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/oom.h>
31 #include <linux/notifier.h>
32 #include <linux/topology.h>
33 #include <linux/sysctl.h>
34 #include <linux/cpu.h>
35 #include <linux/cpuset.h>
36 #include <linux/memory_hotplug.h>
37 #include <linux/nodemask.h>
38 #include <linux/vmalloc.h>
39 #include <linux/mempolicy.h>
40 #include <linux/stop_machine.h>
41 #include <linux/sort.h>
42 #include <linux/pfn.h>
43 #include <linux/backing-dev.h>
44 #include <linux/fault-inject.h>
45 #include <linux/page-isolation.h>
47 #include <asm/tlbflush.h>
48 #include <asm/div64.h>
52 * Array of node states.
54 nodemask_t node_states
[NR_NODE_STATES
] __read_mostly
= {
55 [N_POSSIBLE
] = NODE_MASK_ALL
,
56 [N_ONLINE
] = { { [0] = 1UL } },
58 [N_NORMAL_MEMORY
] = { { [0] = 1UL } },
60 [N_HIGH_MEMORY
] = { { [0] = 1UL } },
62 [N_CPU
] = { { [0] = 1UL } },
65 EXPORT_SYMBOL(node_states
);
67 unsigned long totalram_pages __read_mostly
;
68 unsigned long totalreserve_pages __read_mostly
;
70 int percpu_pagelist_fraction
;
72 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
73 int pageblock_order __read_mostly
;
76 static void __free_pages_ok(struct page
*page
, unsigned int order
);
79 * results with 256, 32 in the lowmem_reserve sysctl:
80 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
81 * 1G machine -> (16M dma, 784M normal, 224M high)
82 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
83 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
84 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
86 * TBD: should special case ZONE_DMA32 machines here - in those we normally
87 * don't need any ZONE_NORMAL reservation
89 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
90 #ifdef CONFIG_ZONE_DMA
93 #ifdef CONFIG_ZONE_DMA32
102 EXPORT_SYMBOL(totalram_pages
);
104 static char * const zone_names
[MAX_NR_ZONES
] = {
105 #ifdef CONFIG_ZONE_DMA
108 #ifdef CONFIG_ZONE_DMA32
112 #ifdef CONFIG_HIGHMEM
118 int min_free_kbytes
= 1024;
120 unsigned long __meminitdata nr_kernel_pages
;
121 unsigned long __meminitdata nr_all_pages
;
122 static unsigned long __meminitdata dma_reserve
;
124 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
126 * MAX_ACTIVE_REGIONS determines the maxmimum number of distinct
127 * ranges of memory (RAM) that may be registered with add_active_range().
128 * Ranges passed to add_active_range() will be merged if possible
129 * so the number of times add_active_range() can be called is
130 * related to the number of nodes and the number of holes
132 #ifdef CONFIG_MAX_ACTIVE_REGIONS
133 /* Allow an architecture to set MAX_ACTIVE_REGIONS to save memory */
134 #define MAX_ACTIVE_REGIONS CONFIG_MAX_ACTIVE_REGIONS
136 #if MAX_NUMNODES >= 32
137 /* If there can be many nodes, allow up to 50 holes per node */
138 #define MAX_ACTIVE_REGIONS (MAX_NUMNODES*50)
140 /* By default, allow up to 256 distinct regions */
141 #define MAX_ACTIVE_REGIONS 256
145 static struct node_active_region __meminitdata early_node_map
[MAX_ACTIVE_REGIONS
];
146 static int __meminitdata nr_nodemap_entries
;
147 static unsigned long __meminitdata arch_zone_lowest_possible_pfn
[MAX_NR_ZONES
];
148 static unsigned long __meminitdata arch_zone_highest_possible_pfn
[MAX_NR_ZONES
];
149 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
150 static unsigned long __meminitdata node_boundary_start_pfn
[MAX_NUMNODES
];
151 static unsigned long __meminitdata node_boundary_end_pfn
[MAX_NUMNODES
];
152 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
153 unsigned long __initdata required_kernelcore
;
154 static unsigned long __initdata required_movablecore
;
155 unsigned long __meminitdata zone_movable_pfn
[MAX_NUMNODES
];
157 /* movable_zone is the "real" zone pages in ZONE_MOVABLE are taken from */
159 EXPORT_SYMBOL(movable_zone
);
160 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
163 int nr_node_ids __read_mostly
= MAX_NUMNODES
;
164 EXPORT_SYMBOL(nr_node_ids
);
167 int page_group_by_mobility_disabled __read_mostly
;
169 static void set_pageblock_migratetype(struct page
*page
, int migratetype
)
171 set_pageblock_flags_group(page
, (unsigned long)migratetype
,
172 PB_migrate
, PB_migrate_end
);
175 #ifdef CONFIG_DEBUG_VM
176 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
180 unsigned long pfn
= page_to_pfn(page
);
183 seq
= zone_span_seqbegin(zone
);
184 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
186 else if (pfn
< zone
->zone_start_pfn
)
188 } while (zone_span_seqretry(zone
, seq
));
193 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
195 if (!pfn_valid_within(page_to_pfn(page
)))
197 if (zone
!= page_zone(page
))
203 * Temporary debugging check for pages not lying within a given zone.
205 static int bad_range(struct zone
*zone
, struct page
*page
)
207 if (page_outside_zone_boundaries(zone
, page
))
209 if (!page_is_consistent(zone
, page
))
215 static inline int bad_range(struct zone
*zone
, struct page
*page
)
221 static void bad_page(struct page
*page
)
223 printk(KERN_EMERG
"Bad page state in process '%s'\n"
224 KERN_EMERG
"page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
225 KERN_EMERG
"Trying to fix it up, but a reboot is needed\n"
226 KERN_EMERG
"Backtrace:\n",
227 current
->comm
, page
, (int)(2*sizeof(unsigned long)),
228 (unsigned long)page
->flags
, page
->mapping
,
229 page_mapcount(page
), page_count(page
));
231 page
->flags
&= ~(1 << PG_lru
|
241 set_page_count(page
, 0);
242 reset_page_mapcount(page
);
243 page
->mapping
= NULL
;
244 add_taint(TAINT_BAD_PAGE
);
248 * Higher-order pages are called "compound pages". They are structured thusly:
250 * The first PAGE_SIZE page is called the "head page".
252 * The remaining PAGE_SIZE pages are called "tail pages".
254 * All pages have PG_compound set. All pages have their ->private pointing at
255 * the head page (even the head page has this).
257 * The first tail page's ->lru.next holds the address of the compound page's
258 * put_page() function. Its ->lru.prev holds the order of allocation.
259 * This usage means that zero-order pages may not be compound.
262 static void free_compound_page(struct page
*page
)
264 __free_pages_ok(page
, compound_order(page
));
267 static void prep_compound_page(struct page
*page
, unsigned long order
)
270 int nr_pages
= 1 << order
;
272 set_compound_page_dtor(page
, free_compound_page
);
273 set_compound_order(page
, order
);
275 for (i
= 1; i
< nr_pages
; i
++) {
276 struct page
*p
= page
+ i
;
279 p
->first_page
= page
;
283 static void destroy_compound_page(struct page
*page
, unsigned long order
)
286 int nr_pages
= 1 << order
;
288 if (unlikely(compound_order(page
) != order
))
291 if (unlikely(!PageHead(page
)))
293 __ClearPageHead(page
);
294 for (i
= 1; i
< nr_pages
; i
++) {
295 struct page
*p
= page
+ i
;
297 if (unlikely(!PageTail(p
) |
298 (p
->first_page
!= page
)))
304 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
308 VM_BUG_ON((gfp_flags
& (__GFP_WAIT
| __GFP_HIGHMEM
)) == __GFP_HIGHMEM
);
310 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
311 * and __GFP_HIGHMEM from hard or soft interrupt context.
313 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
314 for (i
= 0; i
< (1 << order
); i
++)
315 clear_highpage(page
+ i
);
318 static inline void set_page_order(struct page
*page
, int order
)
320 set_page_private(page
, order
);
321 __SetPageBuddy(page
);
324 static inline void rmv_page_order(struct page
*page
)
326 __ClearPageBuddy(page
);
327 set_page_private(page
, 0);
331 * Locate the struct page for both the matching buddy in our
332 * pair (buddy1) and the combined O(n+1) page they form (page).
334 * 1) Any buddy B1 will have an order O twin B2 which satisfies
335 * the following equation:
337 * For example, if the starting buddy (buddy2) is #8 its order
339 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
341 * 2) Any buddy B will have an order O+1 parent P which
342 * satisfies the following equation:
345 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
347 static inline struct page
*
348 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
350 unsigned long buddy_idx
= page_idx
^ (1 << order
);
352 return page
+ (buddy_idx
- page_idx
);
355 static inline unsigned long
356 __find_combined_index(unsigned long page_idx
, unsigned int order
)
358 return (page_idx
& ~(1 << order
));
362 * This function checks whether a page is free && is the buddy
363 * we can do coalesce a page and its buddy if
364 * (a) the buddy is not in a hole &&
365 * (b) the buddy is in the buddy system &&
366 * (c) a page and its buddy have the same order &&
367 * (d) a page and its buddy are in the same zone.
369 * For recording whether a page is in the buddy system, we use PG_buddy.
370 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
372 * For recording page's order, we use page_private(page).
374 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
377 if (!pfn_valid_within(page_to_pfn(buddy
)))
380 if (page_zone_id(page
) != page_zone_id(buddy
))
383 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
384 BUG_ON(page_count(buddy
) != 0);
391 * Freeing function for a buddy system allocator.
393 * The concept of a buddy system is to maintain direct-mapped table
394 * (containing bit values) for memory blocks of various "orders".
395 * The bottom level table contains the map for the smallest allocatable
396 * units of memory (here, pages), and each level above it describes
397 * pairs of units from the levels below, hence, "buddies".
398 * At a high level, all that happens here is marking the table entry
399 * at the bottom level available, and propagating the changes upward
400 * as necessary, plus some accounting needed to play nicely with other
401 * parts of the VM system.
402 * At each level, we keep a list of pages, which are heads of continuous
403 * free pages of length of (1 << order) and marked with PG_buddy. Page's
404 * order is recorded in page_private(page) field.
405 * So when we are allocating or freeing one, we can derive the state of the
406 * other. That is, if we allocate a small block, and both were
407 * free, the remainder of the region must be split into blocks.
408 * If a block is freed, and its buddy is also free, then this
409 * triggers coalescing into a block of larger size.
414 static inline void __free_one_page(struct page
*page
,
415 struct zone
*zone
, unsigned int order
)
417 unsigned long page_idx
;
418 int order_size
= 1 << order
;
419 int migratetype
= get_pageblock_migratetype(page
);
421 if (unlikely(PageCompound(page
)))
422 destroy_compound_page(page
, order
);
424 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
426 VM_BUG_ON(page_idx
& (order_size
- 1));
427 VM_BUG_ON(bad_range(zone
, page
));
429 __mod_zone_page_state(zone
, NR_FREE_PAGES
, order_size
);
430 while (order
< MAX_ORDER
-1) {
431 unsigned long combined_idx
;
434 buddy
= __page_find_buddy(page
, page_idx
, order
);
435 if (!page_is_buddy(page
, buddy
, order
))
436 break; /* Move the buddy up one level. */
438 list_del(&buddy
->lru
);
439 zone
->free_area
[order
].nr_free
--;
440 rmv_page_order(buddy
);
441 combined_idx
= __find_combined_index(page_idx
, order
);
442 page
= page
+ (combined_idx
- page_idx
);
443 page_idx
= combined_idx
;
446 set_page_order(page
, order
);
448 &zone
->free_area
[order
].free_list
[migratetype
]);
449 zone
->free_area
[order
].nr_free
++;
452 static inline int free_pages_check(struct page
*page
)
454 if (unlikely(page_mapcount(page
) |
455 (page
->mapping
!= NULL
) |
456 (page_count(page
) != 0) |
469 __ClearPageDirty(page
);
471 * For now, we report if PG_reserved was found set, but do not
472 * clear it, and do not free the page. But we shall soon need
473 * to do more, for when the ZERO_PAGE count wraps negative.
475 return PageReserved(page
);
479 * Frees a list of pages.
480 * Assumes all pages on list are in same zone, and of same order.
481 * count is the number of pages to free.
483 * If the zone was previously in an "all pages pinned" state then look to
484 * see if this freeing clears that state.
486 * And clear the zone's pages_scanned counter, to hold off the "all pages are
487 * pinned" detection logic.
489 static void free_pages_bulk(struct zone
*zone
, int count
,
490 struct list_head
*list
, int order
)
492 spin_lock(&zone
->lock
);
493 zone
->all_unreclaimable
= 0;
494 zone
->pages_scanned
= 0;
498 VM_BUG_ON(list_empty(list
));
499 page
= list_entry(list
->prev
, struct page
, lru
);
500 /* have to delete it as __free_one_page list manipulates */
501 list_del(&page
->lru
);
502 __free_one_page(page
, zone
, order
);
504 spin_unlock(&zone
->lock
);
507 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
509 spin_lock(&zone
->lock
);
510 zone
->all_unreclaimable
= 0;
511 zone
->pages_scanned
= 0;
512 __free_one_page(page
, zone
, order
);
513 spin_unlock(&zone
->lock
);
516 static void __free_pages_ok(struct page
*page
, unsigned int order
)
522 for (i
= 0 ; i
< (1 << order
) ; ++i
)
523 reserved
+= free_pages_check(page
+ i
);
527 if (!PageHighMem(page
))
528 debug_check_no_locks_freed(page_address(page
),PAGE_SIZE
<<order
);
529 arch_free_page(page
, order
);
530 kernel_map_pages(page
, 1 << order
, 0);
532 local_irq_save(flags
);
533 __count_vm_events(PGFREE
, 1 << order
);
534 free_one_page(page_zone(page
), page
, order
);
535 local_irq_restore(flags
);
539 * permit the bootmem allocator to evade page validation on high-order frees
541 void fastcall __init
__free_pages_bootmem(struct page
*page
, unsigned int order
)
544 __ClearPageReserved(page
);
545 set_page_count(page
, 0);
546 set_page_refcounted(page
);
552 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
553 struct page
*p
= &page
[loop
];
555 if (loop
+ 1 < BITS_PER_LONG
)
557 __ClearPageReserved(p
);
558 set_page_count(p
, 0);
561 set_page_refcounted(page
);
562 __free_pages(page
, order
);
568 * The order of subdivision here is critical for the IO subsystem.
569 * Please do not alter this order without good reasons and regression
570 * testing. Specifically, as large blocks of memory are subdivided,
571 * the order in which smaller blocks are delivered depends on the order
572 * they're subdivided in this function. This is the primary factor
573 * influencing the order in which pages are delivered to the IO
574 * subsystem according to empirical testing, and this is also justified
575 * by considering the behavior of a buddy system containing a single
576 * large block of memory acted on by a series of small allocations.
577 * This behavior is a critical factor in sglist merging's success.
581 static inline void expand(struct zone
*zone
, struct page
*page
,
582 int low
, int high
, struct free_area
*area
,
585 unsigned long size
= 1 << high
;
591 VM_BUG_ON(bad_range(zone
, &page
[size
]));
592 list_add(&page
[size
].lru
, &area
->free_list
[migratetype
]);
594 set_page_order(&page
[size
], high
);
599 * This page is about to be returned from the page allocator
601 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
603 if (unlikely(page_mapcount(page
) |
604 (page
->mapping
!= NULL
) |
605 (page_count(page
) != 0) |
620 * For now, we report if PG_reserved was found set, but do not
621 * clear it, and do not allocate the page: as a safety net.
