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/config.h>
18 #include <linux/stddef.h>
20 #include <linux/swap.h>
21 #include <linux/interrupt.h>
22 #include <linux/pagemap.h>
23 #include <linux/bootmem.h>
24 #include <linux/compiler.h>
25 #include <linux/kernel.h>
26 #include <linux/module.h>
27 #include <linux/suspend.h>
28 #include <linux/pagevec.h>
29 #include <linux/blkdev.h>
30 #include <linux/slab.h>
31 #include <linux/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>
42 #include <asm/tlbflush.h>
43 #include <asm/div64.h>
47 * MCD - HACK: Find somewhere to initialize this EARLY, or make this
50 nodemask_t node_online_map __read_mostly
= { { [0] = 1UL } };
51 EXPORT_SYMBOL(node_online_map
);
52 nodemask_t node_possible_map __read_mostly
= NODE_MASK_ALL
;
53 EXPORT_SYMBOL(node_possible_map
);
54 unsigned long totalram_pages __read_mostly
;
55 unsigned long totalhigh_pages __read_mostly
;
56 unsigned long totalreserve_pages __read_mostly
;
58 int percpu_pagelist_fraction
;
60 static void __free_pages_ok(struct page
*page
, unsigned int order
);
63 * results with 256, 32 in the lowmem_reserve sysctl:
64 * 1G machine -> (16M dma, 800M-16M normal, 1G-800M high)
65 * 1G machine -> (16M dma, 784M normal, 224M high)
66 * NORMAL allocation will leave 784M/256 of ram reserved in the ZONE_DMA
67 * HIGHMEM allocation will leave 224M/32 of ram reserved in ZONE_NORMAL
68 * HIGHMEM allocation will (224M+784M)/256 of ram reserved in ZONE_DMA
70 * TBD: should special case ZONE_DMA32 machines here - in those we normally
71 * don't need any ZONE_NORMAL reservation
73 int sysctl_lowmem_reserve_ratio
[MAX_NR_ZONES
-1] = { 256, 256, 32 };
75 EXPORT_SYMBOL(totalram_pages
);
78 * Used by page_zone() to look up the address of the struct zone whose
79 * id is encoded in the upper bits of page->flags
81 struct zone
*zone_table
[1 << ZONETABLE_SHIFT
] __read_mostly
;
82 EXPORT_SYMBOL(zone_table
);
84 static char *zone_names
[MAX_NR_ZONES
] = { "DMA", "DMA32", "Normal", "HighMem" };
85 int min_free_kbytes
= 1024;
87 unsigned long __meminitdata nr_kernel_pages
;
88 unsigned long __meminitdata nr_all_pages
;
90 #ifdef CONFIG_DEBUG_VM
91 static int page_outside_zone_boundaries(struct zone
*zone
, struct page
*page
)
95 unsigned long pfn
= page_to_pfn(page
);
98 seq
= zone_span_seqbegin(zone
);
99 if (pfn
>= zone
->zone_start_pfn
+ zone
->spanned_pages
)
101 else if (pfn
< zone
->zone_start_pfn
)
103 } while (zone_span_seqretry(zone
, seq
));
108 static int page_is_consistent(struct zone
*zone
, struct page
*page
)
110 #ifdef CONFIG_HOLES_IN_ZONE
111 if (!pfn_valid(page_to_pfn(page
)))
114 if (zone
!= page_zone(page
))
120 * Temporary debugging check for pages not lying within a given zone.
122 static int bad_range(struct zone
*zone
, struct page
*page
)
124 if (page_outside_zone_boundaries(zone
, page
))
126 if (!page_is_consistent(zone
, page
))
133 static inline int bad_range(struct zone
*zone
, struct page
*page
)
139 static void bad_page(struct page
*page
)
141 printk(KERN_EMERG
"Bad page state in process '%s'\n"
142 KERN_EMERG
"page:%p flags:0x%0*lx mapping:%p mapcount:%d count:%d\n"
143 KERN_EMERG
"Trying to fix it up, but a reboot is needed\n"
144 KERN_EMERG
"Backtrace:\n",
145 current
->comm
, page
, (int)(2*sizeof(unsigned long)),
146 (unsigned long)page
->flags
, page
->mapping
,
147 page_mapcount(page
), page_count(page
));
149 page
->flags
&= ~(1 << PG_lru
|
159 set_page_count(page
, 0);
160 reset_page_mapcount(page
);
161 page
->mapping
= NULL
;
162 add_taint(TAINT_BAD_PAGE
);
166 * Higher-order pages are called "compound pages". They are structured thusly:
168 * The first PAGE_SIZE page is called the "head page".
170 * The remaining PAGE_SIZE pages are called "tail pages".
172 * All pages have PG_compound set. All pages have their ->private pointing at
173 * the head page (even the head page has this).
175 * The first tail page's ->lru.next holds the address of the compound page's
176 * put_page() function. Its ->lru.prev holds the order of allocation.
177 * This usage means that zero-order pages may not be compound.
180 static void free_compound_page(struct page
*page
)
182 __free_pages_ok(page
, (unsigned long)page
[1].lru
.prev
);
185 static void prep_compound_page(struct page
*page
, unsigned long order
)
188 int nr_pages
= 1 << order
;
190 page
[1].lru
.next
= (void *)free_compound_page
; /* set dtor */
191 page
[1].lru
.prev
= (void *)order
;
192 for (i
= 0; i
< nr_pages
; i
++) {
193 struct page
*p
= page
+ i
;
195 __SetPageCompound(p
);
196 set_page_private(p
, (unsigned long)page
);
200 static void destroy_compound_page(struct page
*page
, unsigned long order
)
203 int nr_pages
= 1 << order
;
205 if (unlikely((unsigned long)page
[1].lru
.prev
!= order
))
208 for (i
= 0; i
< nr_pages
; i
++) {
209 struct page
*p
= page
+ i
;
211 if (unlikely(!PageCompound(p
) |
212 (page_private(p
) != (unsigned long)page
)))
214 __ClearPageCompound(p
);
218 static inline void prep_zero_page(struct page
*page
, int order
, gfp_t gfp_flags
)
222 BUG_ON((gfp_flags
& (__GFP_WAIT
| __GFP_HIGHMEM
)) == __GFP_HIGHMEM
);
224 * clear_highpage() will use KM_USER0, so it's a bug to use __GFP_ZERO
225 * and __GFP_HIGHMEM from hard or soft interrupt context.
227 BUG_ON((gfp_flags
& __GFP_HIGHMEM
) && in_interrupt());
228 for (i
= 0; i
< (1 << order
); i
++)
229 clear_highpage(page
+ i
);
233 * function for dealing with page's order in buddy system.
234 * zone->lock is already acquired when we use these.
235 * So, we don't need atomic page->flags operations here.
237 static inline unsigned long page_order(struct page
*page
)
239 return page_private(page
);
242 static inline void set_page_order(struct page
*page
, int order
)
244 set_page_private(page
, order
);
245 __SetPageBuddy(page
);
248 static inline void rmv_page_order(struct page
*page
)
250 __ClearPageBuddy(page
);
251 set_page_private(page
, 0);
255 * Locate the struct page for both the matching buddy in our
256 * pair (buddy1) and the combined O(n+1) page they form (page).
258 * 1) Any buddy B1 will have an order O twin B2 which satisfies
259 * the following equation:
261 * For example, if the starting buddy (buddy2) is #8 its order
263 * B2 = 8 ^ (1 << 1) = 8 ^ 2 = 10
265 * 2) Any buddy B will have an order O+1 parent P which
266 * satisfies the following equation:
269 * Assumption: *_mem_map is contiguous at least up to MAX_ORDER
271 static inline struct page
*
272 __page_find_buddy(struct page
*page
, unsigned long page_idx
, unsigned int order
)
274 unsigned long buddy_idx
= page_idx
^ (1 << order
);
276 return page
+ (buddy_idx
- page_idx
);
279 static inline unsigned long
280 __find_combined_index(unsigned long page_idx
, unsigned int order
)
282 return (page_idx
& ~(1 << order
));
286 * This function checks whether a page is free && is the buddy
287 * we can do coalesce a page and its buddy if
288 * (a) the buddy is not in a hole &&
289 * (b) the buddy is in the buddy system &&
290 * (c) a page and its buddy have the same order &&
291 * (d) a page and its buddy are in the same zone.
293 * For recording whether a page is in the buddy system, we use PG_buddy.
294 * Setting, clearing, and testing PG_buddy is serialized by zone->lock.
296 * For recording page's order, we use page_private(page).
298 static inline int page_is_buddy(struct page
*page
, struct page
*buddy
,
301 #ifdef CONFIG_HOLES_IN_ZONE
302 if (!pfn_valid(page_to_pfn(buddy
)))
306 if (page_zone_id(page
) != page_zone_id(buddy
))
309 if (PageBuddy(buddy
) && page_order(buddy
) == order
) {
310 BUG_ON(page_count(buddy
) != 0);
317 * Freeing function for a buddy system allocator.
319 * The concept of a buddy system is to maintain direct-mapped table
320 * (containing bit values) for memory blocks of various "orders".
321 * The bottom level table contains the map for the smallest allocatable
322 * units of memory (here, pages), and each level above it describes
323 * pairs of units from the levels below, hence, "buddies".
