2 * linux/mm/page_alloc.c
4 * Manages the free list, the system allocates free pages here.
5 * Note that kmalloc() lives in slab.c
7 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
8 * Swap reorganised 29.12.95, Stephen Tweedie
9 * Support of BIGMEM added by Gerhard Wichert, Siemens AG, July 1999
10 * Reshaped it to be a zoned allocator, Ingo Molnar, Red Hat, 1999
11 * Discontiguous memory support, Kanoj Sarcar, SGI, Nov 1999
12 * Zone balancing, Kanoj Sarcar, SGI, Jan 2000
13 * Per cpu hot/cold page lists, bulk allocation, Martin J. Bligh, Sept 2002
14 * (lots of bits borrowed from Ingo Molnar & Andrew Morton)
17 #include <linux/stddef.h>
19 #include <linux/swap.h>
20 #include <linux/interrupt.h>
21 #include <linux/pagemap.h>
22 #include <linux/bootmem.h>
23 #include <linux/compiler.h>
24 #include <linux/kernel.h>
25 #include <linux/module.h>
26 #include <linux/suspend.h>
27 #include <linux/pagevec.h>
28 #include <linux/blkdev.h>
29 #include <linux/slab.h>
30 #include <linux/notifier.h>
31 #include <linux/topology.h>
32 #include <linux/sysctl.h>
33 #include <linux/cpu.h>
34 #include <linux/cpuset.h>
35 #include <linux/memory_hotplug.h>
36 #include <linux/nodemask.h>
37 #include <linux/vmalloc.h>
38 #include <linux/mempolicy.h>
39 #include <linux/stop_machine.h>
41 #include <asm/tlbflush.h>
42 #include <asm/div64.h>
46 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
49 nodemask_t node_online_map __read_mostly
= { { [0] = 1UL } };
50 EXPORT_SYMBOL(node_online_map
);
51 nodemask_t node_possible_map __read_mostly
= NODE_MASK_ALL
;
52 EXPORT_SYMBOL(node_possible_map
);
53 unsigned long totalram_pages __read_mostly
;
54 unsigned long totalreserve_pages __read_mostly
;
56 int percpu_pagelist_fraction
;
58 static void __free_pages_ok(struct page
*page
, unsigned int order
);
61 * results with 256, 32 in the lowmem_reserve sysctl:
62 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
63 * 1G machine -> (16M dma, 784M normal, 224M high)
64 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
65 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
66 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
68 * TBD: should special case ZONE_DMA32 machines here - in those we normally
69 * don't need any ZONE_NORMAL reservation
71 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = {
73 #ifdef CONFIG_ZONE_DMA32
79 EXPORT_SYMBOL(totalram_pages
);
82 * Used by page_zone() to look up the address of the struct zone whose
83 * id is encoded in the upper bits of page->flags
85 struct zone
*zone_table
[1 << ZONETABLE_SHIFT
] __read_mostly
;
86 EXPORT_SYMBOL(zone_table
);
88 static char *zone_names
[MAX_NR_ZONES
] = {
90 #ifdef CONFIG_ZONE_DMA32
97 int min_free_kbytes
= 1024;
99 unsigned long __meminitdata nr_kernel_pages
;
100 unsigned long __meminitdata nr_all_pages
;
102 #ifdef CONFIG_DEBUG_VM
103 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
107 unsigned long pfn
= page_to_pfn(page
);
110 seq
= zone_span_seqbegin(zone
);
111 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
113 else if (pfn
< zone
->zone_start_pfn
)
115 } while (zone_span_seqretry(zone
, seq
));
120 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
122 #ifdef CONFIG_HOLES_IN_ZONE
123 if (!pfn_valid(page_to_pfn(page
)))
126 if (zone
!= page_zone(page
))
132 * Temporary debugging check for pages not lying within a given zone.
134 static int bad_range(struct zone
*zone
, struct page
*page
)
136 if (page_outside_zone_boundaries(zone
, page
))
138 if (!page_is_consistent(zone
, page
))
144 static inline int bad_range(struct zone
*zone
, struct page
*page
)
150 static void bad_page(struct page
*page
)
152 printk(KERN_EMERG
"Bad page state in process '%s'\n"
153 KERN_EMERG
"page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
154 KERN_EMERG
"Trying to fix it up, but a reboot is needed\n"
155 KERN_EMERG
"Backtrace:\n",
156 current
->comm
, page
, (int)(2*sizeof(unsigned long)),
157 (unsigned long)page
->flags
, page
->mapping
,
158 page_mapcount(page
), page_count(page
));
160 page
->flags
&= ~(1 << PG_lru
|
170 set_page_count(page
, 0);
171 reset_page_mapcount(page
);
172 page
->mapping
= NULL
;
173 add_taint(TAINT_BAD_PAGE
);
177 * Higher-order pages are called "compound pages". They are structured thusly:
179 * The first PAGE_SIZE page is called the "head page".
181 * The remaining PAGE_SIZE pages are called "tail pages".
183 * All pages have PG_compound set. All pages have their ->private pointing at
184 * the head page (even the head page has this).
186 * The first tail page's ->lru.next holds the address of the compound page's
187 * put_page() function. Its ->lru.prev holds the order of allocation.
188 * This usage means that zero-order pages may not be compound.
191 static void free_compound_page(struct page
*page
)
193 __free_pages_ok(page
, (unsigned long)page
[1].lru
.prev
);
196 static void prep_compound_page(struct page
*page
, unsigned long order
)
199 int nr_pages
= 1 << order
;
201 page
[1].lru
.next
= (void *)free_compound_page
; /* set dtor */
202 page
[1].lru
.prev
= (void *)order
;
203 for (i
= 0; i
< nr_pages
; i
++) {
204 struct page
*p
= page
+ i
;
206 __SetPageCompound(p
);
207 set_page_private(p
, (unsigned long)page
);
211 static void destroy_compound_page(struct page
*page
, unsigned long order
)
214 int nr_pages
= 1 << order
;
216 if (unlikely((unsigned long)page
[1].lru
.prev
!= order
))
219 for (i
= 0; i
< nr_pages
; i
++) {
220 struct page
*p
= page
+ i
;
222 if (unlikely(!PageCompound(p
) |
223 (page_private(p
) != (unsigned long)page
)))
225 __ClearPageCompound(p
);
229 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
233 VM_BUG_ON((gfp_flags
& (__GFP_WAIT
| __GFP_HIGHMEM
)) == __GFP_HIGHMEM
);
235 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
236 * and __GFP_HIGHMEM from hard or soft interrupt context.
238 VM_BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
239 for (i
= 0; i
< (1 << order
); i
++)
240 clear_highpage(page
+ i
);
244 * function for dealing with page's order in buddy system.
245 * zone->lock is already acquired when we use these.
246 * So, we don't need atomic page->flags operations here.
248 static inline unsigned long page_order(struct page
*page
)
250 return page_private(page
);
253 static inline void set_page_order(struct page
*page
, int order
)
255 set_page_private(page
, order
);
256 __SetPageBuddy(page
);
259 static inline void rmv_page_order(struct page
*page
)
261 __ClearPageBuddy(page
);
262 set_page_private(page
, 0);
266 * Locate the struct page for both the matching buddy in our
267 * pair (buddy1) and the combined O(n+1) page they form (page).
269 * 1) Any buddy B1 will have an order O twin B2 which satisfies
270 * the following equation:
272 * For example, if the starting buddy (buddy2) is #8 its order
274 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
276 * 2) Any buddy B will have an order O+1 parent P which
277 * satisfies the following equation:
280 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
282 static inline struct page
*
283 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
285 unsigned long buddy_idx
= page_idx
^ (1 << order
);
287 return page
+ (buddy_idx
- page_idx
);
290 static inline unsigned long
291 __find_combined_index(unsigned long page_idx
, unsigned int order
)
293 return (page_idx
& ~(1 << order
));
297 * This function checks whether a page is free && is the buddy
298 * we can do coalesce a page and its buddy if
299 * (a) the buddy is not in a hole &&
300 * (b) the buddy is in the buddy system &&
301 * (c) a page and its buddy have the same order &&
302 * (d) a page and its buddy are in the same zone.