623 if (PageReserved(page
))
626 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
| 1 << PG_readahead
|
627 1 << PG_referenced
| 1 << PG_arch_1
|
628 1 << PG_owner_priv_1
| 1 << PG_mappedtodisk
);
629 set_page_private(page
, 0);
630 set_page_refcounted(page
);
632 arch_alloc_page(page
, order
);
633 kernel_map_pages(page
, 1 << order
, 1);
635 if (gfp_flags
& __GFP_ZERO
)
636 prep_zero_page(page
, order
, gfp_flags
);
638 if (order
&& (gfp_flags
& __GFP_COMP
))
639 prep_compound_page(page
, order
);
645 * Go through the free lists for the given migratetype and remove
646 * the smallest available page from the freelists
648 static struct page
*__rmqueue_smallest(struct zone
*zone
, unsigned int order
,
651 unsigned int current_order
;
652 struct free_area
* area
;
655 /* Find a page of the appropriate size in the preferred list */
656 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
657 area
= &(zone
->free_area
[current_order
]);
658 if (list_empty(&area
->free_list
[migratetype
]))
661 page
= list_entry(area
->free_list
[migratetype
].next
,
663 list_del(&page
->lru
);
664 rmv_page_order(page
);
666 __mod_zone_page_state(zone
, NR_FREE_PAGES
, - (1UL << order
));
667 expand(zone
, page
, order
, current_order
, area
, migratetype
);
676 * This array describes the order lists are fallen back to when
677 * the free lists for the desirable migrate type are depleted
679 static int fallbacks
[MIGRATE_TYPES
][MIGRATE_TYPES
-1] = {
680 [MIGRATE_UNMOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
681 [MIGRATE_RECLAIMABLE
] = { MIGRATE_UNMOVABLE
, MIGRATE_MOVABLE
, MIGRATE_RESERVE
},
682 [MIGRATE_MOVABLE
] = { MIGRATE_RECLAIMABLE
, MIGRATE_UNMOVABLE
, MIGRATE_RESERVE
},
683 [MIGRATE_RESERVE
] = { MIGRATE_RESERVE
, MIGRATE_RESERVE
, MIGRATE_RESERVE
}, /* Never used */
687 * Move the free pages in a range to the free lists of the requested type.
688 * Note that start_page and end_pages are not aligned on a pageblock
689 * boundary. If alignment is required, use move_freepages_block()
691 int move_freepages(struct zone
*zone
,
692 struct page
*start_page
, struct page
*end_page
,
699 #ifndef CONFIG_HOLES_IN_ZONE
701 * page_zone is not safe to call in this context when
702 * CONFIG_HOLES_IN_ZONE is set. This bug check is probably redundant
703 * anyway as we check zone boundaries in move_freepages_block().
704 * Remove at a later date when no bug reports exist related to
705 * grouping pages by mobility
707 BUG_ON(page_zone(start_page
) != page_zone(end_page
));
710 for (page
= start_page
; page
<= end_page
;) {
711 if (!pfn_valid_within(page_to_pfn(page
))) {
716 if (!PageBuddy(page
)) {
721 order
= page_order(page
);
722 list_del(&page
->lru
);
724 &zone
->free_area
[order
].free_list
[migratetype
]);
726 pages_moved
+= 1 << order
;
732 int move_freepages_block(struct zone
*zone
, struct page
*page
, int migratetype
)
734 unsigned long start_pfn
, end_pfn
;
735 struct page
*start_page
, *end_page
;
737 start_pfn
= page_to_pfn(page
);
738 start_pfn
= start_pfn
& ~(pageblock_nr_pages
-1);
739 start_page
= pfn_to_page(start_pfn
);
740 end_page
= start_page
+ pageblock_nr_pages
- 1;
741 end_pfn
= start_pfn
+ pageblock_nr_pages
- 1;
743 /* Do not cross zone boundaries */
744 if (start_pfn
< zone
->zone_start_pfn
)
746 if (end_pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
749 return move_freepages(zone
, start_page
, end_page
, migratetype
);
752 /* Return the page with the lowest PFN in the list */
753 static struct page
*min_page(struct list_head
*list
)
755 unsigned long min_pfn
= -1UL;
756 struct page
*min_page
= NULL
, *page
;;
758 list_for_each_entry(page
, list
, lru
) {
759 unsigned long pfn
= page_to_pfn(page
);
769 /* Remove an element from the buddy allocator from the fallback list */
770 static struct page
*__rmqueue_fallback(struct zone
*zone
, int order
,
771 int start_migratetype
)
773 struct free_area
* area
;
778 /* Find the largest possible block of pages in the other list */
779 for (current_order
= MAX_ORDER
-1; current_order
>= order
;
781 for (i
= 0; i
< MIGRATE_TYPES
- 1; i
++) {
782 migratetype
= fallbacks
[start_migratetype
][i
];
784 /* MIGRATE_RESERVE handled later if necessary */
785 if (migratetype
== MIGRATE_RESERVE
)
788 area
= &(zone
->free_area
[current_order
]);
789 if (list_empty(&area
->free_list
[migratetype
]))
792 /* Bias kernel allocations towards low pfns */
793 page
= list_entry(area
->free_list
[migratetype
].next
,
795 if (unlikely(start_migratetype
!= MIGRATE_MOVABLE
))
796 page
= min_page(&area
->free_list
[migratetype
]);
800 * If breaking a large block of pages, move all free
801 * pages to the preferred allocation list. If falling
802 * back for a reclaimable kernel allocation, be more
803 * agressive about taking ownership of free pages
805 if (unlikely(current_order
>= (pageblock_order
>> 1)) ||
806 start_migratetype
== MIGRATE_RECLAIMABLE
) {
808 pages
= move_freepages_block(zone
, page
,
811 /* Claim the whole block if over half of it is free */
812 if (pages
>= (1 << (pageblock_order
-1)))
813 set_pageblock_migratetype(page
,
816 migratetype
= start_migratetype
;
819 /* Remove the page from the freelists */
820 list_del(&page
->lru
);
821 rmv_page_order(page
);
822 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
825 if (current_order
== pageblock_order
)
826 set_pageblock_migratetype(page
,
829 expand(zone
, page
, order
, current_order
, area
, migratetype
);
834 /* Use MIGRATE_RESERVE rather than fail an allocation */
835 return __rmqueue_smallest(zone
, order
, MIGRATE_RESERVE
);
839 * Do the hard work of removing an element from the buddy allocator.
840 * Call me with the zone->lock already held.
842 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
,
847 page
= __rmqueue_smallest(zone
, order
, migratetype
);
850 page
= __rmqueue_fallback(zone
, order
, migratetype
);
856 * Obtain a specified number of elements from the buddy allocator, all under
857 * a single hold of the lock, for efficiency. Add them to the supplied list.
858 * Returns the number of new pages which were placed at *list.
860 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
861 unsigned long count
, struct list_head
*list
,
866 spin_lock(&zone
->lock
);
867 for (i
= 0; i
< count
; ++i
) {
868 struct page
*page
= __rmqueue(zone
, order
, migratetype
);
869 if (unlikely(page
== NULL
))
871 list_add(&page
->lru
, list
);
872 set_page_private(page
, migratetype
);
874 spin_unlock(&zone
->lock
);
880 * Called from the vmstat counter updater to drain pagesets of this
881 * currently executing processor on remote nodes after they have
884 * Note that this function must be called with the thread pinned to
885 * a single processor.
887 void drain_zone_pages(struct zone
*zone
, struct per_cpu_pages
*pcp
)
892 local_irq_save(flags
);
893 if (pcp
->count
>= pcp
->batch
)
894 to_drain
= pcp
->batch
;
896 to_drain
= pcp
->count
;
897 free_pages_bulk(zone
, to_drain
, &pcp
->list
, 0);
898 pcp
->count
-= to_drain
;
899 local_irq_restore(flags
);
903 static void __drain_pages(unsigned int cpu
)
909 for_each_zone(zone
) {
910 struct per_cpu_pageset
*pset
;
912 if (!populated_zone(zone
))
915 pset
= zone_pcp(zone
, cpu
);
916 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
917 struct per_cpu_pages
*pcp
;
920 local_irq_save(flags
);
921 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
923 local_irq_restore(flags
);
928 #ifdef CONFIG_HIBERNATION
930 void mark_free_pages(struct zone
*zone
)
932 unsigned long pfn
, max_zone_pfn
;
935 struct list_head
*curr
;
937 if (!zone
->spanned_pages
)
940 spin_lock_irqsave(&zone
->lock
, flags
);
942 max_zone_pfn
= zone
->zone_start_pfn
+ zone
->spanned_pages
;
943 for (pfn
= zone
->zone_start_pfn
; pfn
< max_zone_pfn
; pfn
++)
944 if (pfn_valid(pfn
)) {
945 struct page
*page
= pfn_to_page(pfn
);
947 if (!swsusp_page_is_forbidden(page
))
948 swsusp_unset_page_free(page
);
951 for_each_migratetype_order(order
, t
) {
952 list_for_each(curr
, &zone
->free_area
[order
].free_list
[t
]) {
955 pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
956 for (i
= 0; i
< (1UL << order
); i
++)
957 swsusp_set_page_free(pfn_to_page(pfn
+ i
));
960 spin_unlock_irqrestore(&zone
->lock
, flags
);
962 #endif /* CONFIG_PM */
965 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
967 void drain_local_pages(void)
971 local_irq_save(flags
);
972 __drain_pages(smp_processor_id());
973 local_irq_restore(flags
);
976 void smp_drain_local_pages(void *arg
)
982 * Spill all the per-cpu pages from all CPUs back into the buddy allocator
984 void drain_all_local_pages(void)
988 local_irq_save(flags
);
989 __drain_pages(smp_processor_id());
990 local_irq_restore(flags
);
992 smp_call_function(smp_drain_local_pages
, NULL
, 0, 1);
996 * Free a 0-order page
998 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
)
1000 struct zone
*zone
= page_zone(page
);
1001 struct per_cpu_pages
*pcp
;
1002 unsigned long flags
;
1005 page
->mapping
= NULL
;
1006 if (free_pages_check(page
))
1009 if (!PageHighMem(page
))
1010 debug_check_no_locks_freed(page_address(page
), PAGE_SIZE
);
1011 arch_free_page(page
, 0);
1012 kernel_map_pages(page
, 1, 0);
1014 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
1015 local_irq_save(flags
);
1016 __count_vm_event(PGFREE
);
1017 list_add(&page
->lru
, &pcp
->list
);
1018 set_page_private(page
, get_pageblock_migratetype(page
));
1020 if (pcp
->count
>= pcp
->high
) {
1021 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
1022 pcp
->count
-= pcp
->batch
;
1024 local_irq_restore(flags
);
1028 void fastcall
free_hot_page(struct page
*page
)
1030 free_hot_cold_page(page
, 0);
1033 void fastcall
free_cold_page(struct page
*page
)
1035 free_hot_cold_page(page
, 1);
1039 * split_page takes a non-compound higher-order page, and splits it into
1040 * n (1<<order) sub-pages: page[0..n]
1041 * Each sub-page must be freed individually.
1043 * Note: this is probably too low level an operation for use in drivers.
1044 * Please consult with lkml before using this in your driver.
1046 void split_page(struct page
*page
, unsigned int order
)
1050 VM_BUG_ON(PageCompound(page
));
1051 VM_BUG_ON(!page_count(page
));
1052 for (i
= 1; i
< (1 << order
); i
++)
1053 set_page_refcounted(page
+ i
);
1057 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
1058 * we cheat by calling it from here, in the order > 0 path. Saves a branch
1061 static struct page
*buffered_rmqueue(struct zonelist
*zonelist
,
1062 struct zone
*zone
, int order
, gfp_t gfp_flags
)
1064 unsigned long flags
;
1066 int cold
= !!(gfp_flags
& __GFP_COLD
);
1068 int migratetype
= allocflags_to_migratetype(gfp_flags
);
1072 if (likely(order
== 0)) {
1073 struct per_cpu_pages
*pcp
;
1075 pcp
= &zone_pcp(zone
, cpu
)->pcp
[cold
];
1076 local_irq_save(flags
);
1078 pcp
->count
= rmqueue_bulk(zone
, 0,
1079 pcp
->batch
, &pcp
->list
, migratetype
);
1080 if (unlikely(!pcp
->count
))
1084 /* Find a page of the appropriate migrate type */
1085 list_for_each_entry(page
, &pcp
->list
, lru
)
1086 if (page_private(page
) == migratetype
)
1089 /* Allocate more to the pcp list if necessary */
1090 if (unlikely(&page
->lru
== &pcp
->list
)) {
1091 pcp
->count
+= rmqueue_bulk(zone
, 0,
1092 pcp
->batch
, &pcp
->list
, migratetype
);
1093 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
1096 list_del(&page
->lru
);
1099 spin_lock_irqsave(&zone
->lock
, flags
);
1100 page
= __rmqueue(zone
, order
, migratetype
);
1101 spin_unlock(&zone
->lock
);
1106 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
1107 zone_statistics(zonelist
, zone
);
1108 local_irq_restore(flags
);
1111 VM_BUG_ON(bad_range(zone
, page
));
1112 if (prep_new_page(page
, order
, gfp_flags
))
1117 local_irq_restore(flags
);
1122 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
1123 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
1124 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
1125 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
1126 #define ALLOC_HARDER 0x10 /* try to alloc harder */
1127 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
1128 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
1130 #ifdef CONFIG_FAIL_PAGE_ALLOC
1132 static struct fail_page_alloc_attr
{
1133 struct fault_attr attr
;
1135 u32 ignore_gfp_highmem
;
1136 u32 ignore_gfp_wait
;
1139 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1141 struct dentry
*ignore_gfp_highmem_file
;
1142 struct dentry
*ignore_gfp_wait_file
;
1143 struct dentry
*min_order_file
;
1145 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1147 } fail_page_alloc
= {
1148 .attr
= FAULT_ATTR_INITIALIZER
,
1149 .ignore_gfp_wait
= 1,
1150 .ignore_gfp_highmem
= 1,
1154 static int __init
setup_fail_page_alloc(char *str
)
1156 return setup_fault_attr(&fail_page_alloc
.attr
, str
);
1158 __setup("fail_page_alloc=", setup_fail_page_alloc
);
1160 static int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1162 if (order
< fail_page_alloc
.min_order
)
1164 if (gfp_mask
& __GFP_NOFAIL
)
1166 if (fail_page_alloc
.ignore_gfp_highmem
&& (gfp_mask
& __GFP_HIGHMEM
))
1168 if (fail_page_alloc
.ignore_gfp_wait
&& (gfp_mask
& __GFP_WAIT
))
1171 return should_fail(&fail_page_alloc
.attr
, 1 << order
);
1174 #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS
1176 static int __init
fail_page_alloc_debugfs(void)
1178 mode_t mode
= S_IFREG
| S_IRUSR
| S_IWUSR
;
1182 err
= init_fault_attr_dentries(&fail_page_alloc
.attr
,
1186 dir
= fail_page_alloc
.attr
.dentries
.dir
;
1188 fail_page_alloc
.ignore_gfp_wait_file
=
1189 debugfs_create_bool("ignore-gfp-wait", mode
, dir
,
1190 &fail_page_alloc
.ignore_gfp_wait
);
1192 fail_page_alloc
.ignore_gfp_highmem_file
=
1193 debugfs_create_bool("ignore-gfp-highmem", mode
, dir
,
1194 &fail_page_alloc
.ignore_gfp_highmem
);
1195 fail_page_alloc
.min_order_file
=
1196 debugfs_create_u32("min-order", mode
, dir
,
1197 &fail_page_alloc
.min_order
);
1199 if (!fail_page_alloc
.ignore_gfp_wait_file
||
1200 !fail_page_alloc
.ignore_gfp_highmem_file
||
1201 !fail_page_alloc
.min_order_file
) {
1203 debugfs_remove(fail_page_alloc
.ignore_gfp_wait_file
);
1204 debugfs_remove(fail_page_alloc
.ignore_gfp_highmem_file
);
1205 debugfs_remove(fail_page_alloc
.min_order_file
);
1206 cleanup_fault_attr_dentries(&fail_page_alloc
.attr
);
1212 late_initcall(fail_page_alloc_debugfs
);
1214 #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */
1216 #else /* CONFIG_FAIL_PAGE_ALLOC */
1218 static inline int should_fail_alloc_page(gfp_t gfp_mask
, unsigned int order
)
1223 #endif /* CONFIG_FAIL_PAGE_ALLOC */
1226 * Return 1 if free pages are above 'mark'. This takes into account the order
1227 * of the allocation.
1229 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
1230 int classzone_idx
, int alloc_flags
)
1232 /* free_pages my go negative - that's OK */
1234 long free_pages
= zone_page_state(z
, NR_FREE_PAGES
) - (1 << order
) + 1;
1237 if (alloc_flags
& ALLOC_HIGH
)
1239 if (alloc_flags
& ALLOC_HARDER
)
1242 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
1244 for (o
= 0; o
< order
; o
++) {
1245 /* At the next order, this order's pages become unavailable */
1246 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
1248 /* Require fewer higher order pages to be free */
1251 if (free_pages
<= min
)
1259 * zlc_setup - Setup for "zonelist cache". Uses cached zone data to
1260 * skip over zones that are not allowed by the cpuset, or that have
1261 * been recently (in last second) found to be nearly full. See further
1262 * comments in mmzone.h. Reduces cache footprint of zonelist scans
1263 * that have to skip over alot of full or unallowed zones.
