913 * 2 = order by (-zonetype, [node] distance)
1915 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1916 * the same zonelist. So only NUMA can configure this param.
1918 #define ZONELIST_ORDER_DEFAULT 0
1919 #define ZONELIST_ORDER_NODE 1
1920 #define ZONELIST_ORDER_ZONE 2
1922 /* zonelist order in the kernel.
1923 * set_zonelist_order() will set this to NODE or ZONE.
1925 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1926 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
1930 /* The value user specified ....changed by config */
1931 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1932 /* string for sysctl */
1933 #define NUMA_ZONELIST_ORDER_LEN 16
1934 char numa_zonelist_order
[16] = "default";
1937 * interface for configure zonelist ordering.
1938 * command line option "numa_zonelist_order"
1939 * = "[dD]efault - default, automatic configuration.
1940 * = "[nN]ode - order by node locality, then by zone within node
1941 * = "[zZ]one - order by zone, then by locality within zone
1944 static int __parse_numa_zonelist_order(char *s
)
1946 if (*s
== 'd' || *s
== 'D') {
1947 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1948 } else if (*s
== 'n' || *s
== 'N') {
1949 user_zonelist_order
= ZONELIST_ORDER_NODE
;
1950 } else if (*s
== 'z' || *s
== 'Z') {
1951 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
1954 "Ignoring invalid numa_zonelist_order value: "
1961 static __init
int setup_numa_zonelist_order(char *s
)
1964 return __parse_numa_zonelist_order(s
);
1967 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
1970 * sysctl handler for numa_zonelist_order
1972 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
1973 struct file
*file
, void __user
*buffer
, size_t *length
,
1976 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
1980 strncpy(saved_string
, (char*)table
->data
,
1981 NUMA_ZONELIST_ORDER_LEN
);
1982 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
1986 int oldval
= user_zonelist_order
;
1987 if (__parse_numa_zonelist_order((char*)table
->data
)) {
1989 * bogus value. restore saved string
1991 strncpy((char*)table
->data
, saved_string
,
1992 NUMA_ZONELIST_ORDER_LEN
);
1993 user_zonelist_order
= oldval
;
1994 } else if (oldval
!= user_zonelist_order
)
1995 build_all_zonelists();
2001 #define MAX_NODE_LOAD (num_online_nodes())
2002 static int node_load
[MAX_NUMNODES
];
2005 * find_next_best_node - find the next node that should appear in a given node's fallback list
2006 * @node: node whose fallback list we're appending
2007 * @used_node_mask: nodemask_t of already used nodes
2009 * We use a number of factors to determine which is the next node that should
2010 * appear on a given node's fallback list. The node should not have appeared
2011 * already in @node's fallback list, and it should be the next closest node
2012 * according to the distance array (which contains arbitrary distance values
2013 * from each node to each node in the system), and should also prefer nodes
2014 * with no CPUs, since presumably they'll have very little allocation pressure
2015 * on them otherwise.
2016 * It returns -1 if no node is found.
2018 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2021 int min_val
= INT_MAX
;
2024 /* Use the local node if we haven't already */
2025 if (!node_isset(node
, *used_node_mask
)) {
2026 node_set(node
, *used_node_mask
);
2030 for_each_node_state(n
, N_HIGH_MEMORY
) {
2033 /* Don't want a node to appear more than once */
2034 if (node_isset(n
, *used_node_mask
))
2037 /* Use the distance array to find the distance */
2038 val
= node_distance(node
, n
);
2040 /* Penalize nodes under us ("prefer the next node") */
2043 /* Give preference to headless and unused nodes */
2044 tmp
= node_to_cpumask(n
);
2045 if (!cpus_empty(tmp
))
2046 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2048 /* Slight preference for less loaded node */
2049 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2050 val
+= node_load
[n
];
2052 if (val
< min_val
) {
2059 node_set(best_node
, *used_node_mask
);
2066 * Build zonelists ordered by node and zones within node.
2067 * This results in maximum locality--normal zone overflows into local
2068 * DMA zone, if any--but risks exhausting DMA zone.
2070 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2074 struct zonelist
*zonelist
;
2076 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2077 zonelist
= pgdat
->node_zonelists
+ i
;
2078 for (j
= 0; zonelist
->zones
[j
] != NULL
; j
++)
2080 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
2081 zonelist
->zones
[j
] = NULL
;
2086 * Build gfp_thisnode zonelists
2088 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2092 struct zonelist
*zonelist
;
2094 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2095 zonelist
= pgdat
->node_zonelists
+ MAX_NR_ZONES
+ i
;
2096 j
= build_zonelists_node(pgdat
, zonelist
, 0, i
);
2097 zonelist
->zones
[j
] = NULL
;
2102 * Build zonelists ordered by zone and nodes within zones.
2103 * This results in conserving DMA zone[s] until all Normal memory is
2104 * exhausted, but results in overflowing to remote node while memory
2105 * may still exist in local DMA zone.
2107 static int node_order
[MAX_NUMNODES
];
2109 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2113 int zone_type
; /* needs to be signed */
2115 struct zonelist
*zonelist
;
2117 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2118 zonelist
= pgdat
->node_zonelists
+ i
;
2120 for (zone_type
= i
; zone_type
>= 0; zone_type
--) {
2121 for (j
= 0; j
< nr_nodes
; j
++) {
2122 node
= node_order
[j
];
2123 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2124 if (populated_zone(z
)) {
2125 zonelist
->zones
[pos
++] = z
;
2126 check_highest_zone(zone_type
);
2130 zonelist
->zones
[pos
] = NULL
;
2134 static int default_zonelist_order(void)
2137 unsigned long low_kmem_size
,total_size
;
2141 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2142 * If they are really small and used heavily, the system can fall
2143 * into OOM very easily.
2144 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2146 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2149 for_each_online_node(nid
) {
2150 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2151 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2152 if (populated_zone(z
)) {
2153 if (zone_type
< ZONE_NORMAL
)
2154 low_kmem_size
+= z
->present_pages
;
2155 total_size
+= z
->present_pages
;
2159 if (!low_kmem_size
|| /* there are no DMA area. */
2160 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2161 return ZONELIST_ORDER_NODE
;
2163 * look into each node's config.
2164 * If there is a node whose DMA/DMA32 memory is very big area on
2165 * local memory, NODE_ORDER may be suitable.
2167 average_size
= total_size
/
2168 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2169 for_each_online_node(nid
) {
2172 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2173 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2174 if (populated_zone(z
)) {
2175 if (zone_type
< ZONE_NORMAL
)
2176 low_kmem_size
+= z
->present_pages
;
2177 total_size
+= z
->present_pages
;
2180 if (low_kmem_size
&&
2181 total_size
> average_size
&& /* ignore small node */
2182 low_kmem_size
> total_size
* 70/100)
2183 return ZONELIST_ORDER_NODE
;
2185 return ZONELIST_ORDER_ZONE
;
2188 static void set_zonelist_order(void)
2190 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2191 current_zonelist_order
= default_zonelist_order();
2193 current_zonelist_order
= user_zonelist_order
;
2196 static void build_zonelists(pg_data_t
*pgdat
)
2200 nodemask_t used_mask
;
2201 int local_node
, prev_node
;
2202 struct zonelist
*zonelist
;
2203 int order
= current_zonelist_order
;
2205 /* initialize zonelists */
2206 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2207 zonelist
= pgdat
->node_zonelists
+ i
;
2208 zonelist
->zones
[0] = NULL
;
2211 /* NUMA-aware ordering of nodes */
2212 local_node
= pgdat
->node_id
;
2213 load
= num_online_nodes();
2214 prev_node
= local_node
;
2215 nodes_clear(used_mask
);
2217 memset(node_load
, 0, sizeof(node_load
));
2218 memset(node_order
, 0, sizeof(node_order
));
2221 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2222 int distance
= node_distance(local_node
, node
);
2225 * If another node is sufficiently far away then it is better
2226 * to reclaim pages in a zone before going off node.
2228 if (distance
> RECLAIM_DISTANCE
)
2229 zone_reclaim_mode
= 1;
2232 * We don't want to pressure a particular node.