324 * At a high level, all that happens here is marking the table entry
325 * at the bottom level available, and propagating the changes upward
326 * as necessary, plus some accounting needed to play nicely with other
327 * parts of the VM system.
328 * At each level, we keep a list of pages, which are heads of continuous
329 * free pages of length of (1 << order) and marked with PG_buddy. Page's
330 * order is recorded in page_private(page) field.
331 * So when we are allocating or freeing one, we can derive the state of the
332 * other. That is, if we allocate a small block, and both were
333 * free, the remainder of the region must be split into blocks.
334 * If a block is freed, and its buddy is also free, then this
335 * triggers coalescing into a block of larger size.
340 static inline void __free_one_page(struct page
*page
,
341 struct zone
*zone
, unsigned int order
)
343 unsigned long page_idx
;
344 int order_size
= 1 << order
;
346 if (unlikely(PageCompound(page
)))
347 destroy_compound_page(page
, order
);
349 page_idx
= page_to_pfn(page
) & ((1 << MAX_ORDER
) - 1);
351 BUG_ON(page_idx
& (order_size
- 1));
352 BUG_ON(bad_range(zone
, page
));
354 zone
->free_pages
+= order_size
;
355 while (order
< MAX_ORDER
-1) {
356 unsigned long combined_idx
;
357 struct free_area
*area
;
360 buddy
= __page_find_buddy(page
, page_idx
, order
);
361 if (!page_is_buddy(page
, buddy
, order
))
362 break; /* Move the buddy up one level. */
364 list_del(&buddy
->lru
);
365 area
= zone
->free_area
+ order
;
367 rmv_page_order(buddy
);
368 combined_idx
= __find_combined_index(page_idx
, order
);
369 page
= page
+ (combined_idx
- page_idx
);
370 page_idx
= combined_idx
;
373 set_page_order(page
, order
);
374 list_add(&page
->lru
, &zone
->free_area
[order
].free_list
);
375 zone
->free_area
[order
].nr_free
++;
378 static inline int free_pages_check(struct page
*page
)
380 if (unlikely(page_mapcount(page
) |
381 (page
->mapping
!= NULL
) |
382 (page_count(page
) != 0) |
396 __ClearPageDirty(page
);
398 * For now, we report if PG_reserved was found set, but do not
399 * clear it, and do not free the page. But we shall soon need
400 * to do more, for when the ZERO_PAGE count wraps negative.
402 return PageReserved(page
);
406 * Frees a list of pages.
407 * Assumes all pages on list are in same zone, and of same order.
408 * count is the number of pages to free.
410 * If the zone was previously in an "all pages pinned" state then look to
411 * see if this freeing clears that state.
413 * And clear the zone's pages_scanned counter, to hold off the "all pages are
414 * pinned" detection logic.
416 static void free_pages_bulk(struct zone
*zone
, int count
,
417 struct list_head
*list
, int order
)
419 spin_lock(&zone
->lock
);
420 zone
->all_unreclaimable
= 0;
421 zone
->pages_scanned
= 0;
425 BUG_ON(list_empty(list
));
426 page
= list_entry(list
->prev
, struct page
, lru
);
427 /* have to delete it as __free_one_page list manipulates */
428 list_del(&page
->lru
);
429 __free_one_page(page
, zone
, order
);
431 spin_unlock(&zone
->lock
);
434 static void free_one_page(struct zone
*zone
, struct page
*page
, int order
)
437 list_add(&page
->lru
, &list
);
438 free_pages_bulk(zone
, 1, &list
, order
);
441 static void __free_pages_ok(struct page
*page
, unsigned int order
)
447 arch_free_page(page
, order
);
448 if (!PageHighMem(page
))
449 debug_check_no_locks_freed(page_address(page
),
452 for (i
= 0 ; i
< (1 << order
) ; ++i
)
453 reserved
+= free_pages_check(page
+ i
);
457 kernel_map_pages(page
, 1 << order
, 0);
458 local_irq_save(flags
);
459 __mod_page_state(pgfree
, 1 << order
);
460 free_one_page(page_zone(page
), page
, order
);
461 local_irq_restore(flags
);
465 * permit the bootmem allocator to evade page validation on high-order frees
467 void fastcall __init
__free_pages_bootmem(struct page
*page
, unsigned int order
)
470 __ClearPageReserved(page
);
471 set_page_count(page
, 0);
472 set_page_refcounted(page
);
478 for (loop
= 0; loop
< BITS_PER_LONG
; loop
++) {
479 struct page
*p
= &page
[loop
];
481 if (loop
+ 1 < BITS_PER_LONG
)
483 __ClearPageReserved(p
);
484 set_page_count(p
, 0);
487 set_page_refcounted(page
);
488 __free_pages(page
, order
);
494 * The order of subdivision here is critical for the IO subsystem.
495 * Please do not alter this order without good reasons and regression
496 * testing. Specifically, as large blocks of memory are subdivided,
497 * the order in which smaller blocks are delivered depends on the order
498 * they're subdivided in this function. This is the primary factor
499 * influencing the order in which pages are delivered to the IO
500 * subsystem according to empirical testing, and this is also justified
501 * by considering the behavior of a buddy system containing a single
502 * large block of memory acted on by a series of small allocations.
503 * This behavior is a critical factor in sglist merging's success.
507 static inline void expand(struct zone
*zone
, struct page
*page
,
508 int low
, int high
, struct free_area
*area
)
510 unsigned long size
= 1 << high
;
516 BUG_ON(bad_range(zone
, &page
[size
]));
517 list_add(&page
[size
].lru
, &area
->free_list
);
519 set_page_order(&page
[size
], high
);
524 * This page is about to be returned from the page allocator
526 static int prep_new_page(struct page
*page
, int order
, gfp_t gfp_flags
)
528 if (unlikely(page_mapcount(page
) |
529 (page
->mapping
!= NULL
) |
530 (page_count(page
) != 0) |
546 * For now, we report if PG_reserved was found set, but do not
547 * clear it, and do not allocate the page: as a safety net.
549 if (PageReserved(page
))
552 page
->flags
&= ~(1 << PG_uptodate
| 1 << PG_error
|
553 1 << PG_referenced
| 1 << PG_arch_1
|
554 1 << PG_checked
| 1 << PG_mappedtodisk
);
555 set_page_private(page
, 0);
556 set_page_refcounted(page
);
557 kernel_map_pages(page
, 1 << order
, 1);
559 if (gfp_flags
& __GFP_ZERO
)
560 prep_zero_page(page
, order
, gfp_flags
);
562 if (order
&& (gfp_flags
& __GFP_COMP
))
563 prep_compound_page(page
, order
);
569 * Do the hard work of removing an element from the buddy allocator.
570 * Call me with the zone->lock already held.
572 static struct page
*__rmqueue(struct zone
*zone
, unsigned int order
)
574 struct free_area
* area
;
575 unsigned int current_order
;
578 for (current_order
= order
; current_order
< MAX_ORDER
; ++current_order
) {
579 area
= zone
->free_area
+ current_order
;
580 if (list_empty(&area
->free_list
))
583 page
= list_entry(area
->free_list
.next
, struct page
, lru
);
584 list_del(&page
->lru
);
585 rmv_page_order(page
);
587 zone
->free_pages
-= 1UL << order
;
588 expand(zone
, page
, order
, current_order
, area
);
596 * Obtain a specified number of elements from the buddy allocator, all under
597 * a single hold of the lock, for efficiency. Add them to the supplied list.
598 * Returns the number of new pages which were placed at *list.
600 static int rmqueue_bulk(struct zone
*zone
, unsigned int order
,
601 unsigned long count
, struct list_head
*list
)
605 spin_lock(&zone
->lock
);
606 for (i
= 0; i
< count
; ++i
) {
607 struct page
*page
= __rmqueue(zone
, order
);
608 if (unlikely(page
== NULL
))
610 list_add_tail(&page
->lru
, list
);
612 spin_unlock(&zone
->lock
);
618 * Called from the slab reaper to drain pagesets on a particular node that
619 * belong to the currently executing processor.
620 * Note that this function must be called with the thread pinned to
621 * a single processor.
623 void drain_node_pages(int nodeid
)
628 for (z
= 0; z
< MAX_NR_ZONES
; z
++) {
629 struct zone
*zone
= NODE_DATA(nodeid
)->node_zones
+ z
;
630 struct per_cpu_pageset
*pset
;
632 pset
= zone_pcp(zone
, smp_processor_id());
633 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
634 struct per_cpu_pages
*pcp
;
638 local_irq_save(flags
);
639 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
641 local_irq_restore(flags
);
648 #if defined(CONFIG_PM) || defined(CONFIG_HOTPLUG_CPU)
649 static void __drain_pages(unsigned int cpu
)
655 for_each_zone(zone
) {
656 struct per_cpu_pageset
*pset
;
658 pset
= zone_pcp(zone
, cpu
);
659 for (i
= 0; i
< ARRAY_SIZE(pset
->pcp
); i
++) {
660 struct per_cpu_pages
*pcp
;
663 local_irq_save(flags
);
664 free_pages_bulk(zone
, pcp
->count
, &pcp
->list
, 0);
666 local_irq_restore(flags
);
670 #endif /* CONFIG_PM || CONFIG_HOTPLUG_CPU */
674 void mark_free_pages(struct zone
*zone
)
676 unsigned long zone_pfn
, flags
;
678 struct list_head
*curr
;
680 if (!zone
->spanned_pages
)
683 spin_lock_irqsave(&zone
->lock
, flags
);
684 for (zone_pfn
= 0; zone_pfn
< zone
->spanned_pages
; ++zone_pfn
)
685 ClearPageNosaveFree(pfn_to_page(zone_pfn
+ zone
->zone_start_pfn
));
687 for (order
= MAX_ORDER
- 1; order
>= 0; --order
)
688 list_for_each(curr
, &zone
->free_area
[order
].free_list
) {
689 unsigned long start_pfn
, i
;
691 start_pfn
= page_to_pfn(list_entry(curr
, struct page
, lru
));
693 for (i
=0; i
< (1<<order
); i
++)
694 SetPageNosaveFree(pfn_to_page(start_pfn
+i
));
696 spin_unlock_irqrestore(&zone
->lock
, flags
);
700 * Spill all of this CPU's per-cpu pages back into the buddy allocator.