1265 * If the zonelist cache is present in the passed in zonelist, then
1266 * returns a pointer to the allowed node mask (either the current
1267 * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
1269 * If the zonelist cache is not available for this zonelist, does
1270 * nothing and returns NULL.
1272 * If the fullzones BITMAP in the zonelist cache is stale (more than
1273 * a second since last zap'd) then we zap it out (clear its bits.)
1275 * We hold off even calling zlc_setup, until after we've checked the
1276 * first zone in the zonelist, on the theory that most allocations will
1277 * be satisfied from that first zone, so best to examine that zone as
1278 * quickly as we can.
1280 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1282 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1283 nodemask_t
*allowednodes
; /* zonelist_cache approximation */
1285 zlc
= zonelist
->zlcache_ptr
;
1289 if (jiffies
- zlc
->last_full_zap
> 1 * HZ
) {
1290 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
1291 zlc
->last_full_zap
= jiffies
;
1294 allowednodes
= !in_interrupt() && (alloc_flags
& ALLOC_CPUSET
) ?
1295 &cpuset_current_mems_allowed
:
1296 &node_states
[N_HIGH_MEMORY
];
1297 return allowednodes
;
1301 * Given 'z' scanning a zonelist, run a couple of quick checks to see
1302 * if it is worth looking at further for free memory:
1303 * 1) Check that the zone isn't thought to be full (doesn't have its
1304 * bit set in the zonelist_cache fullzones BITMAP).
1305 * 2) Check that the zones node (obtained from the zonelist_cache
1306 * z_to_n[] mapping) is allowed in the passed in allowednodes mask.
1307 * Return true (non-zero) if zone is worth looking at further, or
1308 * else return false (zero) if it is not.
1310 * This check -ignores- the distinction between various watermarks,
1311 * such as GFP_HIGH, GFP_ATOMIC, PF_MEMALLOC, ... If a zone is
1312 * found to be full for any variation of these watermarks, it will
1313 * be considered full for up to one second by all requests, unless
1314 * we are so low on memory on all allowed nodes that we are forced
1315 * into the second scan of the zonelist.
1317 * In the second scan we ignore this zonelist cache and exactly
1318 * apply the watermarks to all zones, even it is slower to do so.
1319 * We are low on memory in the second scan, and should leave no stone
1320 * unturned looking for a free page.
1322 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zone
**z
,
1323 nodemask_t
*allowednodes
)
1325 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1326 int i
; /* index of *z in zonelist zones */
1327 int n
; /* node that zone *z is on */
1329 zlc
= zonelist
->zlcache_ptr
;
1333 i
= z
- zonelist
->zones
;
1336 /* This zone is worth trying if it is allowed but not full */
1337 return node_isset(n
, *allowednodes
) && !test_bit(i
, zlc
->fullzones
);
1341 * Given 'z' scanning a zonelist, set the corresponding bit in
1342 * zlc->fullzones, so that subsequent attempts to allocate a page
1343 * from that zone don't waste time re-examining it.
1345 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zone
**z
)
1347 struct zonelist_cache
*zlc
; /* cached zonelist speedup info */
1348 int i
; /* index of *z in zonelist zones */
1350 zlc
= zonelist
->zlcache_ptr
;
1354 i
= z
- zonelist
->zones
;
1356 set_bit(i
, zlc
->fullzones
);
1359 #else /* CONFIG_NUMA */
1361 static nodemask_t
*zlc_setup(struct zonelist
*zonelist
, int alloc_flags
)
1366 static int zlc_zone_worth_trying(struct zonelist
*zonelist
, struct zone
**z
,
1367 nodemask_t
*allowednodes
)
1372 static void zlc_mark_zone_full(struct zonelist
*zonelist
, struct zone
**z
)
1375 #endif /* CONFIG_NUMA */
1378 * get_page_from_freelist goes through the zonelist trying to allocate
1381 static struct page
*
1382 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
,
1383 struct zonelist
*zonelist
, int alloc_flags
)
1386 struct page
*page
= NULL
;
1387 int classzone_idx
= zone_idx(zonelist
->zones
[0]);
1389 nodemask_t
*allowednodes
= NULL
;/* zonelist_cache approximation */
1390 int zlc_active
= 0; /* set if using zonelist_cache */
1391 int did_zlc_setup
= 0; /* just call zlc_setup() one time */
1392 enum zone_type highest_zoneidx
= -1; /* Gets set for policy zonelists */
1396 * Scan zonelist, looking for a zone with enough free.
1397 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1399 z
= zonelist
->zones
;
1403 * In NUMA, this could be a policy zonelist which contains
1404 * zones that may not be allowed by the current gfp_mask.
1405 * Check the zone is allowed by the current flags
1407 if (unlikely(alloc_should_filter_zonelist(zonelist
))) {
1408 if (highest_zoneidx
== -1)
1409 highest_zoneidx
= gfp_zone(gfp_mask
);
1410 if (zone_idx(*z
) > highest_zoneidx
)
1414 if (NUMA_BUILD
&& zlc_active
&&
1415 !zlc_zone_worth_trying(zonelist
, z
, allowednodes
))
1418 if ((alloc_flags
& ALLOC_CPUSET
) &&
1419 !cpuset_zone_allowed_softwall(zone
, gfp_mask
))
1422 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
1424 if (alloc_flags
& ALLOC_WMARK_MIN
)
1425 mark
= zone
->pages_min
;
1426 else if (alloc_flags
& ALLOC_WMARK_LOW
)
1427 mark
= zone
->pages_low
;
1429 mark
= zone
->pages_high
;
1430 if (!zone_watermark_ok(zone
, order
, mark
,
1431 classzone_idx
, alloc_flags
)) {
1432 if (!zone_reclaim_mode
||
1433 !zone_reclaim(zone
, gfp_mask
, order
))
1434 goto this_zone_full
;
1438 page
= buffered_rmqueue(zonelist
, zone
, order
, gfp_mask
);
1443 zlc_mark_zone_full(zonelist
, z
);
1445 if (NUMA_BUILD
&& !did_zlc_setup
) {
1446 /* we do zlc_setup after the first zone is tried */
1447 allowednodes
= zlc_setup(zonelist
, alloc_flags
);
1451 } while (*(++z
) != NULL
);
1453 if (unlikely(NUMA_BUILD
&& page
== NULL
&& zlc_active
)) {
1454 /* Disable zlc cache for second zonelist scan */
1462 * This is the 'heart' of the zoned buddy allocator.
1464 struct page
* fastcall
1465 __alloc_pages(gfp_t gfp_mask
, unsigned int order
,
1466 struct zonelist
*zonelist
)
1468 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
1471 struct reclaim_state reclaim_state
;
1472 struct task_struct
*p
= current
;
1475 int did_some_progress
;
1477 might_sleep_if(wait
);
1479 if (should_fail_alloc_page(gfp_mask
, order
))
1483 z
= zonelist
->zones
; /* the list of zones suitable for gfp_mask */
1485 if (unlikely(*z
== NULL
)) {
1487 * Happens if we have an empty zonelist as a result of
1488 * GFP_THISNODE being used on a memoryless node
1493 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1494 zonelist
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
1499 * GFP_THISNODE (meaning __GFP_THISNODE, __GFP_NORETRY and
1500 * __GFP_NOWARN set) should not cause reclaim since the subsystem
1501 * (f.e. slab) using GFP_THISNODE may choose to trigger reclaim
1502 * using a larger set of nodes after it has established that the
1503 * allowed per node queues are empty and that nodes are
1506 if (NUMA_BUILD
&& (gfp_mask
& GFP_THISNODE
) == GFP_THISNODE
)
1509 for (z
= zonelist
->zones
; *z
; z
++)
1510 wakeup_kswapd(*z
, order
);
1513 * OK, we're below the kswapd watermark and have kicked background
1514 * reclaim. Now things get more complex, so set up alloc_flags according
1515 * to how we want to proceed.
1517 * The caller may dip into page reserves a bit more if the caller
1518 * cannot run direct reclaim, or if the caller has realtime scheduling
1519 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
1520 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
1522 alloc_flags
= ALLOC_WMARK_MIN
;
1523 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
1524 alloc_flags
|= ALLOC_HARDER
;
1525 if (gfp_mask
& __GFP_HIGH
)
1526 alloc_flags
|= ALLOC_HIGH
;
1528 alloc_flags
|= ALLOC_CPUSET
;
1531 * Go through the zonelist again. Let __GFP_HIGH and allocations
1532 * coming from realtime tasks go deeper into reserves.
1534 * This is the last chance, in general, before the goto nopage.
1535 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
1536 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
1538 page
= get_page_from_freelist(gfp_mask
, order
, zonelist
, alloc_flags
);
1542 /* This allocation should allow future memory freeing. */
1545 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
1546 && !in_interrupt()) {
1547 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
1549 /* go through the zonelist yet again, ignoring mins */
1550 page
= get_page_from_freelist(gfp_mask
, order
,
1551 zonelist
, ALLOC_NO_WATERMARKS
);
1554 if (gfp_mask
& __GFP_NOFAIL
) {
1555 congestion_wait(WRITE
, HZ
/50);
1562 /* Atomic allocations - we can't balance anything */
1568 /* We now go into synchronous reclaim */
1569 cpuset_memory_pressure_bump();
1570 p
->flags
|= PF_MEMALLOC
;
1571 reclaim_state
.reclaimed_slab
= 0;
1572 p
->reclaim_state
= &reclaim_state
;
1574 did_some_progress
= try_to_free_pages(zonelist
->zones
, order
, gfp_mask
);
1576 p
->reclaim_state
= NULL
;
1577 p
->flags
&= ~PF_MEMALLOC
;
1582 drain_all_local_pages();
1584 if (likely(did_some_progress
)) {
1585 page
= get_page_from_freelist(gfp_mask
, order
,
1586 zonelist
, alloc_flags
);
1589 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1591 * Go through the zonelist yet one more time, keep
1592 * very high watermark here, this is only to catch
1593 * a parallel oom killing, we must fail if we're still
1594 * under heavy pressure.
1596 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1597 zonelist
, ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1601 /* The OOM killer will not help higher order allocs so fail */
1602 if (order
> PAGE_ALLOC_COSTLY_ORDER
)
1605 out_of_memory(zonelist
, gfp_mask
, order
);
1610 * Don't let big-order allocations loop unless the caller explicitly
1611 * requests that. Wait for some write requests to complete then retry.
1613 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1614 * <= 3, but that may not be true in other implementations.
1617 if (!(gfp_mask
& __GFP_NORETRY
)) {
1618 if ((order
<= PAGE_ALLOC_COSTLY_ORDER
) ||
1619 (gfp_mask
& __GFP_REPEAT
))
1621 if (gfp_mask
& __GFP_NOFAIL
)
1625 congestion_wait(WRITE
, HZ
/50);
1630 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1631 printk(KERN_WARNING
"%s: page allocation failure."
1632 " order:%d, mode:0x%x\n",
1633 p
->comm
, order
, gfp_mask
);
1641 EXPORT_SYMBOL(__alloc_pages
);
1644 * Common helper functions.
1646 fastcall
unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1649 page
= alloc_pages(gfp_mask
, order
);
1652 return (unsigned long) page_address(page
);
1655 EXPORT_SYMBOL(__get_free_pages
);
1657 fastcall
unsigned long get_zeroed_page(gfp_t gfp_mask
)
1662 * get_zeroed_page() returns a 32-bit address, which cannot represent
1665 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1667 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1669 return (unsigned long) page_address(page
);
1673 EXPORT_SYMBOL(get_zeroed_page
);
1675 void __pagevec_free(struct pagevec
*pvec
)
1677 int i
= pagevec_count(pvec
);
1680 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1683 fastcall
void __free_pages(struct page
*page
, unsigned int order
)
1685 if (put_page_testzero(page
)) {
1687 free_hot_page(page
);
1689 __free_pages_ok(page
, order
);
1693 EXPORT_SYMBOL(__free_pages
);
1695 fastcall
void free_pages(unsigned long addr
, unsigned int order
)
1698 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1699 __free_pages(virt_to_page((void *)addr
), order
);
1703 EXPORT_SYMBOL(free_pages
);
1705 static unsigned int nr_free_zone_pages(int offset
)
1707 /* Just pick one node, since fallback list is circular */
1708 pg_data_t
*pgdat
= NODE_DATA(numa_node_id());
1709 unsigned int sum
= 0;
1711 struct zonelist
*zonelist
= pgdat
->node_zonelists
+ offset
;
1712 struct zone
**zonep
= zonelist
->zones
;
1715 for (zone
= *zonep
++; zone
; zone
= *zonep
++) {
1716 unsigned long size
= zone
->present_pages
;
1717 unsigned long high
= zone
->pages_high
;
1726 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1728 unsigned int nr_free_buffer_pages(void)
1730 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1732 EXPORT_SYMBOL_GPL(nr_free_buffer_pages
);
1735 * Amount of free RAM allocatable within all zones
1737 unsigned int nr_free_pagecache_pages(void)
1739 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER_MOVABLE
));
1742 static inline void show_node(struct zone
*zone
)
1745 printk("Node %d ", zone_to_nid(zone
));
1748 void si_meminfo(struct sysinfo
*val
)
1750 val
->totalram
= totalram_pages
;
1752 val
->freeram
= global_page_state(NR_FREE_PAGES
);
1753 val
->bufferram
= nr_blockdev_pages();
1754 val
->totalhigh
= totalhigh_pages
;
1755 val
->freehigh
= nr_free_highpages();
1756 val
->mem_unit
= PAGE_SIZE
;
1759 EXPORT_SYMBOL(si_meminfo
);
1762 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1764 pg_data_t
*pgdat
= NODE_DATA(nid
);
1766 val
->totalram
= pgdat
->node_present_pages
;
1767 val
->freeram
= node_page_state(nid
, NR_FREE_PAGES
);
1768 #ifdef CONFIG_HIGHMEM
1769 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1770 val
->freehigh
= zone_page_state(&pgdat
->node_zones
[ZONE_HIGHMEM
],
1776 val
->mem_unit
= PAGE_SIZE
;
1780 #define K(x) ((x) << (PAGE_SHIFT-10))
1783 * Show free area list (used inside shift_scroll-lock stuff)
1784 * We also calculate the percentage fragmentation. We do this by counting the
1785 * memory on each free list with the exception of the first item on the list.
1787 void show_free_areas(void)
1792 for_each_zone(zone
) {
1793 if (!populated_zone(zone
))
1797 printk("%s per-cpu:\n", zone
->name
);
1799 for_each_online_cpu(cpu
) {
1800 struct per_cpu_pageset
*pageset
;
1802 pageset
= zone_pcp(zone
, cpu
);
1804 printk("CPU %4d: Hot: hi:%5d, btch:%4d usd:%4d "
1805 "Cold: hi:%5d, btch:%4d usd:%4d\n",
1806 cpu
, pageset
->pcp
[0].high
,
1807 pageset
->pcp
[0].batch
, pageset
->pcp
[0].count
,
1808 pageset
->pcp
[1].high
, pageset
->pcp
[1].batch
,
1809 pageset
->pcp
[1].count
);
1813 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu unstable:%lu\n"
1814 " free:%lu slab:%lu mapped:%lu pagetables:%lu bounce:%lu\n",
1815 global_page_state(NR_ACTIVE
),
1816 global_page_state(NR_INACTIVE
),
1817 global_page_state(NR_FILE_DIRTY
),
1818 global_page_state(NR_WRITEBACK
),
1819 global_page_state(NR_UNSTABLE_NFS
),
1820 global_page_state(NR_FREE_PAGES
),
1821 global_page_state(NR_SLAB_RECLAIMABLE
) +
1822 global_page_state(NR_SLAB_UNRECLAIMABLE
),
1823 global_page_state(NR_FILE_MAPPED
),
1824 global_page_state(NR_PAGETABLE
),
1825 global_page_state(NR_BOUNCE
));
1827 for_each_zone(zone
) {
1830 if (!populated_zone(zone
))
1842 " pages_scanned:%lu"
1843 " all_unreclaimable? %s"
1846 K(zone_page_state(zone
, NR_FREE_PAGES
)),
1849 K(zone
->pages_high
),
1850 K(zone_page_state(zone
, NR_ACTIVE
)),
1851 K(zone_page_state(zone
, NR_INACTIVE
)),
1852 K(zone
->present_pages
),
1853 zone
->pages_scanned
,
1854 (zone
->all_unreclaimable
? "yes" : "no")
1856 printk("lowmem_reserve[]:");
1857 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1858 printk(" %lu", zone
->lowmem_reserve
[i
]);
1862 for_each_zone(zone
) {
1863 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1865 if (!populated_zone(zone
))
1869 printk("%s: ", zone
->name
);
1871 spin_lock_irqsave(&zone
->lock
, flags
);
1872 for (order
= 0; order
< MAX_ORDER
; order
++) {
1873 nr
[order
] = zone
->free_area
[order
].nr_free
;
1874 total
+= nr
[order
] << order
;
1876 spin_unlock_irqrestore(&zone
->lock
, flags
);
1877 for (order
= 0; order
< MAX_ORDER
; order
++)
1878 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1879 printk("= %lukB\n", K(total
));
1882 show_swap_cache_info();
1886 * Builds allocation fallback zone lists.