2233 * So adding penalty to the first node in same
2234 * distance group to make it round-robin.
2236 if (distance
!= node_distance(local_node
, prev_node
))
2237 node_load
[node
] = load
;
2241 if (order
== ZONELIST_ORDER_NODE
)
2242 build_zonelists_in_node_order(pgdat
, node
);
2244 node_order
[j
++] = node
; /* remember order */
2247 if (order
== ZONELIST_ORDER_ZONE
) {
2248 /* calculate node order -- i.e., DMA last! */
2249 build_zonelists_in_zone_order(pgdat
, j
);
2252 build_thisnode_zonelists(pgdat
);
2255 /* Construct the zonelist performance cache - see further mmzone.h */
2256 static void build_zonelist_cache(pg_data_t
*pgdat
)
2260 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2261 struct zonelist
*zonelist
;
2262 struct zonelist_cache
*zlc
;
2265 zonelist
= pgdat
->node_zonelists
+ i
;
2266 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2267 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2268 for (z
= zonelist
->zones
; *z
; z
++)
2269 zlc
->z_to_n
[z
- zonelist
->zones
] = zone_to_nid(*z
);
2274 #else /* CONFIG_NUMA */
2276 static void set_zonelist_order(void)
2278 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2281 static void build_zonelists(pg_data_t
*pgdat
)
2283 int node
, local_node
;
2286 local_node
= pgdat
->node_id
;
2287 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2288 struct zonelist
*zonelist
;
2290 zonelist
= pgdat
->node_zonelists
+ i
;
2292 j
= build_zonelists_node(pgdat
, zonelist
, 0, i
);
2294 * Now we build the zonelist so that it contains the zones
2295 * of all the other nodes.
2296 * We don't want to pressure a particular node, so when
2297 * building the zones for node N, we make sure that the
2298 * zones coming right after the local ones are those from
2299 * node N+1 (modulo N)
2301 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2302 if (!node_online(node
))
2304 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
2306 for (node
= 0; node
< local_node
; node
++) {
2307 if (!node_online(node
))
2309 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
2312 zonelist
->zones
[j
] = NULL
;
2316 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2317 static void build_zonelist_cache(pg_data_t
*pgdat
)
2321 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2322 pgdat
->node_zonelists
[i
].zlcache_ptr
= NULL
;
2325 #endif /* CONFIG_NUMA */
2327 /* return values int ....just for stop_machine_run() */
2328 static int __build_all_zonelists(void *dummy
)
2332 for_each_online_node(nid
) {
2333 pg_data_t
*pgdat
= NODE_DATA(nid
);
2335 build_zonelists(pgdat
);
2336 build_zonelist_cache(pgdat
);
2341 void build_all_zonelists(void)
2343 set_zonelist_order();
2345 if (system_state
== SYSTEM_BOOTING
) {
2346 __build_all_zonelists(NULL
);
2347 cpuset_init_current_mems_allowed();
2349 /* we have to stop all cpus to guaranntee there is no user
2351 stop_machine_run(__build_all_zonelists
, NULL
, NR_CPUS
);
2352 /* cpuset refresh routine should be here */
2354 vm_total_pages
= nr_free_pagecache_pages();
2356 * Disable grouping by mobility if the number of pages in the
2357 * system is too low to allow the mechanism to work. It would be
2358 * more accurate, but expensive to check per-zone. This check is
2359 * made on memory-hotadd so a system can start with mobility
2360 * disabled and enable it later
2362 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
2363 page_group_by_mobility_disabled
= 1;
2365 page_group_by_mobility_disabled
= 0;
2367 printk("Built %i zonelists in %s order, mobility grouping %s. "
2368 "Total pages: %ld\n",
2370 zonelist_order_name
[current_zonelist_order
],
2371 page_group_by_mobility_disabled
? "off" : "on",
2374 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2379 * Helper functions to size the waitqueue hash table.
2380 * Essentially these want to choose hash table sizes sufficiently
2381 * large so that collisions trying to wait on pages are rare.
2382 * But in fact, the number of active page waitqueues on typical
2383 * systems is ridiculously low, less than 200. So this is even
2384 * conservative, even though it seems large.
2386 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2387 * waitqueues, i.e. the size of the waitq table given the number of pages.
2389 #define PAGES_PER_WAITQUEUE 256
2391 #ifndef CONFIG_MEMORY_HOTPLUG
2392 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2394 unsigned long size
= 1;
2396 pages
/= PAGES_PER_WAITQUEUE
;
2398 while (size
< pages
)
2402 * Once we have dozens or even hundreds of threads sleeping
2403 * on IO we've got bigger problems than wait queue collision.
2404 * Limit the size of the wait table to a reasonable size.
2406 size
= min(size
, 4096UL);
2408 return max(size
, 4UL);
2412 * A zone's size might be changed by hot-add, so it is not possible to determine
2413 * a suitable size for its wait_table. So we use the maximum size now.
2415 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2417 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2418 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2419 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2421 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2422 * or more by the traditional way. (See above). It equals:
2424 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2425 * ia64(16K page size) : = ( 8G + 4M)byte.
2426 * powerpc (64K page size) : = (32G +16M)byte.
2428 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2435 * This is an integer logarithm so that shifts can be used later
2436 * to extract the more random high bits from the multiplicative
2437 * hash function before the remainder is taken.
2439 static inline unsigned long wait_table_bits(unsigned long size
)
2444 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2447 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2448 * of blocks reserved is based on zone->pages_min. The memory within the
2449 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2450 * higher will lead to a bigger reserve which will get freed as contiguous
2451 * blocks as reclaim kicks in
2453 static void setup_zone_migrate_reserve(struct zone
*zone
)
2455 unsigned long start_pfn
, pfn
, end_pfn
;
2457 unsigned long reserve
, block_migratetype
;
2459 /* Get the start pfn, end pfn and the number of blocks to reserve */
2460 start_pfn
= zone
->zone_start_pfn
;
2461 end_pfn
= start_pfn
+ zone
->spanned_pages
;
2462 reserve
= roundup(zone
->pages_min
, pageblock_nr_pages
) >>
2465 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
2466 if (!pfn_valid(pfn
))
2468 page
= pfn_to_page(pfn
);
2470 /* Blocks with reserved pages will never free, skip them. */
2471 if (PageReserved(page
))
2474 block_migratetype
= get_pageblock_migratetype(page
);
2476 /* If this block is reserved, account for it */
2477 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
2482 /* Suitable for reserving if this block is movable */
2483 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
2484 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
2485 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
2491 * If the reserve is met and this is a previous reserved block,
2494 if (block_migratetype
== MIGRATE_RESERVE
) {
2495 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2496 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
2502 * Initially all pages are reserved - free ones are freed
2503 * up by free_all_bootmem() once the early boot process is
2504 * done. Non-atomic initialization, single-pass.
2506 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2507 unsigned long start_pfn
, enum memmap_context context
)
2510 unsigned long end_pfn
= start_pfn
+ size
;
2513 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2515 * There can be holes in boot-time mem_map[]s
2516 * handed to this function. They do not
2517 * exist on hotplugged memory.
2519 if (context
== MEMMAP_EARLY
) {
2520 if (!early_pfn_valid(pfn
))
2522 if (!early_pfn_in_nid(pfn
, nid
))
2525 page
= pfn_to_page(pfn
);
2526 set_page_links(page
, zone
, nid
, pfn
);
2527 init_page_count(page
);
2528 reset_page_mapcount(page
);
2529 SetPageReserved(page
);
2532 * Mark the block movable so that blocks are reserved for
2533 * movable at startup. This will force kernel allocations
2534 * to reserve their blocks rather than leaking throughout
2535 * the address space during boot when many long-lived
2536 * kernel allocations are made. Later some blocks near
2537 * the start are marked MIGRATE_RESERVE by
2538 * setup_zone_migrate_reserve()
2540 if ((pfn
& (pageblock_nr_pages
-1)))
2541 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2543 INIT_LIST_HEAD(&page
->lru
);
2544 #ifdef WANT_PAGE_VIRTUAL
2545 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2546 if (!is_highmem_idx(zone
))
2547 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2552 static void __meminit
zone_init_free_lists(struct pglist_data
*pgdat
,
2553 struct zone
*zone
, unsigned long size
)
2556 for_each_migratetype_order(order
, t
) {
2557 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
2558 zone
->free_area
[order
].nr_free
= 0;
2562 #ifndef __HAVE_ARCH_MEMMAP_INIT
2563 #define memmap_init(size, nid, zone, start_pfn) \
2564 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2567 static int __devinit
zone_batchsize(struct zone
*zone
)
2572 * The per-cpu-pages pools are set to around 1000th of the
2573 * size of the zone. But no more than 1/2 of a meg.