304 * For recording whether a page is in the buddy system, we use PG_buddy.
305 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
307 * For recording page's order, we use page_private(page).
309 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
312 #ifdef CONFIG_HOLES_IN_ZONE
313 if (!pfn_valid(page_to_pfn(buddy
)))
317 if (page_zone_id(page
) != page_zone_id(buddy
))
320 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
321 BUG_ON(page_count(buddy
) != 0);
328 * Freeing function for a buddy system allocator.
330 * The concept of a buddy system is to maintain direct-mapped table
331 * (containing bit values) for memory blocks of various "orders".
332 * The bottom level table contains the map for the smallest allocatable
333 * units of memory (here, pages), and each level above it describes
334 * pairs of units from the levels below, hence, "buddies".
335 * At a high level, all that happens here is marking the table entry
336 * at the bottom level available, and propagating the changes upward
337 * as necessary, plus some accounting needed to play nicely with other
338 * parts of the VM system.
339 * At each level, we keep a list of pages, which are heads of continuous
340 * free pages of length of (1 << order) and marked with PG_buddy. Page's
341 * order is recorded in page_private(page) field.
342 * So when we are allocating or freeing one, we can derive the state of the
343 * other. That is, if we allocate a small block, and both were
344 * free, the remainder of the region must be split into blocks.
345 * If a block is freed, and its buddy is also free, then this
346 * triggers coalescing into a block of larger size.
351 static inline void __free_one_page(struct page
*page
,
352 struct zone
*zone
, unsigned int order
)
354 unsigned long page_idx
;
355 int order_size
= 1 << order
;
357 if (unlikely(PageCompound(page
)))
358 destroy_compound_page(page
, order
);
360 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
362 VM_BUG_ON(page_idx
& (order_size
- 1));
363 VM_BUG_ON(bad_range(zone
, page
));
365 zone
->free_pages
+= order_size
;
366 while (order
< MAX_ORDER
-1) {
367 unsigned long combined_idx
;
368 struct free_area
*area
;
371 buddy
= __page_find_buddy(page
, page_idx
, order
);
372 if (!page_is_buddy(page
, buddy
, order
))
373 break; /* Move the buddy up one level. */
375 list_del(&buddy
->lru
);
376 area
= zone
->free_area
+ order
;
378 rmv_page_order(buddy
);
379 combined_idx
= __find_combined_index(page_idx
, order
);
380 page
= page
+ (combined_idx
- page_idx
);
381 page_idx
= combined_idx
;
384 set_page_order(page
, order
);
385 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
);
386 zone
->free_area
[order
].nr_free
++;
389 static inline int free_pages_check(struct page
*page
)
391 if (unlikely(page_mapcount(page
) |
392 (page
->mapping
!= NULL
) |
393 (page_count(page
) != 0) |
407 __ClearPageDirty(page
);
409 * For now, we report if PG_reserved was found set, but do not
410 * clear it, and do not free the page. But we shall soon need
411 * to do more, for when the ZERO_PAGE count wraps negative.
413 return PageReserved(page
);
417 * Frees a list of pages.
418 * Assumes all pages on list are in same zone, and of same order.
419 * count is the number of pages to free.
421 * If the zone was previously in an "all pages pinned" state then look to
422 * see if this freeing clears that state.
424 * And clear the zone's pages_scanned counter, to hold off the "all pages are
425 * pinned" detection logic.
427 static void free_pages_bulk(struct zone
*zone
, int count
,
428 struct list_head
*list
, int order
)
430 spin_lock(&zone
->lock
);
431 zone
->all_unreclaimable
= 0;
432 zone
->pages_scanned
= 0;
436 VM_BUG_ON(list_empty(list
));
437 page
= list_entry(list
->prev
, struct page
, lru
);
438 /* have to delete it as __free_one_page list manipulates */
439 list_del(&page
->lru
);
440 __free_one_page(page
, zone
, order
);
442 spin_unlock(&zone
->lock
);
445 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
448 list_add(&page
->lru
, &list
);
449 free_pages_bulk(zone
, 1, &list
, order
);
452 static void __free_pages_ok(struct page
*page
, unsigned int order
)
458 arch_free_page(page
, order
);
459 if (!PageHighMem(page
))
460 debug_check_no_locks_freed(page_address(page
),
463 for (i
= 0 ; i
< (1 << order
) ; ++i
)
464 reserved
+= free_pages_check(page
+ i
);
468 kernel_map_pages(page
, 1 << order
, 0);
469 local_irq_save(flags
);
470 __count_vm_events(PGFREE
, 1 << order
);
471 free_one_page(page_zone(page
), page
, order
);
472 local_irq_restore(flags
);
476 * permit the bootmem allocator to evade page validation on high-order frees
478 void fastcall __init
__free_pages_bootmem(struct page
*page
, unsigned int order
)
481 __ClearPageReserved(page
);
482 set_page_count(page
, 0);
483 set_page_refcounted(page
);
489 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
490 struct page
*p
= &page
[loop
];
492 if (loop
+ 1 < BITS_PER_LONG
)
494 __ClearPageReserved(p
);
495 set_page_count(p
, 0);
498 set_page_refcounted(page
);
499 __free_pages(page
, order
);
505 * The order of subdivision here is critical for the IO subsystem.
506 * Please do not alter this order without good reasons and regression
507 * testing. Specifically, as large blocks of memory are subdivided,
508 * the order in which smaller blocks are delivered depends on the order
509 * they're subdivided in this function. This is the primary factor
510 * influencing the order in which pages are delivered to the IO
511 * subsystem according to empirical testing, and this is also justified
512 * by considering the behavior of a buddy system containing a single
513 * large block of memory acted on by a series of small allocations.
514 * This behavior is a critical factor in sglist merging's success.
518 static inline void expand(struct zone
*zone
, struct page
*page
,
519 int low
, int high
, struct free_area
*area
)
521 unsigned long size
= 1 << high
;
527 VM_BUG_ON(bad_range(zone
, &page
[size
]));
528 list_add(&page
[size
].lru
, &area
->free_list
);
530 set_page_order(&page
[size
], high
);
535 * This page is about to be returned from the page allocator
537 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
539 if (unlikely(page_mapcount(page
) |
540 (page
->mapping
!= NULL
) |
541 (page_count(page
) != 0) |
557 * For now, we report if PG_reserved was found set, but do not
558 * clear it, and do not allocate the page: as a safety net.
560 if (PageReserved(page
))
563 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
|
564 1 << PG_referenced
| 1 << PG_arch_1
|
565 1 << PG_checked
| 1 << PG_mappedtodisk
);
566 set_page_private(page
, 0);
567 set_page_refcounted(page
);
568 kernel_map_pages(page
, 1 << order
, 1);
570 if (gfp_flags
& __GFP_ZERO
)
571 prep_zero_page(page
, order
, gfp_flags
);
573 if (order
&& (gfp_flags
& __GFP_COMP
))
574 prep_compound_page(page
, order
);
580 * Do the hard work of removing an element from the buddy allocator.
581 * Call me with the zone->lock already held.
583 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
)
585 struct free_area
* area
;
586 unsigned int current_order
;
589 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
590 area
= zone
->free_area
+ current_order
;
591 if (list_empty(&area
->free_list
))
594 page
= list_entry(area
->free_list
.next
, struct page
, lru
);
595 list_del(&page
->lru
);
596 rmv_page_order(page
);
598 zone
->free_pages
-= 1UL << order
;
599 expand(zone
, page
, order
, current_order
, area
);
607 * Obtain a specified number of elements from the buddy allocator, all under
608 * a single hold of the lock, for efficiency. Add them to the supplied list.