702 void drain_local_pages(void)
706 local_irq_save(flags
);
707 __drain_pages(smp_processor_id());
708 local_irq_restore(flags
);
710 #endif /* CONFIG_PM */
712 static void zone_statistics(struct zonelist
*zonelist
, struct zone
*z
, int cpu
)
715 pg_data_t
*pg
= z
->zone_pgdat
;
716 pg_data_t
*orig
= zonelist
->zones
[0]->zone_pgdat
;
717 struct per_cpu_pageset
*p
;
719 p
= zone_pcp(z
, cpu
);
724 zone_pcp(zonelist
->zones
[0], cpu
)->numa_foreign
++;
726 if (pg
== NODE_DATA(numa_node_id()))
734 * Free a 0-order page
736 static void fastcall
free_hot_cold_page(struct page
*page
, int cold
)
738 struct zone
*zone
= page_zone(page
);
739 struct per_cpu_pages
*pcp
;
742 arch_free_page(page
, 0);
745 page
->mapping
= NULL
;
746 if (free_pages_check(page
))
749 kernel_map_pages(page
, 1, 0);
751 pcp
= &zone_pcp(zone
, get_cpu())->pcp
[cold
];
752 local_irq_save(flags
);
753 __inc_page_state(pgfree
);
754 list_add(&page
->lru
, &pcp
->list
);
756 if (pcp
->count
>= pcp
->high
) {
757 free_pages_bulk(zone
, pcp
->batch
, &pcp
->list
, 0);
758 pcp
->count
-= pcp
->batch
;
760 local_irq_restore(flags
);
764 void fastcall
free_hot_page(struct page
*page
)
766 free_hot_cold_page(page
, 0);
769 void fastcall
free_cold_page(struct page
*page
)
771 free_hot_cold_page(page
, 1);
775 * split_page takes a non-compound higher-order page, and splits it into
776 * n (1<<order) sub-pages: page[0..n]
777 * Each sub-page must be freed individually.
779 * Note: this is probably too low level an operation for use in drivers.
780 * Please consult with lkml before using this in your driver.
782 void split_page(struct page
*page
, unsigned int order
)
786 BUG_ON(PageCompound(page
));
787 BUG_ON(!page_count(page
));
788 for (i
= 1; i
< (1 << order
); i
++)
789 set_page_refcounted(page
+ i
);
793 * Really, prep_compound_page() should be called from __rmqueue_bulk(). But
794 * we cheat by calling it from here, in the order > 0 path. Saves a branch
797 static struct page
*buffered_rmqueue(struct zonelist
*zonelist
,
798 struct zone
*zone
, int order
, gfp_t gfp_flags
)
802 int cold
= !!(gfp_flags
& __GFP_COLD
);
807 if (likely(order
== 0)) {
808 struct per_cpu_pages
*pcp
;
810 pcp
= &zone_pcp(zone
, cpu
)->pcp
[cold
];
811 local_irq_save(flags
);
813 pcp
->count
+= rmqueue_bulk(zone
, 0,
814 pcp
->batch
, &pcp
->list
);
815 if (unlikely(!pcp
->count
))
818 page
= list_entry(pcp
->list
.next
, struct page
, lru
);
819 list_del(&page
->lru
);
822 spin_lock_irqsave(&zone
->lock
, flags
);
823 page
= __rmqueue(zone
, order
);
824 spin_unlock(&zone
->lock
);
829 __mod_page_state_zone(zone
, pgalloc
, 1 << order
);
830 zone_statistics(zonelist
, zone
, cpu
);
831 local_irq_restore(flags
);
834 BUG_ON(bad_range(zone
, page
));
835 if (prep_new_page(page
, order
, gfp_flags
))
840 local_irq_restore(flags
);
845 #define ALLOC_NO_WATERMARKS 0x01 /* don't check watermarks at all */
846 #define ALLOC_WMARK_MIN 0x02 /* use pages_min watermark */
847 #define ALLOC_WMARK_LOW 0x04 /* use pages_low watermark */
848 #define ALLOC_WMARK_HIGH 0x08 /* use pages_high watermark */
849 #define ALLOC_HARDER 0x10 /* try to alloc harder */
850 #define ALLOC_HIGH 0x20 /* __GFP_HIGH set */
851 #define ALLOC_CPUSET 0x40 /* check for correct cpuset */
854 * Return 1 if free pages are above 'mark'. This takes into account the order
857 int zone_watermark_ok(struct zone
*z
, int order
, unsigned long mark
,
858 int classzone_idx
, int alloc_flags
)
860 /* free_pages my go negative - that's OK */
861 long min
= mark
, free_pages
= z
->free_pages
- (1 << order
) + 1;
864 if (alloc_flags
& ALLOC_HIGH
)
866 if (alloc_flags
& ALLOC_HARDER
)
869 if (free_pages
<= min
+ z
->lowmem_reserve
[classzone_idx
])
871 for (o
= 0; o
< order
; o
++) {
872 /* At the next order, this order's pages become unavailable */
873 free_pages
-= z
->free_area
[o
].nr_free
<< o
;
875 /* Require fewer higher order pages to be free */
878 if (free_pages
<= min
)
885 * get_page_from_freeliest goes through the zonelist trying to allocate
889 get_page_from_freelist(gfp_t gfp_mask
, unsigned int order
,
890 struct zonelist
*zonelist
, int alloc_flags
)
892 struct zone
**z
= zonelist
->zones
;
893 struct page
*page
= NULL
;
894 int classzone_idx
= zone_idx(*z
);
897 * Go through the zonelist once, looking for a zone with enough free.
898 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
901 if ((alloc_flags
& ALLOC_CPUSET
) &&
902 !cpuset_zone_allowed(*z
, gfp_mask
))
905 if (!(alloc_flags
& ALLOC_NO_WATERMARKS
)) {
907 if (alloc_flags
& ALLOC_WMARK_MIN
)
908 mark
= (*z
)->pages_min
;
909 else if (alloc_flags
& ALLOC_WMARK_LOW
)
910 mark
= (*z
)->pages_low
;
912 mark
= (*z
)->pages_high
;
913 if (!zone_watermark_ok(*z
, order
, mark
,
914 classzone_idx
, alloc_flags
))
915 if (!zone_reclaim_mode
||
916 !zone_reclaim(*z
, gfp_mask
, order
))
920 page
= buffered_rmqueue(zonelist
, *z
, order
, gfp_mask
);
924 } while (*(++z
) != NULL
);
929 * This is the 'heart' of the zoned buddy allocator.
931 struct page
* fastcall
932 __alloc_pages(gfp_t gfp_mask
, unsigned int order
,
933 struct zonelist
*zonelist
)
935 const gfp_t wait
= gfp_mask
& __GFP_WAIT
;
938 struct reclaim_state reclaim_state
;
939 struct task_struct
*p
= current
;
942 int did_some_progress
;
944 might_sleep_if(wait
);
947 z
= zonelist
->zones
; /* the list of zones suitable for gfp_mask */
949 if (unlikely(*z
== NULL
)) {
950 /* Should this ever happen?? */
954 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
955 zonelist
, ALLOC_WMARK_LOW
|ALLOC_CPUSET
);
960 wakeup_kswapd(*z
, order
);
964 * OK, we're below the kswapd watermark and have kicked background
965 * reclaim. Now things get more complex, so set up alloc_flags according
966 * to how we want to proceed.
968 * The caller may dip into page reserves a bit more if the caller
969 * cannot run direct reclaim, or if the caller has realtime scheduling
970 * policy or is asking for __GFP_HIGH memory. GFP_ATOMIC requests will
971 * set both ALLOC_HARDER (!wait) and ALLOC_HIGH (__GFP_HIGH).
973 alloc_flags
= ALLOC_WMARK_MIN
;
974 if ((unlikely(rt_task(p
)) && !in_interrupt()) || !wait
)
975 alloc_flags
|= ALLOC_HARDER
;
976 if (gfp_mask
& __GFP_HIGH
)
977 alloc_flags
|= ALLOC_HIGH
;
979 alloc_flags
|= ALLOC_CPUSET
;
982 * Go through the zonelist again. Let __GFP_HIGH and allocations
983 * coming from realtime tasks go deeper into reserves.