2575 * OK, so we don't know how big the cache is. So guess.
2577 batch
= zone
->present_pages
/ 1024;
2578 if (batch
* PAGE_SIZE
> 512 * 1024)
2579 batch
= (512 * 1024) / PAGE_SIZE
;
2580 batch
/= 4; /* We effectively *= 4 below */
2585 * Clamp the batch to a 2^n - 1 value. Having a power
2586 * of 2 value was found to be more likely to have
2587 * suboptimal cache aliasing properties in some cases.
2589 * For example if 2 tasks are alternately allocating
2590 * batches of pages, one task can end up with a lot
2591 * of pages of one half of the possible page colors
2592 * and the other with pages of the other colors.
2594 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
2599 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2601 struct per_cpu_pages
*pcp
;
2603 memset(p
, 0, sizeof(*p
));
2605 pcp
= &p
->pcp
[0]; /* hot */
2607 pcp
->high
= 6 * batch
;
2608 pcp
->batch
= max(1UL, 1 * batch
);
2609 INIT_LIST_HEAD(&pcp
->list
);
2611 pcp
= &p
->pcp
[1]; /* cold*/
2613 pcp
->high
= 2 * batch
;
2614 pcp
->batch
= max(1UL, batch
/2);
2615 INIT_LIST_HEAD(&pcp
->list
);
2619 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2620 * to the value high for the pageset p.
2623 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2626 struct per_cpu_pages
*pcp
;
2628 pcp
= &p
->pcp
[0]; /* hot list */
2630 pcp
->batch
= max(1UL, high
/4);
2631 if ((high
/4) > (PAGE_SHIFT
* 8))
2632 pcp
->batch
= PAGE_SHIFT
* 8;
2638 * Boot pageset table. One per cpu which is going to be used for all
2639 * zones and all nodes. The parameters will be set in such a way
2640 * that an item put on a list will immediately be handed over to
2641 * the buddy list. This is safe since pageset manipulation is done
2642 * with interrupts disabled.
2644 * Some NUMA counter updates may also be caught by the boot pagesets.
2646 * The boot_pagesets must be kept even after bootup is complete for
2647 * unused processors and/or zones. They do play a role for bootstrapping
2648 * hotplugged processors.
2650 * zoneinfo_show() and maybe other functions do
2651 * not check if the processor is online before following the pageset pointer.
2652 * Other parts of the kernel may not check if the zone is available.
2654 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
2657 * Dynamically allocate memory for the
2658 * per cpu pageset array in struct zone.
2660 static int __cpuinit
process_zones(int cpu
)
2662 struct zone
*zone
, *dzone
;
2663 int node
= cpu_to_node(cpu
);
2665 node_set_state(node
, N_CPU
); /* this node has a cpu */
2667 for_each_zone(zone
) {
2669 if (!populated_zone(zone
))
2672 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
2674 if (!zone_pcp(zone
, cpu
))
2677 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
2679 if (percpu_pagelist_fraction
)
2680 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
2681 (zone
->present_pages
/ percpu_pagelist_fraction
));
2686 for_each_zone(dzone
) {
2687 if (!populated_zone(dzone
))
2691 kfree(zone_pcp(dzone
, cpu
));
2692 zone_pcp(dzone
, cpu
) = NULL
;
2697 static inline void free_zone_pagesets(int cpu
)
2701 for_each_zone(zone
) {
2702 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
2704 /* Free per_cpu_pageset if it is slab allocated */
2705 if (pset
!= &boot_pageset
[cpu
])
2707 zone_pcp(zone
, cpu
) = NULL
;
2711 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
2712 unsigned long action
,
2715 int cpu
= (long)hcpu
;
2716 int ret
= NOTIFY_OK
;
2719 case CPU_UP_PREPARE
:
2720 case CPU_UP_PREPARE_FROZEN
:
2721 if (process_zones(cpu
))
2724 case CPU_UP_CANCELED
:
2725 case CPU_UP_CANCELED_FROZEN
:
2727 case CPU_DEAD_FROZEN
:
2728 free_zone_pagesets(cpu
);
2736 static struct notifier_block __cpuinitdata pageset_notifier
=
2737 { &pageset_cpuup_callback
, NULL
, 0 };
2739 void __init
setup_per_cpu_pageset(void)
2743 /* Initialize per_cpu_pageset for cpu 0.
2744 * A cpuup callback will do this for every cpu
2745 * as it comes online
2747 err
= process_zones(smp_processor_id());
2749 register_cpu_notifier(&pageset_notifier
);
2754 static noinline __init_refok
2755 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
2758 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2762 * The per-page waitqueue mechanism uses hashed waitqueues
2765 zone
->wait_table_hash_nr_entries
=
2766 wait_table_hash_nr_entries(zone_size_pages
);
2767 zone
->wait_table_bits
=
2768 wait_table_bits(zone
->wait_table_hash_nr_entries
);
2769 alloc_size
= zone
->wait_table_hash_nr_entries
2770 * sizeof(wait_queue_head_t
);
2772 if (system_state
== SYSTEM_BOOTING
) {
2773 zone
->wait_table
= (wait_queue_head_t
*)
2774 alloc_bootmem_node(pgdat
, alloc_size
);
2777 * This case means that a zone whose size was 0 gets new memory
2778 * via memory hot-add.
2779 * But it may be the case that a new node was hot-added. In
2780 * this case vmalloc() will not be able to use this new node's
2781 * memory - this wait_table must be initialized to use this new
2782 * node itself as well.
2783 * To use this new node's memory, further consideration will be
2786 zone
->wait_table
= vmalloc(alloc_size
);
2788 if (!zone
->wait_table
)
2791 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
2792 init_waitqueue_head(zone
->wait_table
+ i
);
2797 static __meminit
void zone_pcp_init(struct zone
*zone
)
2800 unsigned long batch
= zone_batchsize(zone
);
2802 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
2804 /* Early boot. Slab allocator not functional yet */
2805 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
2806 setup_pageset(&boot_pageset
[cpu
],0);
2808 setup_pageset(zone_pcp(zone
,cpu
), batch
);
2811 if (zone
->present_pages
)
2812 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
2813 zone
->name
, zone
->present_pages
, batch
);
2816 __meminit
int init_currently_empty_zone(struct zone
*zone
,
2817 unsigned long zone_start_pfn
,
2819 enum memmap_context context
)
2821 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2823 ret
= zone_wait_table_init(zone
, size
);
2826 pgdat
->nr_zones
= zone_idx(zone
) + 1;
2828 zone
->zone_start_pfn
= zone_start_pfn
;
2830 memmap_init(size
, pgdat
->node_id
, zone_idx(zone
), zone_start_pfn
);
2832 zone_init_free_lists(pgdat
, zone
, zone
->spanned_pages
);
2837 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2839 * Basic iterator support. Return the first range of PFNs for a node
2840 * Note: nid == MAX_NUMNODES returns first region regardless of node
2842 static int __meminit
first_active_region_index_in_nid(int nid
)
2846 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2847 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
2854 * Basic iterator support. Return the next active range of PFNs for a node
2855 * Note: nid == MAX_NUMNODES returns next region regardles of node
2857 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
2859 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
2860 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
2866 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2868 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2869 * Architectures may implement their own version but if add_active_range()
2870 * was used and there are no special requirements, this is a convenient
2873 int __meminit
early_pfn_to_nid(unsigned long pfn
)
2877 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2878 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
2879 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2881 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
2882 return early_node_map
[i
].nid
;
2887 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2889 /* Basic iterator support to walk early_node_map[] */
2890 #define for_each_active_range_index_in_nid(i, nid) \
2891 for (i = first_active_region_index_in_nid(nid); i != -1; \
2892 i = next_active_region_index_in_nid(i, nid))
2895 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2896 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2897 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2899 * If an architecture guarantees that all ranges registered with
2900 * add_active_ranges() contain no holes and may be freed, this
2901 * this function may be used instead of calling free_bootmem() manually.