609 * Returns the number of new pages which were placed at *list.
611 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
612 unsigned long count
, struct list_head
*list
)
616 spin_lock(&zone
->lock
);
617 for (i
= 0; i
< count
; ++i
) {
618 struct page
*page
= __rmqueue(zone
, order
);
619 if (unlikely(page
== NULL
))
621 list_add_tail(&page
->lru
, list
);
623 spin_unlock(&zone
->lock
);
629 * Called from the slab reaper to drain pagesets on a particular node that
630 * belong to the currently executing processor.
631 * Note that this function must be called with the thread pinned to
632 * a single processor.
634 void drain_node_pages(int nodeid
)
639 for (z
= 0; z
< MAX_NR_ZONES
; z
++) {
640 struct zone
*zone
= NODE_DATA(nodeid
)->node_zones
+ z
;
641 struct per_cpu_pageset
*pset
;
643 pset
= zone_pcp(zone
, smp_processor_id());
644 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
645 struct per_cpu_pages
*pcp
;
649 local_irq_save(flags
);
650 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
652 local_irq_restore(flags
);
659 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
660 static void __drain_pages(unsigned int cpu
)
666 for_each_zone(zone
) {
667 struct per_cpu_pageset
*pset
;
669 pset
= zone_pcp(zone
, cpu
);
670 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
671 struct per_cpu_pages
*pcp
;
674 local_irq_save(flags
);
675 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
677 local_irq_restore(flags
);
681 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
685 void mark_free_pages(struct zone
*zone
)
687 unsigned long zone_pfn
, flags
;
689 struct list_head
*curr
;
691 if (!zone
->spanned_pages
)
694 spin_lock_irqsave(&zone
->lock
, flags
);
695 for (zone_pfn
= 0; zone_pfn
< zone
->spanned_pages
; ++zone_pfn
)
696 ClearPageNosaveFree(pfn_to_page(zone_pfn
+ zone
->zone_start_pfn
));
698 for (order
= MAX_ORDER
- 1; order
>= 0; --order
)
699 list_for_each(curr
, &zone
->free_area
[order
].free_list
) {
700 unsigned long start_pfn
, i
;
702 start_pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
704 for (i
=0; i
< (1<<order
); i
++)
705 SetPageNosaveFree(pfn_to_page(start_pfn
+i
));
707 spin_unlock_irqrestore(&zone
->lock
, flags
);
711 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
713 void drain_local_pages(void)
717 local_irq_save(flags
);
718 __drain_pages(smp_processor_id());
719 local_irq_restore(flags
);
721 #endif /* CONFIG_PM */
724 * Free a 0-order page
726 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
)
728 struct zone
*zone
= page_zone(page
);
729 struct per_cpu_pages
*pcp
;
732 arch_free_page(page
, 0);
735 page
->mapping
= NULL
;
736 if (free_pages_check(page
))
739 kernel_map_pages(page
, 1, 0);
741 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
742 local_irq_save(flags
);
743 __count_vm_event(PGFREE
);
744 list_add(&page
->lru
, &pcp
->list
);
746 if (pcp
->count
>= pcp
->high
) {
747 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
748 pcp
->count
-= pcp
->batch
;
750 local_irq_restore(flags
);
754 void fastcall
free_hot_page(struct page
*page
)
756 free_hot_cold_page(page
, 0);
759 void fastcall
free_cold_page(struct page
*page
)
761 free_hot_cold_page(page
, 1);
765 * split_page takes a non-compound higher-order page, and splits it into
766 * n (1<<order) sub-pages: page[0..n]
767 * Each sub-page must be freed individually.
769 * Note: this is probably too low level an operation for use in drivers.
770 * Please consult with lkml before using this in your driver.
772 void split_page(struct page
*page
, unsigned int order
)
776 VM_BUG_ON(PageCompound(page
));
777 VM_BUG_ON(!page_count(page
));
778 for (i
= 1; i
< (1 << order
); i
++)
779 set_page_refcounted(page
+ i
);
783 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
784 * we cheat by calling it from here, in the order > 0 path. Saves a branch
787 static struct page
*buffered_rmqueue(struct zonelist
*zonelist
,
788 struct zone
*zone
, int order
, gfp_t gfp_flags
)
792 int cold
= !!(gfp_flags
& __GFP_COLD
);
797 if (likely(order
== 0)) {
798 struct per_cpu_pages
*pcp
;
800 pcp
= &zone_pcp(zone
, cpu
)->pcp
[cold
];
801 local_irq_save(flags
);
803 pcp
->count
+= rmqueue_bulk(zone
, 0,
804 pcp
->batch
, &pcp
->list
);
805 if (unlikely(!pcp
->count
))
808 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
809 list_del(&page
->lru
);
812 spin_lock_irqsave(&zone
->lock
, flags
);
813 page
= __rmqueue(zone
, order
);
814 spin_unlock(&zone
->lock
);
819 __count_zone_vm_events(PGALLOC
, zone
, 1 << order
);
820 zone_statistics(zonelist
, zone
);
821 local_irq_restore(flags
);
824 VM_BUG_ON(bad_range(zone
, page
));
825 if (prep_new_page(page
, order
, gfp_flags
))
830 local_irq_restore(flags
);
835 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
836 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
837 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
838 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
839 #define ALLOC_HARDER 0x10 /* try to alloc harder */
840 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
841 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
844 * Return 1 if free pages are above 'mark'. This takes into account the order
847 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
848 int classzone_idx
, int alloc_flags
)
850 /* free_pages my go negative - that's OK */
851 long min
= mark
, free_pages
= z
->free_pages
- (1 << order
) + 1;
854 if (alloc_flags
& ALLOC_HIGH
)
856 if (alloc_flags
& ALLOC_HARDER
)
859 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
861 for (o
= 0; o
< order
; o
++) {
862 /* At the next order, this order's pages become unavailable */
863 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
865 /* Require fewer higher order pages to be free */
868 if (free_pages
<= min
)
875 * get_page_from_freeliest goes through the zonelist trying to allocate
879 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
,
880 struct zonelist
*zonelist
, int alloc_flags
)
882 struct zone
**z
= zonelist
->zones
;
883 struct page
*page
= NULL
;
884 int classzone_idx
= zone_idx(*z
);
887 * Go through the zonelist once, looking for a zone with enough free.
888 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
891 if ((alloc_flags
& ALLOC_CPUSET
) &&
892 !cpuset_zone_allowed(*z
, gfp_mask
))
895 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
897 if (alloc_flags
& ALLOC_WMARK_MIN
)
898 mark
= (*z
)->pages_min
;
899 else if (alloc_flags
& ALLOC_WMARK_LOW
)
900 mark
= (*z
)->pages_low
;
902 mark
= (*z
)->pages_high
;
903 if (!zone_watermark_ok(*z
, order
, mark
,
904 classzone_idx
, alloc_flags
))
905 if (!zone_reclaim_mode
||
906 !zone_reclaim(*z
, gfp_mask
, order
))
910 page
= buffered_rmqueue(zonelist
, *z
, order
, gfp_mask
);
914 } while (*(++z
) != NULL
);
919 * This is the 'heart' of the zoned buddy allocator.