1888 * Add all populated zones of a node to the zonelist.
1890 static int build_zonelists_node(pg_data_t
*pgdat
, struct zonelist
*zonelist
,
1891 int nr_zones
, enum zone_type zone_type
)
1895 BUG_ON(zone_type
>= MAX_NR_ZONES
);
1900 zone
= pgdat
->node_zones
+ zone_type
;
1901 if (populated_zone(zone
)) {
1902 zonelist
->zones
[nr_zones
++] = zone
;
1903 check_highest_zone(zone_type
);
1906 } while (zone_type
);
1913 * 0 = automatic detection of better ordering.
1914 * 1 = order by ([node] distance, -zonetype)
1915 * 2 = order by (-zonetype, [node] distance)
1917 * If not NUMA, ZONELIST_ORDER_ZONE and ZONELIST_ORDER_NODE will create
1918 * the same zonelist. So only NUMA can configure this param.
1920 #define ZONELIST_ORDER_DEFAULT 0
1921 #define ZONELIST_ORDER_NODE 1
1922 #define ZONELIST_ORDER_ZONE 2
1924 /* zonelist order in the kernel.
1925 * set_zonelist_order() will set this to NODE or ZONE.
1927 static int current_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1928 static char zonelist_order_name
[3][8] = {"Default", "Node", "Zone"};
1932 /* The value user specified ....changed by config */
1933 static int user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1934 /* string for sysctl */
1935 #define NUMA_ZONELIST_ORDER_LEN 16
1936 char numa_zonelist_order
[16] = "default";
1939 * interface for configure zonelist ordering.
1940 * command line option "numa_zonelist_order"
1941 * = "[dD]efault - default, automatic configuration.
1942 * = "[nN]ode - order by node locality, then by zone within node
1943 * = "[zZ]one - order by zone, then by locality within zone
1946 static int __parse_numa_zonelist_order(char *s
)
1948 if (*s
== 'd' || *s
== 'D') {
1949 user_zonelist_order
= ZONELIST_ORDER_DEFAULT
;
1950 } else if (*s
== 'n' || *s
== 'N') {
1951 user_zonelist_order
= ZONELIST_ORDER_NODE
;
1952 } else if (*s
== 'z' || *s
== 'Z') {
1953 user_zonelist_order
= ZONELIST_ORDER_ZONE
;
1956 "Ignoring invalid numa_zonelist_order value: "
1963 static __init
int setup_numa_zonelist_order(char *s
)
1966 return __parse_numa_zonelist_order(s
);
1969 early_param("numa_zonelist_order", setup_numa_zonelist_order
);
1972 * sysctl handler for numa_zonelist_order
1974 int numa_zonelist_order_handler(ctl_table
*table
, int write
,
1975 struct file
*file
, void __user
*buffer
, size_t *length
,
1978 char saved_string
[NUMA_ZONELIST_ORDER_LEN
];
1982 strncpy(saved_string
, (char*)table
->data
,
1983 NUMA_ZONELIST_ORDER_LEN
);
1984 ret
= proc_dostring(table
, write
, file
, buffer
, length
, ppos
);
1988 int oldval
= user_zonelist_order
;
1989 if (__parse_numa_zonelist_order((char*)table
->data
)) {
1991 * bogus value. restore saved string
1993 strncpy((char*)table
->data
, saved_string
,
1994 NUMA_ZONELIST_ORDER_LEN
);
1995 user_zonelist_order
= oldval
;
1996 } else if (oldval
!= user_zonelist_order
)
1997 build_all_zonelists();
2003 #define MAX_NODE_LOAD (num_online_nodes())
2004 static int node_load
[MAX_NUMNODES
];
2007 * find_next_best_node - find the next node that should appear in a given node's fallback list
2008 * @node: node whose fallback list we're appending
2009 * @used_node_mask: nodemask_t of already used nodes
2011 * We use a number of factors to determine which is the next node that should
2012 * appear on a given node's fallback list. The node should not have appeared
2013 * already in @node's fallback list, and it should be the next closest node
2014 * according to the distance array (which contains arbitrary distance values
2015 * from each node to each node in the system), and should also prefer nodes
2016 * with no CPUs, since presumably they'll have very little allocation pressure
2017 * on them otherwise.
2018 * It returns -1 if no node is found.
2020 static int find_next_best_node(int node
, nodemask_t
*used_node_mask
)
2023 int min_val
= INT_MAX
;
2026 /* Use the local node if we haven't already */
2027 if (!node_isset(node
, *used_node_mask
)) {
2028 node_set(node
, *used_node_mask
);
2032 for_each_node_state(n
, N_HIGH_MEMORY
) {
2035 /* Don't want a node to appear more than once */
2036 if (node_isset(n
, *used_node_mask
))
2039 /* Use the distance array to find the distance */
2040 val
= node_distance(node
, n
);
2042 /* Penalize nodes under us ("prefer the next node") */
2045 /* Give preference to headless and unused nodes */
2046 tmp
= node_to_cpumask(n
);
2047 if (!cpus_empty(tmp
))
2048 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
2050 /* Slight preference for less loaded node */
2051 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
2052 val
+= node_load
[n
];
2054 if (val
< min_val
) {
2061 node_set(best_node
, *used_node_mask
);
2068 * Build zonelists ordered by node and zones within node.
2069 * This results in maximum locality--normal zone overflows into local
2070 * DMA zone, if any--but risks exhausting DMA zone.
2072 static void build_zonelists_in_node_order(pg_data_t
*pgdat
, int node
)
2076 struct zonelist
*zonelist
;
2078 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2079 zonelist
= pgdat
->node_zonelists
+ i
;
2080 for (j
= 0; zonelist
->zones
[j
] != NULL
; j
++)
2082 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
2083 zonelist
->zones
[j
] = NULL
;
2088 * Build gfp_thisnode zonelists
2090 static void build_thisnode_zonelists(pg_data_t
*pgdat
)
2094 struct zonelist
*zonelist
;
2096 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2097 zonelist
= pgdat
->node_zonelists
+ MAX_NR_ZONES
+ i
;
2098 j
= build_zonelists_node(pgdat
, zonelist
, 0, i
);
2099 zonelist
->zones
[j
] = NULL
;
2104 * Build zonelists ordered by zone and nodes within zones.
2105 * This results in conserving DMA zone[s] until all Normal memory is
2106 * exhausted, but results in overflowing to remote node while memory
2107 * may still exist in local DMA zone.
2109 static int node_order
[MAX_NUMNODES
];
2111 static void build_zonelists_in_zone_order(pg_data_t
*pgdat
, int nr_nodes
)
2115 int zone_type
; /* needs to be signed */
2117 struct zonelist
*zonelist
;
2119 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2120 zonelist
= pgdat
->node_zonelists
+ i
;
2122 for (zone_type
= i
; zone_type
>= 0; zone_type
--) {
2123 for (j
= 0; j
< nr_nodes
; j
++) {
2124 node
= node_order
[j
];
2125 z
= &NODE_DATA(node
)->node_zones
[zone_type
];
2126 if (populated_zone(z
)) {
2127 zonelist
->zones
[pos
++] = z
;
2128 check_highest_zone(zone_type
);
2132 zonelist
->zones
[pos
] = NULL
;
2136 static int default_zonelist_order(void)
2139 unsigned long low_kmem_size
,total_size
;
2143 * ZONE_DMA and ZONE_DMA32 can be very small area in the sytem.
2144 * If they are really small and used heavily, the system can fall
2145 * into OOM very easily.
2146 * This function detect ZONE_DMA/DMA32 size and confgigures zone order.
2148 /* Is there ZONE_NORMAL ? (ex. ppc has only DMA zone..) */
2151 for_each_online_node(nid
) {
2152 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2153 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2154 if (populated_zone(z
)) {
2155 if (zone_type
< ZONE_NORMAL
)
2156 low_kmem_size
+= z
->present_pages
;
2157 total_size
+= z
->present_pages
;
2161 if (!low_kmem_size
|| /* there are no DMA area. */
2162 low_kmem_size
> total_size
/2) /* DMA/DMA32 is big. */
2163 return ZONELIST_ORDER_NODE
;
2165 * look into each node's config.
2166 * If there is a node whose DMA/DMA32 memory is very big area on
2167 * local memory, NODE_ORDER may be suitable.
2169 average_size
= total_size
/
2170 (nodes_weight(node_states
[N_HIGH_MEMORY
]) + 1);
2171 for_each_online_node(nid
) {
2174 for (zone_type
= 0; zone_type
< MAX_NR_ZONES
; zone_type
++) {
2175 z
= &NODE_DATA(nid
)->node_zones
[zone_type
];
2176 if (populated_zone(z
)) {
2177 if (zone_type
< ZONE_NORMAL
)
2178 low_kmem_size
+= z
->present_pages
;
2179 total_size
+= z
->present_pages
;
2182 if (low_kmem_size
&&
2183 total_size
> average_size
&& /* ignore small node */
2184 low_kmem_size
> total_size
* 70/100)
2185 return ZONELIST_ORDER_NODE
;
2187 return ZONELIST_ORDER_ZONE
;
2190 static void set_zonelist_order(void)
2192 if (user_zonelist_order
== ZONELIST_ORDER_DEFAULT
)
2193 current_zonelist_order
= default_zonelist_order();
2195 current_zonelist_order
= user_zonelist_order
;
2198 static void build_zonelists(pg_data_t
*pgdat
)
2202 nodemask_t used_mask
;
2203 int local_node
, prev_node
;
2204 struct zonelist
*zonelist
;
2205 int order
= current_zonelist_order
;
2207 /* initialize zonelists */
2208 for (i
= 0; i
< MAX_ZONELISTS
; i
++) {
2209 zonelist
= pgdat
->node_zonelists
+ i
;
2210 zonelist
->zones
[0] = NULL
;
2213 /* NUMA-aware ordering of nodes */
2214 local_node
= pgdat
->node_id
;
2215 load
= num_online_nodes();
2216 prev_node
= local_node
;
2217 nodes_clear(used_mask
);
2219 memset(node_load
, 0, sizeof(node_load
));
2220 memset(node_order
, 0, sizeof(node_order
));
2223 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
2224 int distance
= node_distance(local_node
, node
);
2227 * If another node is sufficiently far away then it is better
2228 * to reclaim pages in a zone before going off node.
2230 if (distance
> RECLAIM_DISTANCE
)
2231 zone_reclaim_mode
= 1;
2234 * We don't want to pressure a particular node.
2235 * So adding penalty to the first node in same
2236 * distance group to make it round-robin.
2238 if (distance
!= node_distance(local_node
, prev_node
))
2239 node_load
[node
] = load
;
2243 if (order
== ZONELIST_ORDER_NODE
)
2244 build_zonelists_in_node_order(pgdat
, node
);
2246 node_order
[j
++] = node
; /* remember order */
2249 if (order
== ZONELIST_ORDER_ZONE
) {
2250 /* calculate node order -- i.e., DMA last! */
2251 build_zonelists_in_zone_order(pgdat
, j
);
2254 build_thisnode_zonelists(pgdat
);
2257 /* Construct the zonelist performance cache - see further mmzone.h */
2258 static void build_zonelist_cache(pg_data_t
*pgdat
)
2262 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2263 struct zonelist
*zonelist
;
2264 struct zonelist_cache
*zlc
;
2267 zonelist
= pgdat
->node_zonelists
+ i
;
2268 zonelist
->zlcache_ptr
= zlc
= &zonelist
->zlcache
;
2269 bitmap_zero(zlc
->fullzones
, MAX_ZONES_PER_ZONELIST
);
2270 for (z
= zonelist
->zones
; *z
; z
++)
2271 zlc
->z_to_n
[z
- zonelist
->zones
] = zone_to_nid(*z
);
2276 #else /* CONFIG_NUMA */
2278 static void set_zonelist_order(void)
2280 current_zonelist_order
= ZONELIST_ORDER_ZONE
;
2283 static void build_zonelists(pg_data_t
*pgdat
)
2285 int node
, local_node
;
2288 local_node
= pgdat
->node_id
;
2289 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2290 struct zonelist
*zonelist
;
2292 zonelist
= pgdat
->node_zonelists
+ i
;
2294 j
= build_zonelists_node(pgdat
, zonelist
, 0, i
);
2296 * Now we build the zonelist so that it contains the zones
2297 * of all the other nodes.
2298 * We don't want to pressure a particular node, so when
2299 * building the zones for node N, we make sure that the
2300 * zones coming right after the local ones are those from
2301 * node N+1 (modulo N)
2303 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
2304 if (!node_online(node
))
2306 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
2308 for (node
= 0; node
< local_node
; node
++) {
2309 if (!node_online(node
))
2311 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, i
);
2314 zonelist
->zones
[j
] = NULL
;
2318 /* non-NUMA variant of zonelist performance cache - just NULL zlcache_ptr */
2319 static void build_zonelist_cache(pg_data_t
*pgdat
)
2323 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
2324 pgdat
->node_zonelists
[i
].zlcache_ptr
= NULL
;
2327 #endif /* CONFIG_NUMA */
2329 /* return values int ....just for stop_machine_run() */
2330 static int __build_all_zonelists(void *dummy
)
2334 for_each_online_node(nid
) {
2335 pg_data_t
*pgdat
= NODE_DATA(nid
);
2337 build_zonelists(pgdat
);
2338 build_zonelist_cache(pgdat
);
2343 void build_all_zonelists(void)
2345 set_zonelist_order();
2347 if (system_state
== SYSTEM_BOOTING
) {
2348 __build_all_zonelists(NULL
);
2349 cpuset_init_current_mems_allowed();
2351 /* we have to stop all cpus to guaranntee there is no user
2353 stop_machine_run(__build_all_zonelists
, NULL
, NR_CPUS
);
2354 /* cpuset refresh routine should be here */
2356 vm_total_pages
= nr_free_pagecache_pages();
2358 * Disable grouping by mobility if the number of pages in the
2359 * system is too low to allow the mechanism to work. It would be
2360 * more accurate, but expensive to check per-zone. This check is
2361 * made on memory-hotadd so a system can start with mobility
2362 * disabled and enable it later
2364 if (vm_total_pages
< (pageblock_nr_pages
* MIGRATE_TYPES
))
2365 page_group_by_mobility_disabled
= 1;
2367 page_group_by_mobility_disabled
= 0;
2369 printk("Built %i zonelists in %s order, mobility grouping %s. "
2370 "Total pages: %ld\n",
2372 zonelist_order_name
[current_zonelist_order
],
2373 page_group_by_mobility_disabled
? "off" : "on",
2376 printk("Policy zone: %s\n", zone_names
[policy_zone
]);
2381 * Helper functions to size the waitqueue hash table.
2382 * Essentially these want to choose hash table sizes sufficiently
2383 * large so that collisions trying to wait on pages are rare.
2384 * But in fact, the number of active page waitqueues on typical
2385 * systems is ridiculously low, less than 200. So this is even
2386 * conservative, even though it seems large.
2388 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
2389 * waitqueues, i.e. the size of the waitq table given the number of pages.
2391 #define PAGES_PER_WAITQUEUE 256
2393 #ifndef CONFIG_MEMORY_HOTPLUG
2394 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2396 unsigned long size
= 1;
2398 pages
/= PAGES_PER_WAITQUEUE
;
2400 while (size
< pages
)
2404 * Once we have dozens or even hundreds of threads sleeping
2405 * on IO we've got bigger problems than wait queue collision.
2406 * Limit the size of the wait table to a reasonable size.
2408 size
= min(size
, 4096UL);
2410 return max(size
, 4UL);
2414 * A zone's size might be changed by hot-add, so it is not possible to determine
2415 * a suitable size for its wait_table. So we use the maximum size now.