2903 void __init
free_bootmem_with_active_regions(int nid
,
2904 unsigned long max_low_pfn
)
2908 for_each_active_range_index_in_nid(i
, nid
) {
2909 unsigned long size_pages
= 0;
2910 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2912 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
2915 if (end_pfn
> max_low_pfn
)
2916 end_pfn
= max_low_pfn
;
2918 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
2919 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
2920 PFN_PHYS(early_node_map
[i
].start_pfn
),
2921 size_pages
<< PAGE_SHIFT
);
2926 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2927 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
2929 * If an architecture guarantees that all ranges registered with
2930 * add_active_ranges() contain no holes and may be freed, this
2931 * function may be used instead of calling memory_present() manually.
2933 void __init
sparse_memory_present_with_active_regions(int nid
)
2937 for_each_active_range_index_in_nid(i
, nid
)
2938 memory_present(early_node_map
[i
].nid
,
2939 early_node_map
[i
].start_pfn
,
2940 early_node_map
[i
].end_pfn
);
2944 * push_node_boundaries - Push node boundaries to at least the requested boundary
2945 * @nid: The nid of the node to push the boundary for
2946 * @start_pfn: The start pfn of the node
2947 * @end_pfn: The end pfn of the node
2949 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2950 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2951 * be hotplugged even though no physical memory exists. This function allows
2952 * an arch to push out the node boundaries so mem_map is allocated that can
2955 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2956 void __init
push_node_boundaries(unsigned int nid
,
2957 unsigned long start_pfn
, unsigned long end_pfn
)
2959 printk(KERN_DEBUG
"Entering push_node_boundaries(%u, %lu, %lu)\n",
2960 nid
, start_pfn
, end_pfn
);
2962 /* Initialise the boundary for this node if necessary */
2963 if (node_boundary_end_pfn
[nid
] == 0)
2964 node_boundary_start_pfn
[nid
] = -1UL;
2966 /* Update the boundaries */
2967 if (node_boundary_start_pfn
[nid
] > start_pfn
)
2968 node_boundary_start_pfn
[nid
] = start_pfn
;
2969 if (node_boundary_end_pfn
[nid
] < end_pfn
)
2970 node_boundary_end_pfn
[nid
] = end_pfn
;
2973 /* If necessary, push the node boundary out for reserve hotadd */
2974 static void __meminit
account_node_boundary(unsigned int nid
,
2975 unsigned long *start_pfn
, unsigned long *end_pfn
)
2977 printk(KERN_DEBUG
"Entering account_node_boundary(%u, %lu, %lu)\n",
2978 nid
, *start_pfn
, *end_pfn
);
2980 /* Return if boundary information has not been provided */
2981 if (node_boundary_end_pfn
[nid
] == 0)
2984 /* Check the boundaries and update if necessary */
2985 if (node_boundary_start_pfn
[nid
] < *start_pfn
)
2986 *start_pfn
= node_boundary_start_pfn
[nid
];
2987 if (node_boundary_end_pfn
[nid
] > *end_pfn
)
2988 *end_pfn
= node_boundary_end_pfn
[nid
];
2991 void __init
push_node_boundaries(unsigned int nid
,
2992 unsigned long start_pfn
, unsigned long end_pfn
) {}
2994 static void __meminit
account_node_boundary(unsigned int nid
,
2995 unsigned long *start_pfn
, unsigned long *end_pfn
) {}
3000 * get_pfn_range_for_nid - Return the start and end page frames for a node
3001 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3002 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3003 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3005 * It returns the start and end page frame of a node based on information
3006 * provided by an arch calling add_active_range(). If called for a node
3007 * with no available memory, a warning is printed and the start and end
3010 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3011 unsigned long *start_pfn
, unsigned long *end_pfn
)
3017 for_each_active_range_index_in_nid(i
, nid
) {
3018 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3019 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3022 if (*start_pfn
== -1UL)
3025 /* Push the node boundaries out if requested */
3026 account_node_boundary(nid
, start_pfn
, end_pfn
);
3030 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3031 * assumption is made that zones within a node are ordered in monotonic
3032 * increasing memory addresses so that the "highest" populated zone is used
3034 void __init
find_usable_zone_for_movable(void)
3037 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3038 if (zone_index
== ZONE_MOVABLE
)
3041 if (arch_zone_highest_possible_pfn
[zone_index
] >
3042 arch_zone_lowest_possible_pfn
[zone_index
])
3046 VM_BUG_ON(zone_index
== -1);
3047 movable_zone
= zone_index
;
3051 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3052 * because it is sized independant of architecture. Unlike the other zones,
3053 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3054 * in each node depending on the size of each node and how evenly kernelcore
3055 * is distributed. This helper function adjusts the zone ranges
3056 * provided by the architecture for a given node by using the end of the
3057 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3058 * zones within a node are in order of monotonic increases memory addresses
3060 void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3061 unsigned long zone_type
,
3062 unsigned long node_start_pfn
,
3063 unsigned long node_end_pfn
,
3064 unsigned long *zone_start_pfn
,
3065 unsigned long *zone_end_pfn
)
3067 /* Only adjust if ZONE_MOVABLE is on this node */
3068 if (zone_movable_pfn
[nid
]) {
3069 /* Size ZONE_MOVABLE */
3070 if (zone_type
== ZONE_MOVABLE
) {
3071 *zone_start_pfn
= zone_movable_pfn
[nid
];
3072 *zone_end_pfn
= min(node_end_pfn
,
3073 arch_zone_highest_possible_pfn
[movable_zone
]);
3075 /* Adjust for ZONE_MOVABLE starting within this range */
3076 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3077 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3078 *zone_end_pfn
= zone_movable_pfn
[nid
];
3080 /* Check if this whole range is within ZONE_MOVABLE */
3081 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3082 *zone_start_pfn
= *zone_end_pfn
;
3087 * Return the number of pages a zone spans in a node, including holes
3088 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3090 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3091 unsigned long zone_type
,
3092 unsigned long *ignored
)
3094 unsigned long node_start_pfn
, node_end_pfn
;
3095 unsigned long zone_start_pfn
, zone_end_pfn
;
3097 /* Get the start and end of the node and zone */
3098 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3099 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3100 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3101 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3102 node_start_pfn
, node_end_pfn
,
3103 &zone_start_pfn
, &zone_end_pfn
);
3105 /* Check that this node has pages within the zone's required range */
3106 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3109 /* Move the zone boundaries inside the node if necessary */
3110 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3111 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3113 /* Return the spanned pages */
3114 return zone_end_pfn
- zone_start_pfn
;
3118 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3119 * then all holes in the requested range will be accounted for.
3121 unsigned long __meminit
__absent_pages_in_range(int nid
,
3122 unsigned long range_start_pfn
,
3123 unsigned long range_end_pfn
)
3126 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3127 unsigned long start_pfn
;
3129 /* Find the end_pfn of the first active range of pfns in the node */
3130 i
= first_active_region_index_in_nid(nid
);
3134 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3136 /* Account for ranges before physical memory on this node */
3137 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3138 hole_pages
= prev_end_pfn
- range_start_pfn
;
3140 /* Find all holes for the zone within the node */
3141 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3143 /* No need to continue if prev_end_pfn is outside the zone */
3144 if (prev_end_pfn
>= range_end_pfn
)
3147 /* Make sure the end of the zone is not within the hole */
3148 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3149 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3151 /* Update the hole size cound and move on */
3152 if (start_pfn
> range_start_pfn
) {
3153 BUG_ON(prev_end_pfn
> start_pfn
);
3154 hole_pages
+= start_pfn
- prev_end_pfn
;
3156 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3159 /* Account for ranges past physical memory on this node */
3160 if (range_end_pfn
> prev_end_pfn
)
3161 hole_pages
+= range_end_pfn
-
3162 max(range_start_pfn
, prev_end_pfn
);
3168 * absent_pages_in_range - Return number of page frames in holes within a range
3169 * @start_pfn: The start PFN to start searching for holes
3170 * @end_pfn: The end PFN to stop searching for holes
3172 * It returns the number of pages frames in memory holes within a range.