921 struct page
* fastcall
922 __alloc_pages(gfp_t gfp_mask
, unsigned int order
,
923 struct zonelist
*zonelist
)
925 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
928 struct reclaim_state reclaim_state
;
929 struct task_struct
*p
= current
;
932 int did_some_progress
;
934 might_sleep_if(wait
);
937 z
= zonelist
->zones
; /* the list of zones suitable for gfp_mask */
939 if (unlikely(*z
== NULL
)) {
940 /* Should this ever happen?? */
944 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
945 zonelist
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
950 wakeup_kswapd(*z
, order
);
954 * OK, we're below the kswapd watermark and have kicked background
955 * reclaim. Now things get more complex, so set up alloc_flags according
956 * to how we want to proceed.
958 * The caller may dip into page reserves a bit more if the caller
959 * cannot run direct reclaim, or if the caller has realtime scheduling
960 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
961 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
963 alloc_flags
= ALLOC_WMARK_MIN
;
964 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
965 alloc_flags
|= ALLOC_HARDER
;
966 if (gfp_mask
& __GFP_HIGH
)
967 alloc_flags
|= ALLOC_HIGH
;
969 alloc_flags
|= ALLOC_CPUSET
;
972 * Go through the zonelist again. Let __GFP_HIGH and allocations
973 * coming from realtime tasks go deeper into reserves.
975 * This is the last chance, in general, before the goto nopage.
976 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
977 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
979 page
= get_page_from_freelist(gfp_mask
, order
, zonelist
, alloc_flags
);
983 /* This allocation should allow future memory freeing. */
985 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
986 && !in_interrupt()) {
987 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
989 /* go through the zonelist yet again, ignoring mins */
990 page
= get_page_from_freelist(gfp_mask
, order
,
991 zonelist
, ALLOC_NO_WATERMARKS
);
994 if (gfp_mask
& __GFP_NOFAIL
) {
995 blk_congestion_wait(WRITE
, HZ
/50);
1002 /* Atomic allocations - we can't balance anything */
1009 /* We now go into synchronous reclaim */
1010 cpuset_memory_pressure_bump();
1011 p
->flags
|= PF_MEMALLOC
;
1012 reclaim_state
.reclaimed_slab
= 0;
1013 p
->reclaim_state
= &reclaim_state
;
1015 did_some_progress
= try_to_free_pages(zonelist
->zones
, gfp_mask
);
1017 p
->reclaim_state
= NULL
;
1018 p
->flags
&= ~PF_MEMALLOC
;
1022 if (likely(did_some_progress
)) {
1023 page
= get_page_from_freelist(gfp_mask
, order
,
1024 zonelist
, alloc_flags
);
1027 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1029 * Go through the zonelist yet one more time, keep
1030 * very high watermark here, this is only to catch
1031 * a parallel oom killing, we must fail if we're still
1032 * under heavy pressure.
1034 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1035 zonelist
, ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1039 out_of_memory(zonelist
, gfp_mask
, order
);
1044 * Don't let big-order allocations loop unless the caller explicitly
1045 * requests that. Wait for some write requests to complete then retry.
1047 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1048 * <= 3, but that may not be true in other implementations.
1051 if (!(gfp_mask
& __GFP_NORETRY
)) {
1052 if ((order
<= 3) || (gfp_mask
& __GFP_REPEAT
))
1054 if (gfp_mask
& __GFP_NOFAIL
)
1058 blk_congestion_wait(WRITE
, HZ
/50);
1063 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1064 printk(KERN_WARNING
"%s: page allocation failure."
1065 " order:%d, mode:0x%x\n",
1066 p
->comm
, order
, gfp_mask
);
1074 EXPORT_SYMBOL(__alloc_pages
);
1077 * Common helper functions.
1079 fastcall
unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1082 page
= alloc_pages(gfp_mask
, order
);
1085 return (unsigned long) page_address(page
);
1088 EXPORT_SYMBOL(__get_free_pages
);
1090 fastcall
unsigned long get_zeroed_page(gfp_t gfp_mask
)
1095 * get_zeroed_page() returns a 32-bit address, which cannot represent
1098 VM_BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1100 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1102 return (unsigned long) page_address(page
);
1106 EXPORT_SYMBOL(get_zeroed_page
);
1108 void __pagevec_free(struct pagevec
*pvec
)
1110 int i
= pagevec_count(pvec
);
1113 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1116 fastcall
void __free_pages(struct page
*page
, unsigned int order
)
1118 if (put_page_testzero(page
)) {
1120 free_hot_page(page
);
1122 __free_pages_ok(page
, order
);
1126 EXPORT_SYMBOL(__free_pages
);
1128 fastcall
void free_pages(unsigned long addr
, unsigned int order
)
1131 VM_BUG_ON(!virt_addr_valid((void *)addr
));
1132 __free_pages(virt_to_page((void *)addr
), order
);
1136 EXPORT_SYMBOL(free_pages
);
1139 * Total amount of free (allocatable) RAM:
1141 unsigned int nr_free_pages(void)
1143 unsigned int sum
= 0;
1147 sum
+= zone
->free_pages
;
1152 EXPORT_SYMBOL(nr_free_pages
);
1155 unsigned int nr_free_pages_pgdat(pg_data_t
*pgdat
)
1157 unsigned int i
, sum
= 0;
1159 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1160 sum
+= pgdat
->node_zones
[i
].free_pages
;
1166 static unsigned int nr_free_zone_pages(int offset
)
1168 /* Just pick one node, since fallback list is circular */
1169 pg_data_t
*pgdat
= NODE_DATA(numa_node_id());
1170 unsigned int sum
= 0;
1172 struct zonelist
*zonelist
= pgdat
->node_zonelists
+ offset
;
1173 struct zone
**zonep
= zonelist
->zones
;
1176 for (zone
= *zonep
++; zone
; zone
= *zonep
++) {
1177 unsigned long size
= zone
->present_pages
;
1178 unsigned long high
= zone
->pages_high
;
1187 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1189 unsigned int nr_free_buffer_pages(void)
1191 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1195 * Amount of free RAM allocatable within all zones
1197 unsigned int nr_free_pagecache_pages(void)
1199 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER
));
1202 static void show_node(struct zone
*zone
)
1204 printk("Node %d ", zone
->zone_pgdat
->node_id
);
1207 #define show_node(zone) do { } while (0)
1210 void si_meminfo(struct sysinfo
*val
)
1212 val
->totalram
= totalram_pages
;
1214 val
->freeram
= nr_free_pages();
1215 val
->bufferram
= nr_blockdev_pages();
1216 val
->totalhigh
= totalhigh_pages
;
1217 val
->freehigh
= nr_free_highpages();
1218 val
->mem_unit
= PAGE_SIZE
;
1221 EXPORT_SYMBOL(si_meminfo
);
1224 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1226 pg_data_t
*pgdat
= NODE_DATA(nid
);
1228 val
->totalram
= pgdat
->node_present_pages
;
1229 val
->freeram
= nr_free_pages_pgdat(pgdat
);
1230 #ifdef CONFIG_HIGHMEM
1231 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1232 val
->freehigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1237 val
->mem_unit
= PAGE_SIZE
;
1241 #define K(x) ((x) << (PAGE_SHIFT-10))
1244 * Show free area list (used inside shift_scroll-lock stuff)
1245 * We also calculate the percentage fragmentation. We do this by counting the
1246 * memory on each free list with the exception of the first item on the list.