2417 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
2419 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
2420 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
2421 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
2423 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
2424 * or more by the traditional way. (See above). It equals:
2426 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
2427 * ia64(16K page size) : = ( 8G + 4M)byte.
2428 * powerpc (64K page size) : = (32G +16M)byte.
2430 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
2437 * This is an integer logarithm so that shifts can be used later
2438 * to extract the more random high bits from the multiplicative
2439 * hash function before the remainder is taken.
2441 static inline unsigned long wait_table_bits(unsigned long size
)
2446 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
2449 * Mark a number of pageblocks as MIGRATE_RESERVE. The number
2450 * of blocks reserved is based on zone->pages_min. The memory within the
2451 * reserve will tend to store contiguous free pages. Setting min_free_kbytes
2452 * higher will lead to a bigger reserve which will get freed as contiguous
2453 * blocks as reclaim kicks in
2455 static void setup_zone_migrate_reserve(struct zone
*zone
)
2457 unsigned long start_pfn
, pfn
, end_pfn
;
2459 unsigned long reserve
, block_migratetype
;
2461 /* Get the start pfn, end pfn and the number of blocks to reserve */
2462 start_pfn
= zone
->zone_start_pfn
;
2463 end_pfn
= start_pfn
+ zone
->spanned_pages
;
2464 reserve
= roundup(zone
->pages_min
, pageblock_nr_pages
) >>
2467 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
+= pageblock_nr_pages
) {
2468 if (!pfn_valid(pfn
))
2470 page
= pfn_to_page(pfn
);
2472 /* Blocks with reserved pages will never free, skip them. */
2473 if (PageReserved(page
))
2476 block_migratetype
= get_pageblock_migratetype(page
);
2478 /* If this block is reserved, account for it */
2479 if (reserve
> 0 && block_migratetype
== MIGRATE_RESERVE
) {
2484 /* Suitable for reserving if this block is movable */
2485 if (reserve
> 0 && block_migratetype
== MIGRATE_MOVABLE
) {
2486 set_pageblock_migratetype(page
, MIGRATE_RESERVE
);
2487 move_freepages_block(zone
, page
, MIGRATE_RESERVE
);
2493 * If the reserve is met and this is a previous reserved block,
2496 if (block_migratetype
== MIGRATE_RESERVE
) {
2497 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2498 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
2504 * Initially all pages are reserved - free ones are freed
2505 * up by free_all_bootmem() once the early boot process is
2506 * done. Non-atomic initialization, single-pass.
2508 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
2509 unsigned long start_pfn
, enum memmap_context context
)
2512 unsigned long end_pfn
= start_pfn
+ size
;
2515 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
2517 * There can be holes in boot-time mem_map[]s
2518 * handed to this function. They do not
2519 * exist on hotplugged memory.
2521 if (context
== MEMMAP_EARLY
) {
2522 if (!early_pfn_valid(pfn
))
2524 if (!early_pfn_in_nid(pfn
, nid
))
2527 page
= pfn_to_page(pfn
);
2528 set_page_links(page
, zone
, nid
, pfn
);
2529 init_page_count(page
);
2530 reset_page_mapcount(page
);
2531 SetPageReserved(page
);
2534 * Mark the block movable so that blocks are reserved for
2535 * movable at startup. This will force kernel allocations
2536 * to reserve their blocks rather than leaking throughout
2537 * the address space during boot when many long-lived
2538 * kernel allocations are made. Later some blocks near
2539 * the start are marked MIGRATE_RESERVE by
2540 * setup_zone_migrate_reserve()
2542 if ((pfn
& (pageblock_nr_pages
-1)))
2543 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
2545 INIT_LIST_HEAD(&page
->lru
);
2546 #ifdef WANT_PAGE_VIRTUAL
2547 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
2548 if (!is_highmem_idx(zone
))
2549 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
2554 static void __meminit
zone_init_free_lists(struct pglist_data
*pgdat
,
2555 struct zone
*zone
, unsigned long size
)
2558 for_each_migratetype_order(order
, t
) {
2559 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
[t
]);
2560 zone
->free_area
[order
].nr_free
= 0;
2564 #ifndef __HAVE_ARCH_MEMMAP_INIT
2565 #define memmap_init(size, nid, zone, start_pfn) \
2566 memmap_init_zone((size), (nid), (zone), (start_pfn), MEMMAP_EARLY)
2569 static int __devinit
zone_batchsize(struct zone
*zone
)
2574 * The per-cpu-pages pools are set to around 1000th of the
2575 * size of the zone. But no more than 1/2 of a meg.
2577 * OK, so we don't know how big the cache is. So guess.
2579 batch
= zone
->present_pages
/ 1024;
2580 if (batch
* PAGE_SIZE
> 512 * 1024)
2581 batch
= (512 * 1024) / PAGE_SIZE
;
2582 batch
/= 4; /* We effectively *= 4 below */
2587 * Clamp the batch to a 2^n - 1 value. Having a power
2588 * of 2 value was found to be more likely to have
2589 * suboptimal cache aliasing properties in some cases.
2591 * For example if 2 tasks are alternately allocating
2592 * batches of pages, one task can end up with a lot
2593 * of pages of one half of the possible page colors
2594 * and the other with pages of the other colors.
2596 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
2601 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
2603 struct per_cpu_pages
*pcp
;
2605 memset(p
, 0, sizeof(*p
));
2607 pcp
= &p
->pcp
[0]; /* hot */
2609 pcp
->high
= 6 * batch
;
2610 pcp
->batch
= max(1UL, 1 * batch
);
2611 INIT_LIST_HEAD(&pcp
->list
);
2613 pcp
= &p
->pcp
[1]; /* cold*/
2615 pcp
->high
= 2 * batch
;
2616 pcp
->batch
= max(1UL, batch
/2);
2617 INIT_LIST_HEAD(&pcp
->list
);
2621 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
2622 * to the value high for the pageset p.
2625 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
2628 struct per_cpu_pages
*pcp
;
2630 pcp
= &p
->pcp
[0]; /* hot list */
2632 pcp
->batch
= max(1UL, high
/4);
2633 if ((high
/4) > (PAGE_SHIFT
* 8))
2634 pcp
->batch
= PAGE_SHIFT
* 8;
2640 * Boot pageset table. One per cpu which is going to be used for all
2641 * zones and all nodes. The parameters will be set in such a way
2642 * that an item put on a list will immediately be handed over to
2643 * the buddy list. This is safe since pageset manipulation is done
2644 * with interrupts disabled.
2646 * Some NUMA counter updates may also be caught by the boot pagesets.
2648 * The boot_pagesets must be kept even after bootup is complete for
2649 * unused processors and/or zones. They do play a role for bootstrapping
2650 * hotplugged processors.
2652 * zoneinfo_show() and maybe other functions do
2653 * not check if the processor is online before following the pageset pointer.
2654 * Other parts of the kernel may not check if the zone is available.
2656 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
2659 * Dynamically allocate memory for the
2660 * per cpu pageset array in struct zone.
2662 static int __cpuinit
process_zones(int cpu
)
2664 struct zone
*zone
, *dzone
;
2665 int node
= cpu_to_node(cpu
);
2667 node_set_state(node
, N_CPU
); /* this node has a cpu */
2669 for_each_zone(zone
) {
2671 if (!populated_zone(zone
))
2674 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
2676 if (!zone_pcp(zone
, cpu
))
2679 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
2681 if (percpu_pagelist_fraction
)
2682 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
2683 (zone
->present_pages
/ percpu_pagelist_fraction
));
2688 for_each_zone(dzone
) {
2689 if (!populated_zone(dzone
))
2693 kfree(zone_pcp(dzone
, cpu
));
2694 zone_pcp(dzone
, cpu
) = NULL
;
2699 static inline void free_zone_pagesets(int cpu
)
2703 for_each_zone(zone
) {
2704 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
2706 /* Free per_cpu_pageset if it is slab allocated */
2707 if (pset
!= &boot_pageset
[cpu
])
2709 zone_pcp(zone
, cpu
) = NULL
;
2713 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
2714 unsigned long action
,
2717 int cpu
= (long)hcpu
;
2718 int ret
= NOTIFY_OK
;
2721 case CPU_UP_PREPARE
:
2722 case CPU_UP_PREPARE_FROZEN
:
2723 if (process_zones(cpu
))
2726 case CPU_UP_CANCELED
:
2727 case CPU_UP_CANCELED_FROZEN
:
2729 case CPU_DEAD_FROZEN
:
2730 free_zone_pagesets(cpu
);
2738 static struct notifier_block __cpuinitdata pageset_notifier
=
2739 { &pageset_cpuup_callback
, NULL
, 0 };
2741 void __init
setup_per_cpu_pageset(void)
2745 /* Initialize per_cpu_pageset for cpu 0.
2746 * A cpuup callback will do this for every cpu
2747 * as it comes online
2749 err
= process_zones(smp_processor_id());
2751 register_cpu_notifier(&pageset_notifier
);
2756 static noinline __init_refok
2757 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
2760 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2764 * The per-page waitqueue mechanism uses hashed waitqueues
2767 zone
->wait_table_hash_nr_entries
=
2768 wait_table_hash_nr_entries(zone_size_pages
);
2769 zone
->wait_table_bits
=
2770 wait_table_bits(zone
->wait_table_hash_nr_entries
);
2771 alloc_size
= zone
->wait_table_hash_nr_entries
2772 * sizeof(wait_queue_head_t
);
2774 if (system_state
== SYSTEM_BOOTING
) {
2775 zone
->wait_table
= (wait_queue_head_t
*)
2776 alloc_bootmem_node(pgdat
, alloc_size
);
2779 * This case means that a zone whose size was 0 gets new memory
2780 * via memory hot-add.
2781 * But it may be the case that a new node was hot-added. In
2782 * this case vmalloc() will not be able to use this new node's
2783 * memory - this wait_table must be initialized to use this new
2784 * node itself as well.
2785 * To use this new node's memory, further consideration will be
2788 zone
->wait_table
= vmalloc(alloc_size
);
2790 if (!zone
->wait_table
)
2793 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
2794 init_waitqueue_head(zone
->wait_table
+ i
);
2799 static __meminit
void zone_pcp_init(struct zone
*zone
)
2802 unsigned long batch
= zone_batchsize(zone
);
2804 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
2806 /* Early boot. Slab allocator not functional yet */
2807 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
2808 setup_pageset(&boot_pageset
[cpu
],0);
2810 setup_pageset(zone_pcp(zone
,cpu
), batch
);
2813 if (zone
->present_pages
)
2814 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
2815 zone
->name
, zone
->present_pages
, batch
);
2818 __meminit
int init_currently_empty_zone(struct zone
*zone
,
2819 unsigned long zone_start_pfn
,
2821 enum memmap_context context
)
2823 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
2825 ret
= zone_wait_table_init(zone
, size
);
2828 pgdat
->nr_zones
= zone_idx(zone
) + 1;
2830 zone
->zone_start_pfn
= zone_start_pfn
;
2832 memmap_init(size
, pgdat
->node_id
, zone_idx(zone
), zone_start_pfn
);
2834 zone_init_free_lists(pgdat
, zone
, zone
->spanned_pages
);
2839 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
2841 * Basic iterator support. Return the first range of PFNs for a node
2842 * Note: nid == MAX_NUMNODES returns first region regardless of node
2844 static int __meminit
first_active_region_index_in_nid(int nid
)
2848 for (i
= 0; i
< nr_nodemap_entries
; i
++)
2849 if (nid
== MAX_NUMNODES
|| early_node_map
[i
].nid
== nid
)
2856 * Basic iterator support. Return the next active range of PFNs for a node
2857 * Note: nid == MAX_NUMNODES returns next region regardles of node
2859 static int __meminit
next_active_region_index_in_nid(int index
, int nid
)
2861 for (index
= index
+ 1; index
< nr_nodemap_entries
; index
++)
2862 if (nid
== MAX_NUMNODES
|| early_node_map
[index
].nid
== nid
)
2868 #ifndef CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID
2870 * Required by SPARSEMEM. Given a PFN, return what node the PFN is on.
2871 * Architectures may implement their own version but if add_active_range()
2872 * was used and there are no special requirements, this is a convenient
2875 int __meminit
early_pfn_to_nid(unsigned long pfn
)
2879 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
2880 unsigned long start_pfn
= early_node_map
[i
].start_pfn
;
2881 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2883 if (start_pfn
<= pfn
&& pfn
< end_pfn
)
2884 return early_node_map
[i
].nid
;
2889 #endif /* CONFIG_HAVE_ARCH_EARLY_PFN_TO_NID */
2891 /* Basic iterator support to walk early_node_map[] */
2892 #define for_each_active_range_index_in_nid(i, nid) \
2893 for (i = first_active_region_index_in_nid(nid); i != -1; \
2894 i = next_active_region_index_in_nid(i, nid))
2897 * free_bootmem_with_active_regions - Call free_bootmem_node for each active range
2898 * @nid: The node to free memory on. If MAX_NUMNODES, all nodes are freed.
2899 * @max_low_pfn: The highest PFN that will be passed to free_bootmem_node
2901 * If an architecture guarantees that all ranges registered with
2902 * add_active_ranges() contain no holes and may be freed, this
2903 * this function may be used instead of calling free_bootmem() manually.
2905 void __init
free_bootmem_with_active_regions(int nid
,
2906 unsigned long max_low_pfn
)
2910 for_each_active_range_index_in_nid(i
, nid
) {
2911 unsigned long size_pages
= 0;
2912 unsigned long end_pfn
= early_node_map
[i
].end_pfn
;
2914 if (early_node_map
[i
].start_pfn
>= max_low_pfn
)
2917 if (end_pfn
> max_low_pfn
)
2918 end_pfn
= max_low_pfn
;
2920 size_pages
= end_pfn
- early_node_map
[i
].start_pfn
;
2921 free_bootmem_node(NODE_DATA(early_node_map
[i
].nid
),
2922 PFN_PHYS(early_node_map
[i
].start_pfn
),
2923 size_pages
<< PAGE_SHIFT
);
2928 * sparse_memory_present_with_active_regions - Call memory_present for each active range
2929 * @nid: The node to call memory_present for. If MAX_NUMNODES, all nodes will be used.
2931 * If an architecture guarantees that all ranges registered with
2932 * add_active_ranges() contain no holes and may be freed, this
2933 * function may be used instead of calling memory_present() manually.
2935 void __init
sparse_memory_present_with_active_regions(int nid
)
2939 for_each_active_range_index_in_nid(i
, nid
)
2940 memory_present(early_node_map
[i
].nid
,
2941 early_node_map
[i
].start_pfn
,
2942 early_node_map
[i
].end_pfn
);
2946 * push_node_boundaries - Push node boundaries to at least the requested boundary
2947 * @nid: The nid of the node to push the boundary for
2948 * @start_pfn: The start pfn of the node
2949 * @end_pfn: The end pfn of the node
2951 * In reserve-based hot-add, mem_map is allocated that is unused until hotadd
2952 * time. Specifically, on x86_64, SRAT will report ranges that can potentially
2953 * be hotplugged even though no physical memory exists. This function allows
2954 * an arch to push out the node boundaries so mem_map is allocated that can
2957 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
2958 void __init
push_node_boundaries(unsigned int nid
,
2959 unsigned long start_pfn
, unsigned long end_pfn
)
2961 printk(KERN_DEBUG
"Entering push_node_boundaries(%u, %lu, %lu)\n",
2962 nid
, start_pfn
, end_pfn
);
2964 /* Initialise the boundary for this node if necessary */
2965 if (node_boundary_end_pfn
[nid
] == 0)
2966 node_boundary_start_pfn
[nid
] = -1UL;
2968 /* Update the boundaries */
2969 if (node_boundary_start_pfn
[nid
] > start_pfn
)
2970 node_boundary_start_pfn
[nid
] = start_pfn
;
2971 if (node_boundary_end_pfn
[nid
] < end_pfn
)
2972 node_boundary_end_pfn
[nid
] = end_pfn
;
2975 /* If necessary, push the node boundary out for reserve hotadd */
2976 static void __meminit
account_node_boundary(unsigned int nid
,
2977 unsigned long *start_pfn
, unsigned long *end_pfn
)
2979 printk(KERN_DEBUG
"Entering account_node_boundary(%u, %lu, %lu)\n",
2980 nid
, *start_pfn
, *end_pfn
);
2982 /* Return if boundary information has not been provided */
2983 if (node_boundary_end_pfn
[nid
] == 0)
2986 /* Check the boundaries and update if necessary */
2987 if (node_boundary_start_pfn
[nid
] < *start_pfn
)
2988 *start_pfn
= node_boundary_start_pfn
[nid
];
2989 if (node_boundary_end_pfn
[nid
] > *end_pfn
)
2990 *end_pfn
= node_boundary_end_pfn
[nid
];
2993 void __init
push_node_boundaries(unsigned int nid
,
2994 unsigned long start_pfn
, unsigned long end_pfn
) {}
2996 static void __meminit
account_node_boundary(unsigned int nid
,
2997 unsigned long *start_pfn
, unsigned long *end_pfn
) {}
3002 * get_pfn_range_for_nid - Return the start and end page frames for a node
3003 * @nid: The nid to return the range for. If MAX_NUMNODES, the min and max PFN are returned.