3174 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3175 unsigned long end_pfn
)
3177 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3180 /* Return the number of page frames in holes in a zone on a node */
3181 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3182 unsigned long zone_type
,
3183 unsigned long *ignored
)
3185 unsigned long node_start_pfn
, node_end_pfn
;
3186 unsigned long zone_start_pfn
, zone_end_pfn
;
3188 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3189 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3191 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3194 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3195 node_start_pfn
, node_end_pfn
,
3196 &zone_start_pfn
, &zone_end_pfn
);
3197 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3201 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3202 unsigned long zone_type
,
3203 unsigned long *zones_size
)
3205 return zones_size
[zone_type
];
3208 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3209 unsigned long zone_type
,
3210 unsigned long *zholes_size
)
3215 return zholes_size
[zone_type
];
3220 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3221 unsigned long *zones_size
, unsigned long *zholes_size
)
3223 unsigned long realtotalpages
, totalpages
= 0;
3226 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3227 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3229 pgdat
->node_spanned_pages
= totalpages
;
3231 realtotalpages
= totalpages
;
3232 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3234 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3236 pgdat
->node_present_pages
= realtotalpages
;
3237 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3241 #ifndef CONFIG_SPARSEMEM
3243 * Calculate the size of the zone->blockflags rounded to an unsigned long
3244 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3245 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3246 * round what is now in bits to nearest long in bits, then return it in
3249 static unsigned long __init
usemap_size(unsigned long zonesize
)
3251 unsigned long usemapsize
;
3253 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
3254 usemapsize
= usemapsize
>> pageblock_order
;
3255 usemapsize
*= NR_PAGEBLOCK_BITS
;
3256 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3258 return usemapsize
/ 8;
3261 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3262 struct zone
*zone
, unsigned long zonesize
)
3264 unsigned long usemapsize
= usemap_size(zonesize
);
3265 zone
->pageblock_flags
= NULL
;
3267 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3268 memset(zone
->pageblock_flags
, 0, usemapsize
);
3272 static void inline setup_usemap(struct pglist_data
*pgdat
,
3273 struct zone
*zone
, unsigned long zonesize
) {}
3274 #endif /* CONFIG_SPARSEMEM */
3276 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3277 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3278 static inline void __init
set_pageblock_order(unsigned int order
)
3280 /* Check that pageblock_nr_pages has not already been setup */
3281 if (pageblock_order
)
3285 * Assume the largest contiguous order of interest is a huge page.
3286 * This value may be variable depending on boot parameters on IA64
3288 pageblock_order
= order
;
3290 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3292 /* Defined this way to avoid accidently referencing HUGETLB_PAGE_ORDER */
3293 #define set_pageblock_order(x) do {} while (0)
3295 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3298 * Set up the zone data structures:
3299 * - mark all pages reserved
3300 * - mark all memory queues empty
3301 * - clear the memory bitmaps
3303 static void __meminit
free_area_init_core(struct pglist_data
*pgdat
,
3304 unsigned long *zones_size
, unsigned long *zholes_size
)
3307 int nid
= pgdat
->node_id
;
3308 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
3311 pgdat_resize_init(pgdat
);
3312 pgdat
->nr_zones
= 0;
3313 init_waitqueue_head(&pgdat
->kswapd_wait
);
3314 pgdat
->kswapd_max_order
= 0;
3316 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3317 struct zone
*zone
= pgdat
->node_zones
+ j
;
3318 unsigned long size
, realsize
, memmap_pages
;
3320 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
3321 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
3325 * Adjust realsize so that it accounts for how much memory
3326 * is used by this zone for memmap. This affects the watermark
3327 * and per-cpu initialisations
3329 memmap_pages
= (size
* sizeof(struct page
)) >> PAGE_SHIFT
;
3330 if (realsize
>= memmap_pages
) {
3331 realsize
-= memmap_pages
;
3333 " %s zone: %lu pages used for memmap\n",
3334 zone_names
[j
], memmap_pages
);
3337 " %s zone: %lu pages exceeds realsize %lu\n",
3338 zone_names
[j
], memmap_pages
, realsize
);
3340 /* Account for reserved pages */
3341 if (j
== 0 && realsize
> dma_reserve
) {
3342 realsize
-= dma_reserve
;
3343 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
3344 zone_names
[0], dma_reserve
);
3347 if (!is_highmem_idx(j
))
3348 nr_kernel_pages
+= realsize
;
3349 nr_all_pages
+= realsize
;
3351 zone
->spanned_pages
= size
;
3352 zone
->present_pages
= realsize
;
3355 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3357 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3359 zone
->name
= zone_names
[j
];
3360 spin_lock_init(&zone
->lock
);
3361 spin_lock_init(&zone
->lru_lock
);
3362 zone_seqlock_init(zone
);
3363 zone
->zone_pgdat
= pgdat
;
3365 zone
->prev_priority
= DEF_PRIORITY
;
3367 zone_pcp_init(zone
);
3368 INIT_LIST_HEAD(&zone
->active_list
);
3369 INIT_LIST_HEAD(&zone
->inactive_list
);
3370 zone
->nr_scan_active
= 0;
3371 zone
->nr_scan_inactive
= 0;
3372 zap_zone_vm_stats(zone
);
3373 atomic_set(&zone
->reclaim_in_progress
, 0);
3377 set_pageblock_order(HUGETLB_PAGE_ORDER
);
3378 setup_usemap(pgdat
, zone
, size
);
3379 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3380 size
, MEMMAP_EARLY
);
3382 zone_start_pfn
+= size
;
3386 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3388 /* Skip empty nodes */
3389 if (!pgdat
->node_spanned_pages
)
3392 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3393 /* ia64 gets its own node_mem_map, before this, without bootmem */
3394 if (!pgdat
->node_mem_map
) {
3395 unsigned long size
, start
, end
;
3399 * The zone's endpoints aren't required to be MAX_ORDER
3400 * aligned but the node_mem_map endpoints must be in order
3401 * for the buddy allocator to function correctly.
3403 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3404 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3405 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3406 size
= (end
- start
) * sizeof(struct page
);
3407 map
= alloc_remap(pgdat
->node_id
, size
);
3409 map
= alloc_bootmem_node(pgdat
, size
);
3410 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3412 #ifndef CONFIG_NEED_MULTIPLE_NODES
3414 * With no DISCONTIG, the global mem_map is just set as node 0's
3416 if (pgdat
== NODE_DATA(0)) {
3417 mem_map
= NODE_DATA(0)->node_mem_map
;
3418 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3419 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3420 mem_map
-= pgdat
->node_start_pfn
;
3421 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3424 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3427 void __meminit
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
3428 unsigned long *zones_size
, unsigned long node_start_pfn
,
3429 unsigned long *zholes_size
)
3431 pgdat
->node_id
= nid
;
3432 pgdat
->node_start_pfn
= node_start_pfn
;
3433 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3435 alloc_node_mem_map(pgdat
);
3437 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3440 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3442 #if MAX_NUMNODES > 1
3444 * Figure out the number of possible node ids.
3446 static void __init
setup_nr_node_ids(void)
3449 unsigned int highest
= 0;
3451 for_each_node_mask(node
, node_possible_map
)
3453 nr_node_ids
= highest
+ 1;
3456 static inline void setup_nr_node_ids(void)
3462 * add_active_range - Register a range of PFNs backed by physical memory
3463 * @nid: The node ID the range resides on
3464 * @start_pfn: The start PFN of the available physical memory
3465 * @end_pfn: The end PFN of the available physical memory
3467 * These ranges are stored in an early_node_map[] and later used by
3468 * free_area_init_nodes() to calculate zone sizes and holes. If the
3469 * range spans a memory hole, it is up to the architecture to ensure
3470 * the memory is not freed by the bootmem allocator. If possible
3471 * the range being registered will be merged with existing ranges.