1248 void show_free_areas(void)
1250 int cpu
, temperature
;
1251 unsigned long active
;
1252 unsigned long inactive
;
1256 for_each_zone(zone
) {
1258 printk("%s per-cpu:", zone
->name
);
1260 if (!populated_zone(zone
)) {
1266 for_each_online_cpu(cpu
) {
1267 struct per_cpu_pageset
*pageset
;
1269 pageset
= zone_pcp(zone
, cpu
);
1271 for (temperature
= 0; temperature
< 2; temperature
++)
1272 printk("cpu %d %s: high %d, batch %d used:%d\n",
1274 temperature
? "cold" : "hot",
1275 pageset
->pcp
[temperature
].high
,
1276 pageset
->pcp
[temperature
].batch
,
1277 pageset
->pcp
[temperature
].count
);
1281 get_zone_counts(&active
, &inactive
, &free
);
1283 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1284 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1287 global_page_state(NR_FILE_DIRTY
),
1288 global_page_state(NR_WRITEBACK
),
1289 global_page_state(NR_UNSTABLE_NFS
),
1291 global_page_state(NR_SLAB
),
1292 global_page_state(NR_FILE_MAPPED
),
1293 global_page_state(NR_PAGETABLE
));
1295 for_each_zone(zone
) {
1307 " pages_scanned:%lu"
1308 " all_unreclaimable? %s"
1311 K(zone
->free_pages
),
1314 K(zone
->pages_high
),
1316 K(zone
->nr_inactive
),
1317 K(zone
->present_pages
),
1318 zone
->pages_scanned
,
1319 (zone
->all_unreclaimable
? "yes" : "no")
1321 printk("lowmem_reserve[]:");
1322 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1323 printk(" %lu", zone
->lowmem_reserve
[i
]);
1327 for_each_zone(zone
) {
1328 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1331 printk("%s: ", zone
->name
);
1332 if (!populated_zone(zone
)) {
1337 spin_lock_irqsave(&zone
->lock
, flags
);
1338 for (order
= 0; order
< MAX_ORDER
; order
++) {
1339 nr
[order
] = zone
->free_area
[order
].nr_free
;
1340 total
+= nr
[order
] << order
;
1342 spin_unlock_irqrestore(&zone
->lock
, flags
);
1343 for (order
= 0; order
< MAX_ORDER
; order
++)
1344 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1345 printk("= %lukB\n", K(total
));
1348 show_swap_cache_info();
1352 * Builds allocation fallback zone lists.
1354 * Add all populated zones of a node to the zonelist.
1356 static int __meminit
build_zonelists_node(pg_data_t
*pgdat
,
1357 struct zonelist
*zonelist
, int nr_zones
, int zone_type
)
1361 BUG_ON(zone_type
>= MAX_NR_ZONES
);
1364 zone
= pgdat
->node_zones
+ zone_type
;
1365 if (populated_zone(zone
)) {
1366 zonelist
->zones
[nr_zones
++] = zone
;
1367 check_highest_zone(zone_type
);
1371 } while (zone_type
>= 0);
1375 static inline int highest_zone(int zone_bits
)
1377 int res
= ZONE_NORMAL
;
1378 if (zone_bits
& (__force
int)__GFP_HIGHMEM
)
1380 #ifdef CONFIG_ZONE_DMA32
1381 if (zone_bits
& (__force
int)__GFP_DMA32
)
1384 if (zone_bits
& (__force
int)__GFP_DMA
)
1390 #define MAX_NODE_LOAD (num_online_nodes())
1391 static int __meminitdata node_load
[MAX_NUMNODES
];
1393 * find_next_best_node - find the next node that should appear in a given node's fallback list
1394 * @node: node whose fallback list we're appending
1395 * @used_node_mask: nodemask_t of already used nodes
1397 * We use a number of factors to determine which is the next node that should
1398 * appear on a given node's fallback list. The node should not have appeared
1399 * already in @node's fallback list, and it should be the next closest node
1400 * according to the distance array (which contains arbitrary distance values
1401 * from each node to each node in the system), and should also prefer nodes
1402 * with no CPUs, since presumably they'll have very little allocation pressure
1403 * on them otherwise.
1404 * It returns -1 if no node is found.
1406 static int __meminit
find_next_best_node(int node
, nodemask_t
*used_node_mask
)
1409 int min_val
= INT_MAX
;
1412 /* Use the local node if we haven't already */
1413 if (!node_isset(node
, *used_node_mask
)) {
1414 node_set(node
, *used_node_mask
);
1418 for_each_online_node(n
) {
1421 /* Don't want a node to appear more than once */
1422 if (node_isset(n
, *used_node_mask
))
1425 /* Use the distance array to find the distance */
1426 val
= node_distance(node
, n
);
1428 /* Penalize nodes under us ("prefer the next node") */
1431 /* Give preference to headless and unused nodes */
1432 tmp
= node_to_cpumask(n
);
1433 if (!cpus_empty(tmp
))
1434 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
1436 /* Slight preference for less loaded node */
1437 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
1438 val
+= node_load
[n
];
1440 if (val
< min_val
) {
1447 node_set(best_node
, *used_node_mask
);
1452 static void __meminit
build_zonelists(pg_data_t
*pgdat
)
1454 int i
, j
, k
, node
, local_node
;
1455 int prev_node
, load
;
1456 struct zonelist
*zonelist
;
1457 nodemask_t used_mask
;
1459 /* initialize zonelists */
1460 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1461 zonelist
= pgdat
->node_zonelists
+ i
;
1462 zonelist
->zones
[0] = NULL
;
1465 /* NUMA-aware ordering of nodes */
1466 local_node
= pgdat
->node_id
;
1467 load
= num_online_nodes();
1468 prev_node
= local_node
;
1469 nodes_clear(used_mask
);
1470 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
1471 int distance
= node_distance(local_node
, node
);
1474 * If another node is sufficiently far away then it is better
1475 * to reclaim pages in a zone before going off node.
1477 if (distance
> RECLAIM_DISTANCE
)
1478 zone_reclaim_mode
= 1;
1481 * We don't want to pressure a particular node.
1482 * So adding penalty to the first node in same
1483 * distance group to make it round-robin.
1486 if (distance
!= node_distance(local_node
, prev_node
))
1487 node_load
[node
] += load
;
1490 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1491 zonelist
= pgdat
->node_zonelists
+ i
;
1492 for (j
= 0; zonelist
->zones
[j
] != NULL
; j
++);
1494 k
= highest_zone(i
);
1496 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1497 zonelist
->zones
[j
] = NULL
;
1502 #else /* CONFIG_NUMA */
1504 static void __meminit
build_zonelists(pg_data_t
*pgdat
)
1506 int i
, node
, local_node
;
1510 local_node
= pgdat
->node_id
;
1511 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1512 struct zonelist
*zonelist
;
1514 zonelist
= pgdat
->node_zonelists
+ i
;
1517 k
= highest_zone(i
);
1518 j
= build_zonelists_node(pgdat
, zonelist
, j
, k
);
1520 * Now we build the zonelist so that it contains the zones
1521 * of all the other nodes.
1522 * We don't want to pressure a particular node, so when
1523 * building the zones for node N, we make sure that the
1524 * zones coming right after the local ones are those from
1525 * node N+1 (modulo N)
1527 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
1528 if (!node_online(node
))
1530 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1532 for (node
= 0; node
< local_node
; node
++) {
1533 if (!node_online(node
))
1535 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1538 zonelist
->zones
[j
] = NULL
;
1542 #endif /* CONFIG_NUMA */
1544 /* return values int ....just for stop_machine_run() */
1545 static int __meminit
__build_all_zonelists(void *dummy
)
1548 for_each_online_node(nid
)
1549 build_zonelists(NODE_DATA(nid
));
1553 void __meminit
build_all_zonelists(void)
1555 if (system_state
== SYSTEM_BOOTING
) {
1556 __build_all_zonelists(0);
1557 cpuset_init_current_mems_allowed();
1559 /* we have to stop all cpus to guaranntee there is no user
1561 stop_machine_run(__build_all_zonelists
, NULL
, NR_CPUS
);
1562 /* cpuset refresh routine should be here */
1564 vm_total_pages
= nr_free_pagecache_pages();
1565 printk("Built %i zonelists. Total pages: %ld\n",
1566 num_online_nodes(), vm_total_pages
);
1570 * Helper functions to size the waitqueue hash table.