3004 * @start_pfn: Passed by reference. On return, it will have the node start_pfn.
3005 * @end_pfn: Passed by reference. On return, it will have the node end_pfn.
3007 * It returns the start and end page frame of a node based on information
3008 * provided by an arch calling add_active_range(). If called for a node
3009 * with no available memory, a warning is printed and the start and end
3012 void __meminit
get_pfn_range_for_nid(unsigned int nid
,
3013 unsigned long *start_pfn
, unsigned long *end_pfn
)
3019 for_each_active_range_index_in_nid(i
, nid
) {
3020 *start_pfn
= min(*start_pfn
, early_node_map
[i
].start_pfn
);
3021 *end_pfn
= max(*end_pfn
, early_node_map
[i
].end_pfn
);
3024 if (*start_pfn
== -1UL)
3027 /* Push the node boundaries out if requested */
3028 account_node_boundary(nid
, start_pfn
, end_pfn
);
3032 * This finds a zone that can be used for ZONE_MOVABLE pages. The
3033 * assumption is made that zones within a node are ordered in monotonic
3034 * increasing memory addresses so that the "highest" populated zone is used
3036 void __init
find_usable_zone_for_movable(void)
3039 for (zone_index
= MAX_NR_ZONES
- 1; zone_index
>= 0; zone_index
--) {
3040 if (zone_index
== ZONE_MOVABLE
)
3043 if (arch_zone_highest_possible_pfn
[zone_index
] >
3044 arch_zone_lowest_possible_pfn
[zone_index
])
3048 VM_BUG_ON(zone_index
== -1);
3049 movable_zone
= zone_index
;
3053 * The zone ranges provided by the architecture do not include ZONE_MOVABLE
3054 * because it is sized independant of architecture. Unlike the other zones,
3055 * the starting point for ZONE_MOVABLE is not fixed. It may be different
3056 * in each node depending on the size of each node and how evenly kernelcore
3057 * is distributed. This helper function adjusts the zone ranges
3058 * provided by the architecture for a given node by using the end of the
3059 * highest usable zone for ZONE_MOVABLE. This preserves the assumption that
3060 * zones within a node are in order of monotonic increases memory addresses
3062 void __meminit
adjust_zone_range_for_zone_movable(int nid
,
3063 unsigned long zone_type
,
3064 unsigned long node_start_pfn
,
3065 unsigned long node_end_pfn
,
3066 unsigned long *zone_start_pfn
,
3067 unsigned long *zone_end_pfn
)
3069 /* Only adjust if ZONE_MOVABLE is on this node */
3070 if (zone_movable_pfn
[nid
]) {
3071 /* Size ZONE_MOVABLE */
3072 if (zone_type
== ZONE_MOVABLE
) {
3073 *zone_start_pfn
= zone_movable_pfn
[nid
];
3074 *zone_end_pfn
= min(node_end_pfn
,
3075 arch_zone_highest_possible_pfn
[movable_zone
]);
3077 /* Adjust for ZONE_MOVABLE starting within this range */
3078 } else if (*zone_start_pfn
< zone_movable_pfn
[nid
] &&
3079 *zone_end_pfn
> zone_movable_pfn
[nid
]) {
3080 *zone_end_pfn
= zone_movable_pfn
[nid
];
3082 /* Check if this whole range is within ZONE_MOVABLE */
3083 } else if (*zone_start_pfn
>= zone_movable_pfn
[nid
])
3084 *zone_start_pfn
= *zone_end_pfn
;
3089 * Return the number of pages a zone spans in a node, including holes
3090 * present_pages = zone_spanned_pages_in_node() - zone_absent_pages_in_node()
3092 static unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3093 unsigned long zone_type
,
3094 unsigned long *ignored
)
3096 unsigned long node_start_pfn
, node_end_pfn
;
3097 unsigned long zone_start_pfn
, zone_end_pfn
;
3099 /* Get the start and end of the node and zone */
3100 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3101 zone_start_pfn
= arch_zone_lowest_possible_pfn
[zone_type
];
3102 zone_end_pfn
= arch_zone_highest_possible_pfn
[zone_type
];
3103 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3104 node_start_pfn
, node_end_pfn
,
3105 &zone_start_pfn
, &zone_end_pfn
);
3107 /* Check that this node has pages within the zone's required range */
3108 if (zone_end_pfn
< node_start_pfn
|| zone_start_pfn
> node_end_pfn
)
3111 /* Move the zone boundaries inside the node if necessary */
3112 zone_end_pfn
= min(zone_end_pfn
, node_end_pfn
);
3113 zone_start_pfn
= max(zone_start_pfn
, node_start_pfn
);
3115 /* Return the spanned pages */
3116 return zone_end_pfn
- zone_start_pfn
;
3120 * Return the number of holes in a range on a node. If nid is MAX_NUMNODES,
3121 * then all holes in the requested range will be accounted for.
3123 unsigned long __meminit
__absent_pages_in_range(int nid
,
3124 unsigned long range_start_pfn
,
3125 unsigned long range_end_pfn
)
3128 unsigned long prev_end_pfn
= 0, hole_pages
= 0;
3129 unsigned long start_pfn
;
3131 /* Find the end_pfn of the first active range of pfns in the node */
3132 i
= first_active_region_index_in_nid(nid
);
3136 prev_end_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3138 /* Account for ranges before physical memory on this node */
3139 if (early_node_map
[i
].start_pfn
> range_start_pfn
)
3140 hole_pages
= prev_end_pfn
- range_start_pfn
;
3142 /* Find all holes for the zone within the node */
3143 for (; i
!= -1; i
= next_active_region_index_in_nid(i
, nid
)) {
3145 /* No need to continue if prev_end_pfn is outside the zone */
3146 if (prev_end_pfn
>= range_end_pfn
)
3149 /* Make sure the end of the zone is not within the hole */
3150 start_pfn
= min(early_node_map
[i
].start_pfn
, range_end_pfn
);
3151 prev_end_pfn
= max(prev_end_pfn
, range_start_pfn
);
3153 /* Update the hole size cound and move on */
3154 if (start_pfn
> range_start_pfn
) {
3155 BUG_ON(prev_end_pfn
> start_pfn
);
3156 hole_pages
+= start_pfn
- prev_end_pfn
;
3158 prev_end_pfn
= early_node_map
[i
].end_pfn
;
3161 /* Account for ranges past physical memory on this node */
3162 if (range_end_pfn
> prev_end_pfn
)
3163 hole_pages
+= range_end_pfn
-
3164 max(range_start_pfn
, prev_end_pfn
);
3170 * absent_pages_in_range - Return number of page frames in holes within a range
3171 * @start_pfn: The start PFN to start searching for holes
3172 * @end_pfn: The end PFN to stop searching for holes
3174 * It returns the number of pages frames in memory holes within a range.
3176 unsigned long __init
absent_pages_in_range(unsigned long start_pfn
,
3177 unsigned long end_pfn
)
3179 return __absent_pages_in_range(MAX_NUMNODES
, start_pfn
, end_pfn
);
3182 /* Return the number of page frames in holes in a zone on a node */
3183 static unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3184 unsigned long zone_type
,
3185 unsigned long *ignored
)
3187 unsigned long node_start_pfn
, node_end_pfn
;
3188 unsigned long zone_start_pfn
, zone_end_pfn
;
3190 get_pfn_range_for_nid(nid
, &node_start_pfn
, &node_end_pfn
);
3191 zone_start_pfn
= max(arch_zone_lowest_possible_pfn
[zone_type
],
3193 zone_end_pfn
= min(arch_zone_highest_possible_pfn
[zone_type
],
3196 adjust_zone_range_for_zone_movable(nid
, zone_type
,
3197 node_start_pfn
, node_end_pfn
,
3198 &zone_start_pfn
, &zone_end_pfn
);
3199 return __absent_pages_in_range(nid
, zone_start_pfn
, zone_end_pfn
);
3203 static inline unsigned long __meminit
zone_spanned_pages_in_node(int nid
,
3204 unsigned long zone_type
,
3205 unsigned long *zones_size
)
3207 return zones_size
[zone_type
];
3210 static inline unsigned long __meminit
zone_absent_pages_in_node(int nid
,
3211 unsigned long zone_type
,
3212 unsigned long *zholes_size
)
3217 return zholes_size
[zone_type
];
3222 static void __meminit
calculate_node_totalpages(struct pglist_data
*pgdat
,
3223 unsigned long *zones_size
, unsigned long *zholes_size
)
3225 unsigned long realtotalpages
, totalpages
= 0;
3228 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3229 totalpages
+= zone_spanned_pages_in_node(pgdat
->node_id
, i
,
3231 pgdat
->node_spanned_pages
= totalpages
;
3233 realtotalpages
= totalpages
;
3234 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
3236 zone_absent_pages_in_node(pgdat
->node_id
, i
,
3238 pgdat
->node_present_pages
= realtotalpages
;
3239 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
,
3243 #ifndef CONFIG_SPARSEMEM
3245 * Calculate the size of the zone->blockflags rounded to an unsigned long
3246 * Start by making sure zonesize is a multiple of pageblock_order by rounding
3247 * up. Then use 1 NR_PAGEBLOCK_BITS worth of bits per pageblock, finally
3248 * round what is now in bits to nearest long in bits, then return it in
3251 static unsigned long __init
usemap_size(unsigned long zonesize
)
3253 unsigned long usemapsize
;
3255 usemapsize
= roundup(zonesize
, pageblock_nr_pages
);
3256 usemapsize
= usemapsize
>> pageblock_order
;
3257 usemapsize
*= NR_PAGEBLOCK_BITS
;
3258 usemapsize
= roundup(usemapsize
, 8 * sizeof(unsigned long));
3260 return usemapsize
/ 8;
3263 static void __init
setup_usemap(struct pglist_data
*pgdat
,
3264 struct zone
*zone
, unsigned long zonesize
)
3266 unsigned long usemapsize
= usemap_size(zonesize
);
3267 zone
->pageblock_flags
= NULL
;
3269 zone
->pageblock_flags
= alloc_bootmem_node(pgdat
, usemapsize
);
3270 memset(zone
->pageblock_flags
, 0, usemapsize
);
3274 static void inline setup_usemap(struct pglist_data
*pgdat
,
3275 struct zone
*zone
, unsigned long zonesize
) {}
3276 #endif /* CONFIG_SPARSEMEM */
3278 #ifdef CONFIG_HUGETLB_PAGE_SIZE_VARIABLE
3279 /* Initialise the number of pages represented by NR_PAGEBLOCK_BITS */
3280 static inline void __init
set_pageblock_order(unsigned int order
)
3282 /* Check that pageblock_nr_pages has not already been setup */
3283 if (pageblock_order
)
3287 * Assume the largest contiguous order of interest is a huge page.
3288 * This value may be variable depending on boot parameters on IA64
3290 pageblock_order
= order
;
3292 #else /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3294 /* Defined this way to avoid accidently referencing HUGETLB_PAGE_ORDER */
3295 #define set_pageblock_order(x) do {} while (0)
3297 #endif /* CONFIG_HUGETLB_PAGE_SIZE_VARIABLE */
3300 * Set up the zone data structures:
3301 * - mark all pages reserved
3302 * - mark all memory queues empty
3303 * - clear the memory bitmaps
3305 static void __meminit
free_area_init_core(struct pglist_data
*pgdat
,
3306 unsigned long *zones_size
, unsigned long *zholes_size
)
3309 int nid
= pgdat
->node_id
;
3310 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
3313 pgdat_resize_init(pgdat
);
3314 pgdat
->nr_zones
= 0;
3315 init_waitqueue_head(&pgdat
->kswapd_wait
);
3316 pgdat
->kswapd_max_order
= 0;
3318 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
3319 struct zone
*zone
= pgdat
->node_zones
+ j
;
3320 unsigned long size
, realsize
, memmap_pages
;
3322 size
= zone_spanned_pages_in_node(nid
, j
, zones_size
);
3323 realsize
= size
- zone_absent_pages_in_node(nid
, j
,
3327 * Adjust realsize so that it accounts for how much memory
3328 * is used by this zone for memmap. This affects the watermark
3329 * and per-cpu initialisations
3331 memmap_pages
= (size
* sizeof(struct page
)) >> PAGE_SHIFT
;
3332 if (realsize
>= memmap_pages
) {
3333 realsize
-= memmap_pages
;
3335 " %s zone: %lu pages used for memmap\n",
3336 zone_names
[j
], memmap_pages
);
3339 " %s zone: %lu pages exceeds realsize %lu\n",
3340 zone_names
[j
], memmap_pages
, realsize
);
3342 /* Account for reserved pages */
3343 if (j
== 0 && realsize
> dma_reserve
) {
3344 realsize
-= dma_reserve
;
3345 printk(KERN_DEBUG
" %s zone: %lu pages reserved\n",
3346 zone_names
[0], dma_reserve
);
3349 if (!is_highmem_idx(j
))
3350 nr_kernel_pages
+= realsize
;
3351 nr_all_pages
+= realsize
;
3353 zone
->spanned_pages
= size
;
3354 zone
->present_pages
= realsize
;
3357 zone
->min_unmapped_pages
= (realsize
*sysctl_min_unmapped_ratio
)
3359 zone
->min_slab_pages
= (realsize
* sysctl_min_slab_ratio
) / 100;
3361 zone
->name
= zone_names
[j
];
3362 spin_lock_init(&zone
->lock
);
3363 spin_lock_init(&zone
->lru_lock
);
3364 zone_seqlock_init(zone
);
3365 zone
->zone_pgdat
= pgdat
;
3367 zone
->prev_priority
= DEF_PRIORITY
;
3369 zone_pcp_init(zone
);
3370 INIT_LIST_HEAD(&zone
->active_list
);
3371 INIT_LIST_HEAD(&zone
->inactive_list
);
3372 zone
->nr_scan_active
= 0;
3373 zone
->nr_scan_inactive
= 0;
3374 zap_zone_vm_stats(zone
);
3375 atomic_set(&zone
->reclaim_in_progress
, 0);
3379 set_pageblock_order(HUGETLB_PAGE_ORDER
);
3380 setup_usemap(pgdat
, zone
, size
);
3381 ret
= init_currently_empty_zone(zone
, zone_start_pfn
,
3382 size
, MEMMAP_EARLY
);
3384 zone_start_pfn
+= size
;
3388 static void __init_refok
alloc_node_mem_map(struct pglist_data
*pgdat
)
3390 /* Skip empty nodes */
3391 if (!pgdat
->node_spanned_pages
)
3394 #ifdef CONFIG_FLAT_NODE_MEM_MAP
3395 /* ia64 gets its own node_mem_map, before this, without bootmem */
3396 if (!pgdat
->node_mem_map
) {
3397 unsigned long size
, start
, end
;
3401 * The zone's endpoints aren't required to be MAX_ORDER
3402 * aligned but the node_mem_map endpoints must be in order
3403 * for the buddy allocator to function correctly.
3405 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
3406 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
3407 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
3408 size
= (end
- start
) * sizeof(struct page
);
3409 map
= alloc_remap(pgdat
->node_id
, size
);
3411 map
= alloc_bootmem_node(pgdat
, size
);
3412 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
3414 #ifndef CONFIG_NEED_MULTIPLE_NODES
3416 * With no DISCONTIG, the global mem_map is just set as node 0's
3418 if (pgdat
== NODE_DATA(0)) {
3419 mem_map
= NODE_DATA(0)->node_mem_map
;
3420 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3421 if (page_to_pfn(mem_map
) != pgdat
->node_start_pfn
)
3422 mem_map
-= pgdat
->node_start_pfn
;
3423 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3426 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
3429 void __meminit
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
3430 unsigned long *zones_size
, unsigned long node_start_pfn
,
3431 unsigned long *zholes_size
)
3433 pgdat
->node_id
= nid
;
3434 pgdat
->node_start_pfn
= node_start_pfn
;
3435 calculate_node_totalpages(pgdat
, zones_size
, zholes_size
);
3437 alloc_node_mem_map(pgdat
);
3439 free_area_init_core(pgdat
, zones_size
, zholes_size
);
3442 #ifdef CONFIG_ARCH_POPULATES_NODE_MAP
3444 #if MAX_NUMNODES > 1
3446 * Figure out the number of possible node ids.