985 * This is the last chance, in general, before the goto nopage.
986 * Ignore cpuset if GFP_ATOMIC (!wait) rather than fail alloc.
987 * See also cpuset_zone_allowed() comment in kernel/cpuset.c.
989 page
= get_page_from_freelist(gfp_mask
, order
, zonelist
, alloc_flags
);
993 /* This allocation should allow future memory freeing. */
995 if (((p
->flags
& PF_MEMALLOC
) || unlikely(test_thread_flag(TIF_MEMDIE
)))
996 && !in_interrupt()) {
997 if (!(gfp_mask
& __GFP_NOMEMALLOC
)) {
999 /* go through the zonelist yet again, ignoring mins */
1000 page
= get_page_from_freelist(gfp_mask
, order
,
1001 zonelist
, ALLOC_NO_WATERMARKS
);
1004 if (gfp_mask
& __GFP_NOFAIL
) {
1005 blk_congestion_wait(WRITE
, HZ
/50);
1012 /* Atomic allocations - we can't balance anything */
1019 /* We now go into synchronous reclaim */
1020 cpuset_memory_pressure_bump();
1021 p
->flags
|= PF_MEMALLOC
;
1022 reclaim_state
.reclaimed_slab
= 0;
1023 p
->reclaim_state
= &reclaim_state
;
1025 did_some_progress
= try_to_free_pages(zonelist
->zones
, gfp_mask
);
1027 p
->reclaim_state
= NULL
;
1028 p
->flags
&= ~PF_MEMALLOC
;
1032 if (likely(did_some_progress
)) {
1033 page
= get_page_from_freelist(gfp_mask
, order
,
1034 zonelist
, alloc_flags
);
1037 } else if ((gfp_mask
& __GFP_FS
) && !(gfp_mask
& __GFP_NORETRY
)) {
1039 * Go through the zonelist yet one more time, keep
1040 * very high watermark here, this is only to catch
1041 * a parallel oom killing, we must fail if we're still
1042 * under heavy pressure.
1044 page
= get_page_from_freelist(gfp_mask
|__GFP_HARDWALL
, order
,
1045 zonelist
, ALLOC_WMARK_HIGH
|ALLOC_CPUSET
);
1049 out_of_memory(zonelist
, gfp_mask
, order
);
1054 * Don't let big-order allocations loop unless the caller explicitly
1055 * requests that. Wait for some write requests to complete then retry.
1057 * In this implementation, __GFP_REPEAT means __GFP_NOFAIL for order
1058 * <= 3, but that may not be true in other implementations.
1061 if (!(gfp_mask
& __GFP_NORETRY
)) {
1062 if ((order
<= 3) || (gfp_mask
& __GFP_REPEAT
))
1064 if (gfp_mask
& __GFP_NOFAIL
)
1068 blk_congestion_wait(WRITE
, HZ
/50);
1073 if (!(gfp_mask
& __GFP_NOWARN
) && printk_ratelimit()) {
1074 printk(KERN_WARNING
"%s: page allocation failure."
1075 " order:%d, mode:0x%x\n",
1076 p
->comm
, order
, gfp_mask
);
1084 EXPORT_SYMBOL(__alloc_pages
);
1087 * Common helper functions.
1089 fastcall
unsigned long __get_free_pages(gfp_t gfp_mask
, unsigned int order
)
1092 page
= alloc_pages(gfp_mask
, order
);
1095 return (unsigned long) page_address(page
);
1098 EXPORT_SYMBOL(__get_free_pages
);
1100 fastcall
unsigned long get_zeroed_page(gfp_t gfp_mask
)
1105 * get_zeroed_page() returns a 32-bit address, which cannot represent
1108 BUG_ON((gfp_mask
& __GFP_HIGHMEM
) != 0);
1110 page
= alloc_pages(gfp_mask
| __GFP_ZERO
, 0);
1112 return (unsigned long) page_address(page
);
1116 EXPORT_SYMBOL(get_zeroed_page
);
1118 void __pagevec_free(struct pagevec
*pvec
)
1120 int i
= pagevec_count(pvec
);
1123 free_hot_cold_page(pvec
->pages
[i
], pvec
->cold
);
1126 fastcall
void __free_pages(struct page
*page
, unsigned int order
)
1128 if (put_page_testzero(page
)) {
1130 free_hot_page(page
);
1132 __free_pages_ok(page
, order
);
1136 EXPORT_SYMBOL(__free_pages
);
1138 fastcall
void free_pages(unsigned long addr
, unsigned int order
)
1141 BUG_ON(!virt_addr_valid((void *)addr
));
1142 __free_pages(virt_to_page((void *)addr
), order
);
1146 EXPORT_SYMBOL(free_pages
);
1149 * Total amount of free (allocatable) RAM:
1151 unsigned int nr_free_pages(void)
1153 unsigned int sum
= 0;
1157 sum
+= zone
->free_pages
;
1162 EXPORT_SYMBOL(nr_free_pages
);
1165 unsigned int nr_free_pages_pgdat(pg_data_t
*pgdat
)
1167 unsigned int i
, sum
= 0;
1169 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1170 sum
+= pgdat
->node_zones
[i
].free_pages
;
1176 static unsigned int nr_free_zone_pages(int offset
)
1178 /* Just pick one node, since fallback list is circular */
1179 pg_data_t
*pgdat
= NODE_DATA(numa_node_id());
1180 unsigned int sum
= 0;
1182 struct zonelist
*zonelist
= pgdat
->node_zonelists
+ offset
;
1183 struct zone
**zonep
= zonelist
->zones
;
1186 for (zone
= *zonep
++; zone
; zone
= *zonep
++) {
1187 unsigned long size
= zone
->present_pages
;
1188 unsigned long high
= zone
->pages_high
;
1197 * Amount of free RAM allocatable within ZONE_DMA and ZONE_NORMAL
1199 unsigned int nr_free_buffer_pages(void)
1201 return nr_free_zone_pages(gfp_zone(GFP_USER
));
1205 * Amount of free RAM allocatable within all zones
1207 unsigned int nr_free_pagecache_pages(void)
1209 return nr_free_zone_pages(gfp_zone(GFP_HIGHUSER
));
1212 #ifdef CONFIG_HIGHMEM
1213 unsigned int nr_free_highpages (void)
1216 unsigned int pages
= 0;
1218 for_each_online_pgdat(pgdat
)
1219 pages
+= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1226 static void show_node(struct zone
*zone
)
1228 printk("Node %d ", zone
->zone_pgdat
->node_id
);
1231 #define show_node(zone) do { } while (0)
1234 void si_meminfo(struct sysinfo
*val
)
1236 val
->totalram
= totalram_pages
;
1238 val
->freeram
= nr_free_pages();
1239 val
->bufferram
= nr_blockdev_pages();
1240 #ifdef CONFIG_HIGHMEM
1241 val
->totalhigh
= totalhigh_pages
;
1242 val
->freehigh
= nr_free_highpages();
1247 val
->mem_unit
= PAGE_SIZE
;
1250 EXPORT_SYMBOL(si_meminfo
);
1253 void si_meminfo_node(struct sysinfo
*val
, int nid
)
1255 pg_data_t
*pgdat
= NODE_DATA(nid
);
1257 val
->totalram
= pgdat
->node_present_pages
;
1258 val
->freeram
= nr_free_pages_pgdat(pgdat
);
1259 val
->totalhigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].present_pages
;
1260 val
->freehigh
= pgdat
->node_zones
[ZONE_HIGHMEM
].free_pages
;
1261 val
->mem_unit
= PAGE_SIZE
;
1265 #define K(x) ((x) << (PAGE_SHIFT-10))
1268 * Show free area list (used inside shift_scroll-lock stuff)
1269 * We also calculate the percentage fragmentation. We do this by counting the
1270 * memory on each free list with the exception of the first item on the list.
1272 void show_free_areas(void)
1274 struct page_state ps
;
1275 int cpu
, temperature
;
1276 unsigned long active
;
1277 unsigned long inactive
;
1281 for_each_zone(zone
) {
1283 printk("%s per-cpu:", zone
->name
);
1285 if (!populated_zone(zone
)) {
1291 for_each_online_cpu(cpu
) {
1292 struct per_cpu_pageset
*pageset
;
1294 pageset
= zone_pcp(zone
, cpu
);
1296 for (temperature
= 0; temperature
< 2; temperature
++)
1297 printk("cpu %d %s: high %d, batch %d used:%d\n",
1299 temperature
? "cold" : "hot",
1300 pageset
->pcp
[temperature
].high
,
1301 pageset
->pcp
[temperature
].batch
,
1302 pageset
->pcp
[temperature
].count
);
1306 get_page_state(&ps
);
1307 get_zone_counts(&active
, &inactive
, &free
);
1309 printk("Free pages: %11ukB (%ukB HighMem)\n",
1311 K(nr_free_highpages()));
1313 printk("Active:%lu inactive:%lu dirty:%lu writeback:%lu "
1314 "unstable:%lu free:%u slab:%lu mapped:%lu pagetables:%lu\n",
1321 global_page_state(NR_SLAB
),
1322 global_page_state(NR_FILE_MAPPED
),
1323 ps
.nr_page_table_pages
);
1325 for_each_zone(zone
) {
1337 " pages_scanned:%lu"
1338 " all_unreclaimable? %s"
1341 K(zone
->free_pages
),
1344 K(zone
->pages_high
),
1346 K(zone
->nr_inactive
),
1347 K(zone
->present_pages
),
1348 zone
->pages_scanned
,
1349 (zone
->all_unreclaimable
? "yes" : "no")
1351 printk("lowmem_reserve[]:");
1352 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1353 printk(" %lu", zone
->lowmem_reserve
[i
]);
1357 for_each_zone(zone
) {
1358 unsigned long nr
[MAX_ORDER
], flags
, order
, total
= 0;
1361 printk("%s: ", zone
->name
);
1362 if (!populated_zone(zone
)) {
1367 spin_lock_irqsave(&zone
->lock
, flags
);
1368 for (order
= 0; order
< MAX_ORDER
; order
++) {
1369 nr
[order
] = zone
->free_area
[order
].nr_free
;
1370 total
+= nr
[order
] << order
;
1372 spin_unlock_irqrestore(&zone
->lock
, flags
);
1373 for (order
= 0; order
< MAX_ORDER
; order
++)
1374 printk("%lu*%lukB ", nr
[order
], K(1UL) << order
);
1375 printk("= %lukB\n", K(total
));
1378 show_swap_cache_info();
1382 * Builds allocation fallback zone lists.