1571 * Essentially these want to choose hash table sizes sufficiently
1572 * large so that collisions trying to wait on pages are rare.
1573 * But in fact, the number of active page waitqueues on typical
1574 * systems is ridiculously low, less than 200. So this is even
1575 * conservative, even though it seems large.
1577 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1578 * waitqueues, i.e. the size of the waitq table given the number of pages.
1580 #define PAGES_PER_WAITQUEUE 256
1582 #ifndef CONFIG_MEMORY_HOTPLUG
1583 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
1585 unsigned long size
= 1;
1587 pages
/= PAGES_PER_WAITQUEUE
;
1589 while (size
< pages
)
1593 * Once we have dozens or even hundreds of threads sleeping
1594 * on IO we've got bigger problems than wait queue collision.
1595 * Limit the size of the wait table to a reasonable size.
1597 size
= min(size
, 4096UL);
1599 return max(size
, 4UL);
1603 * A zone's size might be changed by hot-add, so it is not possible to determine
1604 * a suitable size for its wait_table. So we use the maximum size now.
1606 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
1608 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
1609 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
1610 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
1612 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
1613 * or more by the traditional way. (See above). It equals:
1615 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
1616 * ia64(16K page size) : = ( 8G + 4M)byte.
1617 * powerpc (64K page size) : = (32G +16M)byte.
1619 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
1626 * This is an integer logarithm so that shifts can be used later
1627 * to extract the more random high bits from the multiplicative
1628 * hash function before the remainder is taken.
1630 static inline unsigned long wait_table_bits(unsigned long size
)
1635 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1637 static void __init
calculate_zone_totalpages(struct pglist_data
*pgdat
,
1638 unsigned long *zones_size
, unsigned long *zholes_size
)
1640 unsigned long realtotalpages
, totalpages
= 0;
1643 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1644 totalpages
+= zones_size
[i
];
1645 pgdat
->node_spanned_pages
= totalpages
;
1647 realtotalpages
= totalpages
;
1649 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1650 realtotalpages
-= zholes_size
[i
];
1651 pgdat
->node_present_pages
= realtotalpages
;
1652 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
, realtotalpages
);
1657 * Initially all pages are reserved - free ones are freed
1658 * up by free_all_bootmem() once the early boot process is
1659 * done. Non-atomic initialization, single-pass.
1661 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
1662 unsigned long start_pfn
)
1665 unsigned long end_pfn
= start_pfn
+ size
;
1668 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
1669 if (!early_pfn_valid(pfn
))
1671 page
= pfn_to_page(pfn
);
1672 set_page_links(page
, zone
, nid
, pfn
);
1673 init_page_count(page
);
1674 reset_page_mapcount(page
);
1675 SetPageReserved(page
);
1676 INIT_LIST_HEAD(&page
->lru
);
1677 #ifdef WANT_PAGE_VIRTUAL
1678 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1679 if (!is_highmem_idx(zone
))
1680 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
1685 void zone_init_free_lists(struct pglist_data
*pgdat
, struct zone
*zone
,
1689 for (order
= 0; order
< MAX_ORDER
; order
++) {
1690 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
);
1691 zone
->free_area
[order
].nr_free
= 0;
1695 #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1696 void zonetable_add(struct zone
*zone
, int nid
, enum zone_type zid
,
1697 unsigned long pfn
, unsigned long size
)
1699 unsigned long snum
= pfn_to_section_nr(pfn
);
1700 unsigned long end
= pfn_to_section_nr(pfn
+ size
);
1703 zone_table
[ZONETABLE_INDEX(nid
, zid
)] = zone
;
1705 for (; snum
<= end
; snum
++)
1706 zone_table
[ZONETABLE_INDEX(snum
, zid
)] = zone
;
1709 #ifndef __HAVE_ARCH_MEMMAP_INIT
1710 #define memmap_init(size, nid, zone, start_pfn) \
1711 memmap_init_zone((size), (nid), (zone), (start_pfn))
1714 static int __cpuinit
zone_batchsize(struct zone
*zone
)
1719 * The per-cpu-pages pools are set to around 1000th of the
1720 * size of the zone. But no more than 1/2 of a meg.
1722 * OK, so we don't know how big the cache is. So guess.
1724 batch
= zone
->present_pages
/ 1024;
1725 if (batch
* PAGE_SIZE
> 512 * 1024)
1726 batch
= (512 * 1024) / PAGE_SIZE
;
1727 batch
/= 4; /* We effectively *= 4 below */
1732 * Clamp the batch to a 2^n - 1 value. Having a power
1733 * of 2 value was found to be more likely to have
1734 * suboptimal cache aliasing properties in some cases.
1736 * For example if 2 tasks are alternately allocating
1737 * batches of pages, one task can end up with a lot
1738 * of pages of one half of the possible page colors
1739 * and the other with pages of the other colors.
1741 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
1746 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
1748 struct per_cpu_pages
*pcp
;
1750 memset(p
, 0, sizeof(*p
));
1752 pcp
= &p
->pcp
[0]; /* hot */
1754 pcp
->high
= 6 * batch
;
1755 pcp
->batch
= max(1UL, 1 * batch
);
1756 INIT_LIST_HEAD(&pcp
->list
);
1758 pcp
= &p
->pcp
[1]; /* cold*/
1760 pcp
->high
= 2 * batch
;
1761 pcp
->batch
= max(1UL, batch
/2);
1762 INIT_LIST_HEAD(&pcp
->list
);
1766 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1767 * to the value high for the pageset p.
1770 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
1773 struct per_cpu_pages
*pcp
;
1775 pcp
= &p
->pcp
[0]; /* hot list */
1777 pcp
->batch
= max(1UL, high
/4);
1778 if ((high
/4) > (PAGE_SHIFT
* 8))
1779 pcp
->batch
= PAGE_SHIFT
* 8;
1785 * Boot pageset table. One per cpu which is going to be used for all
1786 * zones and all nodes. The parameters will be set in such a way
1787 * that an item put on a list will immediately be handed over to
1788 * the buddy list. This is safe since pageset manipulation is done
1789 * with interrupts disabled.
1791 * Some NUMA counter updates may also be caught by the boot pagesets.
1793 * The boot_pagesets must be kept even after bootup is complete for
1794 * unused processors and/or zones. They do play a role for bootstrapping
1795 * hotplugged processors.
1797 * zoneinfo_show() and maybe other functions do
1798 * not check if the processor is online before following the pageset pointer.
1799 * Other parts of the kernel may not check if the zone is available.
1801 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
1804 * Dynamically allocate memory for the
1805 * per cpu pageset array in struct zone.
1807 static int __cpuinit
process_zones(int cpu
)
1809 struct zone
*zone
, *dzone
;
1811 for_each_zone(zone
) {
1813 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
1814 GFP_KERNEL
, cpu_to_node(cpu
));
1815 if (!zone_pcp(zone
, cpu
))
1818 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
1820 if (percpu_pagelist_fraction
)
1821 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
1822 (zone
->present_pages
/ percpu_pagelist_fraction
));
1827 for_each_zone(dzone
) {
1830 kfree(zone_pcp(dzone
, cpu
));
1831 zone_pcp(dzone
, cpu
) = NULL
;
1836 static inline void free_zone_pagesets(int cpu
)
1840 for_each_zone(zone
) {
1841 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
1843 /* Free per_cpu_pageset if it is slab allocated */
1844 if (pset
!= &boot_pageset
[cpu
])
1846 zone_pcp(zone
, cpu
) = NULL
;
1850 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
1851 unsigned long action
,
1854 int cpu
= (long)hcpu
;
1855 int ret
= NOTIFY_OK
;
1858 case CPU_UP_PREPARE
:
1859 if (process_zones(cpu
))
1862 case CPU_UP_CANCELED
:
1864 free_zone_pagesets(cpu
);
1872 static struct notifier_block __cpuinitdata pageset_notifier
=
1873 { &pageset_cpuup_callback
, NULL
, 0 };
1875 void __init
setup_per_cpu_pageset(void)
1879 /* Initialize per_cpu_pageset for cpu 0.