3473 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3474 unsigned long end_pfn
)
3478 printk(KERN_DEBUG
"Entering add_active_range(%d, %lu, %lu) "
3479 "%d entries of %d used\n",
3480 nid
, start_pfn
, end_pfn
,
3481 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3483 /* Merge with existing active regions if possible */
3484 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3485 if (early_node_map
[i
].nid
!= nid
)
3488 /* Skip if an existing region covers this new one */
3489 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3490 end_pfn
<= early_node_map
[i
].end_pfn
)
3493 /* Merge forward if suitable */
3494 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3495 end_pfn
> early_node_map
[i
].end_pfn
) {
3496 early_node_map
[i
].end_pfn
= end_pfn
;
3500 /* Merge backward if suitable */
3501 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3502 end_pfn
>= early_node_map
[i
].start_pfn
) {
3503 early_node_map
[i
].start_pfn
= start_pfn
;
3508 /* Check that early_node_map is large enough */
3509 if (i
>= MAX_ACTIVE_REGIONS
) {
3510 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3511 MAX_ACTIVE_REGIONS
);
3515 early_node_map
[i
].nid
= nid
;
3516 early_node_map
[i
].start_pfn
= start_pfn
;
3517 early_node_map
[i
].end_pfn
= end_pfn
;
3518 nr_nodemap_entries
= i
+ 1;
3522 * shrink_active_range - Shrink an existing registered range of PFNs
3523 * @nid: The node id the range is on that should be shrunk
3524 * @old_end_pfn: The old end PFN of the range
3525 * @new_end_pfn: The new PFN of the range
3527 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3528 * The map is kept at the end physical page range that has already been
3529 * registered with add_active_range(). This function allows an arch to shrink
3530 * an existing registered range.
3532 void __init
shrink_active_range(unsigned int nid
, unsigned long old_end_pfn
,
3533 unsigned long new_end_pfn
)
3537 /* Find the old active region end and shrink */
3538 for_each_active_range_index_in_nid(i
, nid
)
3539 if (early_node_map
[i
].end_pfn
== old_end_pfn
) {
3540 early_node_map
[i
].end_pfn
= new_end_pfn
;
3546 * remove_all_active_ranges - Remove all currently registered regions
3548 * During discovery, it may be found that a table like SRAT is invalid
3549 * and an alternative discovery method must be used. This function removes
3550 * all currently registered regions.
3552 void __init
remove_all_active_ranges(void)
3554 memset(early_node_map
, 0, sizeof(early_node_map
));
3555 nr_nodemap_entries
= 0;
3556 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3557 memset(node_boundary_start_pfn
, 0, sizeof(node_boundary_start_pfn
));
3558 memset(node_boundary_end_pfn
, 0, sizeof(node_boundary_end_pfn
));
3559 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3562 /* Compare two active node_active_regions */
3563 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3565 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3566 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3568 /* Done this way to avoid overflows */
3569 if (arange
->start_pfn
> brange
->start_pfn
)
3571 if (arange
->start_pfn
< brange
->start_pfn
)
3577 /* sort the node_map by start_pfn */
3578 static void __init
sort_node_map(void)
3580 sort(early_node_map
, (size_t)nr_nodemap_entries
,
3581 sizeof(struct node_active_region
),
3582 cmp_node_active_region
, NULL
);
3585 /* Find the lowest pfn for a node */
3586 unsigned long __init
find_min_pfn_for_node(unsigned long nid
)
3589 unsigned long min_pfn
= ULONG_MAX
;
3591 /* Assuming a sorted map, the first range found has the starting pfn */
3592 for_each_active_range_index_in_nid(i
, nid
)
3593 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
3595 if (min_pfn
== ULONG_MAX
) {
3597 "Could not find start_pfn for node %lu\n", nid
);
3605 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3607 * It returns the minimum PFN based on information provided via
3608 * add_active_range().
3610 unsigned long __init
find_min_pfn_with_active_regions(void)
3612 return find_min_pfn_for_node(MAX_NUMNODES
);
3616 * find_max_pfn_with_active_regions - Find the maximum PFN registered
3618 * It returns the maximum PFN based on information provided via
3619 * add_active_range().
3621 unsigned long __init
find_max_pfn_with_active_regions(void)
3624 unsigned long max_pfn
= 0;
3626 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3627 max_pfn
= max(max_pfn
, early_node_map
[i
].end_pfn
);
3633 * early_calculate_totalpages()
3634 * Sum pages in active regions for movable zone.
3635 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3637 static unsigned long __init
early_calculate_totalpages(void)
3640 unsigned long totalpages
= 0;
3642 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3643 unsigned long pages
= early_node_map
[i
].end_pfn
-
3644 early_node_map
[i
].start_pfn
;
3645 totalpages
+= pages
;
3647 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
3653 * Find the PFN the Movable zone begins in each node. Kernel memory
3654 * is spread evenly between nodes as long as the nodes have enough
3655 * memory. When they don't, some nodes will have more kernelcore than
3658 void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
3661 unsigned long usable_startpfn
;
3662 unsigned long kernelcore_node
, kernelcore_remaining
;
3663 unsigned long totalpages
= early_calculate_totalpages();
3664 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
3667 * If movablecore was specified, calculate what size of
3668 * kernelcore that corresponds so that memory usable for
3669 * any allocation type is evenly spread. If both kernelcore
3670 * and movablecore are specified, then the value of kernelcore
3671 * will be used for required_kernelcore if it's greater than
3672 * what movablecore would have allowed.
3674 if (required_movablecore
) {
3675 unsigned long corepages
;
3678 * Round-up so that ZONE_MOVABLE is at least as large as what
3679 * was requested by the user
3681 required_movablecore
=
3682 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
3683 corepages
= totalpages
- required_movablecore
;
3685 required_kernelcore
= max(required_kernelcore
, corepages
);
3688 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3689 if (!required_kernelcore
)
3692 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3693 find_usable_zone_for_movable();
3694 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
3697 /* Spread kernelcore memory as evenly as possible throughout nodes */
3698 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3699 for_each_node_state(nid
, N_HIGH_MEMORY
) {
3701 * Recalculate kernelcore_node if the division per node
3702 * now exceeds what is necessary to satisfy the requested
3703 * amount of memory for the kernel
3705 if (required_kernelcore
< kernelcore_node
)
3706 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3709 * As the map is walked, we track how much memory is usable
3710 * by the kernel using kernelcore_remaining. When it is
3711 * 0, the rest of the node is usable by ZONE_MOVABLE
3713 kernelcore_remaining
= kernelcore_node
;
3715 /* Go through each range of PFNs within this node */
3716 for_each_active_range_index_in_nid(i
, nid
) {
3717 unsigned long start_pfn
, end_pfn
;
3718 unsigned long size_pages
;
3720 start_pfn
= max(early_node_map
[i
].start_pfn
,
3721 zone_movable_pfn
[nid
]);
3722 end_pfn
= early_node_map
[i
].end_pfn
;
3723 if (start_pfn
>= end_pfn
)
3726 /* Account for what is only usable for kernelcore */
3727 if (start_pfn
< usable_startpfn
) {
3728 unsigned long kernel_pages
;
3729 kernel_pages
= min(end_pfn
, usable_startpfn
)
3732 kernelcore_remaining
-= min(kernel_pages
,
3733 kernelcore_remaining
);
3734 required_kernelcore
-= min(kernel_pages
,
3735 required_kernelcore
);
3737 /* Continue if range is now fully accounted */
3738 if (end_pfn
<= usable_startpfn
) {
3741 * Push zone_movable_pfn to the end so
3742 * that if we have to rebalance
3743 * kernelcore across nodes, we will
3744 * not double account here
3746 zone_movable_pfn
[nid
] = end_pfn
;
3749 start_pfn
= usable_startpfn
;
3753 * The usable PFN range for ZONE_MOVABLE is from
3754 * start_pfn->end_pfn. Calculate size_pages as the
3755 * number of pages used as kernelcore
3757 size_pages
= end_pfn
- start_pfn
;
3758 if (size_pages
> kernelcore_remaining
)
3759 size_pages
= kernelcore_remaining
;
3760 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
3763 * Some kernelcore has been met, update counts and
3764 * break if the kernelcore for this node has been
3767 required_kernelcore
-= min(required_kernelcore
,
3769 kernelcore_remaining
-= size_pages
;
3770 if (!kernelcore_remaining
)
3776 * If there is still required_kernelcore, we do another pass with one
3777 * less node in the count. This will push zone_movable_pfn[nid] further
3778 * along on the nodes that still have memory until kernelcore is
3782 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
3785 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3786 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
3787 zone_movable_pfn
[nid
] =
3788 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
3791 /* Any regular memory on that node ? */
3792 static void check_for_regular_memory(pg_data_t
*pgdat
)
3794 #ifdef CONFIG_HIGHMEM
3795 enum zone_type zone_type
;
3797 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
3798 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
3799 if (zone
->present_pages
)
3800 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
3806 * free_area_init_nodes - Initialise all pg_data_t and zone data
3807 * @max_zone_pfn: an array of max PFNs for each zone
3809 * This will call free_area_init_node() for each active node in the system.