1384 * Add all populated zones of a node to the zonelist.
1386 static int __meminit
build_zonelists_node(pg_data_t
*pgdat
,
1387 struct zonelist
*zonelist
, int nr_zones
, int zone_type
)
1391 BUG_ON(zone_type
> ZONE_HIGHMEM
);
1394 zone
= pgdat
->node_zones
+ zone_type
;
1395 if (populated_zone(zone
)) {
1396 #ifndef CONFIG_HIGHMEM
1397 BUG_ON(zone_type
> ZONE_NORMAL
);
1399 zonelist
->zones
[nr_zones
++] = zone
;
1400 check_highest_zone(zone_type
);
1404 } while (zone_type
>= 0);
1408 static inline int highest_zone(int zone_bits
)
1410 int res
= ZONE_NORMAL
;
1411 if (zone_bits
& (__force
int)__GFP_HIGHMEM
)
1413 if (zone_bits
& (__force
int)__GFP_DMA32
)
1415 if (zone_bits
& (__force
int)__GFP_DMA
)
1421 #define MAX_NODE_LOAD (num_online_nodes())
1422 static int __meminitdata node_load
[MAX_NUMNODES
];
1424 * find_next_best_node - find the next node that should appear in a given node's fallback list
1425 * @node: node whose fallback list we're appending
1426 * @used_node_mask: nodemask_t of already used nodes
1428 * We use a number of factors to determine which is the next node that should
1429 * appear on a given node's fallback list. The node should not have appeared
1430 * already in @node's fallback list, and it should be the next closest node
1431 * according to the distance array (which contains arbitrary distance values
1432 * from each node to each node in the system), and should also prefer nodes
1433 * with no CPUs, since presumably they'll have very little allocation pressure
1434 * on them otherwise.
1435 * It returns -1 if no node is found.
1437 static int __meminit
find_next_best_node(int node
, nodemask_t
*used_node_mask
)
1440 int min_val
= INT_MAX
;
1443 /* Use the local node if we haven't already */
1444 if (!node_isset(node
, *used_node_mask
)) {
1445 node_set(node
, *used_node_mask
);
1449 for_each_online_node(n
) {
1452 /* Don't want a node to appear more than once */
1453 if (node_isset(n
, *used_node_mask
))
1456 /* Use the distance array to find the distance */
1457 val
= node_distance(node
, n
);
1459 /* Penalize nodes under us ("prefer the next node") */
1462 /* Give preference to headless and unused nodes */
1463 tmp
= node_to_cpumask(n
);
1464 if (!cpus_empty(tmp
))
1465 val
+= PENALTY_FOR_NODE_WITH_CPUS
;
1467 /* Slight preference for less loaded node */
1468 val
*= (MAX_NODE_LOAD
*MAX_NUMNODES
);
1469 val
+= node_load
[n
];
1471 if (val
< min_val
) {
1478 node_set(best_node
, *used_node_mask
);
1483 static void __meminit
build_zonelists(pg_data_t
*pgdat
)
1485 int i
, j
, k
, node
, local_node
;
1486 int prev_node
, load
;
1487 struct zonelist
*zonelist
;
1488 nodemask_t used_mask
;
1490 /* initialize zonelists */
1491 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1492 zonelist
= pgdat
->node_zonelists
+ i
;
1493 zonelist
->zones
[0] = NULL
;
1496 /* NUMA-aware ordering of nodes */
1497 local_node
= pgdat
->node_id
;
1498 load
= num_online_nodes();
1499 prev_node
= local_node
;
1500 nodes_clear(used_mask
);
1501 while ((node
= find_next_best_node(local_node
, &used_mask
)) >= 0) {
1502 int distance
= node_distance(local_node
, node
);
1505 * If another node is sufficiently far away then it is better
1506 * to reclaim pages in a zone before going off node.
1508 if (distance
> RECLAIM_DISTANCE
)
1509 zone_reclaim_mode
= 1;
1512 * We don't want to pressure a particular node.
1513 * So adding penalty to the first node in same
1514 * distance group to make it round-robin.
1517 if (distance
!= node_distance(local_node
, prev_node
))
1518 node_load
[node
] += load
;
1521 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1522 zonelist
= pgdat
->node_zonelists
+ i
;
1523 for (j
= 0; zonelist
->zones
[j
] != NULL
; j
++);
1525 k
= highest_zone(i
);
1527 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1528 zonelist
->zones
[j
] = NULL
;
1533 #else /* CONFIG_NUMA */
1535 static void __meminit
build_zonelists(pg_data_t
*pgdat
)
1537 int i
, j
, k
, node
, local_node
;
1539 local_node
= pgdat
->node_id
;
1540 for (i
= 0; i
< GFP_ZONETYPES
; i
++) {
1541 struct zonelist
*zonelist
;
1543 zonelist
= pgdat
->node_zonelists
+ i
;
1546 k
= highest_zone(i
);
1547 j
= build_zonelists_node(pgdat
, zonelist
, j
, k
);
1549 * Now we build the zonelist so that it contains the zones
1550 * of all the other nodes.
1551 * We don't want to pressure a particular node, so when
1552 * building the zones for node N, we make sure that the
1553 * zones coming right after the local ones are those from
1554 * node N+1 (modulo N)
1556 for (node
= local_node
+ 1; node
< MAX_NUMNODES
; node
++) {
1557 if (!node_online(node
))
1559 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1561 for (node
= 0; node
< local_node
; node
++) {
1562 if (!node_online(node
))
1564 j
= build_zonelists_node(NODE_DATA(node
), zonelist
, j
, k
);
1567 zonelist
->zones
[j
] = NULL
;
1571 #endif /* CONFIG_NUMA */
1573 /* return values int ....just for stop_machine_run() */
1574 static int __meminit
__build_all_zonelists(void *dummy
)
1577 for_each_online_node(nid
)
1578 build_zonelists(NODE_DATA(nid
));
1582 void __meminit
build_all_zonelists(void)
1584 if (system_state
== SYSTEM_BOOTING
) {
1585 __build_all_zonelists(0);
1586 cpuset_init_current_mems_allowed();
1588 /* we have to stop all cpus to guaranntee there is no user
1590 stop_machine_run(__build_all_zonelists
, NULL
, NR_CPUS
);
1591 /* cpuset refresh routine should be here */
1593 vm_total_pages
= nr_free_pagecache_pages();
1594 printk("Built %i zonelists. Total pages: %ld\n",
1595 num_online_nodes(), vm_total_pages
);
1599 * Helper functions to size the waitqueue hash table.
1600 * Essentially these want to choose hash table sizes sufficiently
1601 * large so that collisions trying to wait on pages are rare.
1602 * But in fact, the number of active page waitqueues on typical
1603 * systems is ridiculously low, less than 200. So this is even
1604 * conservative, even though it seems large.
1606 * The constant PAGES_PER_WAITQUEUE specifies the ratio of pages to
1607 * waitqueues, i.e. the size of the waitq table given the number of pages.
1609 #define PAGES_PER_WAITQUEUE 256
1611 #ifndef CONFIG_MEMORY_HOTPLUG
1612 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
1614 unsigned long size
= 1;
1616 pages
/= PAGES_PER_WAITQUEUE
;
1618 while (size
< pages
)
1622 * Once we have dozens or even hundreds of threads sleeping
1623 * on IO we've got bigger problems than wait queue collision.
1624 * Limit the size of the wait table to a reasonable size.
1626 size
= min(size
, 4096UL);
1628 return max(size
, 4UL);
1632 * A zone's size might be changed by hot-add, so it is not possible to determine
1633 * a suitable size for its wait_table. So we use the maximum size now.
1635 * The max wait table size = 4096 x sizeof(wait_queue_head_t). ie:
1637 * i386 (preemption config) : 4096 x 16 = 64Kbyte.
1638 * ia64, x86-64 (no preemption): 4096 x 20 = 80Kbyte.