1880 * A cpuup callback will do this for every cpu
1881 * as it comes online
1883 err
= process_zones(smp_processor_id());
1885 register_cpu_notifier(&pageset_notifier
);
1891 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
1894 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1898 * The per-page waitqueue mechanism uses hashed waitqueues
1901 zone
->wait_table_hash_nr_entries
=
1902 wait_table_hash_nr_entries(zone_size_pages
);
1903 zone
->wait_table_bits
=
1904 wait_table_bits(zone
->wait_table_hash_nr_entries
);
1905 alloc_size
= zone
->wait_table_hash_nr_entries
1906 * sizeof(wait_queue_head_t
);
1908 if (system_state
== SYSTEM_BOOTING
) {
1909 zone
->wait_table
= (wait_queue_head_t
*)
1910 alloc_bootmem_node(pgdat
, alloc_size
);
1913 * This case means that a zone whose size was 0 gets new memory
1914 * via memory hot-add.
1915 * But it may be the case that a new node was hot-added. In
1916 * this case vmalloc() will not be able to use this new node's
1917 * memory - this wait_table must be initialized to use this new
1918 * node itself as well.
1919 * To use this new node's memory, further consideration will be
1922 zone
->wait_table
= (wait_queue_head_t
*)vmalloc(alloc_size
);
1924 if (!zone
->wait_table
)
1927 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
1928 init_waitqueue_head(zone
->wait_table
+ i
);
1933 static __meminit
void zone_pcp_init(struct zone
*zone
)
1936 unsigned long batch
= zone_batchsize(zone
);
1938 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
1940 /* Early boot. Slab allocator not functional yet */
1941 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
1942 setup_pageset(&boot_pageset
[cpu
],0);
1944 setup_pageset(zone_pcp(zone
,cpu
), batch
);
1947 if (zone
->present_pages
)
1948 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
1949 zone
->name
, zone
->present_pages
, batch
);
1952 __meminit
int init_currently_empty_zone(struct zone
*zone
,
1953 unsigned long zone_start_pfn
,
1956 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1958 ret
= zone_wait_table_init(zone
, size
);
1961 pgdat
->nr_zones
= zone_idx(zone
) + 1;
1963 zone
->zone_start_pfn
= zone_start_pfn
;
1965 memmap_init(size
, pgdat
->node_id
, zone_idx(zone
), zone_start_pfn
);
1967 zone_init_free_lists(pgdat
, zone
, zone
->spanned_pages
);
1973 * Set up the zone data structures:
1974 * - mark all pages reserved
1975 * - mark all memory queues empty
1976 * - clear the memory bitmaps
1978 static void __meminit
free_area_init_core(struct pglist_data
*pgdat
,
1979 unsigned long *zones_size
, unsigned long *zholes_size
)
1982 int nid
= pgdat
->node_id
;
1983 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
1986 pgdat_resize_init(pgdat
);
1987 pgdat
->nr_zones
= 0;
1988 init_waitqueue_head(&pgdat
->kswapd_wait
);
1989 pgdat
->kswapd_max_order
= 0;
1991 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
1992 struct zone
*zone
= pgdat
->node_zones
+ j
;
1993 unsigned long size
, realsize
;
1995 realsize
= size
= zones_size
[j
];
1997 realsize
-= zholes_size
[j
];
1999 if (!is_highmem_idx(j
))
2000 nr_kernel_pages
+= realsize
;
2001 nr_all_pages
+= realsize
;
2003 zone
->spanned_pages
= size
;
2004 zone
->present_pages
= realsize
;
2006 zone
->min_unmapped_ratio
= (realsize
*sysctl_min_unmapped_ratio
)
2009 zone
->name
= zone_names
[j
];
2010 spin_lock_init(&zone
->lock
);
2011 spin_lock_init(&zone
->lru_lock
);
2012 zone_seqlock_init(zone
);
2013 zone
->zone_pgdat
= pgdat
;
2014 zone
->free_pages
= 0;
2016 zone
->temp_priority
= zone
->prev_priority
= DEF_PRIORITY
;
2018 zone_pcp_init(zone
);
2019 INIT_LIST_HEAD(&zone
->active_list
);
2020 INIT_LIST_HEAD(&zone
->inactive_list
);
2021 zone
->nr_scan_active
= 0;
2022 zone
->nr_scan_inactive
= 0;
2023 zone
->nr_active
= 0;
2024 zone
->nr_inactive
= 0;
2025 zap_zone_vm_stats(zone
);
2026 atomic_set(&zone
->reclaim_in_progress
, 0);
2030 zonetable_add(zone
, nid
, j
, zone_start_pfn
, size
);
2031 ret
= init_currently_empty_zone(zone
, zone_start_pfn
, size
);
2033 zone_start_pfn
+= size
;
2037 static void __init
alloc_node_mem_map(struct pglist_data
*pgdat
)
2039 /* Skip empty nodes */
2040 if (!pgdat
->node_spanned_pages
)
2043 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2044 /* ia64 gets its own node_mem_map, before this, without bootmem */
2045 if (!pgdat
->node_mem_map
) {
2046 unsigned long size
, start
, end
;
2050 * The zone's endpoints aren't required to be MAX_ORDER
2051 * aligned but the node_mem_map endpoints must be in order
2052 * for the buddy allocator to function correctly.
2054 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
2055 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
2056 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
2057 size
= (end
- start
) * sizeof(struct page
);
2058 map
= alloc_remap(pgdat
->node_id
, size
);
2060 map
= alloc_bootmem_node(pgdat
, size
);
2061 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
2063 #ifdef CONFIG_FLATMEM
2065 * With no DISCONTIG, the global mem_map is just set as node 0's
2067 if (pgdat
== NODE_DATA(0))
2068 mem_map
= NODE_DATA(0)->node_mem_map
;
2070 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
2073 void __meminit
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
2074 unsigned long *zones_size
, unsigned long node_start_pfn
,
2075 unsigned long *zholes_size
)
2077 pgdat
->node_id
= nid
;
2078 pgdat
->node_start_pfn
= node_start_pfn
;
2079 calculate_zone_totalpages(pgdat
, zones_size
, zholes_size
);
2081 alloc_node_mem_map(pgdat
);
2083 free_area_init_core(pgdat
, zones_size
, zholes_size
);
2086 #ifndef CONFIG_NEED_MULTIPLE_NODES
2087 static bootmem_data_t contig_bootmem_data
;
2088 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
2090 EXPORT_SYMBOL(contig_page_data
);
2093 void __init
free_area_init(unsigned long *zones_size
)
2095 free_area_init_node(0, NODE_DATA(0), zones_size
,
2096 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
2099 #ifdef CONFIG_HOTPLUG_CPU
2100 static int page_alloc_cpu_notify(struct notifier_block
*self
,
2101 unsigned long action
, void *hcpu
)
2103 int cpu
= (unsigned long)hcpu
;
2105 if (action
== CPU_DEAD
) {
2106 local_irq_disable();
2108 vm_events_fold_cpu(cpu
);
2110 refresh_cpu_vm_stats(cpu
);
2114 #endif /* CONFIG_HOTPLUG_CPU */
2116 void __init
page_alloc_init(void)
2118 hotcpu_notifier(page_alloc_cpu_notify
, 0);
2122 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
2123 * or min_free_kbytes changes.