3448 static void __init
setup_nr_node_ids(void)
3451 unsigned int highest
= 0;
3453 for_each_node_mask(node
, node_possible_map
)
3455 nr_node_ids
= highest
+ 1;
3458 static inline void setup_nr_node_ids(void)
3464 * add_active_range - Register a range of PFNs backed by physical memory
3465 * @nid: The node ID the range resides on
3466 * @start_pfn: The start PFN of the available physical memory
3467 * @end_pfn: The end PFN of the available physical memory
3469 * These ranges are stored in an early_node_map[] and later used by
3470 * free_area_init_nodes() to calculate zone sizes and holes. If the
3471 * range spans a memory hole, it is up to the architecture to ensure
3472 * the memory is not freed by the bootmem allocator. If possible
3473 * the range being registered will be merged with existing ranges.
3475 void __init
add_active_range(unsigned int nid
, unsigned long start_pfn
,
3476 unsigned long end_pfn
)
3480 printk(KERN_DEBUG
"Entering add_active_range(%d, %lu, %lu) "
3481 "%d entries of %d used\n",
3482 nid
, start_pfn
, end_pfn
,
3483 nr_nodemap_entries
, MAX_ACTIVE_REGIONS
);
3485 /* Merge with existing active regions if possible */
3486 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3487 if (early_node_map
[i
].nid
!= nid
)
3490 /* Skip if an existing region covers this new one */
3491 if (start_pfn
>= early_node_map
[i
].start_pfn
&&
3492 end_pfn
<= early_node_map
[i
].end_pfn
)
3495 /* Merge forward if suitable */
3496 if (start_pfn
<= early_node_map
[i
].end_pfn
&&
3497 end_pfn
> early_node_map
[i
].end_pfn
) {
3498 early_node_map
[i
].end_pfn
= end_pfn
;
3502 /* Merge backward if suitable */
3503 if (start_pfn
< early_node_map
[i
].end_pfn
&&
3504 end_pfn
>= early_node_map
[i
].start_pfn
) {
3505 early_node_map
[i
].start_pfn
= start_pfn
;
3510 /* Check that early_node_map is large enough */
3511 if (i
>= MAX_ACTIVE_REGIONS
) {
3512 printk(KERN_CRIT
"More than %d memory regions, truncating\n",
3513 MAX_ACTIVE_REGIONS
);
3517 early_node_map
[i
].nid
= nid
;
3518 early_node_map
[i
].start_pfn
= start_pfn
;
3519 early_node_map
[i
].end_pfn
= end_pfn
;
3520 nr_nodemap_entries
= i
+ 1;
3524 * shrink_active_range - Shrink an existing registered range of PFNs
3525 * @nid: The node id the range is on that should be shrunk
3526 * @old_end_pfn: The old end PFN of the range
3527 * @new_end_pfn: The new PFN of the range
3529 * i386 with NUMA use alloc_remap() to store a node_mem_map on a local node.
3530 * The map is kept at the end physical page range that has already been
3531 * registered with add_active_range(). This function allows an arch to shrink
3532 * an existing registered range.
3534 void __init
shrink_active_range(unsigned int nid
, unsigned long old_end_pfn
,
3535 unsigned long new_end_pfn
)
3539 /* Find the old active region end and shrink */
3540 for_each_active_range_index_in_nid(i
, nid
)
3541 if (early_node_map
[i
].end_pfn
== old_end_pfn
) {
3542 early_node_map
[i
].end_pfn
= new_end_pfn
;
3548 * remove_all_active_ranges - Remove all currently registered regions
3550 * During discovery, it may be found that a table like SRAT is invalid
3551 * and an alternative discovery method must be used. This function removes
3552 * all currently registered regions.
3554 void __init
remove_all_active_ranges(void)
3556 memset(early_node_map
, 0, sizeof(early_node_map
));
3557 nr_nodemap_entries
= 0;
3558 #ifdef CONFIG_MEMORY_HOTPLUG_RESERVE
3559 memset(node_boundary_start_pfn
, 0, sizeof(node_boundary_start_pfn
));
3560 memset(node_boundary_end_pfn
, 0, sizeof(node_boundary_end_pfn
));
3561 #endif /* CONFIG_MEMORY_HOTPLUG_RESERVE */
3564 /* Compare two active node_active_regions */
3565 static int __init
cmp_node_active_region(const void *a
, const void *b
)
3567 struct node_active_region
*arange
= (struct node_active_region
*)a
;
3568 struct node_active_region
*brange
= (struct node_active_region
*)b
;
3570 /* Done this way to avoid overflows */
3571 if (arange
->start_pfn
> brange
->start_pfn
)
3573 if (arange
->start_pfn
< brange
->start_pfn
)
3579 /* sort the node_map by start_pfn */
3580 static void __init
sort_node_map(void)
3582 sort(early_node_map
, (size_t)nr_nodemap_entries
,
3583 sizeof(struct node_active_region
),
3584 cmp_node_active_region
, NULL
);
3587 /* Find the lowest pfn for a node */
3588 unsigned long __init
find_min_pfn_for_node(unsigned long nid
)
3591 unsigned long min_pfn
= ULONG_MAX
;
3593 /* Assuming a sorted map, the first range found has the starting pfn */
3594 for_each_active_range_index_in_nid(i
, nid
)
3595 min_pfn
= min(min_pfn
, early_node_map
[i
].start_pfn
);
3597 if (min_pfn
== ULONG_MAX
) {
3599 "Could not find start_pfn for node %lu\n", nid
);
3607 * find_min_pfn_with_active_regions - Find the minimum PFN registered
3609 * It returns the minimum PFN based on information provided via
3610 * add_active_range().
3612 unsigned long __init
find_min_pfn_with_active_regions(void)
3614 return find_min_pfn_for_node(MAX_NUMNODES
);
3618 * find_max_pfn_with_active_regions - Find the maximum PFN registered
3620 * It returns the maximum PFN based on information provided via
3621 * add_active_range().
3623 unsigned long __init
find_max_pfn_with_active_regions(void)
3626 unsigned long max_pfn
= 0;
3628 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3629 max_pfn
= max(max_pfn
, early_node_map
[i
].end_pfn
);
3635 * early_calculate_totalpages()
3636 * Sum pages in active regions for movable zone.
3637 * Populate N_HIGH_MEMORY for calculating usable_nodes.
3639 static unsigned long __init
early_calculate_totalpages(void)
3642 unsigned long totalpages
= 0;
3644 for (i
= 0; i
< nr_nodemap_entries
; i
++) {
3645 unsigned long pages
= early_node_map
[i
].end_pfn
-
3646 early_node_map
[i
].start_pfn
;
3647 totalpages
+= pages
;
3649 node_set_state(early_node_map
[i
].nid
, N_HIGH_MEMORY
);
3655 * Find the PFN the Movable zone begins in each node. Kernel memory
3656 * is spread evenly between nodes as long as the nodes have enough
3657 * memory. When they don't, some nodes will have more kernelcore than
3660 void __init
find_zone_movable_pfns_for_nodes(unsigned long *movable_pfn
)
3663 unsigned long usable_startpfn
;
3664 unsigned long kernelcore_node
, kernelcore_remaining
;
3665 unsigned long totalpages
= early_calculate_totalpages();
3666 int usable_nodes
= nodes_weight(node_states
[N_HIGH_MEMORY
]);
3669 * If movablecore was specified, calculate what size of
3670 * kernelcore that corresponds so that memory usable for
3671 * any allocation type is evenly spread. If both kernelcore
3672 * and movablecore are specified, then the value of kernelcore
3673 * will be used for required_kernelcore if it's greater than
3674 * what movablecore would have allowed.
3676 if (required_movablecore
) {
3677 unsigned long corepages
;
3680 * Round-up so that ZONE_MOVABLE is at least as large as what
3681 * was requested by the user
3683 required_movablecore
=
3684 roundup(required_movablecore
, MAX_ORDER_NR_PAGES
);
3685 corepages
= totalpages
- required_movablecore
;
3687 required_kernelcore
= max(required_kernelcore
, corepages
);
3690 /* If kernelcore was not specified, there is no ZONE_MOVABLE */
3691 if (!required_kernelcore
)
3694 /* usable_startpfn is the lowest possible pfn ZONE_MOVABLE can be at */
3695 find_usable_zone_for_movable();
3696 usable_startpfn
= arch_zone_lowest_possible_pfn
[movable_zone
];
3699 /* Spread kernelcore memory as evenly as possible throughout nodes */
3700 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3701 for_each_node_state(nid
, N_HIGH_MEMORY
) {
3703 * Recalculate kernelcore_node if the division per node
3704 * now exceeds what is necessary to satisfy the requested
3705 * amount of memory for the kernel
3707 if (required_kernelcore
< kernelcore_node
)
3708 kernelcore_node
= required_kernelcore
/ usable_nodes
;
3711 * As the map is walked, we track how much memory is usable
3712 * by the kernel using kernelcore_remaining. When it is
3713 * 0, the rest of the node is usable by ZONE_MOVABLE
3715 kernelcore_remaining
= kernelcore_node
;
3717 /* Go through each range of PFNs within this node */
3718 for_each_active_range_index_in_nid(i
, nid
) {
3719 unsigned long start_pfn
, end_pfn
;
3720 unsigned long size_pages
;
3722 start_pfn
= max(early_node_map
[i
].start_pfn
,
3723 zone_movable_pfn
[nid
]);
3724 end_pfn
= early_node_map
[i
].end_pfn
;
3725 if (start_pfn
>= end_pfn
)
3728 /* Account for what is only usable for kernelcore */
3729 if (start_pfn
< usable_startpfn
) {
3730 unsigned long kernel_pages
;
3731 kernel_pages
= min(end_pfn
, usable_startpfn
)
3734 kernelcore_remaining
-= min(kernel_pages
,
3735 kernelcore_remaining
);
3736 required_kernelcore
-= min(kernel_pages
,
3737 required_kernelcore
);
3739 /* Continue if range is now fully accounted */
3740 if (end_pfn
<= usable_startpfn
) {
3743 * Push zone_movable_pfn to the end so
3744 * that if we have to rebalance
3745 * kernelcore across nodes, we will
3746 * not double account here
3748 zone_movable_pfn
[nid
] = end_pfn
;
3751 start_pfn
= usable_startpfn
;
3755 * The usable PFN range for ZONE_MOVABLE is from
3756 * start_pfn->end_pfn. Calculate size_pages as the
3757 * number of pages used as kernelcore
3759 size_pages
= end_pfn
- start_pfn
;
3760 if (size_pages
> kernelcore_remaining
)
3761 size_pages
= kernelcore_remaining
;
3762 zone_movable_pfn
[nid
] = start_pfn
+ size_pages
;
3765 * Some kernelcore has been met, update counts and
3766 * break if the kernelcore for this node has been
3769 required_kernelcore
-= min(required_kernelcore
,
3771 kernelcore_remaining
-= size_pages
;
3772 if (!kernelcore_remaining
)
3778 * If there is still required_kernelcore, we do another pass with one
3779 * less node in the count. This will push zone_movable_pfn[nid] further
3780 * along on the nodes that still have memory until kernelcore is
3784 if (usable_nodes
&& required_kernelcore
> usable_nodes
)
3787 /* Align start of ZONE_MOVABLE on all nids to MAX_ORDER_NR_PAGES */
3788 for (nid
= 0; nid
< MAX_NUMNODES
; nid
++)
3789 zone_movable_pfn
[nid
] =
3790 roundup(zone_movable_pfn
[nid
], MAX_ORDER_NR_PAGES
);
3793 /* Any regular memory on that node ? */
3794 static void check_for_regular_memory(pg_data_t
*pgdat
)
3796 #ifdef CONFIG_HIGHMEM
3797 enum zone_type zone_type
;
3799 for (zone_type
= 0; zone_type
<= ZONE_NORMAL
; zone_type
++) {
3800 struct zone
*zone
= &pgdat
->node_zones
[zone_type
];
3801 if (zone
->present_pages
)
3802 node_set_state(zone_to_nid(zone
), N_NORMAL_MEMORY
);
3808 * free_area_init_nodes - Initialise all pg_data_t and zone data
3809 * @max_zone_pfn: an array of max PFNs for each zone
3811 * This will call free_area_init_node() for each active node in the system.
3812 * Using the page ranges provided by add_active_range(), the size of each
3813 * zone in each node and their holes is calculated. If the maximum PFN
3814 * between two adjacent zones match, it is assumed that the zone is empty.
3815 * For example, if arch_max_dma_pfn == arch_max_dma32_pfn, it is assumed
3816 * that arch_max_dma32_pfn has no pages. It is also assumed that a zone
3817 * starts where the previous one ended. For example, ZONE_DMA32 starts
3818 * at arch_max_dma_pfn.
3820 void __init
free_area_init_nodes(unsigned long *max_zone_pfn
)
3825 /* Sort early_node_map as initialisation assumes it is sorted */
3828 /* Record where the zone boundaries are */
3829 memset(arch_zone_lowest_possible_pfn
, 0,
3830 sizeof(arch_zone_lowest_possible_pfn
));
3831 memset(arch_zone_highest_possible_pfn
, 0,
3832 sizeof(arch_zone_highest_possible_pfn
));
3833 arch_zone_lowest_possible_pfn
[0] = find_min_pfn_with_active_regions();
3834 arch_zone_highest_possible_pfn
[0] = max_zone_pfn
[0];
3835 for (i
= 1; i
< MAX_NR_ZONES
; i
++) {
3836 if (i
== ZONE_MOVABLE
)
3838 arch_zone_lowest_possible_pfn
[i
] =
3839 arch_zone_highest_possible_pfn
[i
-1];
3840 arch_zone_highest_possible_pfn
[i
] =
3841 max(max_zone_pfn
[i
], arch_zone_lowest_possible_pfn
[i
]);
3843 arch_zone_lowest_possible_pfn
[ZONE_MOVABLE
] = 0;
3844 arch_zone_highest_possible_pfn
[ZONE_MOVABLE
] = 0;
3846 /* Find the PFNs that ZONE_MOVABLE begins at in each node */
3847 memset(zone_movable_pfn
, 0, sizeof(zone_movable_pfn
));
3848 find_zone_movable_pfns_for_nodes(zone_movable_pfn
);
3850 /* Print out the zone ranges */
3851 printk("Zone PFN ranges:\n");
3852 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
3853 if (i
== ZONE_MOVABLE
)
3855 printk(" %-8s %8lu -> %8lu\n",
3857 arch_zone_lowest_possible_pfn
[i
],
3858 arch_zone_highest_possible_pfn
[i
]);
3861 /* Print out the PFNs ZONE_MOVABLE begins at in each node */
3862 printk("Movable zone start PFN for each node\n");
3863 for (i
= 0; i
< MAX_NUMNODES
; i
++) {
3864 if (zone_movable_pfn
[i
])
3865 printk(" Node %d: %lu\n", i
, zone_movable_pfn
[i
]);
3868 /* Print out the early_node_map[] */
3869 printk("early_node_map[%d] active PFN ranges\n", nr_nodemap_entries
);
3870 for (i
= 0; i
< nr_nodemap_entries
; i
++)
3871 printk(" %3d: %8lu -> %8lu\n", early_node_map
[i
].nid
,
3872 early_node_map
[i
].start_pfn
,
3873 early_node_map
[i
].end_pfn
);
3875 /* Initialise every node */
3876 setup_nr_node_ids();
3877 for_each_online_node(nid
) {
3878 pg_data_t
*pgdat
= NODE_DATA(nid
);
3879 free_area_init_node(nid
, pgdat
, NULL
,
3880 find_min_pfn_for_node(nid
), NULL
);
3882 /* Any memory on that node */
3883 if (pgdat
->node_present_pages
)
3884 node_set_state(nid
, N_HIGH_MEMORY
);
3885 check_for_regular_memory(pgdat
);
3889 static int __init
cmdline_parse_core(char *p
, unsigned long *core
)
3891 unsigned long long coremem
;
3895 coremem
= memparse(p
, &p
);
3896 *core
= coremem
>> PAGE_SHIFT
;
3898 /* Paranoid check that UL is enough for the coremem value */
3899 WARN_ON((coremem
>> PAGE_SHIFT
) > ULONG_MAX
);
3905 * kernelcore=size sets the amount of memory for use for allocations that
3906 * cannot be reclaimed or migrated.