1639 * ia64, x86-64 (preemption) : 4096 x 24 = 96Kbyte.
1641 * The maximum entries are prepared when a zone's memory is (512K + 256) pages
1642 * or more by the traditional way. (See above). It equals:
1644 * i386, x86-64, powerpc(4K page size) : = ( 2G + 1M)byte.
1645 * ia64(16K page size) : = ( 8G + 4M)byte.
1646 * powerpc (64K page size) : = (32G +16M)byte.
1648 static inline unsigned long wait_table_hash_nr_entries(unsigned long pages
)
1655 * This is an integer logarithm so that shifts can be used later
1656 * to extract the more random high bits from the multiplicative
1657 * hash function before the remainder is taken.
1659 static inline unsigned long wait_table_bits(unsigned long size
)
1664 #define LONG_ALIGN(x) (((x)+(sizeof(long))-1)&~((sizeof(long))-1))
1666 static void __init
calculate_zone_totalpages(struct pglist_data
*pgdat
,
1667 unsigned long *zones_size
, unsigned long *zholes_size
)
1669 unsigned long realtotalpages
, totalpages
= 0;
1672 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1673 totalpages
+= zones_size
[i
];
1674 pgdat
->node_spanned_pages
= totalpages
;
1676 realtotalpages
= totalpages
;
1678 for (i
= 0; i
< MAX_NR_ZONES
; i
++)
1679 realtotalpages
-= zholes_size
[i
];
1680 pgdat
->node_present_pages
= realtotalpages
;
1681 printk(KERN_DEBUG
"On node %d totalpages: %lu\n", pgdat
->node_id
, realtotalpages
);
1686 * Initially all pages are reserved - free ones are freed
1687 * up by free_all_bootmem() once the early boot process is
1688 * done. Non-atomic initialization, single-pass.
1690 void __meminit
memmap_init_zone(unsigned long size
, int nid
, unsigned long zone
,
1691 unsigned long start_pfn
)
1694 unsigned long end_pfn
= start_pfn
+ size
;
1697 for (pfn
= start_pfn
; pfn
< end_pfn
; pfn
++) {
1698 if (!early_pfn_valid(pfn
))
1700 page
= pfn_to_page(pfn
);
1701 set_page_links(page
, zone
, nid
, pfn
);
1702 init_page_count(page
);
1703 reset_page_mapcount(page
);
1704 SetPageReserved(page
);
1705 INIT_LIST_HEAD(&page
->lru
);
1706 #ifdef WANT_PAGE_VIRTUAL
1707 /* The shift won't overflow because ZONE_NORMAL is below 4G. */
1708 if (!is_highmem_idx(zone
))
1709 set_page_address(page
, __va(pfn
<< PAGE_SHIFT
));
1714 void zone_init_free_lists(struct pglist_data
*pgdat
, struct zone
*zone
,
1718 for (order
= 0; order
< MAX_ORDER
; order
++) {
1719 INIT_LIST_HEAD(&zone
->free_area
[order
].free_list
);
1720 zone
->free_area
[order
].nr_free
= 0;
1724 #define ZONETABLE_INDEX(x, zone_nr) ((x << ZONES_SHIFT) | zone_nr)
1725 void zonetable_add(struct zone
*zone
, int nid
, int zid
, unsigned long pfn
,
1728 unsigned long snum
= pfn_to_section_nr(pfn
);
1729 unsigned long end
= pfn_to_section_nr(pfn
+ size
);
1732 zone_table
[ZONETABLE_INDEX(nid
, zid
)] = zone
;
1734 for (; snum
<= end
; snum
++)
1735 zone_table
[ZONETABLE_INDEX(snum
, zid
)] = zone
;
1738 #ifndef __HAVE_ARCH_MEMMAP_INIT
1739 #define memmap_init(size, nid, zone, start_pfn) \
1740 memmap_init_zone((size), (nid), (zone), (start_pfn))
1743 static int __cpuinit
zone_batchsize(struct zone
*zone
)
1748 * The per-cpu-pages pools are set to around 1000th of the
1749 * size of the zone. But no more than 1/2 of a meg.
1751 * OK, so we don't know how big the cache is. So guess.
1753 batch
= zone
->present_pages
/ 1024;
1754 if (batch
* PAGE_SIZE
> 512 * 1024)
1755 batch
= (512 * 1024) / PAGE_SIZE
;
1756 batch
/= 4; /* We effectively *= 4 below */
1761 * Clamp the batch to a 2^n - 1 value. Having a power
1762 * of 2 value was found to be more likely to have
1763 * suboptimal cache aliasing properties in some cases.
1765 * For example if 2 tasks are alternately allocating
1766 * batches of pages, one task can end up with a lot
1767 * of pages of one half of the possible page colors
1768 * and the other with pages of the other colors.
1770 batch
= (1 << (fls(batch
+ batch
/2)-1)) - 1;
1775 inline void setup_pageset(struct per_cpu_pageset
*p
, unsigned long batch
)
1777 struct per_cpu_pages
*pcp
;
1779 memset(p
, 0, sizeof(*p
));
1781 pcp
= &p
->pcp
[0]; /* hot */
1783 pcp
->high
= 6 * batch
;
1784 pcp
->batch
= max(1UL, 1 * batch
);
1785 INIT_LIST_HEAD(&pcp
->list
);
1787 pcp
= &p
->pcp
[1]; /* cold*/
1789 pcp
->high
= 2 * batch
;
1790 pcp
->batch
= max(1UL, batch
/2);
1791 INIT_LIST_HEAD(&pcp
->list
);
1795 * setup_pagelist_highmark() sets the high water mark for hot per_cpu_pagelist
1796 * to the value high for the pageset p.
1799 static void setup_pagelist_highmark(struct per_cpu_pageset
*p
,
1802 struct per_cpu_pages
*pcp
;
1804 pcp
= &p
->pcp
[0]; /* hot list */
1806 pcp
->batch
= max(1UL, high
/4);
1807 if ((high
/4) > (PAGE_SHIFT
* 8))
1808 pcp
->batch
= PAGE_SHIFT
* 8;
1814 * Boot pageset table. One per cpu which is going to be used for all
1815 * zones and all nodes. The parameters will be set in such a way
1816 * that an item put on a list will immediately be handed over to
1817 * the buddy list. This is safe since pageset manipulation is done
1818 * with interrupts disabled.
1820 * Some NUMA counter updates may also be caught by the boot pagesets.
1822 * The boot_pagesets must be kept even after bootup is complete for
1823 * unused processors and/or zones. They do play a role for bootstrapping
1824 * hotplugged processors.
1826 * zoneinfo_show() and maybe other functions do
1827 * not check if the processor is online before following the pageset pointer.
1828 * Other parts of the kernel may not check if the zone is available.
1830 static struct per_cpu_pageset boot_pageset
[NR_CPUS
];
1833 * Dynamically allocate memory for the
1834 * per cpu pageset array in struct zone.
1836 static int __cpuinit
process_zones(int cpu
)
1838 struct zone
*zone
, *dzone
;
1840 for_each_zone(zone
) {
1842 zone_pcp(zone
, cpu
) = kmalloc_node(sizeof(struct per_cpu_pageset
),
1843 GFP_KERNEL
, cpu_to_node(cpu
));
1844 if (!zone_pcp(zone
, cpu
))
1847 setup_pageset(zone_pcp(zone
, cpu
), zone_batchsize(zone
));
1849 if (percpu_pagelist_fraction
)
1850 setup_pagelist_highmark(zone_pcp(zone
, cpu
),
1851 (zone
->present_pages
/ percpu_pagelist_fraction
));
1856 for_each_zone(dzone
) {
1859 kfree(zone_pcp(dzone
, cpu
));
1860 zone_pcp(dzone
, cpu
) = NULL
;
1865 static inline void free_zone_pagesets(int cpu
)
1869 for_each_zone(zone
) {
1870 struct per_cpu_pageset
*pset
= zone_pcp(zone
, cpu
);
1872 zone_pcp(zone
, cpu
) = NULL
;
1877 static int __cpuinit
pageset_cpuup_callback(struct notifier_block
*nfb
,
1878 unsigned long action
,
1881 int cpu
= (long)hcpu
;
1882 int ret
= NOTIFY_OK
;
1885 case CPU_UP_PREPARE
:
1886 if (process_zones(cpu
))
1889 case CPU_UP_CANCELED
:
1891 free_zone_pagesets(cpu
);
1899 static struct notifier_block __cpuinitdata pageset_notifier
=
1900 { &pageset_cpuup_callback
, NULL
, 0 };
1902 void __init
setup_per_cpu_pageset(void)
1906 /* Initialize per_cpu_pageset for cpu 0.
1907 * A cpuup callback will do this for every cpu
1908 * as it comes online
1910 err
= process_zones(smp_processor_id());
1912 register_cpu_notifier(&pageset_notifier
);
1918 int zone_wait_table_init(struct zone
*zone
, unsigned long zone_size_pages
)
1921 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1925 * The per-page waitqueue mechanism uses hashed waitqueues
1928 zone
->wait_table_hash_nr_entries
=
1929 wait_table_hash_nr_entries(zone_size_pages
);
1930 zone
->wait_table_bits
=
1931 wait_table_bits(zone
->wait_table_hash_nr_entries
);
1932 alloc_size
= zone
->wait_table_hash_nr_entries
1933 * sizeof(wait_queue_head_t
);
1935 if (system_state
== SYSTEM_BOOTING
) {
1936 zone
->wait_table
= (wait_queue_head_t
*)
1937 alloc_bootmem_node(pgdat
, alloc_size
);
1940 * This case means that a zone whose size was 0 gets new memory
1941 * via memory hot-add.