3810 * Using the page ranges provided by add_active_range(), the size of each
3811 * zone in each node and their holes is calculated. If the maximum PFN
3812 * between two adjacent zones match, it is assumed that the zone is empty.
3813 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3814 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3815 * starts where the previous one ended. For example, ZONE_DMA32 starts
3816 * at arch_max_dma_pfn.
3818 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
3823 /* Sort early_node_map as initialisation assumes it is sorted */
3826 /* Record where the zone boundaries are */
3827 memset(arch_zone_lowest_possible_pfn
, 0,
3828 sizeof(arch_zone_lowest_possible_pfn
));
3829 memset(arch_zone_highest_possible_pfn
, 0,
3830 sizeof(arch_zone_highest_possible_pfn
));
3831 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
3832 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
3833 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
3834 if (i
== ZONE_MOVABLE
)
3836 arch_zone_lowest_possible_pfn
[i
] =
3837 arch_zone_highest_possible_pfn
[i
-1];
3838 arch_zone_highest_possible_pfn
[i
] =
3839 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
3841 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
3842 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
3844 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3845 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
3846 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
3848 /* Print out the zone ranges */
3849 printk("Zone PFN ranges:\n");
3850 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
3851 if (i
== ZONE_MOVABLE
)
3853 printk(" %-8s %8lu -> %8lu\n",
3855 arch_zone_lowest_possible_pfn
[i
],
3856 arch_zone_highest_possible_pfn
[i
]);
3859 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
3860 printk("Movable zone start PFN for each node\n");
3861 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
3862 if (zone_movable_pfn
[i
])
3863 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
3866 /* Print out the early_node_map[] */
3867 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
3868 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3869 printk(" %3d: %8lu -> %8lu\n", early_node_map
[i
].nid
,
3870 early_node_map
[i
].start_pfn
,
3871 early_node_map
[i
].end_pfn
);
3873 /* Initialise every node */
3874 setup_nr_node_ids();
3875 for_each_online_node(nid
) {
3876 pg_data_t
*pgdat
= NODE_DATA(nid
);
3877 free_area_init_node(nid
, pgdat
, NULL
,
3878 find_min_pfn_for_node(nid
), NULL
);
3880 /* Any memory on that node */
3881 if (pgdat
->node_present_pages
)
3882 node_set_state(nid
, N_HIGH_MEMORY
);
3883 check_for_regular_memory(pgdat
);
3887 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
3889 unsigned long long coremem
;
3893 coremem
= memparse(p
, &p
);
3894 *core
= coremem
>> PAGE_SHIFT
;
3896 /* Paranoid check that UL is enough for the coremem value */
3897 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
3903 * kernelcore=size sets the amount of memory for use for allocations that
3904 * cannot be reclaimed or migrated.
3906 static int __init
cmdline_parse_kernelcore(char *p
)
3908 return cmdline_parse_core(p
, &required_kernelcore
);
3912 * movablecore=size sets the amount of memory for use for allocations that
3913 * can be reclaimed or migrated.
3915 static int __init
cmdline_parse_movablecore(char *p
)
3917 return cmdline_parse_core(p
, &required_movablecore
);
3920 early_param("kernelcore", cmdline_parse_kernelcore
);
3921 early_param("movablecore", cmdline_parse_movablecore
);
3923 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3926 * set_dma_reserve - set the specified number of pages reserved in the first zone
3927 * @new_dma_reserve: The number of pages to mark reserved
3929 * The per-cpu batchsize and zone watermarks are determined by present_pages.
3930 * In the DMA zone, a significant percentage may be consumed by kernel image
3931 * and other unfreeable allocations which can skew the watermarks badly. This
3932 * function may optionally be used to account for unfreeable pages in the
3933 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
3934 * smaller per-cpu batchsize.
3936 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
3938 dma_reserve
= new_dma_reserve
;
3941 #ifndef CONFIG_NEED_MULTIPLE_NODES
3942 static bootmem_data_t contig_bootmem_data
;
3943 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
3945 EXPORT_SYMBOL(contig_page_data
);
3948 void __init
free_area_init(unsigned long *zones_size
)
3950 free_area_init_node(0, NODE_DATA(0), zones_size
,
3951 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
3954 static int page_alloc_cpu_notify(struct notifier_block
*self
,
3955 unsigned long action
, void *hcpu
)
3957 int cpu
= (unsigned long)hcpu
;
3959 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
3960 local_irq_disable();
3962 vm_events_fold_cpu(cpu
);
3964 refresh_cpu_vm_stats(cpu
);
3969 void __init
page_alloc_init(void)
3971 hotcpu_notifier(page_alloc_cpu_notify
, 0);
3975 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
3976 * or min_free_kbytes changes.
3978 static void calculate_totalreserve_pages(void)
3980 struct pglist_data
*pgdat
;
3981 unsigned long reserve_pages
= 0;
3982 enum zone_type i
, j
;
3984 for_each_online_pgdat(pgdat
) {
3985 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
3986 struct zone
*zone
= pgdat
->node_zones
+ i
;
3987 unsigned long max
= 0;
3989 /* Find valid and maximum lowmem_reserve in the zone */
3990 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
3991 if (zone
->lowmem_reserve
[j
] > max
)
3992 max
= zone
->lowmem_reserve
[j
];
3995 /* we treat pages_high as reserved pages. */
3996 max
+= zone
->pages_high
;
3998 if (max
> zone
->present_pages
)
3999 max
= zone
->present_pages
;
4000 reserve_pages
+= max
;
4003 totalreserve_pages
= reserve_pages
;
4007 * setup_per_zone_lowmem_reserve - called whenever
4008 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4009 * has a correct pages reserved value, so an adequate number of
4010 * pages are left in the zone after a successful __alloc_pages().
4012 static void setup_per_zone_lowmem_reserve(void)
4014 struct pglist_data
*pgdat
;
4015 enum zone_type j
, idx
;
4017 for_each_online_pgdat(pgdat
) {
4018 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4019 struct zone
*zone
= pgdat
->node_zones
+ j
;
4020 unsigned long present_pages
= zone
->present_pages
;
4022 zone
->lowmem_reserve
[j
] = 0;
4026 struct zone
*lower_zone
;
4030 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4031 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4033 lower_zone
= pgdat
->node_zones
+ idx
;
4034 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4035 sysctl_lowmem_reserve_ratio
[idx
];
4036 present_pages
+= lower_zone
->present_pages
;
4041 /* update totalreserve_pages */
4042 calculate_totalreserve_pages();
4046 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4048 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4049 * with respect to min_free_kbytes.
4051 void setup_per_zone_pages_min(void)
4053 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4054 unsigned long lowmem_pages
= 0;
4056 unsigned long flags
;
4058 /* Calculate total number of !ZONE_HIGHMEM pages */
4059 for_each_zone(zone
) {
4060 if (!is_highmem(zone
))
4061 lowmem_pages
+= zone
->present_pages
;
4064 for_each_zone(zone
) {
4067 spin_lock_irqsave(&zone
->lru_lock
, flags
);
4068 tmp
= (u64
)pages_min
* zone
->present_pages
;
4069 do_div(tmp
, lowmem_pages
);
4070 if (is_highmem(zone
)) {
4072 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4073 * need highmem pages, so cap pages_min to a small
4076 * The (pages_high-pages_low) and (pages_low-pages_min)
4077 * deltas controls asynch page reclaim, and so should
4078 * not be capped for highmem.
4082 min_pages
= zone
->present_pages
/ 1024;
4083 if (min_pages
< SWAP_CLUSTER_MAX
)
4084 min_pages
= SWAP_CLUSTER_MAX
;
4085 if (min_pages
> 128)
4087 zone
->pages_min
= min_pages
;
4090 * If it's a lowmem zone, reserve a number of pages
4091 * proportionate to the zone's size.
4093 zone
->pages_min
= tmp
;
4096 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
4097 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
4098 setup_zone_migrate_reserve(zone
);
4099 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
4102 /* update totalreserve_pages */
4103 calculate_totalreserve_pages();
4107 * Initialise min_free_kbytes.
4109 * For small machines we want it small (128k min). For large machines
4110 * we want it large (64MB max). But it is not linear, because network
4111 * bandwidth does not increase linearly with machine size. We use
4113 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4114 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4130 static int __init
init_per_zone_pages_min(void)
4132 unsigned long lowmem_kbytes
;
4134 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4136 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4137 if (min_free_kbytes
< 128)
4138 min_free_kbytes
= 128;
4139 if (min_free_kbytes
> 65536)
4140 min_free_kbytes
= 65536;
4141 setup_per_zone_pages_min();
4142 setup_per_zone_lowmem_reserve();
4145 module_init(init_per_zone_pages_min
)
4148 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4149 * that we can call two helper functions whenever min_free_kbytes
4152 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4153 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4155 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
4157 setup_per_zone_pages_min();
4162 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4163 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4168 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4173 zone
->min_unmapped_pages
= (zone
->present_pages
*
4174 sysctl_min_unmapped_ratio
) / 100;
4178 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4179 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4184 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4189 zone
->min_slab_pages
= (zone
->present_pages
*
4190 sysctl_min_slab_ratio
) / 100;
4196 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4197 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4198 * whenever sysctl_lowmem_reserve_ratio changes.
4200 * The reserve ratio obviously has absolutely no relation with the
4201 * pages_min watermarks. The lowmem reserve ratio can only make sense
4202 * if in function of the boot time zone sizes.
4204 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4205 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4207 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4208 setup_per_zone_lowmem_reserve();
4213 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4214 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4215 * can have before it gets flushed back to buddy allocator.
4218 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4219 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4225 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4226 if (!write
|| (ret
== -EINVAL
))
4228 for_each_zone(zone
) {
4229 for_each_online_cpu(cpu
) {
4231 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4232 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
4238 int hashdist
= HASHDIST_DEFAULT
;
4241 static int __init
set_hashdist(char *str
)
4245 hashdist
= simple_strtoul(str
, &str
, 0);
4248 __setup("hashdist=", set_hashdist
);
4252 * allocate a large system hash table from bootmem
4253 * - it is assumed that the hash table must contain an exact power-of-2
4254 * quantity of entries
4255 * - limit is the number of hash buckets, not the total allocation size
4257 void *__init
alloc_large_system_hash(const char *tablename
,
4258 unsigned long bucketsize
,
4259 unsigned long numentries
,
4262 unsigned int *_hash_shift
,
4263 unsigned int *_hash_mask
,
4264 unsigned long limit
)
4266 unsigned long long max
= limit
;
4267 unsigned long log2qty
, size
;
4270 /* allow the kernel cmdline to have a say */
4272 /* round applicable memory size up to nearest megabyte */
4273 numentries
= nr_kernel_pages
;
4274 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4275 numentries
>>= 20 - PAGE_SHIFT
;
4276 numentries
<<= 20 - PAGE_SHIFT
;
4278 /* limit to 1 bucket per 2^scale bytes of low memory */
4279 if (scale
> PAGE_SHIFT
)
4280 numentries
>>= (scale
- PAGE_SHIFT
);
4282 numentries
<<= (PAGE_SHIFT
- scale
);
4284 /* Make sure we've got at least a 0-order allocation.. */
4285 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4286 numentries
= PAGE_SIZE
/ bucketsize
;
4288 numentries
= roundup_pow_of_two(numentries
);
4290 /* limit allocation size to 1/16 total memory by default */
4292 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4293 do_div(max
, bucketsize
);
4296 if (numentries
> max
)
4299 log2qty
= ilog2(numentries
);
4302 size
= bucketsize
<< log2qty
;
4303 if (flags
& HASH_EARLY
)
4304 table
= alloc_bootmem(size
);
4306 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4308 unsigned long order
;
4309 for (order
= 0; ((1UL << order
) << PAGE_SHIFT
) < size
; order
++)
4311 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
4313 * If bucketsize is not a power-of-two, we may free
4314 * some pages at the end of hash table.
4317 unsigned long alloc_end
= (unsigned long)table
+
4318 (PAGE_SIZE
<< order
);
4319 unsigned long used
= (unsigned long)table
+
4321 split_page(virt_to_page(table
), order
);
4322 while (used
< alloc_end
) {
4328 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4331 panic("Failed to allocate %s hash table\n", tablename
);
4333 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4336 ilog2(size
) - PAGE_SHIFT
,
4340 *_hash_shift
= log2qty
;
4342 *_hash_mask
= (1 << log2qty
) - 1;
4347 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4348 struct page
*pfn_to_page(unsigned long pfn
)
4350 return __pfn_to_page(pfn
);
4352 unsigned long page_to_pfn(struct page
*page
)
4354 return __page_to_pfn(page
);
4356 EXPORT_SYMBOL(pfn_to_page
);
4357 EXPORT_SYMBOL(page_to_pfn
);
4358 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4360 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4361 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4364 #ifdef CONFIG_SPARSEMEM
4365 return __pfn_to_section(pfn
)->pageblock_flags
;
4367 return zone
->pageblock_flags
;
4368 #endif /* CONFIG_SPARSEMEM */
4371 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4373 #ifdef CONFIG_SPARSEMEM
4374 pfn
&= (PAGES_PER_SECTION
-1);
4375 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4377 pfn
= pfn
- zone
->zone_start_pfn
;
4378 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4379 #endif /* CONFIG_SPARSEMEM */
4383 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4384 * @page: The page within the block of interest
4385 * @start_bitidx: The first bit of interest to retrieve
4386 * @end_bitidx: The last bit of interest
4387 * returns pageblock_bits flags
4389 unsigned long get_pageblock_flags_group(struct page
*page
,
4390 int start_bitidx
, int end_bitidx
)
4393 unsigned long *bitmap
;
4394 unsigned long pfn
, bitidx
;
4395 unsigned long flags
= 0;
4396 unsigned long value
= 1;
4398 zone
= page_zone(page
);
4399 pfn
= page_to_pfn(page
);
4400 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4401 bitidx
= pfn_to_bitidx(zone
, pfn
);
4403 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4404 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
4411 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4412 * @page: The page within the block of interest
4413 * @start_bitidx: The first bit of interest
4414 * @end_bitidx: The last bit of interest
4415 * @flags: The flags to set
4417 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
4418 int start_bitidx
, int end_bitidx
)
4421 unsigned long *bitmap
;
4422 unsigned long pfn
, bitidx
;
4423 unsigned long value
= 1;
4425 zone
= page_zone(page
);
4426 pfn
= page_to_pfn(page
);
4427 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4428 bitidx
= pfn_to_bitidx(zone
, pfn
);
4430 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4432 __set_bit(bitidx
+ start_bitidx
, bitmap
);
4434 __clear_bit(bitidx
+ start_bitidx
, bitmap
);