3908 static int __init
cmdline_parse_kernelcore(char *p
)
3910 return cmdline_parse_core(p
, &required_kernelcore
);
3914 * movablecore=size sets the amount of memory for use for allocations that
3915 * can be reclaimed or migrated.
3917 static int __init
cmdline_parse_movablecore(char *p
)
3919 return cmdline_parse_core(p
, &required_movablecore
);
3922 early_param("kernelcore", cmdline_parse_kernelcore
);
3923 early_param("movablecore", cmdline_parse_movablecore
);
3925 #endif /* CONFIG_ARCH_POPULATES_NODE_MAP */
3928 * set_dma_reserve - set the specified number of pages reserved in the first zone
3929 * @new_dma_reserve: The number of pages to mark reserved
3931 * The per-cpu batchsize and zone watermarks are determined by present_pages.
3932 * In the DMA zone, a significant percentage may be consumed by kernel image
3933 * and other unfreeable allocations which can skew the watermarks badly. This
3934 * function may optionally be used to account for unfreeable pages in the
3935 * first zone (e.g., ZONE_DMA). The effect will be lower watermarks and
3936 * smaller per-cpu batchsize.
3938 void __init
set_dma_reserve(unsigned long new_dma_reserve
)
3940 dma_reserve
= new_dma_reserve
;
3943 #ifndef CONFIG_NEED_MULTIPLE_NODES
3944 static bootmem_data_t contig_bootmem_data
;
3945 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
3947 EXPORT_SYMBOL(contig_page_data
);
3950 void __init
free_area_init(unsigned long *zones_size
)
3952 free_area_init_node(0, NODE_DATA(0), zones_size
,
3953 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
3956 static int page_alloc_cpu_notify(struct notifier_block
*self
,
3957 unsigned long action
, void *hcpu
)
3959 int cpu
= (unsigned long)hcpu
;
3961 if (action
== CPU_DEAD
|| action
== CPU_DEAD_FROZEN
) {
3962 local_irq_disable();
3964 vm_events_fold_cpu(cpu
);
3966 refresh_cpu_vm_stats(cpu
);
3971 void __init
page_alloc_init(void)
3973 hotcpu_notifier(page_alloc_cpu_notify
, 0);
3977 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
3978 * or min_free_kbytes changes.
3980 static void calculate_totalreserve_pages(void)
3982 struct pglist_data
*pgdat
;
3983 unsigned long reserve_pages
= 0;
3984 enum zone_type i
, j
;
3986 for_each_online_pgdat(pgdat
) {
3987 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
3988 struct zone
*zone
= pgdat
->node_zones
+ i
;
3989 unsigned long max
= 0;
3991 /* Find valid and maximum lowmem_reserve in the zone */
3992 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
3993 if (zone
->lowmem_reserve
[j
] > max
)
3994 max
= zone
->lowmem_reserve
[j
];
3997 /* we treat pages_high as reserved pages. */
3998 max
+= zone
->pages_high
;
4000 if (max
> zone
->present_pages
)
4001 max
= zone
->present_pages
;
4002 reserve_pages
+= max
;
4005 totalreserve_pages
= reserve_pages
;
4009 * setup_per_zone_lowmem_reserve - called whenever
4010 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
4011 * has a correct pages reserved value, so an adequate number of
4012 * pages are left in the zone after a successful __alloc_pages().
4014 static void setup_per_zone_lowmem_reserve(void)
4016 struct pglist_data
*pgdat
;
4017 enum zone_type j
, idx
;
4019 for_each_online_pgdat(pgdat
) {
4020 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
4021 struct zone
*zone
= pgdat
->node_zones
+ j
;
4022 unsigned long present_pages
= zone
->present_pages
;
4024 zone
->lowmem_reserve
[j
] = 0;
4028 struct zone
*lower_zone
;
4032 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
4033 sysctl_lowmem_reserve_ratio
[idx
] = 1;
4035 lower_zone
= pgdat
->node_zones
+ idx
;
4036 lower_zone
->lowmem_reserve
[j
] = present_pages
/
4037 sysctl_lowmem_reserve_ratio
[idx
];
4038 present_pages
+= lower_zone
->present_pages
;
4043 /* update totalreserve_pages */
4044 calculate_totalreserve_pages();
4048 * setup_per_zone_pages_min - called when min_free_kbytes changes.
4050 * Ensures that the pages_{min,low,high} values for each zone are set correctly
4051 * with respect to min_free_kbytes.
4053 void setup_per_zone_pages_min(void)
4055 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
4056 unsigned long lowmem_pages
= 0;
4058 unsigned long flags
;
4060 /* Calculate total number of !ZONE_HIGHMEM pages */
4061 for_each_zone(zone
) {
4062 if (!is_highmem(zone
))
4063 lowmem_pages
+= zone
->present_pages
;
4066 for_each_zone(zone
) {
4069 spin_lock_irqsave(&zone
->lru_lock
, flags
);
4070 tmp
= (u64
)pages_min
* zone
->present_pages
;
4071 do_div(tmp
, lowmem_pages
);
4072 if (is_highmem(zone
)) {
4074 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
4075 * need highmem pages, so cap pages_min to a small
4078 * The (pages_high-pages_low) and (pages_low-pages_min)
4079 * deltas controls asynch page reclaim, and so should
4080 * not be capped for highmem.
4084 min_pages
= zone
->present_pages
/ 1024;
4085 if (min_pages
< SWAP_CLUSTER_MAX
)
4086 min_pages
= SWAP_CLUSTER_MAX
;
4087 if (min_pages
> 128)
4089 zone
->pages_min
= min_pages
;
4092 * If it's a lowmem zone, reserve a number of pages
4093 * proportionate to the zone's size.
4095 zone
->pages_min
= tmp
;
4098 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
4099 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
4100 setup_zone_migrate_reserve(zone
);
4101 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
4104 /* update totalreserve_pages */
4105 calculate_totalreserve_pages();
4109 * Initialise min_free_kbytes.
4111 * For small machines we want it small (128k min). For large machines
4112 * we want it large (64MB max). But it is not linear, because network
4113 * bandwidth does not increase linearly with machine size. We use
4115 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
4116 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
4132 static int __init
init_per_zone_pages_min(void)
4134 unsigned long lowmem_kbytes
;
4136 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
4138 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
4139 if (min_free_kbytes
< 128)
4140 min_free_kbytes
= 128;
4141 if (min_free_kbytes
> 65536)
4142 min_free_kbytes
= 65536;
4143 setup_per_zone_pages_min();
4144 setup_per_zone_lowmem_reserve();
4147 module_init(init_per_zone_pages_min
)
4150 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
4151 * that we can call two helper functions whenever min_free_kbytes
4154 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
4155 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4157 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
4159 setup_per_zone_pages_min();
4164 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
4165 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4170 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4175 zone
->min_unmapped_pages
= (zone
->present_pages
*
4176 sysctl_min_unmapped_ratio
) / 100;
4180 int sysctl_min_slab_ratio_sysctl_handler(ctl_table
*table
, int write
,
4181 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4186 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4191 zone
->min_slab_pages
= (zone
->present_pages
*
4192 sysctl_min_slab_ratio
) / 100;
4198 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
4199 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
4200 * whenever sysctl_lowmem_reserve_ratio changes.
4202 * The reserve ratio obviously has absolutely no relation with the
4203 * pages_min watermarks. The lowmem reserve ratio can only make sense
4204 * if in function of the boot time zone sizes.
4206 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
4207 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4209 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4210 setup_per_zone_lowmem_reserve();
4215 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
4216 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
4217 * can have before it gets flushed back to buddy allocator.
4220 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
4221 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
4227 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
4228 if (!write
|| (ret
== -EINVAL
))
4230 for_each_zone(zone
) {
4231 for_each_online_cpu(cpu
) {
4233 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
4234 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
4240 int hashdist
= HASHDIST_DEFAULT
;
4243 static int __init
set_hashdist(char *str
)
4247 hashdist
= simple_strtoul(str
, &str
, 0);
4250 __setup("hashdist=", set_hashdist
);
4254 * allocate a large system hash table from bootmem
4255 * - it is assumed that the hash table must contain an exact power-of-2
4256 * quantity of entries
4257 * - limit is the number of hash buckets, not the total allocation size
4259 void *__init
alloc_large_system_hash(const char *tablename
,
4260 unsigned long bucketsize
,
4261 unsigned long numentries
,
4264 unsigned int *_hash_shift
,
4265 unsigned int *_hash_mask
,
4266 unsigned long limit
)
4268 unsigned long long max
= limit
;
4269 unsigned long log2qty
, size
;
4272 /* allow the kernel cmdline to have a say */
4274 /* round applicable memory size up to nearest megabyte */
4275 numentries
= nr_kernel_pages
;
4276 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
4277 numentries
>>= 20 - PAGE_SHIFT
;
4278 numentries
<<= 20 - PAGE_SHIFT
;
4280 /* limit to 1 bucket per 2^scale bytes of low memory */
4281 if (scale
> PAGE_SHIFT
)
4282 numentries
>>= (scale
- PAGE_SHIFT
);
4284 numentries
<<= (PAGE_SHIFT
- scale
);
4286 /* Make sure we've got at least a 0-order allocation.. */
4287 if (unlikely((numentries
* bucketsize
) < PAGE_SIZE
))
4288 numentries
= PAGE_SIZE
/ bucketsize
;
4290 numentries
= roundup_pow_of_two(numentries
);
4292 /* limit allocation size to 1/16 total memory by default */
4294 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
4295 do_div(max
, bucketsize
);
4298 if (numentries
> max
)
4301 log2qty
= ilog2(numentries
);
4304 size
= bucketsize
<< log2qty
;
4305 if (flags
& HASH_EARLY
)
4306 table
= alloc_bootmem(size
);
4308 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
4310 unsigned long order
;
4311 for (order
= 0; ((1UL << order
) << PAGE_SHIFT
) < size
; order
++)
4313 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
4315 * If bucketsize is not a power-of-two, we may free
4316 * some pages at the end of hash table.
4319 unsigned long alloc_end
= (unsigned long)table
+
4320 (PAGE_SIZE
<< order
);
4321 unsigned long used
= (unsigned long)table
+
4323 split_page(virt_to_page(table
), order
);
4324 while (used
< alloc_end
) {
4330 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
4333 panic("Failed to allocate %s hash table\n", tablename
);
4335 printk(KERN_INFO
"%s hash table entries: %d (order: %d, %lu bytes)\n",
4338 ilog2(size
) - PAGE_SHIFT
,
4342 *_hash_shift
= log2qty
;
4344 *_hash_mask
= (1 << log2qty
) - 1;
4349 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
4350 struct page
*pfn_to_page(unsigned long pfn
)
4352 return __pfn_to_page(pfn
);
4354 unsigned long page_to_pfn(struct page
*page
)
4356 return __page_to_pfn(page
);
4358 EXPORT_SYMBOL(pfn_to_page
);
4359 EXPORT_SYMBOL(page_to_pfn
);
4360 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */
4362 /* Return a pointer to the bitmap storing bits affecting a block of pages */
4363 static inline unsigned long *get_pageblock_bitmap(struct zone
*zone
,
4366 #ifdef CONFIG_SPARSEMEM
4367 return __pfn_to_section(pfn
)->pageblock_flags
;
4369 return zone
->pageblock_flags
;
4370 #endif /* CONFIG_SPARSEMEM */
4373 static inline int pfn_to_bitidx(struct zone
*zone
, unsigned long pfn
)
4375 #ifdef CONFIG_SPARSEMEM
4376 pfn
&= (PAGES_PER_SECTION
-1);
4377 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4379 pfn
= pfn
- zone
->zone_start_pfn
;
4380 return (pfn
>> pageblock_order
) * NR_PAGEBLOCK_BITS
;
4381 #endif /* CONFIG_SPARSEMEM */
4385 * get_pageblock_flags_group - Return the requested group of flags for the pageblock_nr_pages block of pages
4386 * @page: The page within the block of interest
4387 * @start_bitidx: The first bit of interest to retrieve
4388 * @end_bitidx: The last bit of interest
4389 * returns pageblock_bits flags
4391 unsigned long get_pageblock_flags_group(struct page
*page
,
4392 int start_bitidx
, int end_bitidx
)
4395 unsigned long *bitmap
;
4396 unsigned long pfn
, bitidx
;
4397 unsigned long flags
= 0;
4398 unsigned long value
= 1;
4400 zone
= page_zone(page
);
4401 pfn
= page_to_pfn(page
);
4402 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4403 bitidx
= pfn_to_bitidx(zone
, pfn
);
4405 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4406 if (test_bit(bitidx
+ start_bitidx
, bitmap
))
4413 * set_pageblock_flags_group - Set the requested group of flags for a pageblock_nr_pages block of pages
4414 * @page: The page within the block of interest
4415 * @start_bitidx: The first bit of interest
4416 * @end_bitidx: The last bit of interest
4417 * @flags: The flags to set
4419 void set_pageblock_flags_group(struct page
*page
, unsigned long flags
,
4420 int start_bitidx
, int end_bitidx
)
4423 unsigned long *bitmap
;
4424 unsigned long pfn
, bitidx
;
4425 unsigned long value
= 1;
4427 zone
= page_zone(page
);
4428 pfn
= page_to_pfn(page
);
4429 bitmap
= get_pageblock_bitmap(zone
, pfn
);
4430 bitidx
= pfn_to_bitidx(zone
, pfn
);
4432 for (; start_bitidx
<= end_bitidx
; start_bitidx
++, value
<<= 1)
4434 __set_bit(bitidx
+ start_bitidx
, bitmap
);
4436 __clear_bit(bitidx
+ start_bitidx
, bitmap
);
4440 * This is designed as sub function...plz see page_isolation.c also.
4441 * set/clear page block's type to be ISOLATE.
4442 * page allocater never alloc memory from ISOLATE block.
4445 int set_migratetype_isolate(struct page
*page
)
4448 unsigned long flags
;
4451 zone
= page_zone(page
);
4452 spin_lock_irqsave(&zone
->lock
, flags
);
4454 * In future, more migrate types will be able to be isolation target.
4456 if (get_pageblock_migratetype(page
) != MIGRATE_MOVABLE
)
4458 set_pageblock_migratetype(page
, MIGRATE_ISOLATE
);
4459 move_freepages_block(zone
, page
, MIGRATE_ISOLATE
);
4462 spin_unlock_irqrestore(&zone
->lock
, flags
);
4464 drain_all_local_pages();
4468 void unset_migratetype_isolate(struct page
*page
)
4471 unsigned long flags
;
4472 zone
= page_zone(page
);
4473 spin_lock_irqsave(&zone
->lock
, flags
);
4474 if (get_pageblock_migratetype(page
) != MIGRATE_ISOLATE
)
4476 set_pageblock_migratetype(page
, MIGRATE_MOVABLE
);
4477 move_freepages_block(zone
, page
, MIGRATE_MOVABLE
);
4479 spin_unlock_irqrestore(&zone
->lock
, flags
);
4482 #ifdef CONFIG_MEMORY_HOTREMOVE
4484 * All pages in the range must be isolated before calling this.
4487 __offline_isolated_pages(unsigned long start_pfn
, unsigned long end_pfn
)
4493 unsigned long flags
;
4494 /* find the first valid pfn */
4495 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++)
4500 zone
= page_zone(pfn_to_page(pfn
));
4501 spin_lock_irqsave(&zone
->lock
, flags
);
4503 while (pfn
< end_pfn
) {
4504 if (!pfn_valid(pfn
)) {
4508 page
= pfn_to_page(pfn
);
4509 BUG_ON(page_count(page
));
4510 BUG_ON(!PageBuddy(page
));
4511 order
= page_order(page
);
4512 #ifdef CONFIG_DEBUG_VM
4513 printk(KERN_INFO
"remove from free list %lx %d %lx\n",
4514 pfn
, 1 << order
, end_pfn
);
4516 list_del(&page
->lru
);
4517 rmv_page_order(page
);
4518 zone
->free_area
[order
].nr_free
--;
4519 __mod_zone_page_state(zone
, NR_FREE_PAGES
,
4521 for (i
= 0; i
< (1 << order
); i
++)
4522 SetPageReserved((page
+i
));
4523 pfn
+= (1 << order
);
4525 spin_unlock_irqrestore(&zone
->lock
, flags
);