1942 * But it may be the case that a new node was hot-added. In
1943 * this case vmalloc() will not be able to use this new node's
1944 * memory - this wait_table must be initialized to use this new
1945 * node itself as well.
1946 * To use this new node's memory, further consideration will be
1949 zone
->wait_table
= (wait_queue_head_t
*)vmalloc(alloc_size
);
1951 if (!zone
->wait_table
)
1954 for(i
= 0; i
< zone
->wait_table_hash_nr_entries
; ++i
)
1955 init_waitqueue_head(zone
->wait_table
+ i
);
1960 static __meminit
void zone_pcp_init(struct zone
*zone
)
1963 unsigned long batch
= zone_batchsize(zone
);
1965 for (cpu
= 0; cpu
< NR_CPUS
; cpu
++) {
1967 /* Early boot. Slab allocator not functional yet */
1968 zone_pcp(zone
, cpu
) = &boot_pageset
[cpu
];
1969 setup_pageset(&boot_pageset
[cpu
],0);
1971 setup_pageset(zone_pcp(zone
,cpu
), batch
);
1974 if (zone
->present_pages
)
1975 printk(KERN_DEBUG
" %s zone: %lu pages, LIFO batch:%lu\n",
1976 zone
->name
, zone
->present_pages
, batch
);
1979 __meminit
int init_currently_empty_zone(struct zone
*zone
,
1980 unsigned long zone_start_pfn
,
1983 struct pglist_data
*pgdat
= zone
->zone_pgdat
;
1985 ret
= zone_wait_table_init(zone
, size
);
1988 pgdat
->nr_zones
= zone_idx(zone
) + 1;
1990 zone
->zone_start_pfn
= zone_start_pfn
;
1992 memmap_init(size
, pgdat
->node_id
, zone_idx(zone
), zone_start_pfn
);
1994 zone_init_free_lists(pgdat
, zone
, zone
->spanned_pages
);
2000 * Set up the zone data structures:
2001 * - mark all pages reserved
2002 * - mark all memory queues empty
2003 * - clear the memory bitmaps
2005 static void __meminit
free_area_init_core(struct pglist_data
*pgdat
,
2006 unsigned long *zones_size
, unsigned long *zholes_size
)
2009 int nid
= pgdat
->node_id
;
2010 unsigned long zone_start_pfn
= pgdat
->node_start_pfn
;
2013 pgdat_resize_init(pgdat
);
2014 pgdat
->nr_zones
= 0;
2015 init_waitqueue_head(&pgdat
->kswapd_wait
);
2016 pgdat
->kswapd_max_order
= 0;
2018 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2019 struct zone
*zone
= pgdat
->node_zones
+ j
;
2020 unsigned long size
, realsize
;
2022 realsize
= size
= zones_size
[j
];
2024 realsize
-= zholes_size
[j
];
2026 if (j
< ZONE_HIGHMEM
)
2027 nr_kernel_pages
+= realsize
;
2028 nr_all_pages
+= realsize
;
2030 zone
->spanned_pages
= size
;
2031 zone
->present_pages
= realsize
;
2032 zone
->name
= zone_names
[j
];
2033 spin_lock_init(&zone
->lock
);
2034 spin_lock_init(&zone
->lru_lock
);
2035 zone_seqlock_init(zone
);
2036 zone
->zone_pgdat
= pgdat
;
2037 zone
->free_pages
= 0;
2039 zone
->temp_priority
= zone
->prev_priority
= DEF_PRIORITY
;
2041 zone_pcp_init(zone
);
2042 INIT_LIST_HEAD(&zone
->active_list
);
2043 INIT_LIST_HEAD(&zone
->inactive_list
);
2044 zone
->nr_scan_active
= 0;
2045 zone
->nr_scan_inactive
= 0;
2046 zone
->nr_active
= 0;
2047 zone
->nr_inactive
= 0;
2048 zap_zone_vm_stats(zone
);
2049 atomic_set(&zone
->reclaim_in_progress
, 0);
2053 zonetable_add(zone
, nid
, j
, zone_start_pfn
, size
);
2054 ret
= init_currently_empty_zone(zone
, zone_start_pfn
, size
);
2056 zone_start_pfn
+= size
;
2060 static void __init
alloc_node_mem_map(struct pglist_data
*pgdat
)
2062 /* Skip empty nodes */
2063 if (!pgdat
->node_spanned_pages
)
2066 #ifdef CONFIG_FLAT_NODE_MEM_MAP
2067 /* ia64 gets its own node_mem_map, before this, without bootmem */
2068 if (!pgdat
->node_mem_map
) {
2069 unsigned long size
, start
, end
;
2073 * The zone's endpoints aren't required to be MAX_ORDER
2074 * aligned but the node_mem_map endpoints must be in order
2075 * for the buddy allocator to function correctly.
2077 start
= pgdat
->node_start_pfn
& ~(MAX_ORDER_NR_PAGES
- 1);
2078 end
= pgdat
->node_start_pfn
+ pgdat
->node_spanned_pages
;
2079 end
= ALIGN(end
, MAX_ORDER_NR_PAGES
);
2080 size
= (end
- start
) * sizeof(struct page
);
2081 map
= alloc_remap(pgdat
->node_id
, size
);
2083 map
= alloc_bootmem_node(pgdat
, size
);
2084 pgdat
->node_mem_map
= map
+ (pgdat
->node_start_pfn
- start
);
2086 #ifdef CONFIG_FLATMEM
2088 * With no DISCONTIG, the global mem_map is just set as node 0's
2090 if (pgdat
== NODE_DATA(0))
2091 mem_map
= NODE_DATA(0)->node_mem_map
;
2093 #endif /* CONFIG_FLAT_NODE_MEM_MAP */
2096 void __meminit
free_area_init_node(int nid
, struct pglist_data
*pgdat
,
2097 unsigned long *zones_size
, unsigned long node_start_pfn
,
2098 unsigned long *zholes_size
)
2100 pgdat
->node_id
= nid
;
2101 pgdat
->node_start_pfn
= node_start_pfn
;
2102 calculate_zone_totalpages(pgdat
, zones_size
, zholes_size
);
2104 alloc_node_mem_map(pgdat
);
2106 free_area_init_core(pgdat
, zones_size
, zholes_size
);
2109 #ifndef CONFIG_NEED_MULTIPLE_NODES
2110 static bootmem_data_t contig_bootmem_data
;
2111 struct pglist_data contig_page_data
= { .bdata
= &contig_bootmem_data
};
2113 EXPORT_SYMBOL(contig_page_data
);
2116 void __init
free_area_init(unsigned long *zones_size
)
2118 free_area_init_node(0, NODE_DATA(0), zones_size
,
2119 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
, NULL
);
2122 #ifdef CONFIG_HOTPLUG_CPU
2123 static int page_alloc_cpu_notify(struct notifier_block
*self
,
2124 unsigned long action
, void *hcpu
)
2126 int cpu
= (unsigned long)hcpu
;
2127 unsigned long *src
, *dest
;
2129 if (action
== CPU_DEAD
) {
2132 local_irq_disable();
2135 /* Add dead cpu's page_states to our own. */
2136 dest
= (unsigned long *)&__get_cpu_var(page_states
);
2137 src
= (unsigned long *)&per_cpu(page_states
, cpu
);
2139 for (i
= 0; i
< sizeof(struct page_state
)/sizeof(unsigned long);
2146 refresh_cpu_vm_stats(cpu
);
2150 #endif /* CONFIG_HOTPLUG_CPU */
2152 void __init
page_alloc_init(void)
2154 hotcpu_notifier(page_alloc_cpu_notify
, 0);
2158 * calculate_totalreserve_pages - called when sysctl_lower_zone_reserve_ratio
2159 * or min_free_kbytes changes.
2161 static void calculate_totalreserve_pages(void)
2163 struct pglist_data
*pgdat
;
2164 unsigned long reserve_pages
= 0;
2167 for_each_online_pgdat(pgdat
) {
2168 for (i
= 0; i
< MAX_NR_ZONES
; i
++) {
2169 struct zone
*zone
= pgdat
->node_zones
+ i
;
2170 unsigned long max
= 0;
2172 /* Find valid and maximum lowmem_reserve in the zone */
2173 for (j
= i
; j
< MAX_NR_ZONES
; j
++) {
2174 if (zone
->lowmem_reserve
[j
] > max
)
2175 max
= zone
->lowmem_reserve
[j
];
2178 /* we treat pages_high as reserved pages. */
2179 max
+= zone
->pages_high
;
2181 if (max
> zone
->present_pages
)
2182 max
= zone
->present_pages
;
2183 reserve_pages
+= max
;
2186 totalreserve_pages
= reserve_pages
;
2190 * setup_per_zone_lowmem_reserve - called whenever
2191 * sysctl_lower_zone_reserve_ratio changes. Ensures that each zone
2192 * has a correct pages reserved value, so an adequate number of
2193 * pages are left in the zone after a successful __alloc_pages().
2195 static void setup_per_zone_lowmem_reserve(void)
2197 struct pglist_data
*pgdat
;
2200 for_each_online_pgdat(pgdat
) {
2201 for (j
= 0; j
< MAX_NR_ZONES
; j
++) {
2202 struct zone
*zone
= pgdat
->node_zones
+ j
;
2203 unsigned long present_pages
= zone
->present_pages
;
2205 zone
->lowmem_reserve
[j
] = 0;
2207 for (idx
= j
-1; idx
>= 0; idx
--) {
2208 struct zone
*lower_zone
;
2210 if (sysctl_lowmem_reserve_ratio
[idx
] < 1)
2211 sysctl_lowmem_reserve_ratio
[idx
] = 1;
2213 lower_zone
= pgdat
->node_zones
+ idx
;
2214 lower_zone
->lowmem_reserve
[j
] = present_pages
/
2215 sysctl_lowmem_reserve_ratio
[idx
];
2216 present_pages
+= lower_zone
->present_pages
;
2221 /* update totalreserve_pages */
2222 calculate_totalreserve_pages();
2226 * setup_per_zone_pages_min - called when min_free_kbytes changes. Ensures
2227 * that the pages_{min,low,high} values for each zone are set correctly
2228 * with respect to min_free_kbytes.
2230 void setup_per_zone_pages_min(void)
2232 unsigned long pages_min
= min_free_kbytes
>> (PAGE_SHIFT
- 10);
2233 unsigned long lowmem_pages
= 0;
2235 unsigned long flags
;
2237 /* Calculate total number of !ZONE_HIGHMEM pages */
2238 for_each_zone(zone
) {
2239 if (!is_highmem(zone
))
2240 lowmem_pages
+= zone
->present_pages
;
2243 for_each_zone(zone
) {
2246 spin_lock_irqsave(&zone
->lru_lock
, flags
);
2247 tmp
= (u64
)pages_min
* zone
->present_pages
;
2248 do_div(tmp
, lowmem_pages
);
2249 if (is_highmem(zone
)) {
2251 * __GFP_HIGH and PF_MEMALLOC allocations usually don't
2252 * need highmem pages, so cap pages_min to a small
2255 * The (pages_high-pages_low) and (pages_low-pages_min)
2256 * deltas controls asynch page reclaim, and so should
2257 * not be capped for highmem.
2261 min_pages
= zone
->present_pages
/ 1024;
2262 if (min_pages
< SWAP_CLUSTER_MAX
)
2263 min_pages
= SWAP_CLUSTER_MAX
;
2264 if (min_pages
> 128)
2266 zone
->pages_min
= min_pages
;
2269 * If it's a lowmem zone, reserve a number of pages
2270 * proportionate to the zone's size.
2272 zone
->pages_min
= tmp
;
2275 zone
->pages_low
= zone
->pages_min
+ (tmp
>> 2);
2276 zone
->pages_high
= zone
->pages_min
+ (tmp
>> 1);
2277 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
2280 /* update totalreserve_pages */
2281 calculate_totalreserve_pages();
2285 * Initialise min_free_kbytes.
2287 * For small machines we want it small (128k min). For large machines
2288 * we want it large (64MB max). But it is not linear, because network
2289 * bandwidth does not increase linearly with machine size. We use
2291 * min_free_kbytes = 4 * sqrt(lowmem_kbytes), for better accuracy:
2292 * min_free_kbytes = sqrt(lowmem_kbytes * 16)
2308 static int __init
init_per_zone_pages_min(void)
2310 unsigned long lowmem_kbytes
;
2312 lowmem_kbytes
= nr_free_buffer_pages() * (PAGE_SIZE
>> 10);
2314 min_free_kbytes
= int_sqrt(lowmem_kbytes
* 16);
2315 if (min_free_kbytes
< 128)
2316 min_free_kbytes
= 128;
2317 if (min_free_kbytes
> 65536)
2318 min_free_kbytes
= 65536;
2319 setup_per_zone_pages_min();
2320 setup_per_zone_lowmem_reserve();
2323 module_init(init_per_zone_pages_min
)
2326 * min_free_kbytes_sysctl_handler - just a wrapper around proc_dointvec() so
2327 * that we can call two helper functions whenever min_free_kbytes
2330 int min_free_kbytes_sysctl_handler(ctl_table
*table
, int write
,
2331 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2333 proc_dointvec(table
, write
, file
, buffer
, length
, ppos
);
2334 setup_per_zone_pages_min();
2339 * lowmem_reserve_ratio_sysctl_handler - just a wrapper around
2340 * proc_dointvec() so that we can call setup_per_zone_lowmem_reserve()
2341 * whenever sysctl_lowmem_reserve_ratio changes.
2343 * The reserve ratio obviously has absolutely no relation with the
2344 * pages_min watermarks. The lowmem reserve ratio can only make sense
2345 * if in function of the boot time zone sizes.
2347 int lowmem_reserve_ratio_sysctl_handler(ctl_table
*table
, int write
,
2348 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2350 proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2351 setup_per_zone_lowmem_reserve();
2356 * percpu_pagelist_fraction - changes the pcp->high for each zone on each
2357 * cpu. It is the fraction of total pages in each zone that a hot per cpu pagelist
2358 * can have before it gets flushed back to buddy allocator.
2361 int percpu_pagelist_fraction_sysctl_handler(ctl_table
*table
, int write
,
2362 struct file
*file
, void __user
*buffer
, size_t *length
, loff_t
*ppos
)
2368 ret
= proc_dointvec_minmax(table
, write
, file
, buffer
, length
, ppos
);
2369 if (!write
|| (ret
== -EINVAL
))
2371 for_each_zone(zone
) {
2372 for_each_online_cpu(cpu
) {
2374 high
= zone
->present_pages
/ percpu_pagelist_fraction
;
2375 setup_pagelist_highmark(zone_pcp(zone
, cpu
), high
);
2381 __initdata
int hashdist
= HASHDIST_DEFAULT
;
2384 static int __init
set_hashdist(char *str
)
2388 hashdist
= simple_strtoul(str
, &str
, 0);
2391 __setup("hashdist=", set_hashdist
);
2395 * allocate a large system hash table from bootmem
2396 * - it is assumed that the hash table must contain an exact power-of-2
2397 * quantity of entries
2398 * - limit is the number of hash buckets, not the total allocation size
2400 void *__init
alloc_large_system_hash(const char *tablename
,
2401 unsigned long bucketsize
,
2402 unsigned long numentries
,
2405 unsigned int *_hash_shift
,
2406 unsigned int *_hash_mask
,
2407 unsigned long limit
)
2409 unsigned long long max
= limit
;
2410 unsigned long log2qty
, size
;
2413 /* allow the kernel cmdline to have a say */
2415 /* round applicable memory size up to nearest megabyte */
2416 numentries
= (flags
& HASH_HIGHMEM
) ? nr_all_pages
: nr_kernel_pages
;
2417 numentries
+= (1UL << (20 - PAGE_SHIFT
)) - 1;
2418 numentries
>>= 20 - PAGE_SHIFT
;
2419 numentries
<<= 20 - PAGE_SHIFT
;
2421 /* limit to 1 bucket per 2^scale bytes of low memory */
2422 if (scale
> PAGE_SHIFT
)
2423 numentries
>>= (scale
- PAGE_SHIFT
);
2425 numentries
<<= (PAGE_SHIFT
- scale
);
2427 numentries
= roundup_pow_of_two(numentries
);
2429 /* limit allocation size to 1/16 total memory by default */
2431 max
= ((unsigned long long)nr_all_pages
<< PAGE_SHIFT
) >> 4;
2432 do_div(max
, bucketsize
);
2435 if (numentries
> max
)
2438 log2qty
= long_log2(numentries
);
2441 size
= bucketsize
<< log2qty
;
2442 if (flags
& HASH_EARLY
)
2443 table
= alloc_bootmem(size
);
2445 table
= __vmalloc(size
, GFP_ATOMIC
, PAGE_KERNEL
);
2447 unsigned long order
;
2448 for (order
= 0; ((1UL << order
) << PAGE_SHIFT
) < size
; order
++)
2450 table
= (void*) __get_free_pages(GFP_ATOMIC
, order
);
2452 } while (!table
&& size
> PAGE_SIZE
&& --log2qty
);
2455 panic("Failed to allocate %s hash table\n", tablename
);
2457 printk("%s hash table entries: %d (order: %d, %lu bytes)\n",
2460 long_log2(size
) - PAGE_SHIFT
,
2464 *_hash_shift
= log2qty
;
2466 *_hash_mask
= (1 << log2qty
) - 1;
2471 #ifdef CONFIG_OUT_OF_LINE_PFN_TO_PAGE
2472 struct page
*pfn_to_page(unsigned long pfn
)
2474 return __pfn_to_page(pfn
);
2476 unsigned long page_to_pfn(struct page
*page
)
2478 return __page_to_pfn(page
);
2480 EXPORT_SYMBOL(pfn_to_page
);
2481 EXPORT_SYMBOL(page_to_pfn
);
2482 #endif /* CONFIG_OUT_OF_LINE_PFN_TO_PAGE */