2125 static void calculate_totalreserve_pages(void)
2127 struct pglist_data
*pgdat
;
2128 unsigned long reserve_pages
= 0;
2131 for_each_online_pgdat(pgdat
) {
2132 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2133 struct zone
*zone
= pgdat
->node_zones
+ i
;
2134 unsigned long max
= 0;
2136 /* Find valid and maximum lowmem_reserve in the zone */
2137 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
2138 if (zone
->lowmem_reserve
[j
] > max
)
2139 max
= zone
->lowmem_reserve
[j
];
2142 /* we treat pages_high as reserved pages. */
2143 max
+= zone
->pages_high
;
2145 if (max
> zone
->present_pages
)
2146 max
= zone
->present_pages
;
2147 reserve_pages
+= max
;
2150 totalreserve_pages
= reserve_pages
;
2154 * setup_per_zone_lowmem_reserve - called whenever
2155 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2156 * has a correct pages reserved value, so an adequate number of
2157 * pages are left in the zone after a successful __alloc_pages().
2159 static void setup_per_zone_lowmem_reserve(void)
2161 struct pglist_data
*pgdat
;
2164 for_each_online_pgdat(pgdat
) {
2165 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2166 struct zone
*zone
= pgdat
->node_zones
+ j
;
2167 unsigned long present_pages
= zone
->present_pages
;
2169 zone
->lowmem_reserve
[j
] = 0;
2171 for (idx
= j
-1; idx
>= 0; idx
--) {
2172 struct zone
*lower_zone
;
2174 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
2175 sysctl_lowmem_reserve_ratio
[idx
] = 1;
2177 lower_zone
= pgdat
->node_zones
+ idx
;
2178 lower_zone
->lowmem_reserve
[j
] = present_pages
/
2179 sysctl_lowmem_reserve_ratio
[idx
];
2180 present_pages
+= lower_zone
->present_pages
;
2185 /* update totalreserve_pages */
2186 calculate_totalreserve_pages();
2190 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2191 * that the pages_{min,low,high} values for each zone are set correctly
2192 * with respect to min_free_kbytes.
2194 void setup_per_zone_pages_min(void)
2196 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
2197 unsigned long lowmem_pages
= 0;
2199 unsigned long flags
;
2201 /* Calculate total number of !ZONE_HIGHMEM pages */
2202 for_each_zone(zone
) {
2203 if (!is_highmem(zone
))
2204 lowmem_pages
+= zone
->present_pages
;
2207 for_each_zone(zone
) {
2210 spin_lock_irqsave(&zone
->lru_lock
, flags
);
2211 tmp
= (u64
)pages_min
* zone
->present_pages
;
2212 do_div(tmp
, lowmem_pages
);
2213 if (is_highmem(zone
)) {
2215 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2216 * need highmem pages, so cap pages_min to a small
2219 * The (pages_high-pages_low) and (pages_low-pages_min)
2220 * deltas controls asynch page reclaim, and so should
2221 * not be capped for highmem.
2225 min_pages
= zone
->present_pages
/ 1024;
2226 if (min_pages
< SWAP_CLUSTER_MAX
)
2227 min_pages
= SWAP_CLUSTER_MAX
;
2228 if (min_pages
> 128)
2230 zone
->pages_min
= min_pages
;
2233 * If it's a lowmem zone, reserve a number of pages
2234 * proportionate to the zone's size.
2236 zone
->pages_min
= tmp
;
2239 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
2240 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
2241 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
2244 /* update totalreserve_pages */
2245 calculate_totalreserve_pages();
2249 * Initialise min_free_kbytes.
2251 * For small machines we want it small (128k min). For large machines
2252 * we want it large (64MB max). But it is not linear, because network
2253 * bandwidth does not increase linearly with machine size. We use
2255 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2256 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2272 static int __init
init_per_zone_pages_min(void)
2274 unsigned long lowmem_kbytes
;
2276 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
2278 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
2279 if (min_free_kbytes
< 128)
2280 min_free_kbytes
= 128;
2281 if (min_free_kbytes
> 65536)
2282 min_free_kbytes
= 65536;
2283 setup_per_zone_pages_min();
2284 setup_per_zone_lowmem_reserve();
2287 module_init(init_per_zone_pages_min
)
2290 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2291 * that we can call two helper functions whenever min_free_kbytes
2294 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
2295 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2297 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
2298 setup_per_zone_pages_min();
2303 int sysctl_min_unmapped_ratio_sysctl_handler(ctl_table
*table
, int write
,
2304 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2309 rc
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2314 zone
->min_unmapped_ratio
= (zone
->present_pages
*
2315 sysctl_min_unmapped_ratio
) / 100;
2321 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2322 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2323 * whenever sysctl_lowmem_reserve_ratio changes.
2325 * The reserve ratio obviously has absolutely no relation with the
2326 * pages_min watermarks. The lowmem reserve ratio can only make sense
2327 * if in function of the boot time zone sizes.
2329 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
2330 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2332 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2333 setup_per_zone_lowmem_reserve();
2338 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
2339 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
2340 * can have before it gets flushed back to buddy allocator.
2343 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
2344 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2350 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2351 if (!write
|| (ret
== -EINVAL
))
2353 for_each_zone(zone
) {
2354 for_each_online_cpu(cpu
) {
2356 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
2357 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
2363 int hashdist
= HASHDIST_DEFAULT
;
2366 static int __init
set_hashdist(char *str
)
2370 hashdist
= simple_strtoul(str
, &str
, 0);
2373 __setup("hashdist=", set_hashdist
);
2377 * allocate a large system hash table from bootmem
2378 * - it is assumed that the hash table must contain an exact power-of-2
2379 * quantity of entries
2380 * - limit is the number of hash buckets, not the total allocation size
2382 void *__init
alloc_large_system_hash(const char *tablename
,
2383 unsigned long bucketsize
,
2384 unsigned long numentries
,
2387 unsigned int *_hash_shift
,
2388 unsigned int *_hash_mask
,
2389 unsigned long limit
)
2391 unsigned long long max
= limit
;
2392 unsigned long log2qty
, size
;
2395 /* allow the kernel cmdline to have a say */
2397 /* round applicable memory size up to nearest megabyte */
2398 numentries
= (flags
& HASH_HIGHMEM
) ? nr_all_pages
: nr_kernel_pages
;
2399 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
2400 numentries
>>= 20 - PAGE_SHIFT
;
2401 numentries
<<= 20 - PAGE_SHIFT
;
2403 /* limit to 1 bucket per 2^scale bytes of low memory */
2404 if (scale
> PAGE_SHIFT
)
2405 numentries
>>= (scale
- PAGE_SHIFT
);
2407 numentries
<<= (PAGE_SHIFT
- scale
);
2409 numentries
= roundup_pow_of_two(numentries
);
2411 /* limit allocation size to 1/16 total memory by default */
2413 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
2414 do_div(max
, bucketsize
);
2417 if (numentries
> max
)
2420 log2qty
= long_log2(numentries
);
2423 size
= bucketsize
<< log2qty
;
2424 if (flags
& HASH_EARLY
)
2425 table
= alloc_bootmem(size
);
2427 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
2429 unsigned long order
;
2430 for (order
= 0; ((1UL << order
) << PAGE_SHIFT
) < size
; order
++)
2432 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
2434 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
2437 panic("Failed to allocate %s hash table\n", tablename
);
2439 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2442 long_log2(size
) - PAGE_SHIFT
,
2446 *_hash_shift
= log2qty
;
2448 *_hash_mask
= (1 << log2qty
) - 1;
2453 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
2454 struct page
*pfn_to_page(unsigned long pfn
)
2456 return __pfn_to_page(pfn
);
2458 unsigned long page_to_pfn(struct page
*page
)
2460 return __page_to_pfn(page
);
2462 EXPORT_SYMBOL(pfn_to_page
);
2463 EXPORT_SYMBOL(page_to_pfn
);
2464 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */